Phonologica 1996
Phonologica 1996
Syllables !?
Proceedings of the Eighth International Phonology Meeting Vienna, 1996
edited by John Rennison & Klaus Kühnhammer
Thesus The Hague
1999
Contents
Preface
Contributors
Pier Marco Bertinetto Psycholinguistic evidence for syllable geometry: Italian and beyond 1–28
Wiebke Brockhaus Closet syllables in a syllable-denying framework: A case of déjà vu? 29–48
Grzegorz Dogil, Jörg Mayer & Justus C. Roux Syllables and unencoded speech: Clicks and their accompaniments in Xhosa 49–60
Katarzyna Dziubalska-Kołaczyk Syllables? No! Substantive evidence for the syllable-less beats-&-binding model of phonology 61–87
Contents v
Harry van der Hulst Features, segments and syllables in Radical CV Phonology 8 9 – 111
Iwona Kraska-Szlenk Syllable structure constraints in exceptions 113–131
Michele Loporcaro On possible onsets 133–151
Jean Lowenstamm The beginning of the word 153–166
Krisztina Polgárdi Constraint ranking, Government Licensing and the fate of final empty nuclei 167–182
John R. Rennison Can there be empty phonological elements? On empty heads and empty operators 183–193
David Restle Reduplication as pure constituent copying: In defence of the syllable against the phonemic melody 195–217
Nancy A. Ritter & Robert M. Vago Subsyllabic constituency in Hungarian: Implications for Moraic Phonology and Government Phonology 219–245
Elke Ronneberger-Sibold Ambisyllabic consonants in German: Evidence from dialectal pronunciation of lexical creations 247–271
vi Contents . Grazyna Rowicka On trochaic Proper Government 273–288
Tobias Scheer On constraints vs. non-circular approaches to word-initial clusters 289–304
Margit Tornéus & Björn Andersson The syllable as a cognitive unit in reading and spelling 307–313
Michael Völtz The syntax of syllables: Why syllables are not different 315–321
Laura Walsh Dickey Syllable count and Tzeltal segmental allomorphy 323–334
Shohei Yoshida Inter-nuclear relations in Arabic 335–354
Yuko Yoshida Phonological domains and binary inter-nuclear licensing for pitch accent 355–371
Preface
The announcement for the Eighth International Phonology Meeting ran:
The choice of the theme Syllables!? stems from the dissatisfaction of many phonologists with the status of what up to now has been called “the syllable”. In several of the newer phonological theories, there is no provision for the syllable as a central or even relevant phonological concept. On the other hand, concepts of a lower level (such as onset, nucleus, beat, etc.) and of higher levels (foot, stress group, etc.) have been the focus of increased interest. The aim of this meeting is there- fore to bring together phonologists concerned with establishing new theoretical constructs to replace traditional syllables — whether they bear the name “syllable” or not.
This volume contains a fairly representative selection of the types of re- sponse received, ranging from minor refinements of traditional views to completely new approaches. We would like to thank all the contributors to volume for their co- operation in the mutual reviewing process and for their prompt revisions and replies to queries. Thanks also to the people at HAG for their moral and logistic support at all stages of producing this volume. The Meeting received grants from the (then) Austrian Ministry of Science, Transport and the Arts, the Österreichische Forschungsgemeinschaft, and the Mayor of Vienna. The Ninth International Phonology Meeting is scheduled for 2000.
Vienna, October 1998 John Rennison Klaus Kühnhammer
Contributors
Björn Andersson Harry van der Hulst Department of psychology University of Leiden / HIL Umeå University Leiden University S-901 87 Umeå PO Box 9515 Sweden NL-2300 RA Leiden [email protected] The Netherlands [email protected] Pier Marco Bertinetto Scuola Normale Superiore Iwona Kraska-Szlenk I-56126 Pisa Institute of Oriental Studies Italy University of Warsaw [email protected] Krakowskie Przedmies´cie 26/28 PL-00-927 Warszawa Wiebke Brockhaus Poland [email protected] Dept. of German University of Manchester Oxford Road Michele Loporcaro Manchester M13 9PL Romanisches Seminar England Platenstr. 32 [email protected] CH-8028 Zürich Switzerland Grzegorz Dogil [email protected] Institut für Maschinelle Sprachverarbeitung Jean Lowenstamm — Experimentelle Phonetik — UFRL Azenbergerstraße 12 Université Paris 7 D-70174 Stuttgart 2 pl. Jussieu Germany F-75005 Paris [email protected] Frankreich [email protected] Katarzyna Dziubalska-Kołaczyk School of English Jörg Mayer Adam Mickiewicz University Institut für Maschinelle Sprachverarbeitung al. Niepodleglosci 4 — Experimentelle Phonetik — PL-61-874 Poznan´ Azenbergerstraße 12 Poland D-70174 Stuttgart [email protected] Germany [email protected]
x Contributors . Krisztina Polgárdi Grazyna Rowicka Institute of English and American Studies University of Leiden / HIL József Attila University Postbus 9515 H-6722 Szeged NL-2300 RA Leiden Egyetem u. 2. The Netherlands Hungary [email protected] [email protected] Tobias Scheer John Rennison Université de Nice Sophia Antipolis Institut für Sprachwissenschaft Département de Linguistique Berggasse 11 98, Bd Edouard Herriot A-1090 Wien B.P. 3209 Austria F-06000 Nice Cédex 3 [email protected] France [email protected] David Restle Schellingstr. 3 RG Margit Tornéus D-80799 München Department of psychology Germany Umeå University [email protected] S-901 87 Umeå Sweden [email protected] Robert Vago 14 Library Lane Plainview, NY 11803 Michael Völtz U.S.A. Am Eckbusch 39/103 [email protected] D-42113 Wuppertal Germany [email protected] Nancy Ritter University of Leiden / HIL Laura Walsh Dickey Postbus 9515 Max-Planck Institut für Psycholinguistik NL-2300 RA Leiden The Netherlands Postbus 310 NL-6500 AH Nijmegen [email protected] The Netherlands [email protected] Elke Ronneberger-Sibold Veit-Pogner-Straße 24 Shohei Yoshida D-81927 München Yokohama National University Germany International Student Center [email protected] 79-1 Tokiwadai, Hodogaya-ku Yokohama, 240-0067 Justus Roux Japan [email protected] Research Unit for Experimental Phonology Department of African Languages Yuko Yoshida University of Stellenbosch Institute for Language and Culture Private Bag X1 Doshisha University 7602 Matieland Kyotanabe City, Kyoto, 6100394 South Africa Japan [email protected] [email protected]
Psycholinguistic evidence for syllable geometry Italian and beyond
Pier Marco Bertinetto
Abstract
This paper addresses issues of syllable structure. Section 1 is a plea for the complementarity and necessary convergence of theoretical and experimental research. Section 2 suggests that, on purely theoretical grounds, the issue of the internal organization of the syllable may receive conflicting answers. Section 3 briefly reviews previous experimental investigations of English, showing that in this language the syllable clearly appears to have a right-branching structure. Section 4 presents evidence The Hague: Thesus. 1–28. Thesus.The Hague: 1–28. . that the alleged universality of right branching in syllable geometry has exceptions; indeed, Korean seems to exhibit left branching. Section 5 reviews experimental work performed by the author and co-workers,
Syllables !? demonstrating that the syllable structure of Italian is but weakly right- branching; Italian may be regarded, in contrast to English, as a language with a “low profile” geometry. Section 6 provides further evidence for the above findings, analysing the break point of natural lexical blends, which turns out to be quite different in English and Italian. In the theoretical Phonologica 1996: Phonologica 1996: discussion in section 7, it is suggested that the two main facets of the syllable (peak and margins) play a different role in speech processing. The individuation of syllable peaks seems to be a necessary precondition, and in this restricted sense the syllable is likely to be a fundamental unit of the phonology of natural languages. Syllable margins, on the other hand, become relevant only at a much shallower level, where other factors also come into play (low-level ones, such as phonotactics, and higher-order ones, such as stress and prosodic domains higher than the syllable). 0.
1. On the complementarity of theoretical and experimental approaches
One important facet of the long-standing debate concerning the syllable is
John R. Rennison & Klaus Kühnhammer, eds. (1999). eds. R. Rennison & Klaus Kühnhammer, John (1999). the problem of its internal structure. This may turn out to be obvious for
2 Pier Marco Bertinetto those holding the view that the syllable is one of the constitutive units of phonology (possibly the most fundamental, as authors such as Kahn (1976) or Borowsky (1986) posit). Yet, even those who do not share that view, possibly denying any role whatsoever to the syllable in the phonological component (Dziubalska-Kołaczyk, 1995), will have to address the problem of the internal geometry of the syllable, possibly only to show that it is the result of more fundamental phonological features. There are two possible approaches to the issue of the hierarchical struc- turing of the syllable. One is based on language-internal evidence, the other on external (above all, psycholinguistic) evidence. The two approaches are obviously complementary, and should ideally converge to produce a satis- factory solution. The formulation of linguistically valid hypotheses, which is the specific goal of theoretical investigation, is of course the input of ex- perimental research, for there is no sense in collecting experimental data (either spontaneously elicited or experimentally induced), without the guidance of detailed hypotheses. However, experimentation is not merely ancillary to theorizing; experimental research is often necessary to shed light on alternative solutions envisaged by competing theories. Thus, the relationship between theoretical and experimental work turns out to be a dialectical one: experimental investigation is guided by theory, but may help refining it, and in turn feeds further goals. On the other hand, experimenta- tion is not self-sufficient, for its results are often difficult to interpret. The experimental technique is itself a factor of the output and has to be carefully scrutinized in a cross-methodological way, in order to weed out the possible artefacts of the procedure employed, or to isolate any disturbing elements (Derwing, 1997). The availability of converging results, on the theoretical and experimental side, is ultimately the best guarantee of their validity.
2. Ambiguous evidence for syllable geometry
The case at issue is a good example of this complementarity. Consider the language-internal evidence for the hierarchical organization of the syllable. This may be of two sorts: (a) distributional; (b) processual. Both provide evidence for internal structuring, but this evidence turns out to be rather ambiguous, in that virtually any constituent of the syllable may receive support. First, take distributional evidence. This is the primary source of data, yielding non-trivial phonotactic generalizations such as those concerning: (i) the varying weight of the individual constituents of the syllable (in many
Psycholinguistic evidence for syllable geometry 3 languages — e.g. Italian or Japanese —, more contrasts are available in On than in Co);1 (ii) the varying number of segments admissible in each con- stituent (e.g. some languages do not admit Co, or — cf. again Italian and Japanese — normally admit no more than one segment in this position, etc.); (iii) the possible structuring role of specific features (e.g. in English the maximum number of segments in Co is two, unless the point of articulation is coronal). However, distributional evidence may be tricky. First, there are occasional gaps which partly distort the overall picture (Treiman, 1988). Second, not all observed regularities are synchronically meaningful for syllable structure (Bertinetto, in prep. b). Moreover, and most embarrass- ingly, the question arises as to the proper interpretation of the facts. Is phonotactics the consequence or rather the source of syllabic organization? If the source, then phonotactics must obey other, more primordial con- straints than the syllable, as a number of scholars have indeed suggested (Dziubalska-Kołaczyk, 1995 and references therein). I know of no simple answer to this dilemma. Now consider the processual evidence, which constitutes the very core of phonological investigation. Here the proofs for the conspicuous role of the syllable and of its different constituents abound. As to the syllable, take for instance: (α) the different vocalic inventory in open vs. closed syllables, also reflected in historical developments (e.g. Loporcaro, 1996 with respect to several Italian dialects); (β) the various phonological processes studied by scholars such as Kahn (1976), Borowsky (1986), Vennemann (1988a) etc.; (γ) the contrast ‘sanfter vs. scharfer Silbenschnitt’ originally proposed by Trubeckoj (1939) and recently revived by Vennemann (1991; 1994). As to syllable constituents, there is evidence that they all may play an important role. The role of the Bo is assessed, e.g. by: (a,i) place assimilations such as English ‘velar softening’ or Italian palatalization of velars before front vowels; (a,ii) isochronic relationships between On and Nu in languages such as Japanese (Kubozono, 1995). The role of Rh is assessed e.g. by: (b,i) nasali- zation of vowels before nasal Co; (b,ii) isochronic relationships between Nu and Co in languages such as Swedish or Italian (in careful pronunciation). The problem here is that one and the same language may often provide contradictory evidence. One such language is Italian, quoted in connection with both the Bo and the Rh (cf. points (a.i) and (b.ii)). Indeed, this seems to be the rule rather than the exception, as is abundantly demonstrated by Clements & Keyser (1983); and this argument has been used by Vennemann (1988a) to suggest that the specific organization reached by the syllable in a given language is the result of conflicting forces, which implement a
1 I shall make use of the following abbreviations: Co = coda, Nu = nucleus, On = onset, Bo = body (i.e. On+Nu), Rh = rhyme (i.e. Nu+Co).
4 Pier Marco Bertinetto situation of potential instability, possibly leading to phonetic changes. In ad- dition, one often observes phonological processes involving intersyllabic sequences, such as place assimilations in nasals before heterosyllabic consonants. Thus it is not the case that the various syllable constituents delimit the domain of phonological processes. And even where such pro??cesses tend to involve one specific constituent, the evidence is not decisive, because one cannot easily rule out the hypothesis that the reason for this tendency is to be sought outside the syllable proper. One might therefore conclude, as some scholars indeed have, that the syllable is not a primary unit of the phonological component. While I admit that this is a viable solution, I would like to defer the discussion of this point to a later section of the paper, and turn now to the other kind of evidence, namely external (psycholinguistic) evidence.
3. Experimental investigations of syllable structure
Such evidence has mostly come, in recent years, from a body of carefully controlled experiments performed by a number of researchers (e.g. Rebecca Treiman and Bruce Derwing). Although these techniques of investi- gation have also been exploited for other purposes (particularly for studying the problem of intersegmental cohesion), in this paper I shall only consider experiments bearing on the issue of the internal hierarchy of the syllable. In the case of English, to which these experiments were first applied, the syllable structure proved to be right-branching. That is, of the three possibili- ties depicted below, the one appropriate to represent the internal hierarchy of the English syllable is (1c).
(1) (a) Flat (b) Left-branching (c) Right-branching F F F
On Nu Co Bo Co On Rh
On Nu Nu Co
The evidence for this comes from a variety of sources. The following is a short survey. Treiman et al. (1982) and Treiman (1983) have shown that the attraction be- tween Nu and Co is stronger than the attraction between Nu and On. Inter- estingly, this particular aspect of phonological competence is refined over time, up to the age of adolescence (Treiman, 1985; Derwing et al., 1988). This concurs with the finding that the Rh is more easily detachable than the Bo (Fowler et al., 1993). By contrast, internal cohesion has been proved to be
Psycholinguistic evidence for syllable geometry 5 stronger within a multisegmental On than within a multisegmental Co (Treiman, 1983; 1986), and again this tendency becomes stronger over time (Derwing et al., 1988). In a substitution task, Treiman (1983: experiment 4) has shown that when taught alternative strategies, English subjects perform significantly better if they manipulate the entire Rh instead of the Co alone (experiment 3), or the On alone instead of the entire Bo. In two phoneme monitoring tasks, Treiman et al. (1982) have shown that the initial consonant is detected signifi- cantly faster when embedded in CV or CVC sequences than when em- bedded in a CCV sequence. Besides replicating the preceding result with children between ages 4;6 and 6;5 (which involved the use of an appropriate technique entailing puppets with funny names), Treiman (1985) has shown that children between ages 7 and 9;4 are very sensitive to syllabic structure. The task performed demonstrates that the substitution of the first two segments is significantly easier in CCV than in CVC sequences, while the substitution of the last two segments is significantly easier in CVC than in CCV sequences. When performed by adults (Treiman, 1986: exp. 7), the substitution task yielded still sharper results. Additional support for the right-branching model comes from Fowler (1987), who asked her subjects to exchange either the initial or the final consonant in two simultaneously presented CVC nonsense words. Perform- ance was faster and more accurate for exchanges of initial consonants (coinciding with the On) than for exchanges of final consonants (coinciding with the Co). Finally, Berg (1989; 1991) observes that English speech error corpora show that On errors are far more common than Co errors, even more so after normalization of the data (i.e. after taking into account the different fre- quencies of these positions). He remarks that a prevalence of On errors has also been reported for other Germanic languages, such as German, Dutch and Swedish. Although the references cited represent only a part of the available litera- ture, they make it sufficiently clear that the syllable in English has a strong right-branching tendency. The results of several experiments, using different methods, converge towards the same outcome. Even a potentially embar- rassing objection put forward by Davis (1989) has been shown to be irrele- vant. A convinced supporter of the flat structure model of the syllable, Davis argued that most experimental results gathered on the internal structure of the syllable are misguiding, inasmuch as they wrongly suggest a predomi- nance of the On+Rh structure, without considering the interfering influence of other structural properties. In particular, Davis observed that much of the argument on which the right-branching model is based can be explained away by the fact that, in the experiments to which he refers, the On
6 Pier Marco Bertinetto consonant is also the word-initial consonant. Now, we know, from English speech error corpora, that word-initial consonants are more liable than any other consonant to undergo disruption, and this is true of other languages as well (Berg, 1991 cites other Germanic languages such as German, Dutch and Swedish). Thus, the fact that subjects perform better when manipulating word-initial rather than non-initial consonants (possibly final, when the test word is monosyllabic), does not constitute conclusive evidence with respect to the issue of syllable geometry. This may simply be due to the contribution of higher-order prosodic factors, which undoubtedly play a substantial role in speech performance. I shall call the specific factor referred to by Davis the ‘marginality’ factor (in the sense of ‘close to the margins of the word’) Immediately counterattacked by Fudge (1989) on theoretical grounds, Davis’ objection demanded nevertheless appropriate experimental scrutiny. The challenge was taken up by Fowler et al. (1993; cf. also Treiman et al., 1995). These authors have shown that, for English at least, the hypothesis put forward by Davis does at first sight seem relevant. When subjects were asked to perform On, Co, Bo or Rh substitutions in one of two visually pre- sented disyllables of the type C1VC2C3VC4, they were consistently faster and more accurate when performing the task on the first than when on the second syllable (cf. experiment 3). However, with trisyllables of the type C1VC2VC3C4VC5, a substitution task respectively concerning C2, C3, C2V or VC3 showed a clear advantage of On over Co, and of Rh over Bo (experiment 4). The same result was replicated with oral presentation of the stimuli and without time-limit (experiment 5). It thus appears that when the marginality factor is minimized, as in the middle portion of a polysyllable, the internal hierarchy of the syllable turns out to be a very effective predictor of sub- jects’ behaviour.
4. Possible exceptions to the Universal Rhyme Hypothesis
Although the experimental evidence for right-branching syllable structure in English is overwhelming, the significance of this finding would not be particularly compelling if it turned out that all languages behave in the same way, as the advocates of the so-called ‘Universal Rhyme Hypothesis’ sug- gest. If this were the case, the objection might be raised that this result depended on some hidden experimental artefact, rather than being a true reflexion of the internal organization of the syllable as such. Actually, one reassuring datum is the fact that the evidence gathered on English stems from a variety of experimental techniques; this in itself reduces the risk of artefactuality to a tolerable minimum. But the real solution to this dilemma is the availability of conflicting cross-linguistic data.
Psycholinguistic evidence for syllable geometry 7
Fortunately, this seems to be the case. Experimental investigations car- ried out on Korean have repeatedly demonstrated that in this language the structure of the syllable is likely to be left-branching. In a series of word- blending experiments, Derwing et al. (1993) and Wiebe & Derwing (1994) have shown that, both in blending preference tasks and in blend-learning tasks, the type Bo1+Co2 significantly prevail over the type On1+Rh2 (where the integers refer to ‘Word1’ and ‘Word2’). The same result was obtained via the method of ‘sound similarity judgements’, in which subjects are asked to rate the degree of similarity of two acoustic stimuli. As it happens, Korean speakers reliably considered pairs of stimuli like /mot/ and /mop/ more similar than /mot/ and /pot/, contrary to the tendency of English speakers (Yoon, 1994). In fact, one and the same set of Korean CVC words was very differently judged by Koreans and English speaking subjects (Yoon & Derwing, 1994). What is particularly reassuring about these data is that the same tech- niques have been applied to different languages, yielding different results. The blending technique, for instance, yielded an On1+Rh2 preference with Taiwanese and English speaking subjects (Wiebe & Derwing, 1994). Equally, the sound similarity judgements method produced very different results in the languages to which it was applied. For instance, it may be remarked that for Japanese speakers the most salient unit appeared to be the mora (Derwing & Wiebe, 1994), a result consistent with Otake et al.’s (1993) finding in a monitoring task. However, the mora did not emerge as a relevant con- stituent in the case of English, Korean, Taiwanese or Arabic, where other elements emerged instead: the segment first and foremost in all cases, but also specific constituents of the syllable, according to a hierarchy differing from language to language. For Arabic speakers, in particular, one phono- logical constituent that emerged very clearly, obtaining in fact the same score as the segment, was the C-tier (Derwing & Nearey, 1994). The latter result might raise suspicion, in that (as Derwing et al. (1992) note) vowels are not normally represented in Arabic spelling. Thus, the preferences of subjects might be guided, at least in part, by writing conven- tions, rather than by the internalized phonological component alone. The same objection might in fact be voiced with respect to the Korean ‘hangul’ and the Japanese ‘hiragana’ and ‘katakana’ writing systems, which often give visual prominence to the Bo over and above other constituents (Yoon & Derwing, 1994). As Derwing (1997) judiciously points out, the orthographic factor needs to be addressed directly in future experiments, in order to rule out its potentially disruptive effects. Nevertheless, the results obtained from sound similarity judgements are somewhat reassuring, at least in the sense that they seem to be by and large independent of the specific type of writing system employed in the various languages. Consider the following fact
8 Pier Marco Bertinetto
(Derwing & Nearey, 1994). The Korean ‘hangul’ is said to be a ‘syllabic’ system, although (as noted above) it should best be characterized as a body- based system. Yet, in similarity judgements Korean subjects attribute prominence to the segment just as English speakers, who employ a so- called ‘phonetic’ alphabet. Thus, although the impact of orthography may not simply be dismissed, there are grounds for believing that it does not condition the behaviour of speakers beyond a reasonable limit. In addition to this, it is fairly clear that these experimental findings are supported by a number of other findings, which lend them further credibil- ity. In the case of Korean and Japanese, in particular, converging data are yielded by speech errors, natural (i.e. not experimentally induced) blends and language games. For instance, Kubozono (1989) notes that in Japanese speech errors, Rh errors do not prevail over Bo errors as they do in English, and the same occurs in Korean according to Cheon (1980). Kubozono also points out that Japanese word blends (both in word-formation and in spon- taneous errors) tend to be of the type Bo1+Co2, rather than On1+Rh2, while English shows the opposite tendency (as an example of word-formation blend, consider smog < smoke + fog; see §6 for further details). Finally, there are Korean language games involving the insertion of elements after the Bo of CVC syllables (Gim, 1987), a strategy that seems to be unknown to English. It should also be mentioned that a blending preference study conducted by Kubozono (1995) with English monosyllabic words produced opposite results with English natives and Japanese subjects, with the latter again expressing a strong preference for the Bo1+Co2 combination. Thus, Korean is not the only language for which it has been experimentally shown that the syllable may have a left-branching structure.2 But even if this were the only exception to the alleged ‘Universal Rhyme Hypothesis’, its very existence would compel us to check the situation in other languages. Needless to say, English, a typical right-branching language, is typologically very different from Korean at most levels, including the prosodic level which concerns us here.3 It is therefore interesting to see what happens in Italian, which is
2 Kubozono (1995) reports also the results of a pilot experiment with two Italian speakers and nine speakers of Mexican Spanish, producing blends out of the same English monosylla- bles. In both cases, left-branching structure prevailed. However, due to the small population of both tests, their significance is negligible. 3 Interestingly, Korean and Japanese, two languages which (as noted above) seem to behave alike in a number of situations relevant to our problem, are structured in a way that is per- fectly symmetrical with respect to the Indo-European languages even at the syntactic level, e.g. with complementizers at the end of the clause rather than at the beginning (Sells, 1995). This is clearly another instance of left-branching. Thus, it is possible that the contrast to be observed at the prosodic level is but an aspect of a more basic property of these languages.
Psycholinguistic evidence for syllable geometry 9 typologically much closer to English than to Korean, and whose syllable structure might therefore be expected to be right-branching.4
5. Experiments on the internal organization of the syllable in Italian
A series of five experiments were carried out with Italian subjects. A selec- tion of the main findings is reported below (for a full-fledged description, see Bertinetto et al., in prep.). Each experiment consisted of a variable number of tasks (from 10 to 13). These were of two types: substitutions and blendings. Table 1 indicates the tasks performed in each experiment and
4 Analysing a small corpus of Finnish spontaneous errors, Niemi & Laine (1997) claim that in this language there is evidence for a major division point after the first CV sequence of the word. This may at first sight appear to be a variant of the Bo+Co hypothesis, but in fact it differs from it because the initial sequence CV does not always correspond to a Bo, nor does it necessarily include a whole nucleus, since V may here represent the first half of a long vowel. The authors thus claim that Finnish provides evidence for a different prosodic type, mora-based (in some sense of this word) rather than syllable-based. Alternatively, to adopt Beckman’s (1995) hypothesis concerning Japanese, Finnish provides evidence of a language where the ‘optimal’ syllable type CV plays the role of a cognitive point of reference for the speakers. This is undoubtedly an appealing hypothesis. Nevertheless, on closer examination these data turn out to be best interpretable in terms of the Bo+Co hypothesis. Consider the fol- lowing examples: (i) luistelu-tyyli tuli > luistelu-tuuli tyli (ii) romantiikka ja realismi > remantiikka ja roalismi (iii) lypsää lehmiä > lepsää lyhmiä Leaving aside for the moment case (i), the crucial fact appears to be the interpretation of the vowel sequence /ea/ in (ii). If this is interpreted as a true diphthong, then one is forced to admit that the relevant unit is the first CV sequence, rather than a specific syllable con- stituent, for the sequence CV would correspond to the On in the second word of (ii), but to the Bo in the remaining words of both (ii) and (iii). However, Finnish vowel sequences appear to be much more like hiatuses than like true diphthongs, for none of the two vowels under- goes a proper gliding, not even in terms of duration values. If this is so, then the only rele- vant unit is the Bo in both (ii) and (iii). The same applies to (i), with the difference that in this case there is reassignment of quantity. But this may easily be explained by the prosodic structure of tyyli tuli: C V V C V C V C V | \ / | | | | | | t y l i t u l i The elements that move are the first two segments, i.e. the Bo. As soon as they reach their new site, they are immediately assigned the appropriate quantity values, specified on the CV tier. Note that if the movable elements were the first CV sequence of the skeleton tier, dragging the corresponding segmental units, one would obtain the impossible word tuyli. To remedy this, one might of course posit a low-level harmony readjustment, but the solution advocated here seems to be more consistent with the facts, since the segmental units ap- pear to be the ultimate elements involved in the interference.
10 Pier Marco Bertinetto
no. structural description scheme experiments and results substitution tasks I II III IV V 1 On in monosyllable CVC 86.66 84.13 74.05
3 On1 'CV-CVC 92.06 72.78 76.98 2 Co in monosyllable CVC 97.91 80.95 77.64
4 Co2 'CV-CVC 97.08 91.67 90.35 5 Bo in monosyllable CVC 90.83 94.84 75.88 6 Rh in monosyllable CVC 98.75 95.25 89.91
7 Bo1 'CVC-CV 69.05 65.77 8 Bo2 'CV-CVC 74.16 68.65 53.06 9 Rh1 'CVC-CV 82.91 70.63 72.81 10 Rh2 'CV-CVC 91 . 25 86.11 76.32 11 syllable1 'CV-CVC 81 . 25 12 syllable2 'CVC-CV 96.03 13 ‘free’ On2 'CV-CVC 67.08 68.65 58.33 63.33 14 ‘bound’ On2 'CVC-CV 88.09 81 . 57 79.36 15 Co1 'CVC-CV 88.75 79.36 82.90 72.62 16 Bo2 (stress on syllable2) CV-'CVC-CV 68.65 17 Rh2 (stress on syllable2) CV-'CVC-CV 63.49 18 Bo2 (stress on syllable3) CV-CVC-'CV 57.14 19 Rh2 (stress on syllable3) CV-CVC-'CV 71 . 03 20 ‘bound’ Bo2 (stress on CVC-CVC-'CV 53.57 syllable3) 21 Rh1 (stress on syllable3) CVC-CV-'CV 76.98 22 On2 in open syllables 'CV-CV 90.07 disyllable
23 On2 in open syllables CV-'CV-CV 69.44 trisyllable
24 On1 in open syllables 'CV-CV 78.57 disyllable
25 On1 (closed syllable) 'CVC-CV 84.12 blend learning 26 Bo + Co CVC+CVC > CV+C 70.63 71.43 27 On + Rh CVC+CVC > C+VC 65.87 62.30 blend preference 28 input: CVC+CVC output: C+VC (vs. CV+C) 75.72 29 input: 'CVCCV+'CVCCV output: 'C+VCCV 55.14 (vs. 'CV+CCV ) Table 1 Tasks performed in a series of experiments on the geometry of the syllable in Italian with percent figures of correct responses (the integers refer to the position of the syllable in the word, e.g. Bo2 = body of the second syllable).
Psycholinguistic evidence for syllable geometry 11 the percent results obtained. The materials consisted of a series of non- sense words, carefully chosen in order to conform to the phonotactic constraints of Italian. In order to avoid unnecessary complications, all internal codas consisted of sonorants; no geminates appeared in the ex- perimental list. The choice of nonsense materials was dictated by the need to avoid possible biases caused by the uncontrolled factor of lexical fre- quency, which is known to have a heavy impact on speakers’ performance.5 In the substitution tasks, subjects had to produce a verbal response in reaction to a verbal input. The underlined C and V units in the substitutions section of Table 1 indicate the elements that had to be replaced in each task; the appropriate interpretation in terms of syllable structure is indi- cated on the left-hand side of each row. The phoneme (or sequence of phonemes) to be inserted was shown on the screen of the computer, posi- tioned in front of the subject, just before the auditory presentation of the target stimulus. For instance, if in task 1 subjects saw the character
5 It is reassuring to observe that, according to Treiman (1986), there is no statistical difference between real and nonsense words in this type of experimentation. So, whatever the cogni- tive operation activated by these tasks, it does not seem to differ for linguistic and non- linguistic materials.
12 Pier Marco Bertinetto preference’, in which, by pressing the appropriate key, the subjects had to choose between the two visually suggested blending strategies relating to two auditorily presented words (tasks 28–29). In this case the statistical measure was obtained by comparing the respective percentages yielded by the two types of blending strategy.6 On the assumption (based on coarse-grained typological considerations) that Italian should behave more like English than like Korean, the segment(s) substitution tasks were expected to yield the results listed in (2).
(2) Expectation A: On should be easier to replace than Co; Expectation B: Rh should be easier to replace than Bo; Expectations (A–B) may be condensed as in formula (3).
(3) On = Rh > Co = Bo where = stands for “behaves like”, and > for “has a behavioural advan- tage over”
As to blending tasks, the prediction consistent with the above expectations was as in (4).
(4) On1+Rh2 > Bo1+Co2
If we also take into consideration the ‘marginality’ factor, as defined in §3 according to suggestions by Davis (1989), we may add the following predic- tions (actually, expectation D in (5) was not directly put forward by Davis, but looks like a reasonable extension of his assumption C).
(5) Expectation C: Initial segments should be easier to replace than final ones; Expectation D: Marginal (i.e. initial or final segments) should be easier to replace than medial ones.
To sum up, the general expectation was that both the substitution and the blending tasks would provide evidence that Italian is oriented towards
6 Needless to say, the order of presentation of the two types of output was balanced. More generally, in all five experiments the relevant factors were balanced as carefully as possible in order to minimize experimental biases. Care was taken, for instance, to compose ex- perimental sublists to be presented to different subsets of the subjects’ population, in which the sequence of the tasks, and the succession of the stimuli within each task, was duly inverted. These and several other experimental details are reported in Bertinetto et al. (in prep.). Here I only concentrate on the main results.
Psycholinguistic evidence for syllable geometry 13 an On+Rh structure. The actual results were quite different from the ex- pected ones, indeed in some sense even surprising, with the sole exception of assumption D, which received strong confirmation. There follows a summary of the most important results concerning assumptions A–C. The data relating to the contrast between initial and final segments turned out to be consistently at odds with assumption C. In Italian there appears to be a very strong advantage for final elements over initial ones. This may be gathered in the significantly better performance on tasks 2, 4, 6 and 10 over tasks 1, 3, 5 and 7 (both in the overall comparison and in pair- wise comparisons between related tasks; see Table 1). Italian apparently differs dramatically in this respect from English. Interestingly, the evidence of speech errors supports this conclusion. According to data collected by Magno Caldognetto et al. (to appear), Italian errors localized on the first syllable are only a third of the total (33%), a figure that remains below statis- tical significance even when the different number of initial vs. non-initial syllables is taken into account.7 It is interesting to observe that Italian behaves in this respect very much like Spanish, and contrasts with all the Germanic languages listed in Berg (1991). Furthermore, it is interesting to note that this outcome seems to relate to the fact that, in both Italian and Spanish, stressed syllables do not attract significantly more errors than unstressed ones, again in contrast to Germanic languages (Berg, 1991). The figure for Italian is 38% (Magno Caldognetto et al., to appear); see fn. 7 for the Spanish and German data. While the interpretation of the latter datum is not easy, it is clear that the effect of higher-order prosodic factors, evidently at work in all languages, may produce very different results. One remarkable consequence of this is to be observed in a type of datum that emerges in substitution tasks. As it happens, the length of the stimulus had a strong effect on the results. On the one hand, there was altogether a clear advantage, in strictly comparable tasks, of monosyllables over
7 Since the paper from which these data are taken is not yet available, I have to rely here on personal communication from the first author. In a previously published study, based on a smaller corpus (Magno Caldognetto et al., 1987), a higher percentage (48%) was reported. In order to have a term of comparison, it is useful to cite the data reported in Berg (1991: 284, 290–291) for Spanish and German. It should be noted that ‘word-initial errors’ is not the same as errors localized on the first syllable; thus, these figures are not strictly comparable with those relating to Italian. Nevertheless, they may suffice as a first approximation. Spanish German word-initial errors 27.3% (significant) 80.9% (significant) errors on stressed syllables 40% (not significant) 55% (significant) Interestingly, the percentage of first syllable errors raises decidedly in disfluencies pro- duced by Italian subjects. Magno Caldognetto & Zmarich (1995) report that 74% of Italian disfluences hit the first syllable, a figure corresponding to 22.5% of all initial syllables (against 4.9% of non-initial syllables).
14 Pier Marco Bertinetto disyllables, and of the latter over trisyllables; on the other hand, and most importantly, monosyllables and disyllables yielded diverging results in On vs. Bo comparisons (tasks 1 and 3 vs. 5 and 7) as contrasted to Co vs. Rh comparisons (tasks 2 and 4 vs. 6 and 10). In other words, collapsing the data of experiments I, II and IV, there was an advantage of longer constituents (Bo and Rh) in monosyllables and of shorter constituents (On and Co) in disyl- lables. The results were thus markedly dependent on the length of the stimuli, and this was confirmed (minor details aside) in more than one experimental session. While the reason for this is not at all clear, it must be admitted that higher-order prosodic factors (to be understood in the broad sense) enter into play, strongly interacting with the syllabic tier. Indeed, other effects of such higher-order factors were to be observed in our data, as will become clear in the following description of results. Let us now consider the diverging behaviour of On vs. Co, and Rh vs. Bo (cf. assumptions A–B in (2) above). Since the comparison of these constitu- ents in external positions (cf. tasks 1 to 7 and 10) is obscured by the striking word-offset advantage which is clearly at work in Italian, we have to look for evidence relating to internal positions. As to the comparison between On and Co (cf. tasks 13,14,15), the situation appears to be rather confusing. The Co constituent significantly prevails in tasks 15 vs. 13 in all three experiments in which this test was made, whereas the On prevails (but not significantly) in tasks 14 vs. 15. Thus, no definite direction seems to emerge. To add to the confusion, it turns out that task 14 significantly prevails over task 13, al- though in task 14 the On of the second syllable is (so to say) ‘bound’ within a consonant cluster, because of the presence of the Co of the previous syl- lable. The prediction was rather that the ‘free’ On of task 13 would be fav- oured. Evidently, words exhibiting the structure /'CVCCV/ enjoy an advan- tage for Italian speakers over words of the type /'CVCVC/, possibly a reflex of the higher frequency of words of the former type. I shall return to this below. As to the comparison between Rh and Bo (cf. tasks 8,9,16,17,18), Rh can indeed be seen to prevail over Bo in tasks 8 vs. 9, a contrast that was signifi- cant in two out of the three experimental sessions in which this test was run. However, since this comparison again involves a contrast between the two word types indicated above (/'CVCCV/ vs. /'CVCVC/), it is possible that this outcome reflects higher-order factors concerning the preferred structure of Italian words, rather than the different syllabic targets involved. In trisyllabic words, the situation again looks fairly confusing because of the interaction with stress, another higher-order prosodic factor. In fact, while Rh signifi- cantly prevails over Bo in tasks 19 vs. 18, the reverse happens (although the result is not statistically significant) in tasks 16 vs. 17. Thus, once again no robust conclusion may be drawn.
Psycholinguistic evidence for syllable geometry 15
In order to check the influence of the word structure factor, a number of tasks were performed on the same syllabic constituent (the On) embedded in different word types. Two series of comparisons were performed: (a) word-internally (cf. tasks 13,14,22,23); (b) word-initially (cf. tasks 3,24,25). In the first control series, unsurprisingly, task 22 yielded a significantly better result with respect to both tasks 13 and 14, for the last two involve longer words. There was also an advantage of task 14 over 23, while task 23 yielded an advantage over task 13, despite the fact that words of the type /CV'CVCV/ are one phoneme (and one syllable) longer than words of the type /'CVCVC/. However, the last two comparisons were statistically non-significant. In the second control series, concerning the On in word-initial position, task 24 performed better than 3, and the same happened for 25 over 3 and even 25 over 24, despite the possible disadvantage of a longer word type; but alto- gether no statistically significant result emerged. To sum up, although most of the comparisons did not reach significance, there is some initial indica- tion concerning: (a) the relative disadvantage for Italian speakers of the word type /'CVCVC/ (tasks 3 and 13); (b) the relative prominence of the word type /'CVCCV/ (tasks 14 and 25), except for the predictable superiority of the type /'CVCV/ (tasks 22 and 24), but only when the substitution was per- formed on the second syllable (tasks 14 vs. 22). Further control on a larger population of subjects should be performed to provide confirmation for the tendencies observed. The data just reported suggest that when Italian subjects perform sub- stitution tasks of the sort considered here, syllabic structure is not the only factor involved, nor even the most important one. Among the relevant factors, we should consider at least the following, in decreasing order of importance: (i) word-offset advantage, (ii) word structure, (iii) stress. Given the results obtained in experimental tasks of this sort, syllable structure seems to be slightly more effective than stress, but less effective than factors (i-ii), in influencing subjects’ behaviour. This puts severe limits on the actual relevance of the internal hierarchy of the syllable in Italian phonology, as opposed to what we observe in comparable experiments in English or Korean. Yet, this should come as no surprise, given the experimental results described in Pierrehumbert & Nair (1995) and Finney et al. (1996) concerning the behaviour of English subjects, or the discussion in Beckman (1995) relating to Japanese. As it happens, even in languages (like English) where it has already been experimentally proven that the internal organization of the syllable has a strong impact on the outcome of psycholinguistic experimen- tation, the responses of the subjects are at least to some extent influenced by the general prosodic structure of the language. In particular, according to Finney and co-workers, stress is a powerful predictor of subjects’ behaviour; and according to Beckman, the foot seems to play a dominant role in
16 Pier Marco Bertinetto
English, just as the mora does in Japanese (also according to the testimony of Kubozono, 1989; 1995 and Otake et al., 1993). This is not to say, however, that syllable structure plays no role at all in Italian. When higher-order prosodic factors are neutralized, the effect of syllable structure does emerge. This may be seen in the results concerning blendings. In these tasks, the interfering effect of word-structure is possibly neutralized, for both input and output stimuli have the same structure. As to the factor ‘marginality’, however, given the strong word-offset advantage to be observed in Italian, the two constituents which should prevail in the blending of monosyllables (cf. tasks 26,27,28) are Co and Rh; and among these two, we might expect that the Rh should be favoured, according to what happens in substitution tasks (see the comparison of tasks 2 vs. 6). Nevertheless, it turns out that when Italian subjects are allowed enough time to give their responses, as in tasks 26–27, the two blending types On1+Rh2 and Bo1+Co2 yield essentially the same results. On the other hand, when the response time is cut down to no more than 1.5 sec., as in task 28, 8 the advantage of the On1+Rh2 solution emerges in a striking way. Thus, despite its weak role in the overall design of the phonological component, syllable structure does have an effect in Italian. Yet, its role is indeed non pervasive, as is proven by the results of task 29, a blend preference task performed on disyllabic stimuli. In this case, although the type On1+Rh2 numerically prevails over the type Bo1+Co2, the difference is not significant, suggesting that either the short-term memory burden was too heavy, or the presence of a more complex word structure obscured the role of the syllabic component.
6. The structure of Italian and English natural blends
It may be useful, at this point, to reflect on the validity of the experimental results obtained through artificial manipulations of the language. There is an important question that we should ask ourselves, namely: Do we know the exact cognitive significance of these results? The honest answer is, regrettably, “No”. Note that the tasks employed in this sort of experimentation bear some resemblance to ‘natural’ (i.e. historically developed) language-games, devel- oped by virtually any linguistic community with playful or cryptic purposes.
8 This outcome may also have been enhanced by the possible superiority of the blend preference technique over the blend learning one. However, since no direct comparison between the two methods, using the same response time, was performed, the answer to this question should be deferred.
Psycholinguistic evidence for syllable geometry 17
However, although these games are also artificial constructs, they at least have the support of a living community of speakers. In addition, they have (supposedly) acclimatized themselves to the prosodic system of the host language, so that they may yield legitimate information regarding the struc- ture of this component. By contrast, the experimental techniques described above are often expressly devised in defiance of the inherent prosodic tendencies of the language under scrutiny. Thus, one should always be aware that these findings may only be interpreted indirectly. Indeed, we do not know which particular cognitive capacity has been solicited, and what relation may exist between this capacity and the ones active in the normal process of speech production and comprehension. Thus, the ultimate relevance of these results needs confirmation by other types of data. Nevertheless, the situation is not as confusing as it may appear. First, the very fact that the same experimental techniques yield different results in different languages seems to suggest that some crucial target has been hit, which authorizes interesting speculations concerning the prosodic com- ponent of those languages. In fact, the data obtained are strictly behav- ioural, inasmuch as they exhibit the competence of the speakers in actual linguistic performance (albeit of a somewhat unusual sort), rather than as elicited through metalinguistic judgements. In addition, one should keep in mind that although the blending tasks themselves are performed within an artificial (i.e. experimental) situation, they do bear resemblance with events which occur naturally in speech errors and even in word-formation pro??cesses widely exploited by natural languages. Thus, at least with the blending tasks, the observable contrasts between the various languages investigated may be considered to reflect some kind of spontaneous linguistic behaviour. In support of this observation, we may have a look at Italian natural blends, with an eye at the syllabic interpretation of the switching point between Word1 and Word2. There are actually two main classes of blends in Italian.9 The formations of the first class resemble the usual blends com- bining the initial part of Word1 with the final part of Word2 (possibly including the whole of it). The second class includes a variety of acronyms, build by putting together fragments of the initial portions of successive words. The blends in (6) provide an example of each class.
(6) Coldiretti < coltivatori + diretti (class I) Minculpop < Ministero (della) Cultura Popolare (class II)
9 See Thornton (1993) and Lehrer (1996) for a detailed classification of the possible types of blend.
18 Pier Marco Bertinetto
Since class I includes the type which resembles most the sort of blending exploited in the experiments reported on in §5, in what follows I shall dis- regard class II (see Bertinetto, in prep. a for a full-fledged description of lexical blends in Italian and other languages with respect to syllable struc- ture). The analysis of a corpus of 79 Italian formations yields the situation de- picted in the table in (7) (where Sy stands for ‘syllable’, while the other abbreviations are as above). In the examples, underlined characters indicate the relevant fragments of words entering into the blend, while bold charac- ters indicate the portion of the blend that exhibits the relevant type of switching, viewed in terms of syllable structure. Note finally that in the cases marked ‘On+Rh(=Sy)’, the second fragment corresponds to a ‘naked syl- lable’; thus, the rhyme here coincides with the entire syllable. As may be gathered from (7), the preferred switching point of Italian blends is localized at the boundary of two whole syllables. This type alone accounts for half of the data. As a matter of fact, a whole syllable is also involved, as the second part of the blend, in the other two main types. The other major switching point occurs after the On of Word1. However, this does not constitute a proof of an On+Rh structure for, apart from the cases in which it coincides with a whole syllable, the Rh as such intervenes in not more than 3 blends (e.g. zebrallo < zebra + cavallo, which is by the way the only such example with univocal interpretation). This is notable, because the Rh constituent might obviously have appeared in many more cases, consid- ering the frequency of closed syllables in the corpus. Ultimately, the type On+Sy appears to be exploited simply in order to give rise to closed syl- lables out of open ones (namely, the last On of the relevant portion of Word1 is turned into a Co in the blend; see the example in (7)). This enhances the conclusion that the preferred switching point in Italian blends is the syllable boundary; when this is not the original situation, it may nevertheless turn out to be the final product of the blending process.10 The picture emerging from the analysis of this Italian corpus is fairly con- sistent with the experimental results reported above. Italian blends do not seem to indicate a neat preference for either the On+Rh or the Bo+Co struc- ture. This datum stands in striking contrast to the situation described by Kubozono (1989) for English and Japanese, where On+Rh and Bo+Co, respec- tively, clearly prevail.
10 I did not consider examples involving intervocalic /sC/ clusters, whose syllabic interpreta- tion in Italian is unclear (Bertinetto, in prep. b). However, this cluster occurred in only one case.
Psycholinguistic evidence for syllable geometry 19
(7) Most frequent types of Italian blends with respect to syllable structure Type of splitting Percent distribution Example Sy + Sy 40 (50.6%) Corsera cor.rie.re + se.ra On + Sy 16 (20.2%) polstrada po.li.zi.a + stra.da On + Rh(=Sy) 15 (18.9%) Farmitalia far.ma.ci + I.ta.lia minor types 8 (10.1%)
In order to verify Kubozono’s assertion concerning English blends, I also made an inquiry into a corpus of 44 such formations taken from Lehrer (1996). This yields an interesting comparison with the Italian data. Indeed, the situation in English appears to be not only significantly different from, but also not even directly comparable to Italian. The main difference is that a remarkable proportion of English blends exhibit overlap between Word1 and Word2. Consider for instance: boatel > boat + hotel, which may be either Sy+Rh or Bo+Sy due to the consonant that the two words have in common. In fact, in the corpus of English blends considered, 18 examples (namely 41%) exhibit overlapping, while the corresponding proportion in the Italian corpus is much smaller (only 4 examples, i.e. 3.5%). The table in (8) presents an overview of the data, neglecting the 3 overlapping examples which do not lend themselves to any clear-cut interpretation. The percentages are pro- vided for the ‘univocal’ and ‘total’ columns; the latter encompasses the 26 univocal divisions and the 28 ambiguous ones, corresponding to 15 exam- ples with double interpretation (two of which present one division with uncertain interpretation; see Bertinetto (in prep. a) for further details).11
11 To clarify: a) Examples with univocal interpretation: 26 (See the examples in (8).) b) Examples with ambiguous interpretation: 15 (E.g. Spanglish < Spanish + English.) c) Examples with unclear interpretation: 3 (E.g. swingle < single + week.) The examples with overlapping are 18 (b + c). In set (b), there are two examples presenting one division not amenable to clear interpretation in terms of syllable structure, (e.g. sportianity < sport + Christianity, corresponding to the type Sy + Rh, but possibly also to an undefinable type if /t/ is taken from Word2). This yields a total of 28 interpretable divisions in set (b), instead of the theoretically conceivable 30 divisions. Note that in my classification I have been conservative, for the velar /n/ in Spanglish would suggest a Sy + Co division, rather than On + Sy. However, velarization may also be regarded as an automatic restoration process.
20 Pier Marco Bertinetto
(8) Most frequent types of English blends with respect to syllable structure Type of splitting Distribution Example Totals Univocal Sy + Sy 11 (42.3%) busnapper bus + kidnapper 17 Ambiguous Sy + Sy 6 guestimate (also Bo+Co) (31.4%) guess + estimate Univocal On + Rh 12 (46.1%) brunch breakfast + lunch 17 Ambiguous On + Rh 5 Bisquick (also Sy+Sy) (31.4%) biscuit + quick Univocal Sy + Rh 2 (7.6%) shortalls short + overalls 7 Ambiguous Sy + Rh 5 sloburb (also Bo+Sy) (12.9%) slob + suburb Univocal Bo + Sy 1 (3.8%) infomercial information + commercial 7 Ambiguous Bo + Sy 6 chatire (also Sy+Rh) (12.9%) chat + satire Ambiguous Minor types 6 6 (11.1%)
As may be seen in (8), the proportion of the On+Rh type is considerably higher than in the Italian corpus, which includes only 3 such cases (two of which with ambiguous interpretation). In addition, the Bo constituent emerges as such in only one example (which is furthermore devoid of significance given the independent existence of the clipping info), whereas in the remaining cases the Bo emerges in only one of the two possible divi- sions of the blend. Now, since it turns out that there is a high degree of intersection between the types Bo+Sy and Sy+Rh, and since the Rh enjoys a much stronger status as an independently existing constituent, I believe we can justifiably claim that these English data provide strong additional support for the right-branching structure of the syllable in this language, while at the same time endorsing the conclusion that the internal organiza- tion of the syllable in Italian may indeed be significantly different.
7. On the role of the syllable
In the preceding sections I discussed evidence bearing on the following facts:
Psycholinguistic evidence for syllable geometry 21
(a) The internal organization of the syllable may clearly emerge in psycho- linguistic experimentation. (b) Different languages may yield conflicting results in comparable experi- mental tasks. Indeed, while English has repeatedly been proven to be an On+Rh language (§3), there are strong indications that Korean might be an example of the Bo+Co type (§4), whereas Italian seems at best to be weakly oriented towards the On+Rh type (§5). Apparently, there is not just one type of syllabic hierarchy; and it is even possible that in some lan- guages the syllable presents no internal hierarchy, as predicted by the flat structure model shown in (1a) in §3. (c) Encouraging data emerge from the comparison of experimental results with results stemming from naturally occurring forms of linguistic be- haviour, such as speech errors and blends generated for word formation purposes (§5–6). There seems to be a convergence of structural tenden- cies in these diverse domains. (d) Higher-order prosodic factors, such as marginality, stress, word struc- ture, syllable structure, prosodic domains larger than the syllable (e.g. the foot) or smaller (e.g. the mora), undoubtedly have a strong influence on the overall behaviour of the speakers. The different modulation of these factors in different languages, shapes the prosodic component in the way attested by speakers in both their spontaneous and experimentally elic- ited behaviour. Further questions that deserve special attention, in the light of the above discussion, are the following: (A) What are the reasons for the varying importance of the internal organiza- tion of the syllable in different languages? Specifically, why do some lan- guages appear to present a very low syllabic profile? (B) What is the role of the syllable in natural languages, especially with respect to the contrast ‘high’ vs. ‘low profile’ in syllable organization? In the remainder of this paper, I shall attempt to examine these two points, although, at the present stage, the answer can only be rather speculative. As to point (A), one might suggest that isolating languages provide a fairly obvious candidate for a language type where the internal hierarchy of the syllable does not seem to play a major role, for words tend to coincide with syllables exhibiting a rather simple structure. In such languages, the speaker does not need to develop sophisticated procedures for the segmentation of the speech chain, other than what is necessary to recognize the individual words. However, the experimental investigations carried out so far by Wang & Derwing (1993) do not appear to confirm this hypothesis, at least not in its stronger version. Clearly, further inquiry is needed.
22 Pier Marco Bertinetto
As to the evidence pointing towards a low profile of the internal hierarchy of the syllable in Italian, one might suggest that this is related to the rather simple syllabic structure of Italian in comparison with, say, English. The argument might go as follows. Due to its rather elementary syllabic struc- ture, Italian does not need to develop an elaborate processing strategy to help the speaker assemble the speech chain into chunks of segments conforming to the phonotactics of the language, ultimately guiding her/him in the process of lexical recognition. By contrast, a language like English, which exhibits a much more complicated syllable structure, might be in need of establishing precisely this sort of mechanism, based on a rigid internal hierarchy of syllabic constituents. On this view, the prominent hierarchy of the English syllable would then be a consequence of the more complicated phonotactics of the language. In other words, since English, as opposed to Italian, allows fairly complex sequences of segments, it should only be natural that these should develop a subtler scale of intersegmental cohesion, with the result that the reciprocal attractions would be more finely graded, giving rise to the observed behaviour in terms of separability of syllabic constituents. Thus, a rigid syllabic hierarchy would ultimately be the consequence of the general phonotactic properties of the language, rather than the underlying principle around which phonotactics should be organ- ized. It is easy to see that the solution sketched here may easily be recon- ciled with the theoretical views of the now growing body of scholars con- tending that the syllable should be regarded as an epiphenomenon of more fundamental properties (cf. Vennemann, 1988a,b; 1994; Clements, 1990; Dziubalska-Kołaczyk, 1995; Dziubalska-Kołaczyk & Dressler, 1995; Ohala & Kawasaki-Fukumori, 1997). There is however an important caveat to consider. If this hypothesis is correct, why should a language like Korean, with a syllabic structure not significantly more complex than that of Italian, present a strong hierarchical arrangement, whereas Italian apparently does not? We are of course not attracted by the highly unlikely hypothesis that languages with a simple syllabic structure correlate either with left-branching or with flat structure, while languages with a complex syllabic structure would require right- branching geometry. Although the reason for the structural contrast be- tween Korean and English is not obvious, one would wish to correlate this feature with other formal properties of the grammar of these languages (see fn. 3), rather than with gratuitous reasons such as the one given above.12
12 One possible way out of this dilemma, compatible with the findings discussed in the present paper, is the following. The type of experimentation in which Korean subjects re- vealed a sharp preference for left-branching geometry is of the same type (namely, blending preference with monosyllables) as that in which Italian speakers also exhibited a clear pref-
Psycholinguistic evidence for syllable geometry 23
In any case, languages like Italian seem to exhibit a sort of ‘variable geometry’ structure at the syllabic level. Depending on the particular effect of the various forces acting upon it, the syllable may assume alternative structural profiles, i.e. different syllabic constituents may in turn be singled out. Consider for instance the different effects of word-final position in mono- and disyllables, of stress in trisyllables, of different word structures, and so on (see §5). To put it somewhat bluntly, the geometry of the Italian syllable might turn out to be flat, left- or right-branching, depending on the pressures acting upon it, although with a preference for the latter solution when no disturbing effects are at play. If this variable geometry model proved to be a valid one, Italian would provide a very good example of Vennemann’s (1988a) view of the syllable, conceived of as a dynamic struc- ture rather than a static template, as is the case in most theories. Indeed, diachronic linguistics shows abundant evidence that languages may reshape their syllable structure in the course of time, suggesting that alternative arrangements are always (latently) available. Let us now consider point (B). The most extreme interpretation of the epiphenomenal view of the syllable recalled above would be one in which the syllable is not considered a basic phonological unit, but the mere reflex of other, more fundamental phonotactic properties. Let us call this the ‘radical’ view. At first sight, this interpretation, which has recently been put forward by Dziubalska-Kołaczyk (1995), looks easy to reconcile with the findings discussed above, both in the case of languages with a low syllabic profile, and in the case of languages with a clear orientation at the syllabic level. This position would certainly be preferable to an alternative one, where the syllable is considered to be a deep component only in languages presenting a definite hierarchical organization. According to the latter view, speakers of a language like English would supposedly gain an advantage by adhering to a right-branching structure, in the sense that this would give more efficiency to the segmentation strategy in the process of word recog- nition. However, this is at odds with the findings of Mehler et al. (1981), ac- cording to which the syllable does not seem to play a role in the segmenta- tion of English words, as opposed to what happens in a language like French (Cutler et al., 1986).13 Thus, the real advantage of a right-branching syllable
erence for a hierarchical organization, although in the opposite direction (see task 28 of Table 1). This leaves intact the possibility that Korean speakers, if confronted with substitu- tion tasks, would exhibit the same kind of behaviour as Italians, i.e. a form of behaviour pointing towards a low syllable profile. Further research is obviously needed. 13 It is not clear, given the not entirely converging evidence stemming from other recent studies, whether this difference is related to the fact that French, as opposed to English, is supposed to be syllable-timed. In any case, the solution of this problem would go far be- yond the scope of this paper.
24 Pier Marco Bertinetto structure for an English speaker, whatever it is, does not seem to be con- nected with strategies of word recognition. Nevertheless, the complete dismissal of the syllable as a basic phonolog- ical component, as argued by the ‘radical’ interpretation sketched above, involves serious problems. We know from a number of studies that speakers are often sensitive to the presence of syllabic units. This has been proven, for instance, in tasks of ‘sound similarity judgements’ performed with Korean subjects (Yoon & Derwing, 1995), in tasks of masked priming of word and picture naming performed with English speaking subjects (Ferrand & Segui, 1996), or in tasks of short-term recalling of word lists performed again with English speaking subjects (Treiman et al., 1994). As the latter authors suggest, although the syllable may not be relevant as an on-line segmenta- tion unit, it may become important at other stages of processing, suppos- edly playing a decisive role in remembering speech and in lexical access. This invites a less radical interpretation of the epiphenomenal view of the syllable. The new position would consist in saying that although fine seg- mentation into syllables is not a primary ingredient in speech processing, the syllable as such remains a basic unit of the phonological component. The apparently contradictory character of this statement dissolves once we realize that, of the two fundamental aspects of the syllable (i.e. the defini- tion of the boundaries and the individuation of the nucleus), only the sec- ond is of core relevance. In other words, the syllable is indeed a basic unit of phonology, but one with no assigned internal structure. The latter should be viewed as the real epiphenomenon, emerging under the pressure of both low-level factors (i.e. phonotactics) and higher-order ones (i.e. stress and prosodic domains other than the syllable). This implies that the level at which we observe subsyllabic clustering is shallower than that at which syllabic nuclei exert their action. Syllable constituents evolve as preferential aggregations, or force fields, whose stability differs from language to lan- guage. They do not exist as primitives of phonological organization, but are instead the result of the complex interplay of the overall structure of the language. Yet, once they come into existence, they may acquire prominence and become part of the speakers’ competence, to the extent that they: (a) may be exploited at some (presumably later) stages of speech processing, and (b) may guide, to some extent, certain forms of linguistic behaviour (such as speech errors, language games, natural and elicited blendings etc.). The emerging of a certain type of syllable geometry in a given language, however, should not be regarded as equivalent to the fixation of a rigid template, with the radical exclusion of any other conceivable aggregation. The phonology of natural languages is a complex organism, allowing many sorts of recombinations. Rather, what we expect to find is a prevailing orientation of syllable-sensitive phonetic regularities and phonological
Psycholinguistic evidence for syllable geometry 25 processes along the predominant direction indicated by the internal geometry to be observed in the given language. On this view, the stronger role played by the subsyllabic organization in English with respect to Italian is likely to be matched by a larger number of syllable-sensitive processes in the former language. Although this statement needs further validation, the relative wealth of phenomena of English phonology to be analysed in rela- tion to syllable structure lends initial support to this hypothesis (cf. the flapping of intervocalic dental stops in certain varieties (Kahn, 1976), /n/ velarization and /h/ deletion (Borowsky, 1986), /l/ velarization etc.). This is likely to be concrete evidence of syllabic structuring for speakers of English, and apparently helps shape their phonological competence. Obviously, further research is needed to strengthen and refine the hy- potheses sketched here. As said at the outset, this kind of enterprise re- quires strict co-operation between theoretical and experimental investiga- tion; the convergence of these two lines of research will help reject mistaken assumptions. In any case, the evidence gathered so far seems strong enough to cast doubt on some assumptions, despite the support they have traditionally enjoyed. In particular, the following two statements should now be taken as a robust starting point:
(i) Syllables DO NOT present the same basic geometry in all natural lan- guages; (ii) Syllable organization is NOT a self-contained structure, independent of higher-order prosodic factors.
References.
Beckman, M. (1995). On blending and the mora: Comments on Kubozono, Phonological and Phonetic Evidence. Papers in Laboratory Phonology IV, edited by Bruce Connell & Amalia Arvaniti. Cambridge: Cambridge University Press. Berg, T.. (1989). Intersegmental cohesiveness. Folia Linguistica 23, 245–280. Berg, T. (1991). Phonological processing in a syllable-timed language with pre-final stress: Evidence from Spanish speech error data. Language and Cognitive Processes 6, 265–301. Bertinetto, P.M. (in prep. a). Lexical blends and syllable structure. Bertinetto, P.M. (in prep. b). La sillabazione dei nessi /sC/ in italiano: Un’eccezione alla tendenza ‘universale’?. Bertinetto, P.M., L. Cioni & M. Agonigi (in prep.). Experimental evidence on the internal organi- zation of the syllable in Italian. Borowsky, Toni (1986). Topics in the Lexical Phonology of English. Amherst: Ph.D. Dissertation. Cheon, S.B. (1980). Lapsus linguaeuy umwunloncek haysek (A phonological interpretation of lapsus linguae). Ene 5, 15–32. Clements, G.N. (1990). The role of the sonority cycle in core syllabification. Papers in Laboratory Phonology I. Between the Grammar and Physics of Speech, edited by J. Kingston & M. Beckman, 283–333. Cambridge: Cambridge University Press.
26 Pier Marco Bertinetto
Clements, G.N. & S.J. Keyser (1983). CV Phonology: a Generative View of the Syllable. Cambridge, Mass.: MIT Press. Cutler, A., J. Mehler, D. Norris & J. Segui (1986). The syllable’s differing role in the segmentation of French and English. Journal of Memory and Language 25, 385–400. Davis, S. (1989). On a non-argument for the rhyme. J. of Linguistics 25, 211–217. Derwing, B.L. (1997). Testing phonological universals in the laboratory. Certamen phonologicum III. Papers from the Third Cortona Phonology Meeting (April 1996), edited by P.M. Bertinetto, L. Gaeta, D. Michaels & G. Jetchev. Torino: Rosenberg & Sellier. Derwing, B.L. & T.M. Nearey (1994). Sound similarity judgements and segment prominence: A cross-linguistic study. Proceedings of the ICSLP, 351–354. Derwing, B.L. & G.E. Wiebe (1994). Syllable, mora or segment? Evidence from global sound similarity judgements in Japanese. Proceedings of the 1994 annual conference of the Canadian Lin- guistic Association, edited by P. Koskinen, 155–163. Toronto Working Papers in Linguistics, Lin- guistic Graduate Course Union. Derwing, B.L., M.L. Dow, T.M. Nearey (1988). Experimenting with syllable structure. ESCOL, 83–94. Derwing, B.L., T.M. Nearey, R.A. Beinert & T.A. Bendrien (1992). On the role of the segment in speech processing by human listeners: Evidence from speech perception and from global sound similarity judgements. Proceedings of the 1992 International Conference on Spoken Language Processing, edited by J.J. Ohala, T.M. Nearey, B.L. Derwing, M.M. Hodge & G.E. Wiebe. Edmonton, Alberta: 289–292. Derwing, B.L., Y.B. Yoon & S.W. Cho (1993). The organization of the Korean syllable: Experi- mental evidence. Japanese/Korean Linguistics, edited by Patricia M. Clancy, Stanford, CA: Center for the Study of Language and Information, 223–238. Dziubalska-Kołaczyk, K. (1995). Phonology without the Syllable. A Study in the Natural Framework. Poznan´: Motivex. Dziubalska-Kołaczyk, K. & W.U. Dressler (1995). Evidence of syllabic writing against the phono- logical syllable. Phonologica 1992, edited by W.U. Dressler, H.C. Luschützky, O.E. Pfeiffer & J.R. Rennison. Torino: Rosenberg & Sellier, 65–76. Ferrand, L. & J. Segui (1996). Masked priming of word and picture naming: The role of syllabic units. Journal of Memory and Language 35, 708–723. Finney, S.A., A. Protopapas & P.D. Eimas (1996). Attentional allocation to syllables in American English. Journal of Memory and Language 35, 893–909. Fowler, C.A. (1987). Consonant-vowel cohesiveness in speech production as revealed by initial and final exchanges. Speech Communication 6, 231–244. Fowler, C.A., R. Treiman & J. Gross (1993). The structure of English syllables and polysyllables. J. of Memory and Language 32, 115–140. Fudge, E. (1989). Syllable structure: A reply to Davis. J. of Linguistics 25, 219–220. Gim, C.G. (1987). A study on syllable structure and some processes in its nucleus in Korean. Mal 12, 25–69. Kahn, D. (1976). Syllable-based Generalizations in English Phonology. Bloomington: Indiana University Linguistics Club. Kubozono, H. (1989). The mora and syllable structure in Japanese: Evidence from speech errors. Language and Speech 32, 249–278. Kubozono, H. (1995). Perceptual evidence for the mora in Japanese. Phonological and Phonetic Evidence. Papers in Laboratory Phonology IV, edited by B. Connell & A. Arvaniti, 141–156. Cambridge: Cambridge University Press. Lehrer, A. (1996). Identifying and interpreting blends: An experimental approach. Cognitive Linguistics 7, 359–390.
Psycholinguistic evidence for syllable geometry 27
Loporcaro, M. (1996). On the analysis of geminates in Standard Italian and Italian dialects. Natural Phonology: The State of the Art, edited by B. Hurch & R.A. Rhodes, 153–188. Berlin: Mouton de Gruyter. Magno-Caldognetto, E., L. Tonelli, K. Vagges & P. Cosi (1987). The organization of constraints in phonological speech errors. Proceedings of the 11th International Congress of Phonetic Sciences, Tallinn, 173–178. Magno-Caldognetto, E. & C. Zmarich (1995). I modelli funzionali psicolinguistici e la generazione delle disfluenze nei balbuzienti. Studi Italiani di Linguistica Teorica e Applicata 24, 479–508. Magno-Caldognetto, E., M. Panzeri & L. Tonelli (to appear). Nessi e geminate nei lapsus. Evidenza esterna per un modello di sillaba in italiano. Atti del XXXI Congresso Internazionale di Studi della SLI: Fonologia e morfologia. Roma: Bulzoni. Mehler, J., J.-Y. Dommergues, U. Frauenfelder & J. Segui (1981). The syllable’s role in speech segmentation. Journal of Verbal Learning and Verbal Behavior 20, 298–305. Niemi, J. & M. Laine (1997). Slips of the tongue as linguistic evidence: Finnish word initial segments and vowel harmony. Folia Linguistica 31 , 161–175. Ohala, J.J. & H. Kawasaki-Fukumori (1997). Alternatives to the sonority hierarchy for explaining segmental sequential constraints. Language and Its Ecology: Essays in memory of Einar Haugen, edited by S. Eliasson & E.H. Jahr, 343–365. Berlin: Mouton de Gruyter. Otake, T., G. Hatano, A. Cutler & J. Mehler (1993). Mora or syllable? Speech segmentation in Japanese. Journal of Memory and Language 32, 258–278. Pierrehumbert, J. & R. Nair (1995). Word games and syllable structure. Language and Speech 38, 77–114. Sells, P. (1995). Korean and Japanese morphology from a lexical perspective. Linguistic Inquiry 26, 277–325. Thornton, A.M. (1993). Italian blends. Natural Morphology. Perspectives for the Nineties, edited by L. Tonelli & W.U. Dressler, 143–156. Padova: Unipress. Treiman, R. (1983). The structure of spoken syllables: Evidence from novel word games. Cognition 15, 49–74. Treiman, R. (1985). Onsets and rimes as units of spoken syllables: Evidence from children. J. of Experimental Child Psychology 39, 161–181. Treiman, R. (1986). The division between onsets and rimes in English syllables. J. of Memory and Language 25, 476–491. Treiman, R. (1988). Distributional constraints and syllable structure in English. J. of Phonetics 16, 221–230. Treiman, R., A. Salasoo, L.M. Slowiaczek & D.B. Pisoni (1982). Effects of syllable structure on adults’ phoneme monitoring performance. Research on Speech Perception, Progress Report no.8, Dept. of Psychology, Indiana University, 63–81. Treiman, R., K. Straub & P. Lavery (1994). Syllabification of bisyllabic nonwords: Evidence from short-term memory errors. Language & Speech 37, 45–60. Treiman, R., C.A. Fowler, J. Gross, D. Berch & S. Weatherston (1995). Syllabic structure or word structure? Evidence for onset and rime units with disyllabic and trisyllabic stimuli. Journal of Memory and Language 34, 132–155. Trubeckoj, N.S. (1939). Grundzüge der Phonologie. Göttingen: Vandenhoeck & Ruprecht. Vennemann, T. (1988a). Preference Laws for Syllable Structure and the Explanation of Sound Change. Berlin: Mouton de Gruyter. Vennemann, T. (1988b). The rule dependence of syllable structure. On Language: Rhetorica, Phonologica, Syntactica. A Festschrift for Rh.P. Stockwell, edited by C. Duncan-Rose & T. Vennemann, 257–283. London: Routledge.
28 Pier Marco Bertinetto
Vennemann, T. (1991). Syllable structure and syllable cut prosodies in Modern Standard German. Certamen phonologicum. Papers from the 1990 Cortona Phonology Meeting, edited by P.M. Bertinetto, M. Kenstowicz & M. Loporcaro, 211–243. Torino: Rosenberg & Sellier. Vennemann, T. (1994). Universelle Nuklearphonologie mit epiphänomenaler Silbenstruktur, Universale phonologische Strukturen und Prozesse, edited by K.H. Ramers, H. Vater, H. Wode, 7–54. Tübingen: Niemeyer. Wang, H.S. & B.L. Derwing (1993). Is Taiwanese a ‘body’ language? Proceedings of the 1993 Annual Conference of the Canadian Linguistic Association, 679–694. Toronto Working Papers in Linguistics, Linguistic Graduate Course Union. Wiebe, G.E. & B.L. Derwing (1994). A forced-choice word blending task for testing intrasyllabic break points in English, Korean and Taiwanese. The Twenty-First LACUS Forum, edited by M.J. Powell, 142–151. Chapel Hill, N.C.: LACUS. Yoon, Y.B. (1994). CV as a phonological unit in Korean. Proceedings of the ICSPL, Yokohama, 483–486. Yoon, Y.B. & B. Derwing (1994). The sound similarity judgement of Korean CVC’s by Korean and English speakers, edited by P. Koskinen. Proceedings of the 1994 annual conference of the Canadian Linguistic Association, 657–665. Toronto Working Papers in Linguistics, Linguistic Graduate Course Union. Yoon, Y.B. & B. Derwing (1995). Syllable saliency in the perception of Korean words. Proceedings of the XXth International Conference of Phonetic Sciences, Stockholm, 602–605.
Closet syllables in a syllable-denying framework A case of déjà vu?
Wiebke Brockhaus
Abstract
This paper addresses an issue which is often seen as highly controversial by phonologists when first coming into contact with Government Phonol- ogy, viz. the absence of a syllable node from this framework. Taking the currently wide-spread view that the syllable is an essential unit of phono- logical structure as its point of departure, the paper briefly compares today’s conception of the syllable with the alternatives which are in place The Hague: Thesus. 29–48. Thesus.The Hague: 29–48.
. in Government Phonology. It then draws some parallels, and notes some crucial differences, between Government Phonology and the one other major generative framework to reject the syllable, viz. that of the Sound
Syllables !? Pattern of English (Chomsky & Halle, 1968). The final part of the paper explores why nasal assimilation and glide formation in Spanish have been taken as phonological events which require reference to the syllable and demonstrates that Government Phonology can offer a straightforward alternative account of the relevant data, suggesting that the syllable may Phonologica 1996: Phonologica 1996: not be essential after all. 0.
1. Introduction
In this paper I consider the role of the syllable in Government Phonology (GP),1 which, it could be argued, is the only fully articulated phonological theory today to deny the existence of a syllable constituent universally. The central issue to be resolved is whether dispensing with the syllable is simply
1 See Brockhaus (1995b), Harris & Lindsey (1995) and Kaye (1995) for useful introductions to all aspects of GP. More detailed or more specialized work in this theory can be found in, for example, Brockhaus (1995a), Charette (1991), Gussmann & Kaye (1993), Harris (1990; 1992;
John R. Rennison & Klaus Kühnhammer, eds. (1999). eds. R. Rennison & Klaus Kühnhammer, John (1999). 1994), Harris & Kaye (1990), Kaye (1989a; 1990; 1992a,b) and Kaye et al. (1985; 1990).
30 Wiebke Brockhaus a retrograde step, which returns phonological theory to the well-discussed problems of SPE-type classical generative phonology (e.g. Chomsky & Halle, 1968) or whether the syllable has in fact become redundant in current work and should, therefore, be eliminated by Occam’s razor. I begin by briefly discussing those constituents which, unlike the syllable, are recognized in GP and gathering some arguments in favour of one of these constituents (the onset). I then turn to the question of whether GP really operates without the syllable, as its proponents claim (see e.g. (1) below), or whether it makes reference to phonological units which are syllables except in name. I explore this issue by drawing parallels between GP and SPE, the one other major generative framework to reject the syl- lable. First, I investigate what the absence of the syllable means for the structure of phonological representations in the two frameworks and then examine the significance of informal and (apparently) formal references to the syllable. Pursuing the parallels between SPE and GP further, I consider data from Spanish which, in the literature of the early 1970s, was cited as compelling evidence in favour of the syllable in phonological theory. Indeed, a syllable-based account of the data can be shown to be more transparent and simpler than a non-syllabic one. However, I demonstrate that a straight- forward GP approach is also available, which suggests that, given the machinery of present-day phonological theory, it is no longer necessary to refer to the syllable.
2. The GP view: rejecting the syllable but recognizing onset, nucleus and rhyme
Government Phonology is “a major contemporary theory of phonology” (Trask, 1996: 161) which is growing and developing very rapidly. Although certain fundamental principles are adhered to by all proponents of GP, it is only to be expected that researchers adopt somewhat differing views of particularly active areas for research. The version of GP on which this paper is based is a somewhat conservative, mainstream one which has its roots in publications of the early 1990s. Because the syllable is rejected as a con- stituent here, as stated in (1), it is obviously necessary for other structures and mechanisms to be in place which handle those aspects of phonological theory assumed to provide justification for the syllable in other frameworks. These aspects are summarized in (2).
(1) Following an idea first proposed by Vergnaud (1982), rather than pos- tulating a syllable constituent, we define a tier of representation which conforms to the pattern (O R)* — that is an arbitrary number of repeti- tions of the pattern O R. (Kaye et al., 1990: 201)
Closet syllables in a syllable-denying framework? 31
(2) [The syllable is an indispensable concept in phonology, for three main reasons: a) It is] a natural domain for the statement of many phonotactic constraints b) Phonological rules are often more simply and insightfully ex- pressed if they explicitly refer to the syllable c) Several phonological processes are best interpreted as methods to ensure that the string of phonological segments is parsable into syllables (Kenstowicz, 1994: 250)
I have shown elsewhere (Brockhaus, 1995a; to appear) that such structures and mechanisms can be successfully employed in GP, but a detailed discus- sion of these would be beyond the scope of the present paper. Suffice it to say that the GP structures include the three constituents shown in (3), and that the mechanisms for phonological processing are governed by, among others, the general Licensing Principle in (4) and the more specific “Coda” and Onset Licensing Principles in (5) and (6). So, GP recognizes onsets, nuclei and rhymes, but not syllables, and many of the functions assumed by syllable structure in other frameworks are handled by licensing relations in GP. (For further details of GP constituents and prin- ciples of grammar, see the references in fn. 1.)
(3) The three GP constituents. (The arrows indicate intra-constituent licensing relations, and heads are emboldened.) a) Onset b) Nucleus c) Rhyme R O O O N N N
x x x x x x x x (4) Licensing Principle (Kaye, 1990: 306) All phonological positions save one must be licensed within a domain. The unlicensed position is the head of this domain. (5) “Coda” Licensing Principle (based on Kaye, 1990: 311) A post-nuclear rhymal position must be licensed by a following onset. (6) Onset Licensing Principle (Harris, 1992: 380) An onset head must be licensed by a following nuclear position.
Although syllable-based and non-syllabic frameworks differ in at least one crucial aspect, they may also share certain levels of structure or at least exhibit similarities which make a comparison meaningful. Such similarities
32 Wiebke Brockhaus exist at the level immediately above the skeletal tier, and I would like to focus on this level in order to discuss some of the reasons which prompted GP researchers to recognize the three constituents in (3). In syllable-based frameworks, this is where we would find syllable- internal structure. What exactly this structure should look like is still very much a matter for debate, as illustrated by the detailed critical survey of a range of proposals which is presented in Blevins (1995). Blevins’s own pro- posal for syllable-internal structure, which is arrived at on the basis of this survey, is shown in (7).
(7) Blevins’s (1995: 216) proposal for syllable-internal structure σ R N
1X1 2X2 3X3 4X4 5X5 d V i m Blevins, like GP, recognizes a nucleus and a rhyme, but, unlike GP, she rejects an onset constituent. For arguments in favour of the first two con- stituents, see her paper and the references therein, but note that the con- ception of the nucleus and the rhyme in GP differs somewhat from Blevins’s.2 Let us focus on the onset, or rather its absence, in Blevins’s account. In- stead of allowing such a constituent to intervene between the segments to the left of the nucleus and the syllable node itself, she attaches these mar- ginal segments, or rather their skeletal positions, directly to the syllable node. This strategy is familiar, particularly from Moraic Theory (e.g. Hayes, 1989). It has a number of disadvantages, though, which have been discussed in detail by Noske (1989; 1992). He argues, in my view convincingly, that in the absence of an onset constituent it is much more difficult, if not impossible, to provide a satisfactory explanation for the insertion of a default con- sonant at the beginning of certain vowel-initial syllables in languages like German. Accounting for the spreading of, say, a high vowel to form a hiatus- breaking glide in languages such as French is equally problematic. To this I would add the following argument. Leaving aside any additional stipulations, the rejection of an onset constituent suggests that, in a
2 In GP, constituents are treated as maximally binary by most researchers (but see Harris, 1994: 77ff. for a notable exception), so that a rhyme constituent dominating three skeletal points, as in (7), would be ill-formed. The final position would be assigned to a following onset instead.
Closet syllables in a syllable-denying framework? 33 structure such as the one shown in (7), the relationship between X1 and X2 has exactly the same status as that between X2 and the rhyme, since all three are daughters of the syllable node. This has a number of implications of which I can mention only one here. Specifically, we would expect the same phonotactic constraints to hold between X1 and X2 on the one hand, and X2 and the rhyme on the other. This, however, is not what we find. Typically, in languages where phonotactic constraints hold at all between the consonants to the left of the nucleus (e.g. Germanic and Romance languages), these constraints are very tight, whereas only quite loose constraints,3 if any, are manifested between the right-most of these consonants and the rhyme. Moving from such internal evidence to external evidence, we also notice that speech errors and lan- guage games (e.g. Bell & Hooper, 1978: 21f.) suggest that X1 and X2 are more closely linked to each other than X2 is to the rhyme. None of this is reflected in (7), but is very readily captured if an onset constituent is recognized, as is the case in GP.
3. SPE and GP
3.1. Is there really no syllable?
Having reviewed some of the reasons for assuming at least an onset con- stituent, we can now ask whether the GP claim that the syllable has no role to play in phonological theory is really adhered to by those working within this framework. I explore this issue by drawing parallels between GP and classical generative phonology, as put forward in SPE. For the latter, I assume the version of the theory which is most faithful to Chomsky & Halle (1968), that is, the claims made in the original publication itself and in some others which further elaborate on it, but change none of the central tenets (e.g. Harris, 1969). I begin by investigating what the rejection of the syllable in GP and SPE means for the representations in the two frameworks. SPE representations consist solely of matrices of paradigmatically un- ordered feature bundles, with no prosodic superstructure (although Chomsky & Halle, 1968 (e.g. p.9) do discuss a phonological phrase), so that the absence of a syllable constituent leaves no gap whatsoever. GP, by contrast, is a multi-dimensional framework with prosodic hier- archies into which one would expect a syllable constituent to be integrated. The diagram in (8) represents a possible GP prosodic hierarchy.
3 See, for example, Clements & Keyser (1983: 20f.) for a survey of phonotactic constraints between onset and rhyme in English.
34 Wiebke Brockhaus
(8) A possible prosodic hierarchy in GP (partly based on Blevins, 1995: 210) PhP Phonological phrase Wd Wd Prosodic word Σ Σ Σ Σ Foot R R R R R R R R O N O N O N O N O N O N O N O N x x x x x x x x x x x x x x
Considering this prosodic tree from the top down, we find it reminiscent of tree structures in syntax, with each constituent successively consisting of other constituents. However, once we arrive below the level of the foot, we encounter constituents which appear to be entirely undominated, viz. onsets. On the assumption that (8) represents a true GP prosodic hierarchy, we can only conclude that the absence of a syllable constituent causes serious problems for the theory, as prosodic hierarchies become ill-formed. But this apparent problem resolves itself on closer inspection. Instead of adopting the hierarchy in (8), GP has an alternative to offer, which is argued for in Harris (1992; 1994: 153f.). This alternative consists of the licensing hierarchy illustrated in (9), where the licensing relations between individual skeletal positions are represented arboreally and, for presentational pur- poses only, the constituent structures are shown by means of labelled brackets.
(9) Licensing hierarchy in tawdry [tO:dVi] (based on Harris, 1992: 384)
[x1]O [x2 x3]N [x4 x5]O [x6]N
t O d V i
Two points emerge from (9). Firstly, far from dangling unattached in the prosodic tree, onsets are fully incorporated into the licensing hierarchy, so that ill-formed structures with undominated constituents do not arise. Secondly, the licensing relation between a nuclear head (e.g. x2 in (9)) and the preceding onset (x1 in our example) establishes a link between the two constituents. This link is of a different type from the one between the mem- bers of a single constituent, with the intra-constituent licensing relation being head-initial and the inter-constituent licensing relation head-final. The
Closet syllables in a syllable-denying framework? 35 distinction between these two licensing relations is indicated in (9) by the arrows which represent the latter appearing higher up in the diagram than those which stand for the former (e.g. between x2 and x3). Although it is still a matter for research where exactly the differences between the two types of licensing relations lie and how these differences are to be accounted for in the theory, it appears that the intra-constituent licensing relation imposes greater phonotactic restrictions on the positions entering into it than the inter-constituent licensing relation does. This means that the strong phono- tactic constraints within branching onsets can be captured and so can the less restrictive ones between onsets and nuclei. Having seen, then, what the absence of the syllable means for the repre- sentations in both GP and SPE, we can now look beyond the representa- tions as such and consider other issues which are raised by dispensing with the syllable. The first that springs to mind is whether the syllable finds its way into the two theories by some route other than the representations. Is it perhaps referred to elsewhere in the theory? Do researchers make formal reference to it in, say, rules (SPE) or principles and parameters (GP)? If not, is the syllable used in an informal way and, if so, does this matter? Let us begin with the latter question. There can be no doubt that the answer to it must be yes for both GP and SPE. There are many informal references to the syllable in the SPE literature (ten mentions of the syllable on p. 40 of SPE alone) and in, particularly older, GP publications (e.g. Charette, 1991: 11f.). Does this mean that the syllable has found its way into these theories after all? In my view, it does not. Such informal use of the term “syllable” is merely a convenient shorthand notation, just as IPA sym- bols are often used in representations when the details of feature trees or matrices (or element trees in GP) are not relevant to the point under discus- sion. What does matter, though, is whether formal reference is made to the syllable. Let me take SPE first. Here, probably the best known instance of apparently formal reference to the syllable is the introduction of the feature “syllabic”, which Chomsky & Halle (1968: 354) define with reference to a syllable peak. Employing such a feature is clearly in conflict with the claim that the syllable has no role to play in phonological theory. However, Chomsky & Halle do not actually commit themselves to it. They consider the possible advantages of this feature and then decide to refrain from adding “syllabic” to their list of features. It is not invoked in any of Chomsky & Halle’s rules, apart from the few on pp. 354f., where the authors explore the usefulness of such a feature. Chomsky & Halle do, however, make frequent reference to weak and strong clusters, which are defined as in (10).
36 Wiebke Brockhaus
(10) Chomsky & Halle’s (1968: 29) initial definitions of weak and strong clusters A weak cluster is a string consisting of a simple vocalic nucleus fol- lowed by no more than one consonant; a strong cluster is a string con- sisting of either a vocalic nucleus followed by two or more consonants or a complex vocalic nucleus followed by any number of consonants.
No doubt, readers will find these definitions strongly reminiscent of light and heavy syllables, which is indeed very much how Chomsky & Halle’s weak and strong clusters function, for example, for the purposes of stress assignment. When deciding whether including weak and strong clusters in phonological rules constitutes formal reference to certain syllable types, we need to bear two points in mind, though. Firstly, as later revisions of (10) reveal (see particularly pp. 82f. of SPE), the correspondence between weak clusters and light syllables is actually not as straightforward as it may appear at first glance. On further consideration of English stress facts, Chomsky & Halle come to the conclusion that it is necessary to “extend the notion ‘weak cluster’ to include a lax vowel fol- lowed by no more than a single consonant followed by an optional liquid or glide” (p. 82). In other words, the emboldened vowel + consonant sequences in algebra and vertebra are weak clusters. But no phonologist would consider [7bV] as a single syllable of English. The vast majority would analyse the schwa as constituting the nucleus of the penultimate syllable and the plosive + rhotic sequence as the onset of the final syllable in each of our two ex- ample words (see e.g. Giegerich, 1992: 168). So, it appears that at least the weak clusters of SPE are distinct from any recognisable syllables, which means that no syllabic notion is being invoked in this particular context. Secondly, the discussion so far has tacitly assumed that any reference to phonological units or sequences which could, from today’s point of view, be construed as forming syllables is tantamount to a reference to the syllable as such. This, however, involves reading between the lines and speculating about the spirit of a particular analysis, neither of which can be an objective scientific strategy. There is no way of demonstrating that Chomsky & Halle had what we would call a heavy syllable in mind when they employed the sequence VC2. All that we can observe and evaluate objectively is the letter of their analysis, and this letter contains only informal references to the syllable, but not formal ones. This argument also holds for the CV se- quences which frequently appear in SPE stress and laxing rules (e.g. Chomsky & Halle, 1968: 46, 79). Turning to GP now, we are faced with very much the same situation. Al- though informal references to the syllable are fairly common, especially in the older GP literature, there are no formal references to it. Occasionally, events affecting entire onset-rhyme sequences have been discussed (e.g. in
Closet syllables in a syllable-denying framework? 37
Brockhaus, 1992: 252, where the deletion of certain such sequences is as- sumed), but these do not involve clandestine references to the syllable either, for the reasons just spelt out. An additional point relevant to GP is that a formal connection between onset and rhyme exists in this theory anyway, taking the form of the inter- constituent licensing relation illustrated in (9), so that some interaction between the two constituents is expected in any case.
3.2. What are the implications for the grammar of rejecting or recognizing the syllable?
Having seen, then, that the syllable really does not have any formal status in SPE or GP, we can now explore the implications that the presence or other- wise of a syllable constituent has for a phonological theory as a whole. These implications are most easily observed when considering the gram- mars of individual languages. One of the toughest tests is likely to be a data set which has been used in the literature to argue in favour of the syllable. The data set I have in mind was, as far as I am aware, first present in Harris’s (1969) account of Mexico City Spanish. It was picked up by Hooper (1972) in her paper on the syllable in phonological theory, where she discusses several phonological events in the light of Harris’s earlier analysis.4 For reasons of space, I can focus on only two of them, viz. nasal assimilation and glide strengthening, but I believe that these are sufficient to shed light on what the presence or absence of the syllable can mean for the simplicity and transparency of any analysis. Specifically, what we are trying to discover in this section is, as Hooper (1972: 53) puts it, whether, as a consequence of rejecting the syllable, “the resulting analyses are more complex, and obscure rather than illuminate the phonological facts”. Let me begin by introducing the facts. Harris (1969) distinguishes be- tween four pronunciation styles in Mexico City Spanish, which differ slightly with regard to the phonological events I discuss here. These styles are defined in (11), and the relevant non-IPA symbols used by both Harris and Hooper (1972) are shown in (12).
(11) Harris’s (1969: 7) hierarchy of pronunciation styles Largo: very slow, deliberate, overprecise; typical of, for example, trying to communicate with a foreigner who has little com- petence in the language or correcting a misunderstanding over a bad telephone connection.
4 See also Vennemann (1972) for a less detailed discussion of these facts from essentially the same point of view.
38 Wiebke Brockhaus
Andante: moderately slow, careful, but natural; typical of, for example, delivering a lecture or teaching a class in a large hall without electronic amplification. Allegretto: moderately fast, casual, colloquial. In many situations one might easily alternate between Andante and Allegretto in mid-discourse or even in mid-sentence. Presto: very fast, completely unguarded. (12) Some non-IPA symbols used by Harris (1969) and Hooper (1972) (definitions based on Harris, 1969: 21f.) Harris Hooper [y] [y] palatal glide [G1] [yˆ] palatal glide, combined with palatal friction [x1] (none) palatal glide, combined with voiceless palatal friction [w] [w] labial-velar glide [Gw] [Gw] labial-velar glide, combined with velar friction = = boundary between prefix and stem | (none) pause, total cessation of phonation
Beginning with nasal assimilation (Harris, 1969: 8–18), we find that nasals exhibit full place assimilation to following obstruents only word-internally in Andante. In Allegretto, though, nasal assimilation applies both word- internally and across a word boundary, as illustrated in (13).5
(13) Nasal place assimilation to obstruents in Andante and Allegretto a) Assimilation word-internally Andante/Allegretto campo [kampo] ‘country’ triunfo [triuMfo] ‘triumph’ anca [a9ka] ‘rump’ b) Assimilation across a word boundary (blocked in Andante) Allegretto Andante (roughly) um beso [um beso] [un beso] ‘a kiss’ un foco [uM foko] [un foko] ‘a focus’ un gato [u9 gato] [un gato] ‘a cat’
Nasal assimilation is triggered not only by obstruents, but also by glides. Assimilation to glides, though, is restricted to application across a word boundary, as illustrated in (14).
5 There is some partial place assimilation across word boundaries in Andante, but I shall ignore this for simplicity’s sake.
Closet syllables in a syllable-denying framework? 39
(14) Nasal place assimilation to glides6 a) Assimilation across a word boundary (Allegretto) un hielo [uN yelo] ‘an ice (cube)’ un huevo [u9 weßo] ‘an egg’ b) No assimilation word-internally nieto [nyeto] ‘grandson’ nuevo [nweßo] ‘new’
Now, Andante exhibits only word-internal nasal assimilation, so that nasals can never assimilate to glides. Allegretto, by contrast, does permit nasal assimilation across a #-boundary, so that this style of speaking has two types of nasal assimilation: (a) assimilation to obstruents (word-internally and across a word boundary) and (b) assimilation to glides (across a word boundary only). Harris (1969: 16ff.) first of all entertains the possibility of expressing (a) and (b) in the separate rules (15) and (16), respectively.
(15) Harris’s (1969: 16) rule of nasal assimilation to obstruents (to be revised) +obstr αcor αcor β ant → / ___ (#) β ant [+ nasal] γ back γ back δ distr δ distr (16) Harris’s (1969: 17) rule of nasal assimilation to glides (later rejected) − cor − ant G → / ___ # [+ nasal] + distr α back α back
He does, however, come to the conclusion that this would not be a satisfac- tory solution. Firstly, rule (16) seems no more insightful than, say, the same rule with reversed feature values for the first three features in the output specification (i.e. [+ cor, + ant, – distr]). Secondly, as both (15) and (16) capture
6 As an anonymous reviewer has pointed out, assimilation processes which apply across a word boundary, but not word-internally, should be treated with some suspicion. I share this view, but would add that suspicion is justified only where it is clear that no differences exist between the two environments other than the presence or absence of a word boundary. As we shall see later, it seems, however, that the status of the glides in (14a) is not the same as that enjoyed by those in (14b), so that the word boundary may turn out to be entirely irrele- vant, with the absence of assimilation in (14b) being due to some other factor.
40 Wiebke Brockhaus a nasal assimilation process, there should be some way of combining them, but this seems far from straightforward. For these reasons, Harris dispenses with (16) altogether and opts for an alternative which is suggested by the fact that certain underlying glides become obstruents in Andante. Glide strengthening (Harris, 1969: 20–37) applies in Andante in, roughly speaking, two types of environment, “postvocalically”7 and initially (follow- ing a word boundary or a prefix boundary). “Postvocalic” glides strengthen only if followed by another glide, a vowel or a pause. Initial glides strengthen if followed by another glide or a vowel. There is virtually no glide strength- ening in Allegretto at all. (It only occurs prepausally, where it is optional.) The two types of strengthening environment in Andante are illustrated in (17), and Harris’s glide strengthening rule (where V represents both vowels and glides) is shown in (18). (Although (17) provides examples of palatals only, glides with a high-back tongue position also undergo strengthening — hence the mention of /w/ in the input specification of rule (18).)
(17) Glide strengthening in Andante a) “Postvocalically” aya [aG1a] ‘governess’ voy [box1] ‘I go’ b) Initially yema [G1ema] ‘yolk’ subyugar [sußG1uGar] ‘to subjugate’ (18) Harris’s (1969: 24) glide strengthening rule8
V y γ1 V → w / = ___ w γ | #
We have just seen that in Andante glides strengthen in several environments, including, as the first form in (17b) illustrates, the environment where glides trigger nasal assimilation across a #-boundary in Allegretto (see (14a)). So, instead of positing the two separate rules in (15) and (16), Harris now has the option of tying nasal assimilation involving glides to the process of glide strengthening (as observable in Andante). In other words, he can state nasal assimilation in a more general way, provided that it is only triggered by
7 The quotation marks here indicate that I am using the term ‘postvocalic’ in the broadest possible sense, which differs rather from the way in which it is usually understood. All it is intended to indicate is that the relevant glide is preceded by a tautomorphemic vowel. No other claims (e.g. about syllabification of the glide into a coda) are implied. 8 This rule is later revised by Harris (e.g. pp. 34f., 186f.), but, as far as I can see, these revisions do not affect any of the points made in the present discussion.
Closet syllables in a syllable-denying framework? 41 those glides which undergo strengthening in Andante. This would be entirely straightforward if Allegretto also had glide strengthening. Then, all one would have to say is that nasals assimilate to a following obstruent (whether derived from an underlying glide or not) across an optional word boundary, as in the revised nasal assimilation rule in (19), which applies after the glide strengthening rule in (18).
(19) Harris’s (1969: 187) revised nasal assimilation rule +obstr αcor ___ β ant ANDANTE αcor γ back β ant δ distr → / [+ nasal] γ back +obstr δ distr αcor ALLEGRETTO ___ (#)β ant γ back δ distr
Unfortunately, this cannot be the whole solution, because word-initial glides do not strengthen in Allegretto. But this wrinkle can be ironed out by adding one more rule to the grammar which turns the obstruents derived from underlying glides by rule (18) back into glides in Allegretto, once they have done their job for nasal assimilation. This additional rule is shown in (20).
(20) Harris’s (1969: 27) rule reversing glide strengthening
γ1 y w → γ w
The problem with this account is that it crucially relies on what Pullum (1976) refers to as Duke-of-York derivations. A Duke-of-York derivation takes the form in (21), where first B is derived from A, and subsequently B is turned back into A by some other rule.
(21) Duke of York derivation (Pullum, 1976: 83) A → B → A
Most linguists would view such derivations with a good deal of suspicion, and Harris himself refers to his own solution as “somewhat inelegant” (p. 27). Hooper (1972) suggests that a much simpler analysis of the same facts can be developed if one refers to the syllable, which appears in the form of
42 Wiebke Brockhaus syllable boundaries ($) in her rules and representations. She proposes the glide strengthening rule in (22), which she claims to be equivalent to Harris’s rule in (18).9
(22) Hooper’s (1972: 528) glide strengthening rule
y ˆy → w w γ / $ ___
It is clear that Hooper’s rule is simpler than Harris’s and that it captures the intuitively appealing insight that only glides at the (left) edge of a syllable strengthen, which, broadly speaking, means that segments in the syllable nucleus do not undergo this process. (But see fn. 9 for a critical observation on her rule.) Intuitiveness, of course, is not a valid criterion for the evalua- tion of grammars. Still, we can find more objective measures of how Hooper’s syllable-based analysis performs. Firstly, Hooper succeeds in simplifying Harris’s glide strengthening rule in (18), and, secondly, by changing the nasal assimilation rule, she also eliminates the Duke-of-York derivations. She achieves this by removing the [+ obstr] specification from this rule. Instead of requiring nasals to assimilate only to obstruents, as Harris does in (19), Hooper’s rule (23) states that nasals assimilate to any segment at the left edge of a syllable, which makes the right predictions, provided that, as Hooper (1972: 539) argues, the rule is restricted to the word level in Andante, but may apply to larger syntactic units in Allegretto.
(23) Hooper’s (1972: 526) nasal assimilation rule α coronal α coronal β anterior β anterior → [+ nasal] γ back / ___ $ γ back δ distributed δ distributed
As far as glide strengthening is concerned, all Hooper needs to say is that Allegretto, where initial glides never strengthen, simply lacks the rule in (22), so that no Duke-of-York derivations arise. It would be hard to dispute that the introduction of the syllable here both simplifies the grammar and sheds light on the motivation of the observed processes, rather than obscuring their nature, as Harris’s analysis does on occasions.
9 This claim is not entirely justifiable, as Hooper’s rule fails to capture cases of postvocalic prepausal glide strengthening, where the glide would be syllable-final in Hooper’s (1972: 527) own view.
Closet syllables in a syllable-denying framework? 43
Having seen how useful the syllable turns out to be for these Spanish data, we can now ask whether a non-syllabic GP account of the same facts would return us to the rather complex and unilluminating state of affairs of SPE. It seems to me that the answer to this question must be no. In fact, the GP approach appears to be as simple and perspicacious as the syllable- based one. It was already suggested during the discussion of Hooper’s analysis that glide strengthening is sensitive to the position of a segment within the syllable. Specifically, it matters whether the relevant segment occupies the (left) syllable margin or a more central position, that is, the syllable peak or nucleus. Even though GP does not recognize the syllable as such, distinctions between onset and nucleus can still be drawn, as illus- trated in the partial representations of two simple clauses in Figure 1. Before taking a more detailed look at Figure 1, let me review briefly how Hooper (1972: 529) would interpret the data shown there. Given her account of nasal assimilation and glide formation, she would have to assume that in (b) the glide is syllable-initial, whereas in (a) the glide is in the syllable nu- cleus. In other words, in (a) the second syllable must begin with the nasal instead, that is, we have [so$n yenas] in (a) and [soN$yemas] in (b). However, this analysis is open to the charge of being ad hoc, unless inde- pendent evidence to back up the differing syllabifications can be found. Hooper (1972: fn. 5) suggests that some speakers place the syllable bound- ary in [so$n yenas] so that it does not coincide with the word boundary. For these speakers, there is no nasal assimilation, as shown in (a). However, there are other speakers who maintain the unity of syllable boundary and word boundary, which would yield the same syllabification for both son hienas and son yemas. In both cases, the second word would begin with a glide, and nasal assimilation would occur for these speakers. It is not clear, however, on what kind of evidence Hooper’s claims about the placement of syllable boundaries are based, so this particular argument may not be entirely compelling. A perhaps more convincing piece of evidence is provided by Harris (1969: 26), who points out that the glide/obstruent at the beginning of a word such as yemas never alternates with a vowel, whereas in a small number of items, including hienas, the initial segment may be realized as either a glide or a shortened vowel. The existence of the short vowel seems to suggest that the palatal segment is indeed in the syllable nucleus rather than in its margin. All of this is reflected in the GP representations in Figure 1, and in the syllable-based account proposed by Hooper. An important difference between the two, however, lies in the fact that Hooper’s approach (as al- ready observed in fn. 9) makes the prediction that syllable-final glides never strengthen. If we consider the second form in (17a), we can see that this
44 Wiebke Brockhaus a) son hienas [son yenas] b) son yemas [soN yemas] O 1 N1 O2 N2 O3 N3 O4 N4 O5 N5 O1 N1 O2 N2 O3 N3 O4 N4 O5 N5 x x x x x x x x x x x x x x x x x x x s o n y e n a s s o n y e m a s
O1 N1 O2 N3 O4 N4 O5 N5 O 1 N1 O2 N2 O3 N3 O4 N4 O5 N5 x x x x x x x x x x x x x x x x x x
s o • ROOT • • n a s s o • ROOT • e m a s
N N ? ? • RES • • RES •
Ι Ι Ι Α Figure 1 Partial GP representations of son hienas [son yenas] ‘they are hyenas’ and son yemas [soN yemas] ‘they are yolks’ (Allegretto) (based on Hooper, 1972: 528) prediction is incorrect. How GP overcomes this problem will be discussed shortly, once the details of Figure 1 have been explained. The first set of representations gives the forms of the verbs and the nouns roughly as they would appear once affixation has been completed. To keep matters as simple as possible, levels of projection above the onset and nuclear constituents (e.g. the rhyme) have been omitted, as have the licens- ing relations between the individual positions. The melodic material ap- pears in the form of IPA symbols, for ease of reading. The second set of representations in Figure 1 illustrates the effects of 10 phonological interpretation. N2 and O3 in (a) have been removed, so that O2 and N3 are now adjacent to each other. No such change takes place in (b), though.11 On the melody tier, I have replaced some of the IPA symbols by the ac- tual atoms of segmental structure (elements) used in GP. Elements are cognitive units which can be thought of as being associated with certain acoustic (and, to some extent, articulatory) patterns. The exact number of
10 The mechanism involved here was first discussed by Gussmann & Kaye (1993), who refer to it as ‘reduction’. 11 The empty nuclear positions in Figure 1b) are licensed by parameter setting. For detailed discussion of the licensing of empty nuclei, see e.g. Kaye (1995).
Closet syllables in a syllable-denying framework? 45 elements needed, their definitions and their relationships with each other are areas of lively debate at the moment.12 Probably the best known among the many competing proposals are those put forward by Harris (1994) and Harris & Lindsey (1995), the spirit of which I shall follow here.
(24) Some GP elements Element Acoustic Articulatory label manifestation salient property unmarked properties I [i] Extremity large spectral palatal non-labial, high, gap tense… A [ä] Mass convergence of non-high non-labial, tense… F1 and F2 ? [?] Edge abrupt spectral occluded glottal… change N [n] Murmur low frequency nasal non-labial… intensity
GP elements, of which no more than nine are likely to be needed univer- sally, can be interpreted in isolation, as shown in the second column of (24), or in combination with others. They are geometrically arranged along the lines of a very simple feature tree (with a basic structure akin to that pro- posed in McCarthy, 1988), and (24) lists those elements, with their articula- tory and acoustic properties, which are required for representing the glides and nasals in Figure 1. The representation of [son yenas] in Figure 1a has the verb-final nasal attached to the skeletal position dominated by O2, while both the glide and the following vowel of the noun are attached to the skeletal position domi- nated by N3, forming a light diphthong. In [soN yemas], the nasal is again attached to an onset position (O2), while the noun-initial glide, in contrast to [yenas], also occupies an onset position (O3). The representations thus show a structural difference between the glides in [yenas] and [yemas]. In the former, the glide occupies a nuclear position, whereas in the latter it occupies an onset slot. If we assume that associa- tions between melodic material and skeletal positions are present in the lexicon, then all we need to do is to restrict glide strengthening (in Andante) to onset positions. This would mean that /yemas/ emerges as [G1emas], whereas /yenas/ remains [yenas]. And this is where the GP account appears to be superior to that proposed by Hooper (1972: 528). Recall that for her, glides strengthen only syllable-initially, which makes the incorrect prediction
12 See, for example, Backley (1993), Jensen (1994), Szigetvári (1994), Cyran (1995), Foltin (1995), Rennison (1995), Ritter (1996) and Brockhaus et al. (1996).
46 Wiebke Brockhaus that there should be no glide strengthening in a form such as voy in (17a). The principles and parameters of GP, by contrast, would require the final seg- ment of this word to occupy an onset position — and glides which are attached to onset positions are subject to strengthening. In other words, the GP approach proposed here would correctly predict the form [box1], while Hooper’s rules would, counter-factually, generate *[boy]. In Allegretto, which is the pronunciation style illustrated in Figure 1, there is, of course, no initial glide strengthening at all, so that neither (a) nor (b) of Figure 1 is affected. By contrast with initial glide strengthening, nasal assimilation does apply in Allegretto, and the mechanism at work is also shown in Figure 1. The broken line originating from the ROOT node of the verb-final nasal in [soN yemas] indicates that this segment shares the RESONANCE node of the segment occupying the nearest onset position to the right (i.e. the noun- initial glide, attached to O3). Nasal assimilation, like glide strengthening, affects only segments which fill non-nuclear positions. If the glide is part of a nuclear constituent, as in Figure 1a, the nasal is unable to share its RESONANCE node, whereas the sharing of the RESONANCE node between two onsets is very common. There are numerous examples from other lan- guages (including English) where such sharing occurs between two onsets or between an onset and a preceding coda position, which is also illustrated by the data in (13). The same event, however, is extremely rare between nuclei and preceding onsets. As in Hooper’s approach, the absence of word- internal nasal assimilation to glides (see (14)) is accounted for by the fact that these word-internal glides occupy nuclear positions, so that their RESONANCE nodes are not available for sharing by the nasal segments. Clearly, the brief account I have just given is neither comprehensive nor authoritative. It simply provides an illustration of some of the possibilities which are available under the strict SPE-approach of Harris (1969), which rejects the syllable, the syllable-boundary framework of Hooper (1972) and the syllable-denying theory of GP, which does, however, recognize onsets, nuclei and rhymes.
4. Conclusion
In this paper, I hope to have shown that, although SPE and especially the earlier GP literature contain informal references to the syllable, there are no grounds for claiming that the syllable has any formal status in either theory, which means that there are no closet syllables. We have also seen that syllabic accounts can work better than strict SPE-type analyses, such as that of Harris (1969), but this should perhaps not be taken as incontrovertible evidence for a syllable constituent per se. Phonological representations have
Closet syllables in a syllable-denying framework? 47 changed out of all recognition since the late 1960s and early 1970s, and it may well be that today such a constituent has become redundant. I would argue that this is indeed the case in a theory such as GP, and that, by Occam’s razor, the syllable must be dispensed with. Obviously, a short paper such as this may not be fully convincing, as there are many other phonological events which seem much more inextri- cably linked with the syllable than nasal assimilation and glide strengthening in Spanish. I have dealt with some of these in detail, including final devoicing in German (see Brockhaus, 1995a), where no reference to the syllable is needed. In any case, I hope to have further contributed to the growing body of evidence showing that the syllable is no longer required.
Acknowledgements
I would like to thank an anonymous reviewer for detailed and insightful comments.
References
Backley, P. (1993). Coronal: The undesirable element. UCL Working Papers in Linguistics 5, 301–323. Bell, A. & J.B. Hooper (1978). Issues and evidence in syllabic phonology. Syllables and segments, edited by A. Bell & J.B. Hooper, 3–22. Amsterdam: North-Holland. Blevins, J. (1995). The syllable in phonological theory. The handbook of phonological theory, edited by J.A. Goldsmith, 206–244. Oxford: Blackwell. Brockhaus, W.G. (1992). Final devoicing: Principles and parameters. PhD dissertation, UCL, University of London. Brockhaus, W.G. (1995a). Final devoicing in the phonology of German. Tübingen: Niemeyer. Brockhaus, W.G. (1995b). Skeletal and suprasegmental structure within Government Phonology. Frontiers of phonology: Atoms, structures, derivations, edited by J. Durand & F. Katamba, 180–221. Harlow: Longman. Brockhaus, W.G. (to appear). The syllable in German: Exploring an alternative. The syllable: Views and facts, edited by H.G. van der Hulst & N.A. Ritter. Berlin: Mouton de Gruyter. Brockhaus, W.G., M. Ingleby & C.R. Chalfont (1996). Acoustic signatures of phonological primes. Ms., University of Manchester and University of Huddersfield. Charette, M. (1991). Conditions on phonological government. Cambridge: Cambridge University Press. Chomsky, N. & M. Halle (1968). The sound pattern of English. New York: Harper & Row. Clements, G.N. & S.J. Keyser (1983). CV phonology: A generative theory of the syllable. Cambridge, Mass.: MIT Press. Cyran, E. (1995). Vocalic elements in phonology: A study in Munster Irish. PhD dissertation, Catholic University of Lublin. Foltin, R. (1995). Elements in onset position: Speculations about the reduction of the number of elements. Ms., University of Vienna. Giegerich, H.J. (1992). English phonology: An introduction. Cambridge: Cambridge University Press. Gussmann, E. & J.D. Kaye (1993). Polish notes from a Dubrovnik café I: The yers. SOAS Working Papers in Linguistics and Phonetics 3, 427–462.
48 Wiebke Brockhaus
Harris, James W. (1969). Spanish phonology. Cambridge, Mass.: MIT Press. Harris, James W. (1985). Spanish diphthongisation and stress: A paradox resolved. Phonology Yearbook 2, 31–44. Harris, J. (1990). Segmental complexity and phonological government. Phonology 7, 255–300. Harris, J. (1992). Licensing inheritance. UCL Working Papers in Linguistics 4, 359–406. Harris, J. (1994). English sound structure. Oxford: Blackwell. Harris, J. & G.A. Lindsey (1995). The elements of phonological representation. Frontiers of phonology: Atoms, structures, derivations, edited by J. Durand & F. Katamba, 33–79. Harlow: Longman. Harris, J. & J.D. Kaye (1990). A tale of two cities: London glottalling and New York City tapping. The Linguistic Review 7, 251–274. Hayes, Bruce (1989). Compensatory lengthening in Moraic Phonology. Linguistic Inquiry 20, 253–306. Hooper, J.B. (1972). The syllable in phonological theory. Language 48, 525–540. Jensen, S. (1994). Is ? an element? Towards a non-segmental phonology. SOAS Working Papers in Linguistics and Phonetics 4, 71–78. Kaye, J.D. (1989a). Government in phonology: The case of Moroccan Arabic. The Linguistic Review 6, 131–159. Kaye, J.D. (1989b). Phonology: A cognitive view. Hillsdale, N.J.: Lawrence Erlbaum Associates. Kaye, J.D. (1990). ‘Coda’ licensing. Phonology 7, 301–330. Kaye, J.D. (1992a). On the interaction of theories of lexical phonology and theories of phonolog- ical phenomena. Phonologica 1988: Proceedings of the Sixth International Phonology Meeting, Krems, 1–4 July 1988, edited by W.U. Dressler, H.C. Luschützky, O.E. Pfeiffer & J.R. Rennison (1992), 141–155. Cambridge: Cambridge University Press. Kaye, J.D. (1992b). Do you believe in magic? The story of s+C sequences. SOAS Working Papers in Linguistics and Phonetics 2, 293–313. Kaye, J.D. (1995). Derivations and interfaces. Frontiers of phonology: Atoms, structures, derivations, edited by J. Durand & F. Katamba, 289–332. Harlow: Longman. Kaye, J.D., J. Lowenstamm & J-R. Vergnaud (1985). The internal structure of phonological elements: A theory of charm and government. Phonology Yearbook 2, 305–328. Kaye, J.D., J. Lowenstamm & J-R. Vergnaud (1990). Constituent structure and government in phonology. Phonology 7, 193–231. Kenstowicz, M.J. (1994). Phonology in generative grammar. Oxford: Blackwell. McCarthy, J.J. (1988). Feature geometry and dependency: A review. Phonetica 45, 84–108. Noske, R. (1989). Syllable geometry. Belgian Journal of Linguistics 4, 119–144. Noske, R. (1992). Moraic vs. constituent syllables. Silbenphonologie des Deutschen, edited by P. Eisenberg, K.H. Ramers & H. Vater, 284–328. Tübingen: Narr. Pullum, G.K. (1976). The Duke of York gambit. Journal of Linguistics 12, 83–102. Ritter, N.A. (1996). The role of asymmetrical headedness in the reduction theory of elements. Ms., New York University. Rennison, J.R. (1996). Wann ist ein Merkmal kein Merkmal? Inhaltliche Aspekte phonologischer Strukturen in der Rektionsphonologie (Government Phonology). Wiener Linguistische Gazette 57–59, 27–38. Szigetvári, Péter (1994). The special nature of coronal consonants. University Degree thesis, Eötvös Loránd University, Budapest. Trask, R.L. (1996). A dictionary of phonetics and phonology. London: Routledge. Vennemann, Theo (1972). On the theory of syllabic phonology. Linguistische Berichte 18, 1–18. Vergnaud, J-R. (1982). On the theoretical bases of phonology. Paper presented at the 5th GLOW Colloquium, Paris.
Syllables and unencoded speech Clicks and their accompaniments in Xhosa
Grzegorz Dogil, Jörg Mayer & Justus C. Roux
Abstract
In this paper we argue that clicks are a prototypical case of unencoded speech. Unlike other speech sounds clicks do not coarticulate with their phonetic environment; they block vowel coarticulation and may not be recovered from the transitional features which are left after editing them. Unencoded sounds as such present a challenge to most theories of the syllable and to all syllable based theories of coarticulation and coproduction. The Hague: Thesus. 49–60. Thesus.The Hague: 49–60. . 0.
Syllables !? 1. Introduction
The results of the past decades of phonetic and phonological research show that speech is not perceived, produced, or neurally processed on a segmen- tal basis. Normal human speech appears to be highly encoded, that is, time- Phonologica 1996: Phonologica 1996: compressed to a degree at which the acoustic and motor properties are shared by all sounds forming larger speech units such as syllables (Liberman, 1996). The acoustic cues of consonants are usually modulations of the formant frequency of their tautosyllabic vowels, and the spectral properties of the vowels are usually encoded in the spectral properties of the consonants of the same syllable. Furthermore, the articulatory targets for segments are never reached, and instead, the articulatory gestures are merged within the larger linguistic constituents, the primary one being the syllable. All natural speech results from complex encoding, which makes it extremely difficult to locate and isolate the invariant articulatory and acous- tic properties of particular segments. Speech encoding has always been considered prime evidence for larger constituents like the syllable. However, the degree of speech encoding may differ for particular speech sounds, speech tasks, linguistic categories coded in speech, and parts of the speech
John R. Rennison & Klaus Kühnhammer, eds. (1999). eds. R. Rennison & Klaus Kühnhammer, John (1999). code. For example, it has been shown that phonetic properties pertaining to
50 Grzegorz Dogil, Jörg Mayer & Justus C. Roux stress (like duration), are encoded within the rime, but not within the onset. Further similar asymmetries have been discovered experimentally, others are being simulated in speech recognition research. There exists, however, at least one category of sounds for which it is safe to say that they are unencoded, or at least much less encoded than any other supralaryngeally articulated elements of the speech code. These sounds are clicks (Ladefoged & Maddieson, 1996: 246–281). In this paper we will deal with some “unencoded” properties of Xhosa clicks with special reference to their coarticulatory properties. We will argue that the fact that clicks are phoneti- cally and phonologically dissociated from the rest of the speech code presents a challenge to theories of syllabic encoding. Furthermore, we will argue that the extreme richness of click accompaniments derives from the resistance of the anterior click articulations to interaction with the rest of the speech code. Accompaniments will be argued to function as links be- tween the clicks and the rest of the speech code.
2. Three extraordinary properties of clicks
A milestone in the description of clicks and their accompaniments was reached with the seminal work of Ladefoged & Traill (1994) (henceforth L&T). While this work primarily focuses on the nature and extent of click articula- tions in general and in the Khoisan languages in particular, it also devotes some attention to various click articulations and their accompaniments found in a Bantu language, Xhosa (L&T: 46–48). Despite the comprehensiveness of this work, a number of issues still remain unresolved, specifically in Xhosa and its sister languages Zulu and Swati. L&T give the impression that clicks are quite ordinary types of sounds, which are not only fairly easy to produce, but also perceptually salient to a degree which should make them “highly favoured consonants in the world’s languages. […] Indeed, we cannot explain why these easy to make and perceptually optimal consonants are found in so few languages“ (L&T: 62). We will argue that the possible reason for this restricted distribution of clicks, as well as many of their other extraordinary features which have been reported in literature, is due to their being phonetically “unencoded“. Before arguing this in detail, however, let us review some of clicks’ extraordinary features.
Syllables and unencoded speech: Clicks and their accompaniments in Xhosa 51
initiatory cl. articulatory cl. influx efflux closure phase suction release phase
Figure 1. Articulatory and acoustic phases in the production of the alveopalatal click [ ! ] in Xhosa
2.1. Velic suction
Much research has been devoted to the articulatory description of clicks in various languages and a relatively clear picture has emerged. The full- fledged production of any click consists of the following three distinct phases:
• The CLOSURE PHASE, which consists of two closures: – the initiatory closure, which is velar for all clicks – the articulatory closure (coronal or labial). • The SUCTION PHASE, during which the air in the cavity between the two closures is rarefied by the downward movement of the tongue body. • The RELEASE PHASE, which consists of two releases: – the articulatory release, which has been called influx in traditional de- scriptions and is believed to determine the click type (L&T 34–35), – the initiatory release, always velar or uvular, called efflux by traditional phonetics and referred to as the click accompaniment in L&T. All these phases are timed in such a way, that an acoustically crisp (cf. Johnson, 1993) and perceptually salient sound is produced (cf. Traill, 1994). The salience is concentrated within the period of the noiseburst (the release phase) itself (cf. Traill, 1994), because clicks appear not to coarticulate with their phonetic context as we will show (cf. Sands, 1992). The schematic version of the articulatory phases and the resulting noisebursts of a Xhosa voiced alveo-palatal click [ ! ] are given in Figure 1.
52 Grzegorz Dogil, Jörg Mayer & Justus C. Roux
2.2. Phonological opacity
Click consonants prove to be highly resistant to phonological processes induced by their environment. Out of approximately 36 click-using lan- guages on record (8 Bantu, 12 Bushman, and 16 Hottentot) there is not a single one in which clicks mutate synchronically to other sounds. This is true both of the indigenous click languages of the Khoisan group and of the Bantu languages into which clicks entered through adoption, not mutation. Synchronically, clicks do not undergo phonological changes, neither in a particular phonetic, phonological or lexical context, nor context independ- ently as a whole class (cf. Dogil & Luschützky, 1990: 33–36). The resistance to context influence and the inability to influence the im- mediate context is also a feature of clicks in diachronic phonology. Traill & Vossen (1996) provide a complete list of diachronic changes of clicks in Khoisan languages and there does not appear to be a single one among them which is phonetically or phonologically conditioned.
The changes we have been examining are context free in the sense that they affect a whole class of sounds no matter what the phonological or phonetic context. (Traill & Vossen, 1996: 30)
This context-free behaviour is to be expected from unencoded sounds. In terms of feature geometry, sounds that do not interact with the context should be maximally and discretely specified. Indeed, all attempts to repre- sent clicks in terms of feature geometry have assigned very complex struc- tures to them (e.g. Keyser & Stevens, 1994: 222, 224; Halle, 1995: 8–12; Sands, 1993). Note, however, that representing clicks in segmental phonology is in itself methodologically dubious. The standard view is that decisions about feature geometry should be grounded in generalizations regarding phono- logical processes. Since no process shows that clicks interact with any other part of the phonology, there is no reason to reflect them formally in the phonology. If they are phonological (segmental) at all, they appear to be highly opaque.
2.3. Phonetic unencoding
Recent acoustic phonetic research on clicks has shown that they have very distinctive noiseburst properties (cf. Traill, 1994; L&T; Traill, forthcoming a,b; Roux et al., 1995). Clicks also have quite unique temporal and intensity fea- tures. This allows the classification of [abrupt], [noisy] and [grave] acoustic
Syllables and unencoded speech: Clicks and their accompaniments in Xhosa 53 features. Hence, most of the features relevant for the identification of a click are concentrated within its own noiseburst. Traill (1994) has also convinc- ingly shown that clicks are perceived as soon as their noiseburst is present. Also, unlike other obstruent consonants, clicks are virtually never identified in syllables from which they had been edited. This points to the conclusion that they are unencoded. This is supported by the measurements of the deltas of F2 and F3 in the vicinity of the click (presented orally in Vienna). Below, we provide experimental evidence for the other aspect of speech unencoding — the blocking of V-to-V coarticulation.
3. Method
3.1. Subjects
A homogeneous group of eight female native Xhosa speakers was chosen for the experiment. All speakers came from the same area (Cape Province), all were of approximately the same age (20–25 years old) and attended the same school (the Teachers Training College in Cape Town), studying to become professional Xhosa teachers. They have no speech, language or hearing impairments. The recordings were made by Phillip Lewis in a sound treated room of the Groote Schuur Hospital (Cape Town) using professional equipment (DAT).
3.2. Stimuli
Stimuli used in the experiment consisted of a set of nonsense words of the form [iCili] ~ [iCuli] with a variation in the C position. In the actual test the lateral [l], the four aspirated stops {bilabial, dental, palatal, velar}, the four ejectives, the bilabial implosive, the four murmured voiced stops {bilabial, dental, palatal, velar}, and the fifteen phonemic Xhosa clicks, given in (1), were placed in this critical position between two vowels [i _ i] and [i _ u].
(1) Xhosa phonemic clicks. ‹c› [|] is a noisy dental click, ‹q› [!] an abrupt alveo-
palatal click, and ‹x› [ | ] an alveolar/lateral click. voiceless ‹c› ‹q› ‹x› aspirated ‹ch› ‹qh› ‹xh› voiced-breathy ‹gc› ‹gq› ‹gx› nasal ‹nc› ‹nq› ‹nx› nasal-breathy ‹ngc› ‹ngq› ‹ngx›
54 Grzegorz Dogil, Jörg Mayer & Justus C. Roux
The vowels were chosen because they represent the extremes in terms of tongue position and lip rounding of the five elements of the Xhosa vowel system. The expectation of the theories of coarticulation is that the vowel [u] will influence the vowel [i] in the [iCu…] context, both on the tongue position plane (the constriction should be further back — i.e. F2 lowered) and on the lip rounding plane (the lips should be less spread and more protruded — i.e. F3 lowered).
3.3. Data collection
The tokens were digitized at a 16 kHz sampling rate and analysed using LPC analysis routines of the Entropic/xwaves program (xspectrum). The analysis used the LPC BURG method, a 20 ms Hamming window, with 20 LPC co- efficients. For each talker and for each token, F2 and F3 values were calculated. For this calculation the formants were measured at four different points: in the middle of the vowels preceding and following the critical consonant, and immediately before and after consonant closure. The mid points of the vowels were defined using two criteria: the highest F1 value read from the LPC spectrum (corresponding to the greatest articu- latory opening — the peak in sonority); and the arithmetical mid value of the duration of the vowel’s steady state. In most cases, both criteria picked the same point. The vocalic edges were defined using three criteria: they were the last/ first analysis frames in which three formants were visible in the spectrogram; they coincided with the changing amplitude envelope in the vocalic portion of the waveform; and they had a residual energy of 10–25% of that found in the steady state of the vowel. In all cases at least two of these three criteria determined the edge points.
4. Data analysis and discussion
In order to test the transparency/opacity of the consonant to its adjacent vowels we calculated the differences between the F2 and F3 in the first [i]’s of [i1 Cili] and [i2 Culi]. In the case of transparency, the difference F2/F3 [i1–i2] is supposed to be large: The back position of the tongue in [u] exerts influ- ence on the front tongue position of [i2] across the consonant, thus lowering F2, and the lip rounding of [u] is transported to [i2], resulting in a significant lowering of F3. In cases where there is a difference in one formant only, one has to claim that the intervening consonant is transparent only on one plane
Syllables and unencoded speech: Clicks and their accompaniments in Xhosa 55
200 148 150 134
10 0 56 47 50 37 37
0 –4 –50 –25 –24 –38 –42 –100 –113 –150 laterals voiced implosive voiceless clicks ejectives stops stops
Figure 2. V-to-V coarticulation across consonant classes in Xhosa: F2/F3 [i1–i2] for [i1 Cili] and [i2 Culi]
(i.e. backness or rounding/protrusion). If there is no difference in the for- mant values in F2/F3 [i1–i2] we have to conclude that the consonant in ques- tion blocks V-to-V coarticulation. Another interesting case of interaction, and actually the one employed most often by our speakers, is the one where the value of F2/F3 [i1–i2] is negative. This is the case when speakers dissimilate the vowels by choosing the particularly front and spread variety of [i] before [Cu]. Our hypothesis is that unencoded speech sounds not only block the coarticulation (coproduction) on the vocalic plane, but also break up the signal into discontinuous and discrete parts. The results of the calculations for various classes of Xhosa consonants are given in the chart in Figure 2. As the chart shows, lateral [l] is transparent to both backness and round- ing (see the high positive difference for both F2 (dark bar) and F3 (light bar) between the [iCi] and [iCu] conditions). Similarly, there is a strong coarticu- lation in backness (cf. F2 differences = dark bars) of voiced and voiceless stops and the implosive. The voiced implosive in Xhosa is bilabial, and hence dissimilatory influence upon the rounding/protrusion feature is to be expected. The lips are particularly active (and tense) in the production of the implosive bilabial, thus it is to be expected that the anticipatory [u] effects
56 Grzegorz Dogil, Jörg Mayer & Justus C. Roux would be eliminated by this action. There is no simple explanation for the opacity of voiceless stops on the rounding/protrusion plane. Note the small negative value (–4.0 Hz) for F3 [i1–i2] across these consonants. The only explanation that we can provide has to do with their durational properties. Xhosa pulmonic voiceless stops are strongly aspirated, and their relatively long duration may override the anticipatory tendency of lip rounding, which is strong in voiced (much shorter) pulmonic stops. Both clicks and ejectives are unencoded according to our criteria. A strong dissimilatory tendency in V-to-V relation is illustrated for both classes of sounds in Figure 2. This tendency is even stronger in ejectives than in clicks. Here, the important difference lies in the phonological en- coding of clicks vs. ejectives. We argued above that clicks are phonologically opaque in the sense that they never interact as a class with other sounds. Ejectives do show such typical natural class behaviour in that they radically change their class property (i.e. glottalic) in specific phonological contexts, as illustrated in the next section.
5. Phonological encoding
The only process in which Xhosa clicks and other obstruents are involved is so-called “nasal strengthening“. By this process obstruents, when following a nasal, accommodate the glottal features of the nasal (i.e. [voice] and [– spread glottis]). Hence, implosives become voiced stops and aspirated stops become ejectives, as shown in (2) and (3) below.
(2) implosives > voiced stops /n + B/ > [mb] -Balisa ‘to narrate’ > imbala ‘history’ (3) aspirated stops > ejectives h /n + p / > [mp’] -phila ‘live’ > imp’ilo ‘health, life’ uphondo ‘horn’ > iimp’ondo ‘horns’ h /n + t / > [nt’] -thanda ‘love (VERB)’ > int’ando ‘love (NOUN)’ -thathu ‘three’ > ezint’athu ‘which are three’ h /n + k / > [9k’] -khulu ‘big’ > iink’ulu ‘the big ones‘
By this same process, clicks, when following a nasal, accommodate the glottal features of the nasal within their accompaniment. The voiceless clicks acquire voicing and the aspirated clicks lose their aspiration. However,
Syllables and unencoded speech: Clicks and their accompaniments in Xhosa 57 they both remain within their general category of clicks. Hence, in our view, this process is restricted to the modification of the efflux (the initiatory gesture, or accompaniment). Consider the examples in (4) and (5).
(4) /n + voiceless click/ > [nasal voiced click] n + c [|] > ngc -cengceleza ‘to recite’ > ingcengceleza ‘recitation’ ucango ‘door’ > iingcango ‘doors’ n + q [ ! ] > ngq -qwenga ‘destroy’ > ingqwenga ‘destroyer’ uqaqa ‘squirrel’ > iingqaqa ‘squirrels’ n + x [ | ] > ngx xaka ‘to cause trouble’ > ingxaki ‘the trouble’ uxande ‘a round house’ > iingxande ‘round houses’ (5) /n + aspirated click/ > [nasal + voiceless click] h 1 n + ch [| ] > nkc -chwela ‘to repair’ > inkcwela ‘an expert in repairing’ uchuku ‘grumbler (SG.)’ > iinkcuku ‘grumbler (PL.)’ h n + qh [ ! ] > nkq -qhuba ‘to drive’ > inkqubo ‘progress’ h n + xh [ | ] > nkx -xhensta ‘to dance’ > inkxentsi ‘expert dancer’ uxhaxhazo ‘to operate’ > iinkxaxhaza ‘surgeon’
The clicks whose accompaniments share the glottal properties with the nasal (i.e. voiced and nasal clicks) remain unchanged, as illustrated in (6) and (7) respectively.
(6) /nasal + voiced click/ > [nasal + voiced click] -gxadazela ‘to tumble’ > umgxadazeli ‘a tumbler’ -gquma ‘to cover’ > imgquma ‘a cover’ (7) /nasal + nasal click/ > [nasal + nasal click] -nkcuku ‘jealous’ > i(n)nkcuku ‘jealousy’ -nkxaso ‘support’ > i(n)nkxaso ‘a support’ -ngxaki ‘problem’ > ingxaki ‘a problem’ -ngqumbo ‘wrath’ > ingqumbo ‘a wrath’
The data in (2)–(3) showed that obstruents (stops, ejectives, implosives) when involved in “nasal strengthening“ change their properties in a fairly
1 Orthographic ‹k› is traditionally used to indicate a break in a segmental sequence, hence, ‹nkc› is phonetically realized as two segments [9] and [|]
58 Grzegorz Dogil, Jörg Mayer & Justus C. Roux radical way. The glottalic implosive changes to a pulmonic voiced stop, and the pulmonic aspirated stops change to glottalic ejectives. The data in (3)– (6) show that clicks are only involved in this process due to their efflux (accompaniment) properties. If these properties agree with the specific properties of a nasal (as in (5)–(6)) nothing needs to be accommodated. If the properties of the efflux disagree with the nasal, the efflux is changed accordingly, and the nasal inherits the place category of the efflux (velar). Our formal account is that the underlying glottal and major class properties of a nasal [– spread glottis, voice, nasal, tense] are superimposed on the (partially specified, deactivated) properties of the stops. Whatever the correct account of this process may be, the process itself shows several interesting things. First of all it shows that clicks are involved in processes in which stops are also involved. Hence, the often expressed impression about phonetic closeness between clicks and other types of stops is supported by this limited set of phonological data. Secondly, it shows that clicks are more resistant to phonological mutation than glottalic and pulmonic stops. Actually, it is only the click accompaniments and the velar effluxes which interact with other parts of the string of speech. The distinctive articulatory releases (the influx noisebursts) are not influenced by the context at all. Hence, in languages in which clicks are not restricted to a specific position there appears to be no phonologically (or phonetically) induced variation of any sort, neither within the clicks nor within their context. The click-forming influx has no phonology to speak of. This is precisely what is expected of sounds which are unencoded in respect to the rest of speech signal. The part of the click which seems to licence sound within the rest of the phonological and phonetic code is the efflux (the accompaniment). We have shown evidence for this from the “nasal strength- ening” data, which is the only synchronic phonological process in which clicks appear to be involved. Further evidence comes from the phonetic coarticulation-blocking prop- erties of clicks. These are illustrated in Figure 3. As can be seen, the type of the accompaniment (efflux) appears to be the decisive parameter. Voiceless and aspirated clicks are the most unencoded (i.e. dissimilating) ones, whereas the voiced and the nasal clicks are simply not transparent. The dissimilatory effect of clicks is stronger on the rounding/protrusion plane than on the backness/frontness plane. The place of articulation and articu- lator type appear to play no role. Necessary and sufficient for the coarticu- lation blocking effect is the presence of the noiseburst. Here again, the accompaniment (efflux) apparently influences the degree of “unencoding” of a click. However, the explanation of this influence has to be left for a sepa- rate study.
Syllables and unencoded speech: Clicks and their accompaniments in Xhosa 59
50
9
0 –4 –2 –11 –24 –24
–50 –42 –38 –51 –51
–73 –85 –100 voiceless aspirated voiced nasal voiced all clicks nasal
Figure 3. V-to-V coarticulation across various click types in Xhosa
References
Dogil, G. & H.C. Luschützky (1990). Notes on sonority and segmental strength. Rivista di Linguistica 2, 3–54. Dogil, G. & J. Roux (in prep.). Notes on unencoded speech: Clicks and their accompaniments in Xhosa. Halle, M. (1995). Feature geometry and feature spreading. Linguistic Inquiry 26, 1–46. Johnson, K. (1993). Acoustic and auditory analysis of Xhosa clicks and pulmonics. UCLA Working Papers in Phonetics 83, 33–45. Keyser, J. & K. Stevens (1994). Feature geometry and the vocal tract. Phonology 11 , 207–236. Ladefoged, P. & Traill, A. (1994). Clicks and their accompaniments. Journal of Phonetics 22, 33–64. Liberman, A. (1996). Speech: A special code. Cambridge, Mass.: MIT Press. Roux, J., G. Dogil & W. Wokurek (1995). Click articulations in Xhosa: New perspectives through Wigner Distribution Analysis. Proceedings of the XIIIth ICPhS Stockholm 2, 574–577. Sands, B. (1992). An acoustic study of Xhosa clicks. Proceedings of the XIIth ICPhS Aix-en-Provence 4, 130–133. Sands, B. (1993). Aperture theory and the representation of clicks in southern African languages. Abstract of a talk given to the Phonology Reading Group, University of Arizona. Traill, A. (1994). The perception of clicks in !Xóo˜. Journal of African Languages and Linguistics 15, 161 –174. Traill, A. (forthcoming a). Design features and the perception of clicks. Proceedings of the Tüzing Khoisan Conference, July 1994. (Quellen zur Khoisan Forschung. Köln: Rüdiger Köppe).
60 Grzegorz Dogil, Jörg Mayer & Justus C. Roux
Traill, A. (forthcoming b). Linguistic phonetic features for clicks: Articulatory, acoustic and perceptual evidence. Papers of the First World Congress of African Linguistics (Kwaluseni, Swaziland), edited by R.K. Herbert. Traill, A. & Vossen, R. (1996). Sound change in the Khoisan languages: New data on click loss and click replacement. Ms., University of the Witwatersrand, Johannesburg.
Syllables? No! Substantive evidence for the syllable-less beats-&-binding model of phonology
Katarzyna Dziubalska-Kołaczyk
Abstract
Among the areas of linguistic evidence researched so far, almost all have been claimed to support the analysis of phonological structure in terms of the units called syllables (beside other phonological units). A variety of approaches have emerged which define the syllable in a theory-specific manner, emphasizing accordingly its unity or its compositionality or its The Hague: Thesus. 61–87. Thesus.The Hague: 61–87. . quantity. This paper investigates whether the putative evidence for the syllable actually does provide support for this unit and it proposes a phonological
Syllables !? organization of sequences in terms of beats and bindings which is more compatible with the above evidence than a phonological organization in terms of syllables. The paper consists of two parts: in the first part an outline of the beats- and-binding model will be presented in a necessarily abridged manner. In Phonologica 1996: Phonologica 1996: the second part, the model will be exercised against traditionally syllable- supportive data. 0.
1. The beats-and-binding model
The beats-and-binding model of phonology (B&B in short) which I propose is founded on the principles of Natural Linguistics (Natural Phonology and Natural Morphology in particular). This means that (a) its predictions and explanations are functionalist and semiotic in nature; (b) all generalizations formulated within the model have the status of universal or language- specific preferences and not absolute rules or laws (cf. Dressler, 1988; 1996), and (c) it draws heavily from so-called external linguistic evidence, which for
John R. Rennison & Klaus Kühnhammer, eds. (1999). eds. R. Rennison & Klaus Kühnhammer, John (1999). a naturalist model is regarded as substantive.
62 Katarzyna Dziubalska-Kol-aczyk
The model distinguishes four levels of phonological structure, given in (1).
(1) Level 0: Rhythmical preferences Level 1: Underlying phonological binding preferences Level 2: Phonotactic preferences Level 3: Articulatory preferences
The levels represent a sequential priority relation among preferences con- cerning rhythm, bindings, phonotactics and articulation respectively. The levels are to be understood not as tiers or cycles, but rather as hierarchically related constraint types which are responsible for the shape of phonological representations. The constraint types are independently motivated by universal principles of human perception and production. The organization of preferences according to levels (to be presented below) is not supposed to reflect any linear processing of structure “level by level”, but rather to serve as a reconstruction of the workings of the hierarchically related preferences/constraints types which are responsible for the complexity of structures. Part of the information encapsulated in the form of constraints needs to be stored in the lexicon. This concerns the outputs of language-specific decisions taken at Levels 2 and 3, i.e. language-specific phonotactics. Phonological preferences interact here with morpholexical ones (see §1.5 below). Level 0 information is universally unmarked and constitutes a prelexical default, while it is manifestable to a limited extent in particular languages. Level 1 information is partly universal, partly type-specific, but it is also purely phonological. Postlexically, a final articulatory implementation of the outputs takes place (Level 3). The implementation is style-sensitive: the less formal the style, the more it is subject to the universal constraints. As stated above, all statements formulated within the model express preferences, and should not be misunderstood to be laws. Consequently, existential statements of the type: “there is a preference” will be used below without any claim to absoluteness. Preferences, on the one hand, follow from universal cognitive, semiotic and functionalist principles, while, on the other, they find support in linguistic evidence. They are fundamentally different from constraints or parameters in two basic respects: they follow deductively from independent universal principles and they are hierarchic, or scalar, in nature.
Syllables? No! Substantive evidence for the beats-&-binding model 63
1.1. Beats and the universal rhythmic preferences: Level 0
1.1.1. The primary rhythm units
The primary rhythm units are feet and their constituents — rhythmical BEATS, as in music. There is a universal preference for two beats per foot: the first beat is preferably strong, the second weak, i.e. they constitute a trochee (a metrically falling accent). The preference for a trochaic foot pattern is well-evidenced (see e.g. Dogil’s (1981: 92) Trochaic Projection Constraint). The preference for a binary foot structure can be subsumed under the general preference for binary paradigmatic and syntagmatic contrasts (Dressler, 1990: 81, 85; 1996), which is neurologically based (cf. the binary choice between presence and absence of a neural firing) and also reflects a semiotic principle of figure-and-ground (cf. Dressler, 1985; Ungerer & Schmid, 1996: 156ff), best realizable through a binary contrast. Again, it is well-evidenced (see e.g. Dogil, 1981: 94; 90f). Also the analogy between rhythm in music, verse and language can undeniably be drawn.
1.1.2. Beats
A beat is realized by a phoneme which is traditionally referred to as a “syllable nucleus”; preferentially, it is a vowel; secondarily, a consonant may acquire the function of beat. A vowel is a better candidate for a beat due to its saliency potential based on its high sonority value and articulatory openness. Therefore, those consonants which possess the latter two fea- tures to a higher extent qualify better for a beat than do others.
1.1.3. Non-beats
In accordance with the semiotic principle of figure-and-ground (cf. Dressler, 1985), a hiatus between two beats is avoided by inserting a NON-BEAT (i.e. a consonant) in between.1 Only in this way do the figures, i.e. beats, receive a necessary ground, i.e. non-beats, in the form of consonants.
1 This is not to suggest that VCV is a resulting sequence. On this level of organization (Level 0) phonological structure consists in the flow of alternating rhythmical units, i.e. beats and non-beats as well as strong and weak beats. No other phonological units or relations are seen on this level. The unmarked status of CV-initial structures results from the preferences of further levels.
64 Katarzyna Dziubalska-Kol-aczyk
So, thanks to the preferences in §1.1.1–3, speech flow consists of beats and non-beats which are phonetically realized by perceptually and articula- torily contrasting sounds — vowels and consonants respectively. This is the most general structural level of phonology, the level of RHYTHMICAL PREFERENCES (henceforth LEVEL 0). The layer of structure resulting from the preferences of this level constitutes the most general underlying representa- tion in all languages (the evidence for which comes inter alia from first lan- guage acquisition, aphasia and phonostylistics).
1.2. Bindings and universal perceptual preferences: Level 1 and Level 2
The universal perceptual preferences operate at two levels, as shown in (2).
(2) Level 1: Underlying phonological BINDING PREFERENCES between beats and non-beats. Level 2: PHONOTACTIC PREFERENCES predicting the preferred actual sonority distances between and among vowels and conso- nants, necessary for the origin and maintenance of consonant clusters.
1.2.1. Level 1
LEVEL 1 is the level of binding preferences, which are based on perceptual principles relevant in speech communication. BINDINGS are responsible for another layer of structure, superimposed on the structure of Level 0. On Level 1 diversification among languages can already be identified: language- type-specific choices are made as to how the binding preferences are realized.
1.2.1.1. Bindings
Beats and non-beats in a sequence (cf. 1.1.3.) are joined by means of bind- ings in a binary fashion. In a BnB sequence (where B stands for a beat and n for a non-beat) there are maximally two bindings, i.e. a {Bn} binding (the non-beat is bound to the preceding beat) and an {nB} binding (the non-beat is bound to the following beat), i.e. B{Bn}{nB}B (braces {} will be used to signify bindings). A beat, however, may potentially stay alone while a non- beat (being a non-rhythmical unit) must be bound to a beat. Thus, in a BnB
Syllables? No! Substantive evidence for the beats-&-binding model 65 sequence, n may be bound to one beat only, i.e. either B{nB}B or B{Bn}B, as shown in (3).
(3) B n B B n B B n B B {Bn} {nB} B B {nB} B B {Bn} B
Sound sequences are combinations of two basic binary bindings: {nB} and {Bn} and, possibly, single beats which may result from cutting a sequence just before or just after a beat. The binarity of the two basic bindings them- selves can be explained by the principle of contrast: in this case, an optimal perceptual contrast (or optimal perceptual distance) between a beat and non-beat), measurable in terms of sonority and best realizable by means of binarity. Bindings have a basis not only in the perception of speech, but also in general perception. For an act of perceiving to take place we need to see, hear, sense or taste a figure against a ground, i.e. we need a binary contrast between the elements to which we are exposed. In other words, we need to perform an act of figure/ground segregation (cf. Ungerer & Schmid, 1996: 156ff). What we then perceive is both a figure and a ground, not as a unit, but as two bound elements, the binding between which is a necessary prerequisite for successful perception. The greater the contrast, the better we perceive, and thus the stronger the attraction between the two elements. This kind of cognitive psychological reasoning leads, of course, only to the most general and basic principle of perception. It is exactly for this reason, however, that the principle is compatible with the idea of phonological bindings, which also operate at the general, underlying phonological level and are, in principle, hearer-friendly.2
1.2.1.2. Functions of beats
After non-beat deletion (BnB → BB), a BB-geminate has a status of a single long beat on Level 0 (single, since on Level 0 only alternating sequences of contrasting beats and non-beats are allowed: thus a BB sequence can only be understood as one beat), whereas on Level 1 the two identical units (B and B) are distinct (and therefore represented). At this point it becomes necessary to distinguish between the rhythmi- cal function of beats on Level 0 and the quantitative function of beats
2 One could argue further for the reality of binary bindings between contrasting elements by reference to a general principle of “attraction of opposites” taken from the natural sciences. For example, opposite potentials (physics) or opposite sexes (biology).
66 Katarzyna Dziubalska-Kol-aczyk
(which potentially enter the bindings) on Level 1. Henceforth, beats on Level 0 vs. Level 1 will be represented notationally for these functions as Br and Bq respectively. Thus, non-beat deletion gives the structure in (4).
(4) Level 0 Br Level 1 Bq Bq
Two neighbouring beats without any trace of a non-beat between them (i.e. no gliding or pre-glottalization) and without any morphological boundary separating them, count thus as one (long) beat on Level 0, i.e. on the level of universal rhythmical preferences. The counting of beats on this level cor- responds to what is usually interpreted as speakers’ intuitions about the number of “syllables”. For example, a trochee on Level 0 counts: “one, two”, while on Level 1 it consists of either two or three beats. Thus, the number of beats on Level 1 does not always correspond to the number of beats on Level 0, as shown in (5).
(5) V: … V or VV … V vs. V … V Level 0 Br … Br = Br … Br Level 1 Bq Bq … Bq =/ Bq … Bq
A diphthong or a long vowel may draw stress on to themselves, precisely because of their double-beat (Bq Bq) phonological representation. Addition- ally, a beat participating in a {Bn} binding may also draw stress on itself, whereas a beat without this binding is unable to do so (though it can receive a lexical or other extra-phonologically conditioned stress). This is so be- cause stress naturally needs richer phonological material to be realized on, through phonetic means of length, pitch and loudness. The above men- tioned cases of phonologically conditioned stress used to be described as cases of stress on “heavy syllables”. “Heaviness” in a beats-and-binding model is expressed by means of a number of beats and bindings (where beats and bindings count as equal) contained within a binary foot, from one beat to another or to a phonological word boundary. For example, in a Bqn cluster (VC) there is one beat and one binding, i.e. count 2, and in a BqBq cluster (V:, VV) there are two beats but no bindings, i.e. also count 2.
(6) Bqn = B +{Bn} = 2 Bq Bq = B + B = 2
Syllables? No! Substantive evidence for the beats-&-binding model 67
1.2.1.3. The strength of bindings
The two bindings differ in strength: the {nB} binding, i.e. the binding of a non-beat to the following beat (preferentially realized by a CV sequence), is always stronger than the {Bn} binding, i.e. the binding of a non-beat to the preceding beat (preferentially realized by a VC sequence).
(7) {nB} > {Bn}
The preference for a {nB} binding to be stronger than a {Bn} binding is hearer-friendly (i.e. perceptually based). An acoustic phonetic basis for the preference is that acoustic modulations in a consonant-vowel transition can be perceived much better than in a vowel-consonant transition. “It is gener- ally the case that the most salient acoustic modulations in a syllable occur near the CV interface” (Ohala & Kawasaki, 1984: 117) and “[a]uditory cues present in CV’s are more robust than those in VC’s” (Ohala & Kawasaki, 1984: 118). Articulatory factors also contribute to the better perception of CV’s: “Since there is a richer, more reliable set of place cues in the CV transition than the VC transition, listeners weight the former more heavily than the latter in deciding what they’ve heard” (Ohala, 1990b: 265). And, on the speaker’s side, he “actively tries to create temporally more well defined, more precise, articulations near the CV as opposed to VC interface” (Ohala & Kawasaki, 1984: 119). The preference for the {nB} binding should not be understood as a pref- erence for CV “syllables”, because the preference for the {nB} binding is also valid in languages which, in traditional terms, do not prefer CV “syllables” (cf. Noske, 1989). There is, of course, a lot of phonetic and phonological evidence for the prototypicality of a CV “syllable”, which in fact refers to the prototypicality of the CV sequence, and thus, {nB} binding. The evidence is summarized in divergent sources, e.g. Bell & Hooper (1978: 8ff), Dogil & Braun (1988: 13f), Dogil & Luschützky (1990: 22f), Edwards & Shriberg (1983), Krupa (1973: 55f), Krupa (1968: 26ff), Pulgram (1970).
1.2.1.4. Sonority
A subjective perceptual measure of contrast between a beat and a nonbeat is constituted by sonority. At the level of phonological bindings (Level 1) beats are uniformly more sonorous than nonbeats. Some example repre- sentations of words on Levels 0 and 1 are given in Figure 1.
68 Katarzyna Dziubalska-Kol-aczyk
[k æ t] cat L.0 n Br n one rhythm beat
L.1 n Bq n one quantity beat {nB} {Bn} two bindings [k O: t] caught L.0 n Br n one rhythm beat
L.1 n Bq Bq n two quantity beats {nB} {Bn} two bindings [m A ð 7] mother L.0 n 'Br n Br two rhythm beats
L.1 n 'Bq n Bq two quantity beats {nB} {Bn}{nB} three bindings [p S e r v a] przerwa Pol. ‘break’ L.0 n n 'Br n n Br two rhythm beats
L.1 n n 'Bq n n Bq two quantity beats {nB}{Bn} {nB} three bindings NB: Bq = quantity beat, Br = rhythm beat, n = non-beat, L.= Level, and {nB}, {Bn} are bindings Figure 1. Examples of words as formally represented on Level 0 and Level 1
1.2.2. Level 2
Preferences concerning actual distances between segments in terms of sonority are responsible for the phonotactic make-up of language-specific structures. Sonority is understood as a default intrinsic property of a phonological segment. At this level sonority becomes a measure of dis- tances between segments. This distance decides the fate of segments in a phonotactic sequence. In objective terms, it is the degree of modulation for several acoustic parameters (amplitude, periodicity, spectral shape, F0; cf. Ohala, 1990a) that determines those distances. As Ohala (1990b) notes, larger modulations have more survival value than smaller ones and therefore will persist in languages. The phonotactic preferences crucially refer to the criterion of the inher- ent sonority of sounds, hierarchized as in (8).
Syllables? No! Substantive evidence for the beats-&-binding model 69
(8) vowels semivowels liquids nasals fricatives plosives affricates
The universal preferences consist in the strength-by-distance relations between segments, measured as distance among the six positions on the above scale (e.g. la — distance of two positions, st — distance of one, ka — distance of five, so ka > la > st). The two exemplary preferences specified in (9) and (10) below hold for particular sequences of segments.
(9) CVC, CVCV, CVCVC, VCV : CV > VC Read: For the sequences CVC, CVCV, CVCVC, VCV, the sonority dis- tance of CV is preferably greater than that of VC.
This means that in the sequencing of alternating consonants and vowels the basic, most general preference of Level 1 (namely for an {nB} binding to be stronger than a {Bn} binding) is at work on Level 2 without modification, though with a possible reinforcement. For instance, in an nBn sequence, {nB} is always stronger than {Bn} on Level 1, and this relation should pref- erably be reinforced on Level 2, so that a sonority distance in CV is also greater than in VC, as e.g. in English can. The preferences of Levels 1 and 2 overlap in the case of the discussed sequences, since there is no reason for them not to (contrary to the cases below). In fact, true reinforcement is possible only in an nBn sequence; in a BnB sequence it is logically impos- sible on Level 2, and in CVCV and CVCVC it is possible only in part of a sequence, to the disadvantage of the rest.
(10) C2C1V : C1V ≤ C1C2
Read: In a C2C1V sequence the sonority distance in C1V is preferably smaller than or equal to the distance in C1C2.
When clusters of consonants appear, the basic strength relationship be- tween bindings on Level 1 needs to be counterbalanced on Level 2 in order to maintain the clusters, i.e. to prevent cluster simplification by reduc- tion (to CV) or epenthesis (to CVCV). According to preference (10), clusters like those in (11) are preferred over clusters in which the sonority distance in C1V is greater than in C1C2, i.e. those in (12) and (13).
(11) Type A: pja, pla, tra, pra, fja, mja, etc. (12) Type B: mla, sna, fna, kna, tsa, pfa, psa, etc. (13) Type C: sta, rta, nsa, mna, xfa, ssa, etc.
70 Katarzyna Dziubalska-Kol-aczyk
The above measure thus involves the relation between sonority distances of pairs of sounds, and will henceforth be referred to as the OPTIMAL SONORITY DISTANCE PRINCIPLE (OSDP). The OSDP allows for the grouping of clusters as preferred (Type A, given in (11)) vs. dispreferred (Types B and C, given in (12) and (13)). It is immediately noticeable that the Sonority Sequencing Principle (see e.g. Selkirk, 1984),3 a traditional syllable-structure constraint, is not a suf- ficient well-formedness criterion since it does not distinguish between the clusters in (11) and (12). Is the Sonority Sequencing Principle a well- formedness criterion at all? Firstly, the fact that the preferred clusters in (11) are sonority-sequenced and cannot be sonority-reversed, as in (13), follows directly (and unidirectionally) from the OSDP (in this case, from the prefer- ence in (10)). Secondly, since the former prediction is unidirectional, it does not follow from it that all sonority-sequenced clusters will be of type A; so the group of dispreferred clusters must include both sonority-sequenced and sonority-reversed ones. Thus, the fact that a cluster is or is not sonority- sequenced remains a descriptive structural observation and does not add to or detract from any predictions already made by the OSDP. However, the Sonority Sequencing Principle’s prediction that sonority- sequenced clusters are well-formed fails on two accounts: firstly, it is too general (not distinguishing between A vs. B) and, secondly, it predicts the clusters in (12) to be better than those in (13). This seems to be a wrong because the violators of sonority-sequencing (type C) may, under language- specific conditions, be more stable or common than the non-violating type B. Evidence for this comes from (i) diachrony: clusters of type B tend to be unstable (e.g. kn-, sn-); (ii) phonetics: clusters of type B like ts-, pf- tend to become affricates, i.e. undergo fusion; (iii) language-specific phonotactics: a language may have clusters like rt-, st- (i.e. of type C), rather than fn-, ml- or sn-, (of the type B); (iv) variation: in a popular pronunciations of Italian psicologia ‘psychology’, the initial cluster becomes sic- or pi(s)sic-. So how does the OSDP account for the relative stability of type C clus- ters? Neither type B nor type C clusters obey preference (10). However, type C clusters contain a preferred realization of an {nB} binding (e.g. ta in rta or sta) and thus obey preference (9). Consequently, they become subject to the most general preference of Level 1 for an {nB} binding, namely (7). As a result, one observes in these clusters processes like devoicing, epenthesis or assimilation which, on the one hand, are perceptually motivated (to come
3 The Sonority Sequencing Principle or Generalization, formulated, for instance, by Selkirk (1984: 116) reads: “In any syllable, there is a segment constituting a sonority peak that is pre- ceded and/or followed by a sequence of segments with progressively decreasing sonority values.”
Syllables? No! Substantive evidence for the beats-&-binding model 71 as close to the preferred {nB} binding as possible) and, on the other hand, lead to the improvement of a final phonetic effect, both perceptually and articulatorily. This is a case of an interlevel conflict solution implemented according to the general tendency for balance, by which a cluster that is bad by Level 2 criteria is “repaired” to function as a good structure by Level 1 criteria. So, while clusters like sn- (type B) satisfy neither of the above mentioned preferences, clusters like rt- or st- (type C) are, paradoxically, nearer to the preference in (7), since they already contain a prototypical {nB} binding, i.e. an {nB} binding enhanced by a maximal sonority distance: stop + vowel.4
1.3. The universal articulatory preferences: Level 3
Two main functions of phonology (to serve clarity of perception and ease of articulation) are reflected in perceptual, hearer-friendly preferences on the one hand, and in articulatory, speaker-friendly preferences on the other. Another level of structure, here called Level 3, is reserved precisely for the speaker-friendly preferences for articulatorily easy phonotactic sequences. According to a well-known study by Janson (1986: 193),5 the favoured combi- nations are those in which the articulators do not have to make extensive movements from the consonant gesture to the vowel gesture. While contrast is an underlying principle on the perceptual levels (cf. the figure-and-ground principle), similarity reigns on the articulatory level (cf. the proximity law).
1.4. The universal tendency for balance in perceptual and articulatory preferences
Conflicts among universal preferences, and especially between hearer- friendly and speaker-friendly preferences, are mediated by the major ten- dency for balance, which is realized on a language-specific level. Thus, despite the existing bindings (hearer-friendly) and preferred sonority dis- tances between segments (hearer-friendly), articulatory adjustments be- tween segments are possible (speaker-friendly), which eventually also serve
4 A further criterion which matters for the survival of a cluster is the actual size of the sonority distances: For example, in sta the CV distance is almost maximally greater (cf. pta) than CC, whereas in rta or kna the proportion is nearer to equal. 5 He found the following sequences to be favoured: dental/alveolar consonant + front vowel; liquid + front vowel; labial consonant + back rounded vowel; velar consonant + back rounded vowel.
72 Katarzyna Dziubalska-Kol-aczyk the hearer. Conflict6 solutions are implemented language-specifically to establish language-specific or typological relationships between bindings, phonotactic preferences and pure articulatory preferences. A clear syn- chronic example of how divergent those solutions may be is phonostylistics: while speaker-friendly preferences take precedence over hearer-friendly ones in casual speech, the relationship is reversed in formal speech. Thus, conflict solutions are inter alia style-sensitive. Language systems and lan- guage types diverge exactly according to the nature of the conflict solutions active in them. The principle of balance ensures that a particular system remains within the structural limits capable of serving linguistic functions of communication and cognition (see Dziubalska-Kołaczyk, 1995: §5.2.6. for a proposed scenario for the structuring of phonology). In order to account for the structure of sequences, phoneticians have tended to assign priority either to articulatory or to acoustic criteria. So, for example, Janson’s (1986) results appear to contradict Kawasaki’s (1982) claim about the primacy of acoustic distinctiveness in the creation of sound sequences. For instance, /w/ and labialized consonants tend not to be followed by a rounded vowel; /j/ and palatalized consonants tend not to be followed by front vowels. Both tendencies are due to insufficient auditory contrast in these sequences. Instead of treating this as a contradiction, one should assume a principle of balance between acoustic and articulatory criteria which leads to the creation of optimal sequences. This approach has been represented in various degrees by Lindblom (1986), Ohala (1990b), Maddieson & Precoda (1991) and Lindblom & Maddieson (1988). The last of these state, for example, that “consonant inventories tend to evolve so as to achieve maximal perceptual distinctiveness at minimum articulatory cost” (Lindblom & Maddieson, 1988: 72). In the present model, the effectiveness and optimality of balanced solu- tions are emphasized, which is clearly possible within the functional ap- proach to phonology advocated by the natural framework.
6 This functional conflict does not imply that the features of manner (stricture) and place (location) are themselves in conflict. In fact, they are complementary in the way they serve the two respective functions of phonology. An interesting proposal relating to this issue comes from Radical cv Phonology (van der Hulst, this volume), which claims that all fea- tures (manner, laryngeal and place) can be decomposed into basic elements — atoms C and V. Segment and syllable receive parallel structural interpretation. The categorial gesture is in the head position of the segment since categorial properties (manner in particular) deter- mine the distribution of segments within the syllable (cf. the sonority-based bindings and OSDP-based phonotactic preferences in the beats-and-binding model, i.e. Levels 1 and 2). The location gesture is formally dependent on the categorial one (e.g. if lateral then coro- nal), but this implicational relationship does not exclude other possibilities (cf. level 3 with its ease-of-articulation motivations).
Syllables? No! Substantive evidence for the beats-&-binding model 73
1.5. Semiotic precedence of words and morphemes over phonological entities
1.5.1. Priority of words and morphemes
Words and morphemes have semiotic priority over phonological entities: a. There is a preference, for different respective reasons, for a word-initial {nB} sequence and a word-final {Bn} sequence.7 b. The particular salience of word onset has priority over, and thus may come into conflict with, other preferences. c. The morphological structure of a word may override binding preferences.
1.5.2. Natural Morphology
As a theoretical background for morphological considerations, I adopt a semiotically based model of Natural Morphology (cf. Dressler, 1990; Dressler et al. 1987). The figure-and-ground principle also works in words, of course. The first consonant of a word is always a figure against the ground of the whole word (the final consonant is also a figure, though to a lesser extent). In a single- consonant word onset, the figure status of the consonant is even enhanced through its contrast with the following vowel, i.e. through the non-beat’s participation in the {nB} binding. In word-initial consonant clusters, e.g. C2C1-, the principle of balance de- cides the extent to which the figure status of C2 overrides the phonotactic preferences (Level 2) for a word-initial position. This depends on the phono- tactics of the initial cluster: a. In a cluster which is structured according to the phonotactic preferences, and thus has chances of surviving as a cluster (e.g. pr-), the figure status of C2 is enhanced. Therefore, if it does come to cluster reduction (e.g. in L1 acquisition), it is C1 rather than C2 that is deleted: prV → pV.
7 A word-initial {nB} guarantees salience through contrast, which is essential in (left-to-right) lexical search, as well as a higher informational load (consonants being more informative than vowels). In a word-final {Bn}, the final consonant serves as a marker of a word bound- ary (closes a word). The above preference is documented, for instance, by the following psycholinguistic evidence: — a predominantly left-to-right directionality of word recognition (see e.g. Gósy, 1991); — recognition of words by subjects before hearing the words completely (Frauenfelder & Lahiri, 1989); — TOT (tip-of-the-tongue) phenomena: Subjects in an experiment were able to recall, among other things, letters at the beginnings and ends of words (cf. Cowan, 1992: 276f).
74 Katarzyna Dziubalska-Kol-aczyk b. In a cluster like kn- or ps-, whose chances of survival are smaller due to the phonotactically unfavourable distribution of distances with respect to the following vowel (Level 2), the basic {nB} binding preference of Level 1 (not counteracted any more by phonotactic preferences of Level 2) pro- motes the non-beat bound to the beat (i.e. the C1 of the cluster) to the status of a word-figure. C2 loses its figure-status, and becomes suscep- tible to deletion, e.g.: knV → nV or psV → sV.
After the deletion, former C1 immediately replaces former C2 as a figure. A similar deletion also occurs with prosodically weak (unstressed) word-initial vowels, which often reduce (L1, phonostylistics, aphasia), and in this way are automatically replaced by strong ones in word-initial position.
2. So-called “evidence for the syllable” and its confrontation with the beats-and-binding model
2.1. Early Middle English vowel quantity changes
One of the most vexing problems of historical phonetics is syllabifica- tion in old languages. Seeing that scholars cannot agree on syllabifi- cation in modern languages, the chances of solving this problem for Gothic, Old Icelandic, or even Middle English are small. (Liberman, 1992: 77)
Early Middle English (henceforth EME) vowel quantity changes have tra- ditionally received a “syllable”-related interpretation. Both lengthenings and shortenings of vowels have been related to the “syllable” domain, by refer- ring to the “syllable” structure, boundaries and/or weight, or to the number of “syllables”, i.e. syllable counting.
2.1.1. Traditional description
Familiar handbook descriptions (e.g. Luick, 1921/40; Berndt, 1960; Fisiak, 1968; Reszkiewicz, 1973) refer to four different quantity changes which affected English vowels between the 9th and 13th century: HOMORGANIC LENGTHENING, SHORTENING BEFORE CONSONANT CLUSTERS, TRISYLLABIC SHORTENING and OPEN SYLLABLE LENGTHENING. HOMORGANIC LENGTHENING is described as a process lengthening short vowels if they were followed by clusters of two (roughly) homorganic con- sonants, as in go:ld, wo:rd, cli:mban, behi:ndan, si:ngan, e:orl, mu:rnan (Ritt, 1994: 1 & Appendix II). The process was marked by diacritics for length in Anglo-
Syllables? No! Substantive evidence for the beats-&-binding model 75
Saxon manuscripts, and is assumed to have taken place 400 years before so-called ME OPEN SYLLABLE LENGTHENING. The second change, SHORTENING BEFORE CONSONANT CLUSTERS, affected long vowels before clusters of more than one consonant (except for homor- ganic clusters and those that could function as “syllable” onsets) in relatively few words, such as kepte, brohte, softe, dust, mist, bledsian, wisdom, siknesse (Luick, 1914/21: 324ff, 392; cf. also Ritt, 1994: Appendix III). The third process, so-called TRISYLLABIC SHORTENING, is believed to have applied to long vowels in antepenultimate “syllables” of words. Luick’s examples are heafodu, linenes, rende, deorlingas, othere etc. (Luick, 1914/21: 328f, 392f; cf. also Ritt, 1994: Appendix IV). Finally, the fourth change, MIDDLE ENGLISH OPEN SYLLABLE LENGTHENING (MEOSL), according to the generally familiar textbook descriptions, affected short stressed vowels in open, i.e. unchecked, “syllables” in “disyllabic” words when there was only one consonant between the stressed and un- stressed vowel (unless it was a cluster of the st, sp, sk type), e.g. ma:ken, we:ven, ho:pen, a:le, be:ver, cha:ste, de:ne, ga:me, ha:ste, ta:ken, va:por, bla:d, co:l, ca:f, ho:l, la:c.
2.1.2. A critical note
Although Luick formulated the four separate rules describing the changes in vowel quantity in the Middle English period, he himself realized the need for a unifying approach to those changes. Since then, many attempts in this direction have been made, from the most radical unification by Ritt (1994), who incorporates all the conditioning factors of both the lengthening and shortening of vowels within a single formula, to less radical explanations of the ‘compensatory’, ‘moraic balance’ or ‘phonetic’ type (cf. Minkova, 1982; Jones, 1989; Minkova & Stockwell, 1992; Liberman, 1992; Vennemann, 1988; Murray, 1988; Ogura, 1987; Dresher & Lahiri, 1991, and most recently Minkova & Stockwell, 1996). What is needed is a unification of these approaches within the framework of a functionalist, non-conventionalist theory which would predict changes in the preferred direction as well as allowing for dysfunctional change. One of the reasons for problems with the unification of the EME quantity changes has been the multifaceted nature and disputability of the “syllable”. For example, since the criterion of “syllable” openness fails to explain so- called open syllable lengthening, one refers to “syllable” weight instead. One allows not only for “ambisyllabic” consonants (and counts them as half a mora), but also for fully (ha[st]e, a[pr]il) and partially (plan[t]e) “ambisyllabic” clusters. Additionally, while nt in plante is partially “ambisyllabic”, nd in bindan is a fully “ambisyllabic” prenasalized stop. Moreover, l in alum is also
76 Katarzyna Dziubalska-Kol-aczyk
“ambisyllabic”, but blocks lengthening, while more sonorant “ambisyllabics” favour it. Not only do the “ambisyllabicity” distinctions themselves seem ad hoc; they do not suffice to account for all aspects of the changes either. Also, the observation that sonorants both favour and disfavour the processes re- mains unaccounted for. The evidence from EME vowel quantity changes is believed to support the notion of “syllable” weight. I will present an alterna- tive account of the changes in the beats-and-binding framework.
2.1.3. Reanalysis of the vowel quantity changes
In accordance with the principles of the B&B model, EME quantity adjust- ments can receive the following interpretation: There are three hierarchically ordered domains of the quantity changes: — rhythmical domain — hearer-friendly, perceptual domain governed by the binding prefer- ences — domain of articulatory conditioning and implementation.
This hierarchy is compatible with (a) the ordering of processes in Natural Phonology (i.e. fortitions before lenitions); (b) hierarchical levels of phono- logical organization in the B&B model (Level 0 — rhythmical preferences > perceptual Levels 1 and 2 — governed directly or indirectly by bindings > Level 3 — articulatory preferences). The overall motivation in the rhythmical domain (Level 0) is isochrony, which is understood as the tendency for feet to be of equal weight. The tendency is realized on Level 1 by keeping the number of beats and bindings within a foot equal. As already specified, this is measured by counting beats and bindings from one beat up to the next or up to a phonological word boundary. Due to the preference for a trochee, a minimal foot-weight is 3, as in VCV = B{nB}B on Level 1, i.e. two beats and one binding. Beat-timed languages will preferably have this foot-weight. Stress-timed languages, however, will have a minimal foot-weight of 4 in a VCV = B{Bn}{nB}B se- quence, i.e. two beats and two bindings. In the process of change in a language from beat-timing towards stress-timing, feet of both weights will occur: weight 3 will persist due to the principle of foot-weight preservation, e.g. B{nB}B → BB{Bn} i.e. VCV → VVC, while weight 4 will develop as char- acteristic of the new (stress-)timing. In the discussed case of ME vowel lengthening, the initial step is made by the speaker, who reduces the binary foot (i.e. deletes a schwa), thereby
Syllables? No! Substantive evidence for the beats-&-binding model 77 compensating the weight within a foot (i.e. of the number of bindings/beats within a foot), as in (14).
(14) B{nB}B → B {nB}…→ BB{Bn} : (3→1→3) or B{Bn}{nB}B → B{Bn}…→ BB{Bn} : (4→2→3) i.e. /CVCV/ → /CVC(V)/ → /CV:C/
One will immediately notice that on the level of bindings (Level 1) a similar compensation will be legitimate in sequences with two intervocalic con- sonants, as in (15).
(15) B{Bn}{nB}B → B{Bn}{n… → BB{Bn}{n… : (4→2→3) i.e. /CVCCV/ → /CVCC(V)/ → /CV:CC/ A reverse case is that of a shortening whereby NEGATIVE COMPENSATION (reduction of foot-weight) towards the preferred number of beats and bind- ings takes place, as in (16).
(16) /CV:CCV/ → /CVCCV/ i.e. BB{Bn}{nB}B → B{Bn}{nB}B : (5 → 4)
This basic balance-of-weight principle of Level 1 undergoes modifications on Levels 2 and 3 (which leads to hierarchical conditioning of the quantity changes) which results in detailed solutions (cf. Dziubalska-Kołaczyk, 1995). However, the principle clearly unites all the quantity changes concerned.
2.2. First language acquisition
Data on child language have repeatedly been claimed to provide evidence for the vital role of the “syllable” in the process of the creation of first lan- guage phonology, and therefore for its role in phonology in general. Two well-known psycholinguistic models of first language acquisition may serve as examples: the GESTURAL MODEL OF EARLY CHILD PHONOLOGY (Goodell & Studdert-Kennedy, 1991) and the PHONOTACTIC CONSTRAINTS MODEL OF ACQUI- SITION (Menn, 1986) which claim that the “syllable” is a primary unit of speech processing and a primitive in L1.
2.2.1. Order of acquisition
The following order of acquisition of phonology is predicted by the B&B model: The child starts with a basic {nB} binding and a foot as his basic
78 Katarzyna Dziubalska-Kol-aczyk phonological unit. In early acquisition, a foot constitutes the only functional unit of speech. First it is a one-beat {nB}-shaped foot, which comes in sequences, most often reduplicated. This gives the impression of a proto- typical beat-timing. Then it is a binary {nB}B{nB}B foot which preferentially develops a falling accent, i.e. it acquires the preferred trochaic pattern. Foot- shaped phonological words and the other binding ({Bn}) develop next. At this stage, trochaic words with a final consonant (CVCVC) appear, the final consonant participating in a {Bn} binding. Only then does a {Bn} binding also function intervocalically, so that the above sequence has the structure: {nB}B{Bn}{nB}B{Bn}, and in single-beat words CVC has the structure {nB}B{Bn}. If the child already uses some CVC “words” before this stage, either they come from reduced {nB}B{nB}B sequences (final beat deletion), or they are successful repetitions of very frequent short words addressed to the child, or imitations of onomatopoeia. Finally, longer phonological words also develop, no longer limited to the size of one foot and involving con- sonant clusters. This order of acquisition is compatible with one of the basic assumptions of both psycholinguistic models mentioned in §2.2 above, namely that children begin with sequences from which, later, segments differentiate themselves. In the present model, the acquisition of bindings is a prerequi- site for the differentiation of consonants and vowels from the stream of speech — but “syllables” are not used.
2.2.2. The data
The data collected from two children M. and K.8 illustrate the order of acqui- sition claimed in 2.1. a. Both children start with a one-beat foot of a shape CV (representing the most preferred binding: {nB}), e.g.: 8 weeks M.& K.: le le, la, wa, gje 0;6 M.: pha, ba, phu, bu; K.: ble ble ble, da da da, ne
8 The children were both girls, observed from 8 weeks up to 2 years 8 months (2;8) of age, with random observation later. The regular observation consisted in tape recordings at monthly intervals and notes whenever any new development was observed.
Syllables? No! Substantive evidence for the beats-&-binding model 79 b. Then, a two-beat foot of the shape CVCV or VCV appears as a result of a combination or sequencing of two (the only available) {nB} bindings or a beat (due to non-beat deletion) and an {nB} binding, e.g.: 0;7 K.: dada, tata, ata, ada, a a 0;8 K.: mama, m:am:a, atha, athe 0;11 M.: tati, kaka, nana, mama This foot is not yet systematically of a trochaic stress pattern, but the tendency to stress the first beat is observable. Once a trochaic foot pat- tern gains in regularity, it remains long enough to be audible e.g. in the regularization of some items towards a trochee, as in 1;11: K.: ['bengu] ‘BMW’ /be em 'vu/, [u'esa] ‘USA’ /u es 'a/, ['pego] ‘Peugeot’ /pe'Zo/ c. Next comes a word with the shape of a two-beat trochaic foot (with reductions towards CVCV), e.g.: 0;11 M.: wawa ‹Mikołaj› 1;0 M.: towa ‹Mikołaj›; mma ‹nie ma› . 1;1 M.: tsjita ‹czytaj/c´›; kaka ‹lalka›; bowa, powa ‹ksia˛zka› 1;2 M.: baji ‹balon›. Simultaneously, when CVC words are attempted, they are still rendered without a {Bn} binding, e.g. 1;1 M.: ta: ‹tak› 1;3 M.: du di du di, du du du ‹daj› 1;4 M.: mu i, mu i ‹mis´›. Gradually, words and phrases longer than a binary foot also appear, still involving only an {nB} binding, e.g. 1;2 M.: koti da ‹kotki dwa› 1;4 M.: e towa ‹gdzie Mikołaj›, kaUa mma ‹kawy nie ma›. d. Next comes a word which is a two-beat trochaic foot and contains a {Bn} binding word-finally, e.g. 1;5 M.: tatuf ‹tatus´›, kodon ‹doka˛d›, po:tam ‹prosze˛›, bo:dan ‹Bogdan›. 1;6 K.: jajok ‹jajko›.
80 Katarzyna Dziubalska-Kol-aczyk e. Only now do one-beat words appear containing a {Bn} binding, which constitute indirect evidence for a word-internal {Bn} binding in two-beat words like those listed in d. above, e.g. 1;6 M.: ma ‹pan› (mani ‹pani› already at 1;5), o ‹słon´›, pe ‹pie˛c´›, pe ‹pies›, haj mama ‹hi mama›. . K.: daj to ‹daj to›, pa ‹spac´›, nok ‹nóz› (but: ta: ‹tak›).
Also longer words (of three-beats, i.e. rhythmically beat + trochee) with a final consonant occur more and more often, e.g. 1;5 M.: kokodij ‹krokodyl› 1;6 M.: papatam ‹przepraszam›, kokowaj ‹Mikołaj›, popotan ‹hipopotam›. f. At this stage, words containing CC medially appear: intervocalic double consonant clusters are already legitimate, since they are licensed by a {Bn} and an {nB} binding respectively. No initial or final clusters appear yet. E.g. 1;6 K.: pi- jdzie ‹przyjdzie› (notice that initial [pS] is simplified!) 1;7 K.: pujka ‹piłka›, fefwo ‹krzesło› (again, initial cluster simplified), bojki ‹bombki› 1;8 K.: otfu ‹otwórz›, palku ‹parku› 1;9 M.: damki ‹trampki› (notice initial cluster reduction to one con- sonant and medial cluster reduction to two consonants). g. Much later come words containing CC initially. They start appearing, at first very rarely, after four or five months of the use of word-internal clus- ters, i.e. around 1;10–1;11, and usually increase in the number quite slowly. Some examples are: 1;10 M.: pSi- Set ‹przyszedł› K.: stali- ‹stary› 1;11 M.: mju:t ‹miód› K.: bluga ‹bluza, bluzka›, but deNtKe ‹zdje˛cie› with reduced initial but maintained internal cluster, tKupkla ‹s´rubka› where all bindings are retained while the initial cluster is reduced at the expense of the medial position. Also word-final clusters start to appear, e.g. 1;11 K.: golk ‹Golf›, kibolk ‹keyboard›.
Syllables? No! Substantive evidence for the beats-&-binding model 81
2.2.3. Recapitulation
As predicted, the stronger binding appears to be acquired before the weaker one, i.e. {nB} before {Bn} (Level 1). The order of acquisition of consonant clusters observed above is also predictable within the beats-and-binding model. Intervocalic, word-internal two-consonant clusters are acquired first, since in this position both consonants participate in bindings with separate beats. Word-initial clusters are acquired before word-final ones due to the saliency of the word-initial position, which makes the child more aware of the initial segments. This saliency, however, is at the same time responsible for reductions in initial position, in which listener-friendly processes operate to regain the preferred binding {nB}. Word-final position, being less promi- nent for the listener, may tolerate more clusters but is simultaneously subject to speaker-friendly processes of simplification (Level 3). The universal preference for a trochaic foot (Level 0) has also been con- firmed by the data. The two Polish children prefer a trochee, although Polish has a lexical penultimate stress and is not a trochaic language (as, for in- stance, English is often cited to be). During the period before the appearance of consonant clusters, universal phonotactic preferences for alternating sequences of consonants and vowels (still governed by Level 1) find full support in the data. Numerous cluster simplifications in the form of reductions, assimilations, substitu- tions, coalescences or metatheses all have the same teleology: to maintain within a foot or within a phonological word, in the first place, {nB} bindings, and, secondarily, {Bn} bindings. The result also satisfies Level 3 preferences, i.e. ease-of-articulation based ones. Here are some examples of the above mentioned simplifications: . . bowa ‹ksia˛zka›, kuka ‹ksia˛zka›, papatam ‹przepraszam›, jeje ‹jeszcze›, jajok ‹jajko›, kako ‹ciastko›, fufu ‹siusiu›, pa ‹spac´›, e ‹gdzie›, maga ‹Magda›, kulika ‹kuleczka›, guka ‹gumka›, puka, pika ‹piłka›, miko ‹mydło›, me:ko ‹mleko›, mika ‹miska›, taki ‹kwiaty›, baja ‹Baba Jaga›, papo ‹spadło›, vos’ ‹włos›, potet ‹poszedł›, pojeje ‹pojechał›, pujaja ‹pojechała›, takok ‹kwiatek›, tak ‹kwiat›, dugi takek je ‹drugi kwiatek jest›, lalola ‹Gwiazdora›, fifinka ‹dziewczynka›, papety ‹skarpety›, atobu ‹autobus›, mokika ‹motylek›, podut ‹samochód›, taki ‹ptaki›, . buka pada ‹głupia Magda›, moto pi: ‹mocno ´pi›,s pekek ‹przeciez›, beduty ‹zepsuty›, ukle f poNeni ‹klucze w spodniach›, eba pogagoga pelikjem ‹trzeba posmarowac´ kremikiem›. Later on, when more complex phonotactics were starting to appear, the binding preferences could still be observed, e.g. 5;0 M.: mi-Sa, mSa ‹msza›, pleK ‹ples´n´›, xuxu ‹Uhu›, xedit ‹Edith›, 3;6 K.: i- mgwa, emgwa ‹mgła›, 4;4 K.:
82 Katarzyna Dziubalska-Kol-aczyk doknie ‹dotknie›, 4;10 K.: ktawt ‹kształt›, up to ca. 5;0 K.: gjazda ‹gwiazda›, kjatek, fkfjatek ‹kwiatek›. In general, binding preferences enable one to account in a principled way for the order of phonological acquisition.9
2.3. Second language acquisition
2.3.1. References to the syllable in L2 acquisition studies and their explanation in terms of the B&B model
The syllable has been claimed to play a role in the acquisition of L2 phonol- ogy as a processing unit, as a unit of timing and as a domain of phonotactic constraints. Below I will limit the discussion to refer only to the first claim. The syllable is treated as a unit of phonological processing in such docu- mented SLA cases as: h-EPENTHESIS (more precisely: prothesis) by Italian learners of German e.g. in Ackermann ['ha.k7r.man], acht [haxt]; beeilen [b7.'haI.l7n]; verursachen [f7r.'hur.sa.x7n] (Hurch, 1982: 276), which is (approximately) hierarcharized according to the degree of stress on the syllable concerned. (In general, 70% of all epentheses in the data take place in word-initial position, 20% are hiatus breaking). ANALOGOUS COPYING, e.g. in ich hasse [hoIç] euch, ['ha.l7] alle Häuser, wir haben ['hangst] Angst, which is causally explained as a fortitive improvement of the syllable structure. METATHESIS, e.g. in einholen ['haI.no.l7n], abheben ['ha.be.b7n], inhalieren [‚hi.na.'li.r7n], überholen [‚hy.b7.'ro.l7n] — as above. GRAPHIC INTERFERENCE, e.g. in [hi:r] ihr, [hu:r] Uhr, ['he:r7] Ehre, is analysed as an improvement from a VC to a CV syllable. The phenomena mentioned above are all conditioned by the {nB} prefer- ence reinforced in a word-initial position, but also in a hiatus position. /h/ is also inserted according to the degree of stress on the following beat, ac- cording to the rich-get-richer principle, i.e. the stronger a position is (esp. through stress), the better it should satisfy the perceptual preferences, which would make it even stronger (since a stressed beat should preferen- tially participate in an {nB} binding). ANALOGOUS COPYING is reinforced articulatorily by a kind of (anticipatory or perseverative) consonant har- mony. METATHESIS takes place between the initial positions of neighbouring
9 I fully agree with Wolfgang Dressler’s remark that this account requires elaboration as well as verification against more systematic and richer data. Such data of two further Polish children are now under my scrutiny for the above purpose.
Syllables? No! Substantive evidence for the beats-&-binding model 83 morphemes. So, the morphemes are clearly salient, but still more salient is the word-initial position, which is therefore reinforced with an {nB} binding by means of metathesis. This changes the initial structure of both mor- phemes, so that the boundary is no longer transparent, while the phonol- ogy of the whole word improves, both initially ({nB}) and internally (since an internal consonant cluster was reduced and a new {nB} established). In the case of the GRAPHIC INTERFERENCE, the process is again a consequence of the {nB} preference. The source of /h/ may indeed be graphic, or it may appear due to analogy with other epentheses.
2.3.2. Segmentation algorithm applied by the subjects in a “syllabification” test10
Within the B&B model predictions can be drawn as to how the make-up of internal clusters (with reference to the sonority distances between the relevant consonants and vowels) will influence their segmentation. For example, one phonotactic preference on Level 2 which is relevant to V1C1C2V2 clusters (cf. the OSDP above and Dziubalska-Kołaczyk, 1995: 77ff) reads as in (17).
(17) V1C1C2V2 : V1C1 ≥ C1C2 < C2V2
Read: In a V1C1C2V2 sequence, the sonority distance C1C2 should preferably be smaller than or equal to the distance V1C1 and smaller than the distance C2V2.
A cluster like /VnkV/ observes the above preference, so it is expected to receive a VC-CV segmentation, due to the preferences of Level 1.11 In a cluster like /VltV/, however, which does not observe the above preference, the relationship between V1C1 and C1C2 corresponds to another phonotactic preference, concerning word-final two-consonantal clusters (VC1C2 : VC1 ≤ C1C2). Therefore, one can predict some segmentations of the type Vlt-V in the subjects’ answers, besides Vl-tV segmentations (grounded both phonologi- cally and orthographically), but no segmentations of the *V-ltV type (or only
10 There is no space here to report fully on the segmentation test performed in order to elicit syllable boundaries. The test concentrated on eliciting segmentations of double and triple intervocalic consonant clusters. 11 The expectation was fulfilled by the data, since 74% of the VCCV clusters were segmented in this way.
84 Katarzyna Dziubalska-Kol-aczyk rarely, when supported by morphology). This prediction is precisely borne out by the data.12 Confirmed by the results of the test, the segmentation algorithm in Figure 2 can be derived, based on the principles of the B&B model. The input to be segmented is filtered through the preferences of Level 2 (unless it is segmented immediately due to complex orthography or transparent morphology). If the filtering shows that the input observes the relevant preference, the output comes under the control of Level 1 preferences, as well as morpheme-oriented ones and orthography. If the filtering at Level 2 shows some violation of the relevant preference, other Level 2 preferences take over and the most adequate candidate sequence for an output is selected. Afterwards, as in the above case, the output comes under the control of Level 1 preferences, morpheme-oriented ones and orthography.
3. Final remarks
In this paper I have outlined the beats-and-binding theory of phonology, embedded in the larger theoretical framework of Natural Phonology and Natural Linguistics. It has been claimed that the B&B theory provides more comprehensive and explanatorily adequate predictions and accounts of the phonological phenomena than a theory relying on the unit syllable (be it primitive or derived). In particular, in the B&B model, the phonotactics of segments in a sequence constitute a necessary consequence of the opera- tion of the binding preferences: The phonotactics counteract preferences by maintaining appropriate sonority distances between segments, which are thus prevented from reducing to the absolute minimum (i.e. CV). Thus, phonotactics do make use of “syllable structure constraints”. Linguistic timing relationships between beats account for what is traditionally called “syllable weight”. “Syllable boundaries” do not, in principle, constitute a phonological issue: Speakers of a language are able to produce pauses in between beats as, inter alia, a side-effect of the operation of the binding preferences. The notion of so-called “syllable contact” (cf. Vennemann, 1988) is also untenable: patterns of consonants are a corollary of binding con- sonants to beats.
12 In 7 words containing the relevant clusters (/lt/, /rg/, /lk/, /rt/, /rk/, /ld/), 20 segmentations (out of 105) were of the Vlt#V type and none of the *V#ltV type.
Syllables? No! Substantive evidence for the beats-&-binding model 85
I/Px
COMPLEX ORTHOGRAPHY LEVEL 2 PREFERENCES TRANSPARENT MORPHOLOGY
O/Px
O/Pnk O/P O/Pl etc.
LEVEL 1 PREFERENCES MORPHOLOGY ORTHOGRAPHY
Figure 2. The segmentation algorithm
The above claims have been juxtaposed with the syllable-based analyses of phonological phenomena witnessed in diachrony, first language acquisi- tion and second language acquisition.13 The B&B treatment of these has been demonstrated. Further development of the B&B model is planned, to formalize the binding and phonotactic preferences, to elaborate the phonotactics in juxtaposition with language-specific data, and to refine its predictions by the treating new types of substantive evidence.
Acknowledgements
I would like to thank all the colleagues who made comments on this paper during the Phonology Meeting, in particular Dafydd Gibbon (to whom my special thanks are due for his thorough reading of the manuscript and enlightening questions and suggestions), Harry van der Hulst, Grzegorz Dogil, Wolfgang Dressler, Elke Ronneberger-Sibold, Anders Erikson, Robert Bannert. They made me more aware of both strengths and weaknesses of my model, and the questions they raised will guide me in my nearest-future research.
13 For the discussion of other sources of internal and external evidence in the light of the B&B theory, see Dziubalska-Kołaczyk (1995).
86 Katarzyna Dziubalska-Kol-aczyk
I would also like to thank the anonymous reviewer whose detailed ques- tions showed me yet another reaction from outside my theoretical frame- work.
Bibliography
Bell, A. & J.B. Hooper (1978). Issues and evidence in syllabic phonology. Syllables and Segments, edited by A. Bell & J.B. Hooper, 3–24. Amsterdam: North-Holland. Berndt, R. (1960). Einführung in das Studium des Mittelenglischen. Halle: Max Niemeyer. Cowan, J.R. (1992). A model of lexical storage. Evidence from second language learners’ orthographic errors. The Linguistics of Literacy, edited by P. Downing, S.D. Lima & M. Noonan, 275–290. Amsterdam: Benjamins. Dogil, G. (1981). Elementary accent systems. Phonologica 1980, edited by W.U. Dressler, O.E. Pfeiffer & J.R. Rennison, 89–99. Institut für Sprachwissenschaft, Innsbruck: Innsbrucker Beiträge zur Sprachwissenschaft. Dogil, G. & G. Braun (1988). The Pivot Model of speech parsing. Wien: Verlag der ÖAW. Dogil, G. & H.C. Luschützky (1990). Notes on sonority and segmental strength. Rivista di Linguistica 2, 3–54. Dresher, B.E. & A. Lahiri (1991). The Germanic foot: metrical coherence in Old English. Linguistic Inquiry 22, 251–286. Dressler, W.U. (1985). Morphonology: the Dynamics of Derivation. Ann Arbor: Karoma. Dressler, W.U. (1988). Preferences vs. strict universals in morphology: word-based rules. Theoretical Morphology, edited by M. Hammond & M. Noonan, 143–154. San Diego: Academic Press. Dressler, W.U. (1990). The cognitive perspective of “naturalist” linguistic models. Cognitive Linguistics 1, 75–98 Dressler, W.U. (1996). Principles of naturalness in phonology and across components. Natural Phonology: The State of the Art. Papers from the Bern Workshop on Natural Phonology, edited by B. Hurch & R. Rhodes, 41–52. Berlin: Mouton de Gruyter. Dressler, W.U., W. Mayerthaler, O. Panagl & W.U. Wurzel (1987). Leitmotifs in Natural Morphology. Amsterdam: Benjamins. Dziubalska-Kołaczyk, K. (1995). Phonology without the syllable. Poznan´: Motivex. Edwards, M.L. & L.D. Shriberg (1983). Phonology. applications in communicative disorders. San Diego: College-Hill Press. Fisiak, J. (1968). A short grammar of Middle English. Warszawa: PWN. Frauenfelder, U.H. & A. Lahiri (1989). Understanding words and word recognition: does phonol- ogy help? Lexical representation and process, edited by W. Marslen-Wilson. Cambridge, Mass.: The MIT Press, 319–341. Goodell, E.W. & M. Studdert-Kennedy (1991). Articulatory organization of early words: from syllables to phoneme. ICPhS 12, 166–169. Gósy, M. (1991). On the process of word identification. Proceedings of LP’90, edited by B. Palek & P. Janota, 372–385. Prague: Charles University Press. Hurch, B. (1982). H. Eine Studie zur kontrastiven Phonetik und Phonologie. Parallela, edited by M. Dardano, W.U. Dressler & G. Held. Tübingen: Narr. 271–280. Janson, T. (1986). Cross-linguistic trends in the frequency of CV sequences. Phonology Yearbook 3, 179–186. Jones, C. (1989). A history of English phonology. London: Longman. Kawasaki, H. (1982). An acoustical basis for universal constraints on sound sequences. Ph.D. dissertation. University of California, Berkeley.
Syllables? No! Substantive evidence for the beats-&-binding model 87
Krupa, V. (1968). The Maori Language. Moscow: ‘Nauka’ Publishing House. Krupa, V. (1973). Polynesian languages: A survey of research. The Hague: Mouton. Liberman, A. (1992). A bird’s-eye view of open syllable lengthening in English and in the other Germanic languages. Nowele 20, 67–87. Lindblom, B. (1986). Phonetic universals in vowel systems. Experimental Phonology, edited by J.J. Ohala & J.J. Jaeger, 13–44. Orlando: Academic Press. Lindblom, B. & I. Maddieson (1988). Phonetic universals in consonant systems. Language, Speech and Mind, edited by L. Hyman & C.N. Li, 62–78. London: Routledge. Luick, Karl (1921/1940). Historische Grammatik der englischen Sprache. Stuttgart: Bernhard Tauchnitz. (Reprinted 1964). Maddieson, I. & K. Precoda (1992). Syllable structure and phonetic models. Phonology 9, 45–60. Menn, L. (1986). Language acquisition, aphasia and phonotactic universals. Markedness, edited by F.R. Eckman, E.A. Moravcsik, & J.R. Wirth, 241–256. New York: Plenum Press. Minkova, D. (1982). The environment for Open Syllable Lengthening in Middle English. Folia Linguistica Historica III, 29–58. Minkova, D. & R.P. Stockwell (1992). Homorganic clusters as moric busters in the history of English: The case of -ld, -nd, -mb. History of Englishes, edited by M. Rissanen, O. Ihalainen, T. Nevalainen & I. Taavitsainen, 191–206. Berlin–NY: Mouton de Gruyter. Murray, R.W (1988). Phonological strength and Early Germanic syllable structure. München: Wilhelm Fink Verlag. Noske, R. (1989). Syllable geometry. Belgian Journal of Linguistics 4, 119–144. Ogura, M. (1987). Historical English phonology. A lexical perspective. Tokyo: Kenkyusha. Ohala, J.J. (1990a). Alternatives to the sonority hierarchy for explaining segmental sequential constraints. The parasession on the syllable in phonetics and phonology. CLS 26, 319–338. Ohala, J.J. (1990b). The phonetics and phonology of aspects of assimilation. Papers in laboratory phonology I, edited by J. Kingston & M. Beckman, 258–275. Cambridge: CUP. Ohala, John J. & H. Kawasaki (1984). Prosodic phonology and phonetics. Phonology Yearbook 11 , 113–129. Pulgram, E. (1970). Syllable, word, nexus, cursus. The Hague: Mouton. Reszkiewicz, A. (1973). A diachronic grammar of Old English. Warszawa: PWN. Ritt, N. (1994). Quantity adjustment: vowel lengthening and shortening in Early Middle English. Cambridge: Cambridge University Press. Selkirk, E.O. (1984). On the major class features and syllable theory. Language sound structure, edited by M. Aronoff & R.T Oehrle, 107–136. Cambridge, Mass.: MIT Press. Ungerer, F. & H-J. Schmid (1996). An introduction to cognitive linguistics. London: Longman. Vennemann, T. (1988). Preference laws for syllable structure and the explanation of sound change. Berlin: Mouton.
Features, segments and syllables in Radical CV Phonology
Harry van der Hulst
Abstract
The original goal of Radical cv Phonology (van der Hulst, 1990; 1991; The Hague: Thesus. 89–111. Thesus.The Hague: 89–111.
. 1994a, b; 1995a) was to derive a set of phonological features and feature classes from general categorization principles, instead of simply stipulate them. The resulting theory offers a specific “feature geometry” without
Syllables !? arbitrary or domain-specific properties (see Den Dikken & van der Hulst, 1988). In this article, section 2 presents a summary of the current state of the theory in the segmental domain. Van der Hulst (1996) extended Radi- cal cv Phonology to the domain of syllable structure, showing how it establishes an explicit formal relation between the representation of the Phonologica 1996: Phonologica 1996: relevant segmental properties (i.e. stricture properties) and syllable structure itself (discussed in section 3.1). Section 3.2 develops the syllable module further discussing the con- sequences of assuming syllable structure to be an inherent property of lexical representations, rather than derived through syllabification. Golston & van der Hulst (to appear) argue for two important conse- quences of this position. Firstly, if syllable structure is inherent, infor- mation about linear order can be suppressed. Secondly, given the correla- tion between the position of terminals in the syllabic structure and the stricture properties of segments in these terminals, we can suppress the formal distinction between syllabic structure and stricture properties. This idea is briefly explored within Radical cv Phonology. In doing so I incorporate the “strict CV approach” (to syllabic structure), proposed in Lowenstamm (1996). 0. John R. Rennison & Klaus Kühnhammer, eds. (1999). eds. R. Rennison & Klaus Kühnhammer, John (1999).
90 Harry van der Hulst
1. Phonological categories in Radical CV phonology
1.1. Introduction
Since Radical cv Phonology (RcvP) is a rather abstract theory, I will first discuss the question of how phonological categories and phonological structure in general, relate to phonetic “reality”. The SPE theory (Chomsky & Halle, 1968) embodies a view that, according to Carr & Burton-Roberts (1996), is fundamentally flawed because it assumes that phonological deriva- tions map phonological representations into the phonetic reality. The “phonetic reality” is, given the articulatory bias of SPE, a set of instructions (i.e. motor commands) to the articulators. In line with this view, input or lexical representations are referred to as “phonological” whereas surface or output representations are called “phonetic”. The phonological and the phonetic level forms the beginning and end point of a continuum that is bridged by the application of phonological rules. Carr & Burton-Roberts argue that this view is untenable if phonological representations are taken as cognitive in nature and phonetic representa- tions as physical events in the world (movements of the articulatory devices or noises) because phonological rules or generalizations cannot change the ontological status of their input. Thus phonological representations that form the input to rules must somehow be already intrinsically phonetic. But it is difficult to understand how cognitive categories can be intrinsically phonetic. The RcvP position on the relation between phonology and phonetics is closely related to the views of Hjemslev and Firth. Fischer-Jorgensen (1975) and Anderson (1985) provide introductions to these views and contain references to the original works. The objects that the theory defines have no intrinsic phonetic properties. Yet, part of our knowledge of language is the ability to relate phonological categories to collections of phonological events. Along Firthian lines I will refer to these events as the exponents of phonological categories. As Ogden (1993) writes, “the idea of exponents embodies the view on what today would be called ‘phonetic interpretation’”. At this point, various questions arise. The first question is whether the set of phonological categories is in some sense universal. Firthians claim that this is not the case. The set of categories that is used in this approach is fully determined by what is convenient in analysing a specific set of data. In short, no specific hypothesis is put forward concerning the set of phonolog- ical categories. This means that Firthian phonology aims at description only. Firthians distinguish certain types of categories, such as paradigmatic (or phonematic) units and syntagmatic (or prosodic) units. But they do not believe that sensible things can be said about, for example, the number of
Features segments and syllables in Radical CV Phonology 91 such units. This is something I would like to dispute. A central claim of RcvP is that there are interesting things to say about the set of phonological categories. The claim is that, corresponding to the various phonetic dimen- sions that are relevant to speech, a definable set of phonological categories appears to exist. Thus, instead of postulating a list of phonetically defined features as part of the human language ability (“UG”), RcvP proposes a universal procedure to produce phonological categories. The second question concerns mentalism. As will be clear from the dis- cussion so far, Firthians do not claim that the analysis of a specific set of data in terms of categories and their exponents constitutes a hypothesis about the knowledge that underlies the production and perception of these data. A Firthian analysis makes no claim to psychological reality. It is not clear to me whether this is a point of principle or merely a sceptical attitude based on the idea that, at present, we have no way of testing whether an analysis is psychologically real. Perhaps the latter. My own perspective is different. If it turns out that the set of phonological categories reveals a systematic structure, the question arises where this structure comes from. Several explanations are possible. One is that the people that devote their careers to studying the sound side of languages (phonologists) are somehow designed to come up with a systematic kind of categorization. A second explanation is that language-learning children necessarily learn the phonological structure of their language in terms of a systematic set of categories and that the phonological analysis reveals this structure. These two explanations do not exclude each other, neither is one necessarily less or more interesting than the other. In both cases the categorization in some sense comes from the human cognitive system. There is of course another explanation as well, which is that the structure that our set of categories seems to have is a direct reflection of the structure of the phonetic dimensions that are relevant to speech. Stevens’ quantal theory is presumably the best example of this line of reasoning (Stevens, 1972). But if it appears that generalizations can be made over sets of phono- logical categories that correspond to different phonetic dimensions, it seems plausible to think that these generalizations, at least in part, do not reflect a structure in the phonetic substance (or as Hjemslev would say: purport), but rather reflect a categorization strategy, i.e. a cognitive capacity or bias to categorize the physical world in a certain way. I will try to show that the set of phonological categories needed for phonological analysis reflects a structure which is unlikely to be fully determined by the phonetics of speech and which I will therefore interpret as a reflection of a cognitive capacity to parse continua into a set of discrete categories. In addition, I will claim that this cognitive capacity exists not only in the mind of the phonol- ogist, but also in the mind of the language learner. Whether this capacity is
92 Harry van der Hulst part of a language-oriented ability or part of a more general cognitive ability cannot be determined before we have examined the relevant literature bearing on other human activities.
1.2. Phonological categorization
By saying that the relation between phonological categories and exponents is arbitrary we do not deny or exclude that the latter are always maximally dispersed within their phonetic space. The motivation for dispersion is clear enough. The function of phonetic forms is to keep apart morphemes or words. Hence, it makes no sense to do this with phonetic forms that are very much alike, both from a production and a perception point of view. Liljencrants & Lindblom (1972) have stressed the important role of disper- sion, but perhaps they overstressed it. Dispersion theory cannot explain why we end up with an identical set of categories for each phonetic space. In the simplest case, then, a phonetic space gives rise to two phonolog- ical categories, which we could label (arbitrarily) A and B (or yin and yang). These categories take as their prototypical exponents dispersed phonetic events. I will refer to these categories as polar categories; for simplicity’s sake I represent the phonetics in terms of simple linear scales, as in (1).
(1) A B Polar categories
Below we will see that there are several phonetic scales, such as manner of articulation (stricture), place of articulation and so on. Thus the phonetic substance is not atomic or holistic. Phonological categorization, so it seems, takes place with reference to each of these scales. The isolation of these scales, then, is itself an act of categorization. Let us call these scales c-categories (“c” standing for feature class). The formation of polar catego- ries takes place within each c-category.
(2) A B (e.g. stricture)
C D (e.g. stricture)
E F (e.g. place)
etc.
I would now like to argue that the total set of polar categories {A, B, C, D, E, F, …} is divided into two natural classes. The basis for this classification lies
Features segments and syllables in Radical CV Phonology 93 in the fact that the speech signal is organized rhythmically. Different nota- tions, shown in (3), have been proposed to represent the basic rhythmic structure of speech.
(3) a. … W S W S W S W S W S … b. … V C V C V C V C V C … c. … R O R O R O R O R O …
S, V or R represent points of “high prominence” and W, C or O represent points of low prominence. I take “prominence” to refer to perceptual sali- ence. The relation between rhythm and phonetic scales is that for each scale one polar category contributes to high prominence whereas the other contributes to low prominence. Thus, such diverse properties as lack of stricture, voicing and pharyngeal place produce a high prominence effect, a peak. To express the functional unity of these exponents RcvP sees them as exponents of polar categories of the same type, i.e. V. Likewise, sets of phonetic events that contribute to troughs in the rhythmic organization are regarded as exponents of the category C. Since the original conception of RcvP was biased toward spoken lan- guage, the polar categories are labelled “V” and “C”, referring to V positions in the rhythmic structure (which are preferably provided with V-type polar categories) and C-positions (which are preferably provided with C-type polar categories). We will see that exponents of V do not necessarily contribute to salience. Sometimes V exponents represent the default option (like low tone, which is a V exponent). It turns out that scales can be ‘parsed’ into more than two categories. A maximal number of four categories seems quite typical, as we will see. I will derive this four-way categorization by assuming that the polar categories can enter into a head–dependent relation, which I represent as in (4a), for which I use (4b) as a shorthand.
(4) a. C C V V
V C
b. C Cv cVc V
Within the set of four categories, C and Cv are referred as C-headed, V and Vc as V-headed. Some feature classes are more instrumental in producing the promi- nence alternation than others. In particular, properties that involve degree
94 Harry van der Hulst of stricture play a key role in this respect. To express this, I represent the stricture polar categories as heads and other polar categories as depend- ents.
(5) … V C V … (cf. (4b)) M
Head Dependent (stricture) …
Stricture properties represent the so-called major classes. Hence the claim I am making here is related to the proposal of McCarthy (1988) to regard major class features as head features. Below, I will argue that the major head–dependent distinction cor- responds to the distinction between stricture (or more generally: manner) and place. Within each of these major feature classes I will distinguish subclasses, also in terms of head–dependent relations. This implies that we replace the phonetic names of all feature classes by the formal notions head and dependent. Every class can be uniquely identified as the head or de- pendent at some level in the structure. With the above concepts we can now define a set of unary phonological features.
(6) a. Phonological Feature = [Scale Type, Scale Category] Scale Types: {STRICTURE, PLACE,…} Scale Categories: {C, Cv, Vc, V} b. Features Exponents [Stricture, C] → stop [Stricture, Cv] → fricative [Stricture, Vc] → sonorant consonant [Stricture, V] → vowel [Place, C] → coronal [Place, Cv] → labial [Place, Vc] → high [Place, V] → low …
Although the categories C, Cv, Vc and V themselves are neutral with respect to specific phonetic properties, they have phonetic ‘meanings’ or ‘exponents’ in their specific instances, which I will indicate in articulatory terms, although each unit is supposed to correspond to an identifiable
Features segments and syllables in Radical CV Phonology 95 acoustic event. For ease of exposition I indicate, where possible, traditional labels to the categories. A look at what several decades of phonological research has taught us reveals that eight features is presumably not enough. I extend the set of features by postulating the following feature geometry in (7).
(7)
Manner’ Place
Phonation Manner Secondary Primary Tone articulation articulation {C,Cv,Vc,C} {C,Cv,Vc,C} stricture substricture Major Place Subplace {C,Cv,Vc,C} {C,Cv,Vc,C} {C,Cv,Vc,C} {C,Cv,Vc,C}
This geometry makes use of a double X-bar organization. Within each class we have a four-way distinction. The system proposed so far generates 24 features. My claim is that this is exactly all that is required for phonological analysis. The table in (8) summarizes the exponents of the 24 unary features that RcvP allows.
(8) Phonological Features and their exponents HEAD (Manner) DEP (Place) Spec Head Compl Spec Head Compl (Phon) (Strict) (SubStr)(SecPla) (MajPla) (SubPla) C high / constr. stop lateral palatal coronal front Cv hi-mi / creaky fric. stri/rho labial labial round Vc lo-mi / aspir. son. cns. nasal dorsal high laminal V low / breathy vowel voiced pharyng. low posterior
Note that the exponents that are listed for each of the four scale categories correspond to traditional features that in a number of cases are known to be ‘related’, relations usually being expressed in the form of redundancy rules. Below I will provide more details about certain aspects of this chart. I conclude this section by arguing why phonological features must be ab- stract, in the sense that they are not intrinsically phonetic. The first argument involves the acquisition of language by deaf people. Today, there can be no doubt that deaf people acquiring a sign language acquire phonological knowledge. The form side of sign languages has an autonomous organization of which the units and smallest buildings blocks
96 Harry van der Hulst do not necessarily bear meaning, i.e. sign languages have a phonological organization. Those who claim that humans are endowed with certain capacities to set up this phonological knowledge of a language cannot seriously believe that these capacities make something available that is intrinsically connected to the properties of the human vocal tract and ear, i.e. a set of ‘actual features’. Rather, it would seem that language learners are equipped with a capacity to set up a system of abstract categories which lend themselves to being externally represented by a system of perceptual events and the appropriate corresponding articulatory gestures. The precise physical properties of the perceptual events and articulatory gestures must be largely irrelevant to the nature of this abstract system because it would otherwise not be possible for hearing and deaf language learners to approach their task on the basis of a single human capacity. The hypothesis that deaf human beings appeal to other capabilities to construct their phonological system than hearing people is both unlikely and theoreti- cally less interesting (cf. van der Hulst, 1993; 1995b for ideas on the repre- sentation of signs which were inspired by RcvP). The necessary abstractness of phonological categories follows not only from the modality argument, however. In fact, before I ever considered the relevance of sign language to this debate, I had already concluded that the system must be abstract, based on the argument that the various feature classes for spoken languages that the tradition hands to us seem to be structured in a specific way. RcvP makes this structure, which I outlined above, explicit.
1.3. Segmental categories
1.3.1. Manner
In this section I focus on distinctions that are traditionally referred to as ‘manner distinctions’ (see van der Hulst, 1995a for a more extensive treat- ment). I will argue that there are two types of manner properties, those that involve degree of stricture and those that involve modifications of the relevant resonating cavities. We will see that the elements that express stricture express so-called major class distinctions at the same time. Our system therefore does not recognize major class features or elements as a distinct class. Categories for resonating cavities express activation or ex- pansion of the nasal and pharyngeal cavities as well as modifications of the oral cavity. The relevant distinctions in both classes can be expressed in terms of the same set of four elements (C, Cv, Vc, V), occurring as heads or dependents, respectively. I am also assuming that the laryngeal class is
Features segments and syllables in Radical CV Phonology 97 added as a specifier to the manner node, as shown in (9). Head features (expression stricture distinctions) are immobile, i.e. non-spreading and local, whereas dependent features (expressing substricture and laryngeal distinc- tions) are, or can be, mobile in terms of spreading or non-local interaction. Such differences in behaviour are in line with general characteristic proper- ties of heads and dependents.
(9) Manner Manner
tone / phonation stricture substricture C high constricted stop lateral Cv high-mid creaky voice fricative strident/rhotic Vc low-mid aspirated sonorant nasal V low breathy voice vowel voice
The traditional notion of stricture is represented here in terms of four categories (cf. (4b)).
(10) Stricture
C V (obstruent) (sonorant) C V C V
C Cv cVc V (stop) (fric.) (sonor.) (vowel)
This four-way distinction is also adopted in van de Weijer (1994) and in Humbert (1995). C-headed stricture involves a relatively high degree of stricture (as in ob- struents), while V-headed stricture involves a low degree of stricture (as in sonorants).
(11) Stricture C = absolute stricture (as in stops) Cv = very close approximation creating turbulence (as in fricatives) Vc = stricture that does not impede the airflow because there is an ‘escape route’ (as in sonorant consonants) V = absence of stricture (as in vowels)
98 Harry van der Hulst
This four-way division corresponds to the four-way stricture division pro- posed in Ladefoged & Maddieson (1995: 370–371). In (10) and (11) we express directly that stops and oral vowels are un- marked with respect to their manner or, more narrowly speaking, stricture: stops are unmarked onsets (as opposed to fricatives) and oral vowels are unmarked nuclei. This is in accordance with the fact that the prototypical and least marked syllable consists of a stop followed by an oral vowel, two segments that are maximally opposed in terms of their stricture properties (cf. Jakobson, 1941; Jakobson, Fant & Halle, 1952). Let us now discuss the interpretation of the four scale categories in de- pendent (complement) position.
(12) Substricture C = lateral modulation of oral cavity (lateral) Cv = central modulation of oral cavity (strident, rhotic) Vc = activation of nasal cavity (nasal) V = activation of pharyngeal cavity (voice)
It appears that the interpretation of the substricture features varies some- what depending on what the (head) stricture is. In (13) I provide the specific interpretations for the four primary strictures with their subcategories.
(13)
stricture substricture resulting segment type C (stop) C (lateral) → lateral affricate Cv (strident) → affricate Vc (nasal) → prenasalized stop V (voice) → voiced stop Cv (fric) C (lateral) → lateral fricative Cv (strident) → strident fricative Vc (nasal) → prenasalized affricate V (voice) → voiced fricative Vc (son) C (lateral) → lateral liquid Cv (rhotic) → rhotic liquid Vc (nasal) → nasal consonant V (voice) → approximant? V (vowel) C (lateral) → tap/flap? Cv (rhotic) → rhotic vowel Vc (nasal) → nasalized vowel V (voice) → ATR/tense?
Features segments and syllables in Radical CV Phonology 99
In a number of cases the resulting representations come close to the rep- resentations that Anderson & Ewen (1987) propose in terms of their phona- tory gesture. A specific problem concerns the fact that voicing can be attributed to sonorant consonants and vowels, suggesting a phonemic distinction that does not seem to exist. (So-called voiceless sonorants and vowels will be interpreted as being aspirated — see below.) Vowels in general seem sensitive only to the nasal attribute. In §2.2 below I will consider the possibility of using the structures in (13) for complex onsets and complex nuclei, suggesting that such complex syllabic constituents are phonologi- cally identical to complex segments which are claimed to occupy non- complex syllabic positions. This may fill some of the gaps for the V-headed structures. A complement V, for example, could be used to represent the tense-lax opposition in sonorants.
1.3.2. Laryngeal categories
In the preceding section we discussed distinctions involving supralaryngeal stricture and concomitant activation of certain cavities. We now turn to the distinctions involving laryngeal activity, which play a role in phonation and tone. Some of the basic ideas that I will explore in this section go back to Halle & Stevens (1971), who propose that the features for phonation, tone and airstream are, at least partly, the same. Within feature geometry, correla- tions between tone and phonation have been explored in Yip (1989; 1993) and, more explicitly in Bao (1991) and Duanmu (1991). The idea behind seeing tone and phonation exponents as complemen- tary is that it can perhaps be maintained that the tonal interpretation arises in nuclear position, while phonation interpretations arise in onset position. This issue needs further investigation. RcvP proposes the categorization in (14) for tonal and phonation cate- gories. In both cases I simply reconstruct well-known categorizations. To make this clear I have again added traditional feature labels to the category distinctions (cf. especially Yip, 1980) and van der Hulst & Snider, 1993).
10 0 Harry van der Hulst
(14) a. Tone
C V high register low register + upper – upper
C V C V stiff slack stiff slack + raised – raised + raised – raised high high-mid low-mid low b. Phonation
C V constricted spread
C V C V – voice + voice – voice + voice glottal creaky aspirated breathy
The two C-headed phonation features presumably have a wide range of phonetic interpretations.
(15) C = ejective, glottalized, voiceless implosive, voiceless Cv = (voiced) implosive, creaky voice, laryngealized, preglottalized
I assume that glottalized obstruents and ejectives fall into the same phonological category (cf. Kenstowicz, 1994). This follows from the survey in Ladefoged & Maddieson (1995: 369), which tells us that these phonation types never contrast in a single language. Lombardi (1991; 1995) draws the same conclusion. To regard implosivity (like ejectivity) as a phonation type is desirable, especially if we follow the proposal in Sagey (1986) to regard clicks as com- plex segments. These proposals, then, make it unnecessary to assume separate “airstream features”. To eliminate such a category was also the goal of Halle & Stevens (1971), and the present proposal shares some char- acteristics with their feature system. I assume, then, that there are no air- stream features like [± ingressive] or [± velaric suction]. Cv represents the more sonorous type of glottalization which may also take various forms, as indicated. Following Ladefoged (1971; 1973), Lombardi (1991; 1995) proposes placing implosives in the same phonological category as consonants which are preglottalized or laryngealized (cf. Greenberg, 1970). I follow this idea, including creaky voice. In languages that have a
Features segments and syllables in Radical CV Phonology 101 contrast between voiceless and voiced implosives, I assume that the former category has C phonation, while the second has Cv. V is taken to represent aspiration. So-called “voiceless” sonorants are interpreted as being aspirated, i.e. as having the same phonological prop- erty as aspirated obstruents. This claim is also made in Anderson & Ewen (1987), who use |O| (aspiration) to represent voiceless sonorants. Cho (1991a,b) and Lombardi (1991; 1995) make the same claim in different models and provide a number of relevant examples.
1.3.3. Place categories
Place features within RcvP were treated in van der Hulst (1994b). The pri- mary categories in the place domain are those in (16) (cf. (6b)).
(16) Place
C V C-place V-place colours aperture
C V C V coronal labial high low
In the place class as a whole, RcvP postulates three subclasses. To a large extent we reconstruct traditional and widely accepted distinctions in this case, too. I repeat in (17) the full set of place features with its geometrical structure.
(17) PLACE
Primary place
secondary articulation major place subplace C palatalized coronal / I front Cv labialized labial / U round Vc dorsalized high laminal V pharyngealized low / A posterior
Note that the major places contain one option that most versions of so- called IUA-theories (Dependency Phonology (Anderson & Ewen, 1987), Government Phonology (Kaye, Lowenstamm & Vergnaud, 1985), Particle Phonology (Schane, 1984)) have excluded, namely high.
102 Harry van der Hulst
In (18) I spell out the structures that receive a primary place interpreta- tion. In this case, we only need the complement categories that have an opposite category type as their head, i.e. front and round apply to V-headed place, while laminal and posterior apply to C-headed place. Of course, it is always better if a proposed system is fully instantiated, but at least the observed restriction is not random. Also, we observe that labials are not subcategorized. This is presumably because the subplace features Vc and V refer to tongue body positions.
(18)
stricture substricture C (coronal) C (…) Cv (…) Vc (laminal) V (posterior) Cv (labial) Vc (high V) C (front) Cv (round) Vc (…) V (…) V (low V) C (front) Cv (round) Vc (…) V (…)
The secondary articulations are relevant to both vowel and consonantal place. This implies, for example, that we can make a distinction between “round” and labialized vowels, which has been argued for by Clements (1992) in order to represent inrounded vs. outrounded vowels, as in Norwegian. The distinction between the secondary articulation dorsalization and pharyngealization has been called into question. This may be related to the marginal status of the Vc feature. The distinction between subplace features and secondary articulations produces a large set of places. It could be argued, given the gaps noted, that the subplace class expresses both place subcategories and secondary articulations. This matter needs further investigation.
1.3.4. Cooccurrence restrictions
When considering the full set of features that RcvP allows, it would seem that the theory specifies a set of possible segments that is far too large. It
Features segments and syllables in Radical CV Phonology 103 seems obvious that RcvP cannot do without cooccurrence restrictions. In the preceding section I mentioned one type of restriction holding for place features: head and complement features must be opposite with respect to headedness. To constrain the inventory of possible segments further, I postulate an agreement relation between the manner and the place class.
(19) Manner → Place V V-headed others C-headed
Thus vowels only take V-headed place structures, while all manners that include a C element (as head or dependent) take C-headed places. The restrictions mentioned here are presumably universal. We will also need language-specific restrictions which will tell us precisely which seg- ment set a language has.
2. Syllable structure in RcvP
2.1. The segment–syllable connection
In this section I show how an RcvP treatment of syllable structure can establish a direct and formally non-arbitrary relation between the labelling of syllabic terminal nodes and the representation of manner (specifically stricture) properties of segments (cf. van der Hulst, 1996). Starting from the syllabic organization in (20b), (20a) seems the obvious CV coding of syllable structure.
(20) a. V b. Syllable C V Onset Rhyme
C V V C x x x x
I will now show how we can derive information from the syllabic CV-labelling regarding the types of segments that ideally appear in each syllabic position. If any relation exists at all, we expect that the onset head favours C-type segments, i.e. segments with a simple C-manner (stops), whereas rhyme heads favour V-manner (vowels). This is obviously and uncontroversially the case. What about the two dependent positions? We know that sonorant consonants (liquids in particular) are preferred onset
104 Harry van der Hulst dependents. This suggests that we read the syllabic label “bottom up”, namely: V (terminal label) + C (onset label) = Vc.
(21) CV-labelling of syllabic terminals V C V
C V V C M M M M C cVc V Cv
This means that we predict (22) to be true.
(22) a. Stops are ideal onset heads. b. Sonorant consonants are ideal onset dependents. c. Vowels are ideal rhyme heads. d. Fricatives are ideal rhyme dependents.
These predictions seem correct except for (22d). A solution to this problem may lie in allowing a nucleus constituent. We can accommodate a nucleus by assuming that the rhyme has a complement–head–specifier structure, which is exactly what we expect given the generality of this particular con- stituent type. The specifier position can be labelled “coda”.
(23) V
V C V
C V V C C M M M M M C cVc V cVc vCv
To arrive at the desired manner categories, bottom-up construction of the category labels has to reckon with the principle in (24) (cf. Anderson & Jones, 1974; Anderson & Ewen, 1987).
(24) Head Switch Principle (HSP) Ab + B = Ba (i.e. Cv + V = Vc)
This principle gives us sonorants in the dependent nuclear position, which is what we want, of course. We predict that Vc sonorants are optimal nuclear dependents. With the exclusion of the second half of long vowels, this is
Features segments and syllables in Radical CV Phonology 105 probably correct. We now predict that fricatives are preferred over stops in the ‘coda’ position. There is some evidence for the claim that fricatives (and, more generally, obstruents) are preferred rhyme specifiers. Fikkert (1994) shows that Dutch children have fricative codas in their first CVC words. Her analysis is that at this stage the nucleus is not yet a branching constituent. This finding receives a natural interpretation within the current proposal. I follow Government Phonology in ruling out syllables that have both a branching nucleus and a branching rhyme (cf. Kaye, Lowenstamm & Vergnaud, 1990). The relevant principle may be that a head cannot have two dependents on the same side. We might expect that the onset also has a complement–head–specifier structure, which according to the bottom-up algorithm and principle (17) has fricatives as preferred complement.
(25) V C V C V
V C V V C C M M M M M M vCv C cVc V cVc vCv s t r o 9 k ‘trunk’
As is well known, many languages allow an ‘extra’ fricative syllable-initially (usually an /s/, the optimal fricative). Therefore the structure in (25) gives us exactly what we want.
2.2. Integrating syllabic structure and manner
A further development of the model outlined here (first suggested to me by Marc Oostendorp) would be to “lift” the manner properties out of the segmental structure and attribute them directly to the syllable terminals. In this case we would have a different syllable structure for /pa/ and /fa/, for example. In the former case the onset head would be labelled with bare C, whereas in the latter case this label would be Cv. The properties of syllabic terminals would then be partly determined by the higher syllabic organiza- tion and partly free, i.e. subject to choices with phonemic consequences. The advantage of this approach is that we do not have to duplicate informa- tion in the form of (near) identical specification of syllabic terminals and manner properties. The characterization of the syllabic terminals would directly determine the manner of articulation of the place-of-articulation
106 Harry van der Hulst properties that they dominate. Thus, the dependence of place of articula- tion on manner would also be directly encoded. Similar ideas have been developed by Chris Golston (cf. Golston & van der Hulst, to appear) and in work in progress by Sean Jensen (in Government Phonology), who also replaces subsegmental manner elements by elaborations of the syllabic organization. This approach entails the view that syllabic organization is “underlying” or inherent (cf. Golston & van der Hulst, to appear). An im- mediate advantage of underlying syllable structure is that we can suppress information about linearization, as was first pointed out in Anderson (1987). Golston & van der Hulst (to appear) discuss internal and external, psycho- linguistic evidence supporting underlying presence of syllable structure. Here I wish to briefly consider the consequences that this claim has for the representation of stricture properties. In §2.1 we noted that a model in which syllable structure and stricture properties are regarded as independ- ent involves a certain amount of redundancy: the syllabic structure encodes the same kinds of broad categories that we encode in terms of stricture properties. Instead of coding, for example, a /p/ as [stricture: C, place: Cv] and subjoining this structure to a syllabic head position which is coded C, as in (26a). We could also subjoin (and hence associate) the place feature directly to the syllable node, as in (26b).
(26) a. C (onset) b. C (onset, stop) M vCv (labial)
C Cv (stop) (labial)
The distinction between stops and fricatives can be made by further specifying the syllabic node with a v subscript. In (27), the syllables /pa/ and /fa/ are represented.
(27) a. V (syll) b. V (syll)
(onset, C V (vowel) (onset, vCv V (vowel) (stop) M M (fric) M M Cv V vCv V (lab) (low) (lab) (low)
We arrive at a situation in which the only real features are place features. The rest is syllable structure. Chris Golston aptly refers to this view as: “stricture is structure”. I wish to end this section with a suggestion that will allow a further com- pression of the system. Lowenstamm (1996) proposed that syllables in all
Features segments and syllables in Radical CV Phonology 107 languages are strictly CV. In that framework, complex onsets and nuclei require a disyllabic representation. Within Government Phonology, con- vincing cases have been presented in which such disyllabic representations are indeed called for, for example to make a distinction between ‘true’ onsets (usually obstruent–liquid sequences) and ‘fake’ onsets (which involve two onsets separated by an empty nucleus). Lowenstamm’s theory no longer allows this distinction, unless we choose to represent ‘true branching onsets’ as arising from specific feature combinations which associate to a single syllabic C-position. A proposal along these lines would associate the syllabic C and V position with a manner node which, if complex, may spell out as a linearized phonetic event.
(28) Syll Manner Manner
Phon O Tone O
Strict Substr Strict Substr M M Place Place … … To simplify matters, I have assumed here that phonation and tone are represented as complementary (cf. §1.3.2). A “branching” onset /pl/ would be represented as in (29).
(29) C (onset)
(stop) C C (lateral) M vCv (labial)
The dependent C encodes laterality. This approach entails that complex onsets like /pl/ and lateral stops do not differ in terms of their representa- tions, because in laterals, too, we combine stophood with laterality. A consequence that emerges is that the distinction between complex onsets and complex segments collapses from a phonological point of view. This in fact has also been proposed in Hirst (1985; 1995), who introduces the term ‘macro-segment’. See also Golston & van der Hulst (to appear) for a tenta- tive set of representations along the lines in (29). This line would require a reconsideration of our treatment of voice as an exponent of a substricture class feature. To represent the contrast between /pl/ and /bl/, we would have to locate voice in the phonation class (as the exponent of V), since the
108 Harry van der Hulst substricture slot in (29) is taken by the category for lateral. These and other consequences will be explored elsewhere.
3. Concluding remarks
In many respects, RcvP, which takes most of its inspiration from Depend- ency Phonology (Anderson & Ewen, 1987), is quite similar to models within the field of feature geometry. The specific notation that I have used throughout this article may conceal this fact. As I pointed out in §1, the grouping of features into laryngeal, stricture and place classes can also be found in Clements (1985). The grouping of stricture features under one node has since been abandoned by most phonologists, but recently the grouping of at least the manner features for sonorants has been reintroduced (cf. Rice & Avery, 1989; Piggott, 1990). The point that manner and major class distinc- tions are expressed hand in hand comes directly from Anderson & Ewen (1987). Geometrical models do not unify these distinctions, although it has been acknowledged that major class features and (some of the) manner features are “head features” (McCarthy, 1988). In the model proposed here, such ideas are incorporated and generalized. My approach to the laryngeal gesture perhaps finds its closest analogues in current geometrical models, even though there is no other proposal which is as much in the spirit of Halle & Stevens (1971) as mine, which pushes their program even further. Finally, the proposal for the place class pushes to the extreme that one feature set is shared by vowels and consonants. In my approach, however, the set does not have to be duplicated for vowels and consonants. My consistent use of the CV notation allows us to develop a unified theory of syllabic and segmental structure. One could narrow down the goal of a phonological theory to stipulating a list of segmental and syllabic categories and the various compatibility and solidarity effects that hold among these categories. One could, however, also ask, why we have the set of phonological categories that we seem to need and why the compatibility and solidarity effects hold. Some assume, right from the start, that all the answers to these questions can be found by studying phonetics. RcvP is an attempt to draw at least some of the ex- planation from the realm of phonology.
Acknowledgement
Thanks to Rob Goedemans for comments on an earlier version of this article.
Features segments and syllables in Radical CV Phonology 109
References
Anderson, J. (1987). The limits of linearity. Explorations in dependency phonology, edited by J. Anderson & J. Durand, 199–220. Dordrecht: Foris. Anderson, J.M. & C.J. Ewen (1987). Principles of dependency phonology. Cambridge: Cambridge University Press. Anderson, J. & C. Jones (1974). Phonological structure and the history of English. Amsterdam: North- Holland. Anderson, S. (1985). Phonology in the 20th century. Chicago: Chicago Press. Bao, Z. (1991). On the nature of tone. Doctoral dissertation, MIT. Carr, P. & N. Burton-Roberts (1996). On speech and natural language. Newcastle and Durham Working Papers in Linguistics 4, 13–40. Cho, Y.Y. (1991a). Voicing is not relevant for sonority. Proceedings of the Berkeley Linguistic Society 17 . Cho, Y.Y. (1991b). ‘Voiceless’ sonorants are aspirates. Paper presented at the annual LSA meeting. Chomsky, N. & M. Halle (1968). The sound pattern of English. New York: Harper & Row. Clements, G.N. (1985). The geometry of phonological features. Phonology Yearbook 2, 225–252. Clements, G.N. (1992). Lieu d’articulation des consonnes et des voyelles: une théorie unifiée. L’architecture des représentations phonologiques, edited by B. Laks & A. Rialland, 101–147. Paris: Editions du CNRS. Duanmu, S. (1991). A formal study of syllable, tone, stress and domain in Chinese languages. Doctoral dissertation, MIT. Dikken, M. den & H.G. van der Hulst (1988). Segmental hierarchitecture. Features, segmental structure and harmony processes, Part 1, edited by H.G. van der Hulst & N.S.H. Smith, 1–59. Dordrecht: Foris publications. Fikkert, P. (1994). On the acquisition of prosodic structure. HIL dissertations #6. The Hague: Holland Academic Graphics. Fischer-Jørgensen, E. (1975). Trends in phonological theory. Copenhagen: Akademisk Forlag. Golston, C. & H.G. van der Hulst (to appear). Stricture and structure. Proceedings of the Tilburg conference on Derivational Residues, edited by B. Hermans & M. van Oostendorp, Dordrecht: Reidel. Greenberg, J.H. (1970). Some generalizations concerning glottalic consonants, especially implosives. International Journal of American Linguistics 36, 123–145. Halle, M. & K. Stevens (1971). A note on laryngeal features. Quarterly Progress Report 101, 198 –213. Research Laboratory of Electronics, MIT. Hirst, D. (1985). Linearization and the single segment hypothesis. Grammatical representations, edited by J. Guéron & H.G. Obenhauer, 87–100. Dordrecht: Foris. Hirst, D. (1995). Macro segments. Paper presented at “Current trends in phonology: Models and methods”, June 19–21, 1995, Royaumont. Hulst, H.G. van der (1988). The geometry of vocalic features. Features, segmental structure and harmony processes, edited by H.G. van der Hulst & N. Smith, 77–125. Dordrecht: Foris. Hulst, H.G. van der (1989). Atoms of segmental structure: components, gestures and depend- ency. Phonology 6, 253–284. Hulst, H.G. van der (1990/1991). The molecular structure of phonological segments. Ms., Leiden University. [The book of segments]. Hulst, H.G. van der (1993a). Les atomes de la structure segmentale: composants, gestes, et pendances. L’architecture des représentations phonologiques, edited by B. Laks & A. Rialland, 255–290. Paris: Editions du CNRS. Hulst, H.G. van der (1993b). Units in the analysis of signs. Phonology 10, 209–242. Hulst, H.G. van der (1994a). An introduction to Radical CV Phonology. SKY 1994: 1994 Yearbook of the Linguistic Association of Finland, edited by S. Shore, S. & M. Vilkuna, 23–56. Helsinki.
110 Harry van der Hulst
Hulst, H.G. van der (1994b). Radical CV Phonology: The locational gesture. University College London Working Papers in Linguistics 6, 439–478. Hulst, H.G. van der (1995a). Radical CV Phonology: The categorial gesture. Frontiers in Phonology, edited by F. Katamba & J. Durand, 80–116. London: Longman. Hulst, H.G. van der (1995b). Head-dependency relations in the representation of signs. Sign language research 1994, edited by H. Bos & T. Schermer, 11–38. Hamburg: Signum. Hulst, H.G. van der (1996). The segment–syllable connection. Current trends in phonology: models and methods, edited by J. Durand & B. Laks, 333–362. Paris: CNRS. Hulst, H.G. van der & K.L. Snider (1993). Issues in the representation of tonal register. The phonology of tone: The representation of tonal register, edited by H.G. van der Hulst & K.L. Snider, 1–27. Berlin: Mouton de Gruyter. Humbert, H. (1995). Phonological segments: Their structure and behaviour. HIL dissertations #12. The Hague: Holland Academic Graphics. Jakobson, R. (1941/1968). Kindersprache, Aphasie und allgemeine Lautgesetze. Translated by A. Keiler. Child language, aphasia and phonological universals. The Hague: Mouton Jakobson, R., C.G.M. Fant & M. Halle (1952). Preliminaries to speech analysis: The distinctive features and their correlates. Second Edition. Cambridge, Massachusetts: MIT Press Kaye, J.D., J. Lowenstamm & J.-R. Vergnaud (1985). The internal structure of phonological representations: A theory of charm and government. Phonology Yearbook 2, 305–328. Kaye, J., J. Lowenstamm & J.R. Vergnaud (1990). Constituent structure and government on phonology. Phonology 7, 193–232. Kenstowicz, M. (1994). Phonology in generative grammar. Cambridge & Oxford: Blackwell. Ladefoged, P. (1971). Preliminaries to linguistic phonetics. Chicago: University of Chicago Press. Ladefoged, P. (1973). The features of the larynx. Journal of Phonetics 1, 73–83. Ladefoged, P. & I. Maddieson (1996). The sounds of the world’s languages. Cambridge & Oxford: Blackwell. Lahiri, A. & V. Evers (1991). Palatalization and coronality. The special status of coronal. Phonology and Phonetics 2, edited by C. Paradis & J. Prunet, 79–98. Academic Press. Liljencrants, J. & B. Lindblom (1972). Numerical simulation of vowel quality systems: The role of perceptual contrast. Language 48, 839–862. Lombardi, L. (1991). Laryngeal features and laryngeal neutralization. PhD dissertation, Univer- sity of Massachusetts, Amherst. Lombardi, L. (1995). Laryngeal features and privativity. The Linguistic Review 12, 35–60. Lowenstamm, J. (1996). CV as the only syllable type. Current trends in phonology: models and methods, edited by J. Durand & B. Laks, 333–362. Paris: CNRS. McCarthy, J.J. (1988). Feature geometry and dependency: A review. Phonetica 43, 84–108. Ogden, R.A. (1993). What Firthian prosodic analysis has to say to us. In: Working papers of the department of Cognitive Science, Edinburgh. Piggott, G.L. (1990). The representation of sonorant features. Ms, McGill University. Rennison, J. (1986). On tridirectional feature systems for vowels. Dependency and non-linear phonology, edited by J. Durand, 281–303. London: Croom Helm. Rice, K.D. & P. Avery (1989). On the interaction between sonorancy and voicing. Toronto Working Papers in Linguistics 10, 65–92. Sagey, E.C. (1986). The representation of features and relations in non-linear phonology. PhD dissertation, MIT. Schane, S. (1984). The fundamentals of particle phonology. Phonology Yearbook 1, 129–155. Stevens, K.N. (1972). The quantal nature of speech: Evidence from articulatory-acoustic data. Human communication: a unified view, edited by E.E. David & P.B. Denes, 51–66. New York: McGraw Hill. Weijer, J. van de (1994). Segmental structure and complex segments. HIL dissertations #7. The Hague: Holland Academic Graphics.
Features segments and syllables in Radical CV Phonology 111
Yip, M. (1980). The tonal phonology of Chinese. PhD dissertation, MIT.. Yip, M. (1989). Contour tones. Phonology 6, 149–174. Yip, M. (1993). Tonal register in East Asian languages. The representation of tonal register, edited by H.G. van der Hulst & K.L. Snider, 245–268. Berlin: Mouton de Gruyter.
Syllable structure constraints in exceptions
Iwona Kraska-Szlenk
Abstract
This paper proposes a formal constraint-based analysis of phonological exceptions. It concentrates on one type of exception which involves exceptionality of whole morphemes, as opposed to exceptionality of specific segments. Such exceptional morphemes can be divided into two sub-types: those which violate some of the constraints normally satisfied in a given language, and those which are apparently subject to more constraints than regular morphemes. Loan-words constitute typical
The Hague: Thesus. 113–131. Thesus.The Hague: 113–131. examples of the first kind and grammatical morphemes are commonly of . the second kind. The linguistic data are drawn from a variety of lan- guages, including English, Polish, Somali, Swahili and Turkish. 0. Syllables !?
1. Theoretical framework
Phonologica 1996: Phonologica 1996: As my theoretical framework, I adopt the version of Optimality Theory (OT) outlined in Russell (1995). An important innovation of this model, with respect to the original OT model (Prince & Smolensky, 1993), is the introduc- tion of the so-called MORPHEMIC CONSTRAINTS, which replace the notion of “input” (“underlying representation”) and eliminate the FAITHFULNESS CONSTRAINTS of the Parse type. Since there is no input, the phonological representation is reduced to only one level, i.e. that of the output (“surface representation”)1 Morphemic constraints are constraints which impose a set of require- ments linking semantic representations of morphemes with particular phonological (possibly also morphological and syntactic) properties. The satisfaction of all morphemic constraints associated with one specific
1 By being strictly a one-level theory, Russell’s approach differs from McCarthy & Prince’s (1995) “correspondence” approach (to which it is very similar in other respects). John R. Rennison & Klaus Kühnhammer, eds. (1999). eds. R. Rennison & Klaus Kühnhammer, John (1999).
114 Iwona Kraska-Szlenk morpheme would be nothing but a mental representation of it which corre- sponds to the notion of input. However, unlike the input in two-level theo- ries, morphemic constraints are not in any sense prior to the output; they simply occupy some position(s) in the constraint hierarchy. An interesting repercussion of morphemes-as-constraints approach is that it makes the output less dependent on its mental representation. In traditional terms, different underlying representations, from the most abstract (maximally underspecified) to the most surface-like (maximally specified), predict the same output as optimal; I return to this issue later in this section. In the case of regular phonology, morphemic (MORPH) constraints occupy the intermediate position between two blocks of phonological (PHON) constraints, as shown in (1) below. The dominant Phon-I constraints are all constraints which are active in a given language, which means that their effects are visible in the form of phonological alternations. Phon-I con- straints filter the requirements of Morph constraints so that the optimal output may only have the structure allowed by Phon-I. Within the dominated Phon-II constraints, two types can be distinguished. Type one consists of *Structure constraints which rule out all excessive structure (features, moras, feet etc.), i.e. the structure not required by higher Phon-I and Morph constraints. Type two includes all the constraints of the universal repertoire which do not play any role in the phonological system of a given language, and hence are “invisible” and may not cause phonological alternations.
(1) Phon-I Constraints ↓ Morph Constraints ↓ Phon-II Constraints
The relatively low ranking of Morph constraints in the hierarchy has some consequences for the issue of underspecification and redundancy in pho- nology. In this model, it is unimportant whether the Morph constraints express only the information that is unpredictable from the higher Phon-I constraints, or also contain some redundant information. If, for example, Phon-I constraints guarantee that all nasals will be voiced, all stops voice- less, there will be an initial foot in every stem etc., these structural require- ments might but need not be repeated by the appropriate Morph con- straints. Therefore, the mental representation expressed by Morph con- straints may vary accordingly across a certain continuum, from the least to the most specified, as shown in (2), without any significance for the optimal output selection. Although the same claim can be expressed in the two-level OT model by altering underlying input forms (e.g. Itô et al., 1995), it has a more natural expression in the present model since the “unimportance”
Syllable structure constraints in exceptions 115 appears low in the constraint hierarchy, just as in any other case of “un- importance” in OT.2
(2) Phon-I: Nas ⇒ Voiced Stop ⇒ Voiceless Align{stem,L,Ft,L} ↓ ↓ ↓ Morph: naTa ~ naTa ~ (nata)
nas nas, + vo nas, + vo; – vo Optimal output: (nata)
The “morphemes-as-constraints” approach provokes the natural ques- tion: If morphemes are expressed as constraints, could different morpheme- specific constraints be differently ranked in the hierarchy with respect to phonological constraints, predicting that phonology would not equally affect the word for ‘dog’ and the word for ‘cat’, etc.? The higher the position of particular morphemic constraints, the less phonology would filter them, and conversely, the lower the position of morphemic constraints, the more phonology would show its impact on the mental representations of given morphemes. Such complex interaction between phonological and mor- phemic constraints would appear as schematized in (3a), with a more con- crete example of a hypothetical language hierarchy in (3b). At first glance, the type of constraint interaction illustrated in (3) seems wrong, because every language has ‘regular’ phonology which affects different words and morphemes in the same fashion. However, it is also true that every language seems to have exceptions which behave exactly as predicted by the constraint schema in (3a). Such exceptions will be dis- cussed momentarily. Anticipating that they exist, the more accurate con- straint schema appears as in (4), in which Morph-Reg stands for morphemic
2 The morphemic constraints in (2) have an abbreviated representational formulation. Strictly speaking, I assume after Russell (1995) that Morph constraints include statements as: “must have such and such segment (feature etc.)”. Notice a potential problem with this formulation of Morph constraints. While epenthesis follows naturally under the ranking: Morph >> *Str, it is not immediately obvious how to account for phonologically conditioned deletion in lan- guages which prefer deletion (of, say, a coda consonant) over epenthesis. The ranking Morph >> *Str must be maintained for such languages, too, to guarantee that morphemes emerge on the surface at all. But this ranking predicts epenthesis (violation of the lower *Str), not deletion (violation of the higher Morph), as more harmonic. I leave it for further research whether this problem should be solved by reformulating the Morph constraints (so that they require particular structure exclusively) or in some other way. In this paper, I adopt a tentative analysis which assumes that epenthesis is prohibited by the constraint against morphologically unsponsored segments, Mseg (McCarthy, 1994; see also McCarthy & Prince, 1994; Kraska-Szlenk, 1995).
116 Iwona Kraska-Szlenk constraints referring to a large class of lexical items considered as phono- logically regular, and all instances of Morph-Exc represent what is consid- ered as exceptions.
(3) a. Phon-1 b. No-Coda ↓ ↓ Morph-a Morph-dog ↓ ↓ Phon-2 Onset ↓ ↓ Morph-b Morph-cat ↓ ↓ Phon-3 Ft-Bin ↓ ↓ Morph-c Morph-man …etc. …etc.
Notice that in this model there is no formal distinction between ‘regular- ity’ and ‘exceptionality’. It is not the case that lexical items that are phono- logically regular share some formal property differentiating them from exceptions; or that exceptions have a special formal property. What distin- guishes the two classes is only the criterion of size: Morph-Reg class typi- cally comprises a large portion of the lexicon while Morph-Exc classes may be limited to a few or even single lexical items. If it happens that different Morph classes are comparable in size and more evenly intermingled with phonological constraints, the decision as to which Morph constraints should be taken as a reference point for ‘regularity’ is not that obvious and more arbitrary (cf. the discussion in §4)
(4) Phon-1 ↓ Morph-Exc-a ↓ Phon-2 ↓ Morph-Exc-b ↓ …… ↓ Morph-Reg ↓ Phon-k ↓ Morph-Exc-n ↓ ……
Syllable structure constraints in exceptions 117
In the following section, I will discuss the problem of loan-words which constitute the most typical examples of Morph-Exc constraints ranked above the core Morph-Reg constraints (cf. Itô & Mester, 1995). In §3, gram- matical morphemes will exemplify exceptionally low ranked Morph con- straints (cf. McCarthy & Prince, 1995). The final section (§4) contains exam- ples of a less typical interaction between Phon and Morph constraints.
2. Disobedience to ‘normal’ phonology
The most common case of exceptional morphemes which violate the regu- lar phonology of a language are loan-words (cf. Itô & Mester, 1995). It is not unusual that loan-words, even if nativized in their inflection, repeatedly ignore a number of phonotactic restrictions imposed on the native mor- phemes. For example in Polish, sequences such as ke, ge, ti, si, zi (and others) are not found except in a limited number of words of foreign origin (e.g. kelner ‘waiter’, gen ‘gene’, tik ‘tic’, sizal ‘sisal’, dinar ‘dinar’). This kind of excep- tionality is explained by placing the ‘foreign’ morphemic constraints (Morph- Foreign) above the appropriate Phon constraints dominating Morph-Reg constraints, as shown in (5) below.
(5) Morph-kelner, gen, tik etc. ↓ *ke, *ge, *ti etc. ↓ Morph-Reg
In the process of nativization, loan-words may accommodate more and more to the regular phonology of a given language, which is schematized as a downward movement of appropriate Morph-Foreign constraints, below some of the phonological constraints, as in (6a). By the time they reach Morph-Reg constraints (6c), their ‘foreignness’ disappears. Variation be- tween the old ‘foreign’ and the new nativized form, which appears as a characteristic feature of the intermediate stage in nativization, is repre- sented as variable ranking between the Morph-Foreign and Phon con- straints, as shown in (6b).
(6) a. Morph-For b. Morph-For, Phon c. Phon ↓ ↓ ↓ Phon Morph-Reg Morph-For / Reg ↓ Morph-Reg
118 Iwona Kraska-Szlenk
Various stages of the process of nativization sketched in (6) can be de- tected in one language at the same point of time. For example in the con- temporary standard Polish, the aforementioned Morph-tik occurs at the very top of the hierarchy since [tik] is the only possible pronunciation; Morph- plastik is unranked with respect to the *ti constraint since the pronunciation of ‘plastic’ varies from the ‘foreign’ [plastik] to the more nativized [plasti- k]; finally, a loan-word as tylda ‘tilde’ (from Spanish tilde) may have only the nativized pronunciation [ti-lda], hence Morph-tylda is dominated by *ti. The adaptation of foreign loan-words into the native vocabulary con- straints sometimes happens so rapidly that we can witness this historical change take place even within months. I recall when the word for ‘computer’ was first borrowed from English into Polish, one could occasionally hear [kompyuter] beside the more nativized [komputer]. After a very short time (probably less than a year), the former pronunciation was completely elimi- nated: the Morph-kompyuter moved very quickly below a constraint against the pyu sequence. Sometimes, foreign words may have quite an opposite impact on the target language when, instead of being nativized, they introduce a new phoneme or phonemic contrast which infiltrates into the native vocabulary. In Swahili, the rich early borrowing from Arabic introduced a phonemic distinction between [l] and [r] (e.g. kalamu ‘pen’ vs. karamu ‘reception’). Al- though still the [r] sound is found mostly in loan-words and the [l] in the native Bantu morphemes, there are a number of Bantu roots in which [l] and [r] freely alternate (e.g. telemka ∼ teremka ‘descend’, loga ∼ roga ‘practice witch- craft’) and a few Bantu roots with the fixed [r] (e.g. ruka, ‘jump; fly’, maporomoko ‘waterfalls’). Another, and perhaps more interesting, example of such foreign ‘in- vasion’ on the native phonological system comes from the Southern Bantu languages which borrowed extremely unusual and difficult to produce click sounds from the neighbouring Khoisan languages (cf. Dogil, Mayer & Roux, this volume). As the [r] in Swahili, clicks penetrated into some original Bantu roots, as in the following examples from Zulu and Xhosa (in which “c” stands for dental click): cima < Proto-Bantu *-tima.3 To reflect the phonemic changes of the above kind in the constraint sys- tem, one might assume that in this case it is an appropriate phonological constraint that ‘moves’ down below the Morph-Reg constraints, as shown in two steps in (7).
3 I am grateful to Simon Donnelley (p.c.) for providing this example for me.
Syllable structure constraints in exceptions 119
(7) a. Morph-For b. Morph-For, Morph-Reg ↓ ↓ Sw. *r; S. Bantu *clicks Sw. *r; S. Bantu *clicks ↓ Morph-Reg
The high ranking of exceptional Morph constraints is a powerful mecha- nism which may account for exceptions involving various phonological aspects. In addition to disrespect for phonotactic constraints of the seg- mental type, as those discussed thus far, disobedience to vowel harmony is another cross-linguistically common ‘exceptional’ behaviour. Thus, lan- guages in which roots harmonize in some feature (back vs. front, + ATR vs. – ATR etc.) typically allow for a handful of disharmonic roots, such as the Turkish examples in (8) which combine vowels of front and back series (Lewis, 1986: 17; see also §4).
(8) Turkish Morph-dahi ‘also’, elma ‘apple’, mikrop ‘microbe’ etc. ↓ Vowel Harmony ↓ Morph-Reg
Syllable structure constraints, such as Onset (“all syllables must have onsets”, Prince & Smolensky, 1993) or *Coda (“syllables must not have codas”, Prince & Smolensky, 1993), may be given as yet another common example of phonological constraints which are exceptionally violated. For example, in Polish, onsetless syllables are generally prohibited morpheme- internally but are tolerated in loan-words, as shown in (9).
(9) Morph-oaza ‘oasis’, kakao ‘cocoa’ etc. ↓ Onset ↓ Morph-Reg
3. Obedience to more than ‘normal’ phonology
As pointed out by McCarthy & Prince (1995: 116ff.), grammatical morphemes tend to obey phonological constraints which are otherwise invisible as
120 Iwona Kraska-Szlenk active constraints in a given language.4 In the present model, cases of this sort arise when Morph-Exc constraint is ranked below some Phon con- straint(s) dominated by Morph-Reg constraints, as shown in (10).
(10) Morph-Reg. ↓ Phon ↓ Morph-Exc
Phonologically conditioned allomorphy provides abundant examples of the constraint schema in (10). English a∼an allomorphy receives a straight- forward account, if the crucial constraint in the intermediate position between the Morph-Reg and Morph-Exc constraints is that of *Coda. English tolerates coda consonants quite freely which motivates the ranking Morph- Reg >> *Coda. It is only one particular consonant, the n of the indefinite determiner which deletes, if predicted to be syllabified in coda. I assume that the morphemic constraints require that the indefinite determiner contain the sequence an (which corresponds to the traditional notion of /an/ being its UR). Ranking of Morph-an below the dominant *Coda predicts that the full form of this morpheme will appear only before a vowel initial word. This is the context where the n of the an may syllabify as onset since nV is a possible syllable in English. If the following word begins with a con- sonant, an attempt of avoiding the coda syllabification of the n by adjoining it to the following onset would result in the nC(C) onset which is ruled out by dominant general syllable structure constraints of English. The attempt of syllabifying the n in onset by introducing an epenthetic syllable peak is ruled out by the dominant Mseg constraint prohibiting epenthesis (see fn. 2). The contrast between the an and a ‘regular’ English word is pictured in (11). Idiosyncratic deletion of a coda consonant of a specific morpheme is cross-linguistically quite common. The data in (12) (from Kraska, 1992 after Saeed, 1985) show an example from Somali. The AUTOBENEFACTIVE suffix -at loses its consonant in word final (i.e. coda) position causing an apparent ‘irregularity’ of the singular imperative in (12a). The expected ‘regular’ form
4 The discussion of allomorphy in this section has been very much inspired by McCarthy & Prince’s (1994; 1995) analysis of phonologically unmarked patterns in reduplication. How- ever, in most of McCarthy & Prince’s examples reduplicants have invariant phonological shapes. Therefore, there is no empirical argument that, for example, a CV reduplicative pattern is indeed a product of constraints, such as Onset, *Coda etc., and does not result from a specific ‘template’. On the contrary, cases of allomorphy discussed here involve alternations which excludes a ‘templatic’ account and provides more direct arguments for the kind of “emergence of the unmarked” advocated by McCarthy & Prince.
Syllable structure constraints in exceptions 121 should contain a complete stem, i.e. it should end in a final d (derived from t by the general process of final voicing). Various other words in Somali surface with a final d: the behaviour of the “autobenefactive” suffix is purely idiosyncratic and may not be attributed to the general phonology (note that the raising of the /a/ to [o] in the singular of (12a) is an independently moti- vated process, see Kraska, 1992). The Somali data follow immediately under an analysis analogous to that of English in (11): Morph-Reg >> *Coda >> Morph-at. (The data in (12b), included for comparison, represent the ‘regular’ imperative with a zero suffix in the singular and -a suffix in the plural.)
(11) Syll-Str Mseg Morph-man *Coda Morph-an man * ma *n! ma.nI *! a.nV a.V *n! a.C *n! an.C *! a.nC *! a.nI.C *! (12) a. joogs-o (*joogs-ad) ‘stop SG.’ joogs-ad-a ‘stop PL.’ b. keen ‘bring SG.’ keen-a ‘bring PL.’
The alternation between the emergence and disappearance of a particu- lar morpheme’s consonant may be caused by a more complex configuration of Phon and Morph constraints. The following discussion of Swahili pro- vides an example. The basic inventory of possible syllables in Swahili includes (C)V, NCV (where N is a homorganic nasal) and a single syllabic nasal (in limited con- texts). The lack of closed syllables argues that *Coda is dominant in the constraint hierarchy. This is further motivated by the very quick adaptation of loan-words to the open syllable requirement, e.g. sampuli ‘sample’, supu ‘soup’, muziki ‘music’. Although there exist non-nativized ‘exceptions’ of the type discussed in the previous section (§2), for the purpose of the discus- sion here it is quite appropriate to treat *Coda as a non-violable constraint in Swahili (strictly speaking, such rare exceptions with coda consonants would be accountable by placing relevant Morph-Exc constraints above *Coda). On the other hand, the Onset constraint is frequently violated on the surface, e.g. endelea ‘continue-IMP.’, Uingereza ‘England’. For the further dis- cussion it is important to note that it is not unusual to find two or even three adjacent onsetless syllables, as in kuoana ‘to marry each other’, maua ‘flow- ers’, aoe ‘let him get married’, etc. I conclude that Onset is ranked low in the
122 Iwona Kraska-Szlenk hierarchy, specifically, below Morph constraints, since neither deletion nor consonantal epenthesis is ever used to resolve the problem of hiatus. The syllable structure conditioned allomorphy characterizes the so-called “extensions” — verbal suffixes which immediately follow the root (or another extension). As illustration, I will use the very productive “applied” extension. When suffixed to a consonant-final stem, the applied extension appears as the sole vowel [i] or [e]; the choice of the appropriate vowel is determined by the stem-controlled harmony to which I return later in this section. When suffixed to a vowel-final stem, the applied extension appears as [li] or [le], as shown in (13). (All examples in (13) and later are given with the final suffix -a.)
(13) a. pik-i-a ‘cook (APPL.)’ sem-e-a ‘say (APPL.)’ b. fungu-li-a ‘open (APPL.)’ ende-le-a ‘continue’ (lexicalized)
The l which appears in the (13b) forms may not be epenthetic: if it were, why don’t we find it in (13a) *pikila, *semela, or in mentioned earlier sequences of three onsetless vowels? Given the Swahili tolerance for onsetless syl- lables, the unattested forms as *funguia, *endeea should not qualify for con- sonantal epenthesis.5 A more plausible analysis is that the l of the applied extension deletes after the consonant-final stem because of the unviolable *Coda constraint (*piklia, *semlea etc.). Under this analysis, however, it is somewhat puzzling why the coda problem is solved by means of the l-deletion and not by applying quite common in Swahili vowel epenthesis (see later for examples). The answer must lie in the relatively low ranking of Morph-li, below Mseg, unlike in the case of other Morph constraints. The partial constraint hierarchy appears as in (14), where I include only the (13a) example; the (13b) forms are straightforward, they satisfy all the constraints of (14). The fact that the optimal form violates Onset is irrelevant; as argued earlier, Onset is quite ‘invisible’ in Swahili phonology — I assume that it is the lowest constraint in the hierarchy (not included in (14)).
(14) *Coda Morph-Reg Mseg Morph-li pik-i-a *! pik-li-a *! pi-li-a *k! pik-i-li-a *!
5 In some Swahili dialects, the hiatus is separated by the homorganic glide, e.g. pikiya, semeya. This further argues that the “l” is not a ‘normal’ epenthetic consonant in Swahili.
Syllable structure constraints in exceptions 123
The ranking Mseg >> Morph-li is crucial to exclude epenthesis, as shown by the last output in (14). The relatively low ranking of Mseg (below Morph- Reg) is motivated by the general tendency in Swahili to resolve the un- acceptable clusters by means of epenthesis. This is especially visible in loan-words in which epenthetic vowels are freely inserted whenever the source word does not comply with the Swahili syllable structure, as shown by the examples in (15); conversely, deletion of the segmental material is almost never used for that purpose.
(15) kitabu ‘book’ (Ar. kitaab) bahari ‘sea’ (Ar. bah- r) riziki ‘profit’ (Ar. rizq) sampuli ‘sample’ (Engl. sample) gilasi ‘glass’ (Engl. glass)
Within the native Swahili lexicon, epenthesis may be argued to be at work in the formation of proximate demonstratives “this” (“these”). As shown by the examples in (16), all members of this category are of the CV(C)V shape. The second syllable is that of the class marker required for the purpose of the grammatical agreement (see the rightmost forms in (16)). The first syl- lable starts with the h which I assume to be the only segmental requirement of the morphemic constraints associated with the meaning of “this” (“these”). I assume that the vowel of the first syllable is epenthetic and its features, ‘copied’ from the following syllable, result from the vowel harmony to which I will return momentarily.
(16) h + class marker class ‘it was / they were’ hawa 2 walikuwa haya 6 yalikuwa hapa 16 palikuwa hii 4/9 ilikuwa hiki 7 kilikuwa huku 15 / 18 kulikuwa huu 3/11 ulikuwa
When the class marker is consonant-initial, syllable structure constraints could be argued to trigger epenthesis, given the unacceptability of se- quences as *hwa, *hya etc. But this explanation cannot be maintained in the case of the vowel-initial markers of classes 4/9 and 3/11, which would then have their expected forms as *hi and *hu, respectively. Therefore, I assume that a true reason for epenthesis is that demonstratives form disyllabic words, a requirement visible elsewhere in Swahili phonology (e.g. disyllabic
124 Iwona Kraska-Szlenk
[m.tu] ‘person’, not monosyllabic *[mtu], but [mto.to] ‘child’, not *[m.to.to]). Formally, such a requirement follows from dominance of two constraints: Ft-Bin, which demands that feet be disyllabic, and Lx=Pw, which demands that lexical word be footed (cf. Prince & Smolensky, 1993; McCarthy & Prince, 1993, among others). In sum, examples, as the loan-words in (15) and the demonstratives in (16), argue that Mseg is a violable constraint in Swahili. Yet, it is not ranked low enough to trigger epenthesis in the case of the applied extension. The conclusion must be then that Morph-li is dominated by Mseg, as shown before in (14). A similar behaviour characterizes two other extensions in Swahili, such as, stative (l)ik/(l)ek and causative (l)ish/(l)esh, (l)iz/(l)ez, as shown by the exam- ples in (17).
(17) a. som-ek-a ‘be readable’ b. som-esh-a ‘teach’ fungu-lik-a ‘be openable’ fungu-lish-a ‘make open’
Notice that the consonant-initial allomorphs of the three extensions, ap- plied, stative and causative, all start with the same consonant l. Is it coinci- dental? Probably not. It is quite likely that the l of one extension was gener- alized onto others by analogy. Cases of this sort are not uncommon. For example in Polish, the preposition (also occurring as a prefix) od(e) ‘from’ historically had a voiceless obstruent t (cf. Russian ot ‘from’) but it became voiced due to the analogy with other prepositions (prefixes) that end in voiced obstruents, such as pod(e) ‘under’, nad(e) ‘above’ etc. As a result, all prepositions (prefixes) in Polish have final voiced obstruents underlyingly. However, I don’t think that the analogy that affected Polish prepositions and, presumably, Swahili extensions, is of any importance for the syn- chronic grammar. As far as Swahili is concerned, it seems more appropriate to simply include the l as part of relevant morphemic constraints. Especially since there is one other productive extension, the “reciprocal” an, which does not have a consonant-initial alternant; hence, pat-an-a “get along” but also andiki-an-a “write to each other” (*andiki-lan-a), o-an-a “marry each other” (*o-lan-a) etc. Another reason for the low ranking of Morph constraints referring to the three (l)-initial extensions is the fact that their vowel alternates between i~e partially harmonizing with the final stem vowel. The i vowel appears when the preceding stem vowel is from the set {a,i,u}, and the e vowel occurs after the stem mid vowel {e,o}. This stem-controlled harmony is very limited in Swahili — it affects only some extensions, never other suffixes, prefixes or roots. However the specific constraints on harmony are formulated (see Cole & Kisseberth, 1994 on general theory of harmony in OT), it is fair to
Syllable structure constraints in exceptions 125 hypothesize that they affect the extensions because of the low ranking of their morphemic constraints. Hence, Morph-Reg >> VH >> Morph-li, lik etc., where “VH” stands for the vowel harmony constraints.6 Recall also that the epenthetic vowel of demonstratives shows complete harmony (i.e. identity) with an adjacent vowel. This is expected because there are no Morph constraints imposed on the epenthetic vowel, hence it may satisfy all VH constraints at no cost. In Turkish, several suffixes occur in alternating forms depending on syl- lable structure (see e.g. Lewis, 1986; Kenstowicz & Kisseberth, 1979; Clements & Keyser, 1983). The three suffixes in (18a) have a CV(C) alternant after the vowel-final stem and a V(C) alternant after consonant-final stem. This distribution is exactly parallel to the case of Swahili extensions dis- cussed previously. Similarly, I assume that the crucial constraint is that of *Coda which dominates respective Morph-si, Morph-ya and Morph-nin constraints. Lexical roots, as well as some other suffixes, including the plural suffix lAr shown for comparison in (18b), are unaffected by *Coda which shows that the relevant Morph constraints are ranked higher in the hierarchy, as illustrated in (19). Note in addition that both types of suffixes in (18) (as well as other mentioned later) harmonize with the final stem vowel in backness/frontness (high vowels also in rounding). For the purpose of this paper it is sufficient to assume that Morph-Suff constraints are dominated by appropriate vowel harmony constraints which, however, do not affect the height of suffixal vowels (cf. Kirchner, 1993). Under the ranking VH >> Morph- Suff, it is irrelevant for the analysis whether the Morph-Suff constraints require that suffixal vowels be fully specified (as the vowel [i] or[i-], or the vowel [a] or [e], respectively), or specified for height only.
(18) a. 3 sg. poss. Genitive Dative b. Plural Gloss oda-si- oda-ni-n oda-ya oda-lar ‘room’ dere-si dere-nin dere-ye dere-ler ‘river’ ki-z--i ki-z-i- n k-z i -a ki-z-lar ‘girl’ el-i el-in el-e el-ler ‘hand’
6 The complete analysis of VH in Swahili is a rather complex issue that goes beyond the scope of this paper. There is one more extension which alternates between u∼o, o is found only after the stem o and u after remaining four vowels {a,e,i,u}. An additional aspect is that the epenthetic vowel in loan-words often harmonizes with the onset consonant, hence u is found after labials and i after coronals. (This type of harmony is absent in extensions but we find traces of it in the shape of some nominalizing suffixes.) On the other hand, when the onset of the epenthetic vowel consists of the placeless h or G, the complete harmony with the neighbouring vowel is observed, e.g. the demonstratives in (16) or loan-words as bahari (Arabic [bah- r]) ‘sea’, roho (Arabic [ruuh- ]) ‘soul’.
126 Iwona Kraska-Szlenk
(19) VH Morph-lar *Coda Morph-si etc. el-i *s el-si *! el-i- *! *s el-ler ** el-er *l! *
The data in (20a) contain possessive suffixes whose initial vowel appears only after consonant-final stems and is not realized after vowel-final stems (likewise, -(I)miz ‘1 PL. POSS.’ and -(I)niz ‘2 PL. POSS.’). Similarly, the suffix of the deverbal nouns and adjectives in (20b) alternates between -ak/-ek and -k (after Lewis, 1986: 223).
(20) 1 sg. poss. 2 sg. poss. gloss a. ki-z--mi ki-z-i- n ‘girl’ el-im el-in ‘hand’ oda-m oda-n ‘room’ dere-m dere-n ‘river’ b. dur- dur-ak ‘to stop; stopping-place’ yed- yed-ek ‘to tow; tow-rope’ tara- tara-k ‘to comb; comb’ ele- ele-k ‘to sift; sieve’
A possible analysis of the data in (20) is to assume that initial vowels of these suffixes are truncated as a result of the Onset constraint crucially dominating Morph-im,in,ak etc., as shown in (21), where irrelevant constraints from (19) are omitted.
(21) Onset Morph-im,in,ak etc. dere-m *I dere-im *!
Notice that under the analysis in (21), Morph-im,in,ak should not be domi- nated by *Coda (contrary to the suffixes of (18)) to prevent the deletion of final consonants of these suffixes, hence, Morph-im,in,ak >> * C o d a . It is important to note that in Turkish, similar to other cases discussed earlier, whenever a phonological constraint imposes a partial deletion of segmental material, it is always the suffix which suffers deletion and not the root (e.g. dere-m not *der-im; el-i not *e-si). Naturally, this fact follows from the crucial ranking: Morph-Reg (Root) >> Morph-Exc (Suff). In addition to suffixes which are subject to vowel harmony, as mentioned thus far, some Turkish suffixes occur in invariant forms with their vowels
Syllable structure constraints in exceptions 127 specified as front or back no matter which vowel precedes them. For exam- ple, the front-vowel suffix -ki may follow a back stem, as in yari-n-ki ‘of yester- day’ (Lewis, 1986: 70), or the back-vowel suffix -da¸s (-ta¸s) may follow a front- vowel stem, as in meslek-ta¸s ‘colleague’ (Lewis, 1986: 64). The complex interaction of Turkish phonological and morphemic con- straints is illustrated in (22) below. Most roots are subject to Vowel Harmony, hence the ranking: VH >> Morph-Roots. I assume that Onset is undominated given the general absence of vowel hiatus in Turkish (and evidence as vowel truncating suffixes).
(22) Onset ↓ Morph-ki, das¸, etc. ↓ Vowel Harmony ↓ Morph-lar etc., Morph-im, etc., Morph-Roots ↓ *Coda ↓ Morph-si etc.
4. Other patterns and concluding remarks
It is remarkable that cross-linguistically obedience to ‘more’ phonology is very common for grammatical morphemes and not so common for lexical morphemes. This may relate to historical changes that affect the former and not the latter class of morphemes. Most typically, grammatical morphemes originate as lexical roots (or words). Once they become used in a grammati- cal function as an auxiliary word, a clitic or an affix, they tend to shorten and lose some of their original phonological material. Perhaps one reason for such shortening is that they simply can afford it without losing lexical con- trasts, while roots are not so free in that respect (e.g. English would, will, have still maintaining most contrasts in their shortened forms ’d, ’ll, ’ve vs. hypo- thetical shortening of roots, as ba
128 Iwona Kraska-Szlenk active constraints. This process is schematized in two steps in (23), where (23a) stage describes the original situation before the process of grammati- calization, and (23b) illustrates the outcome of grammaticalization of a particular morpheme X.
(23) a. Morph-Reg, Morph-X b. Morph-Reg ↓ ↓ Phon Phon ↓ Morph-X
Variation between the two steps in (23) is a common, if not indispensable intermediate stage. Similarly as in the case of nativization discussed earlier in §2, such variation is represented as a variable ranking of Morph-X with respect to Phon, i.e. either (23a) or (23b). Consider, for example, Classical Arabic future tense marker [sawfa] used in variation with its shortened form [saa]. Perhaps the only thing that needs to be said about this variation is that Morph-sawfa is variably ranked with respect to *Coda, either above it, as in the case of other ‘regular’ morphemes, as shown in (24a), or, below it by which as much of its structure is lost as needed to satisfy *Coda7 as shown in (24b).
(24) a. Morph-sawfa >> *Coda; output: [sawfa] b. *Coda >> Morph-sawfa; output: [saa]
A valid question to pose is the following: Are there grammatical mor- phemes which are subject to more than ‘regular’ phonology in the fashion parallel to non-nativized loan-words discussed earlier in §2? If so, their respective Morph constraints would be ranked higher than Morph-Reg. A tentative answer to this question seems to be “yes”.8 Consider, for instance, English suffixes, such as -s, -d or -th. Firstly, it is only such suffixes that may introduce consonant clusters which are atypical for English and violate the ‘regular’ syllable structure constraints. Secondly, when some of such
7 I assume here a dominant constraint against ‘gapped’ morphemes (cf. Contiguity in McCarthy & Prince, 1994; 1995) which rules out a form as *saafa. I also assume that com- pensatory lengthening in [saa] is more harmonic than [sa], since the extra mora contained in the former is a piece of structure required by Morph-sawfa in the realization of the first syl- lable, hence, on the whole, the constraint is better satisfied when this mora is “parsed”. 8 This is contra McCarthy & Prince (1995, ch.6), who argue that such a case is impossible, due to the universal dominance of constraints of Root-faithfulness type over Affix-faithfulness (which is synonymous with the ranking of Morph-Root over Morph-Affix in the present model).
Syllable structure constraints in exceptions 129 clusters are resolved by means of deletion, it does happen that the root con- sonant deletes and the consonant of the affix remains “parsed”. For exam- ple, in the colloquial American pronunciation of ‘asked’ [ast], it is the k of the root which deletes and not the affixal t (although there are other varieties of American English where things may be just reversed). Similarly, in acts [aks], the t deletes, not the s etc. The data are accountable in the straightforward manner under the constraint ranking: Morph-Affix >> Morph-Root. Specifi- cally, Morph-d,s should dominate Morph-ask,act, as shown in the tableau below (where *Cluster is a cover-term for relevant constraints prohibiting clusters as skt#, kts# etc.).
(25) *Cluster Morph-d,-s Morph-ask,act ast *k askt *! ask *!
Perhaps the unusually high ranking of English Morph-d,-s,-th is related to the fact that each of these morphemes is realized by just one segment. If such a segment were deleted in obedience to the higher *Cluster (or other) constraint, as in the last output in (25), there would be no phonetic cues for the presence of a given suffix which would cause a merger of suffixed and non-suffixed words (present and past tense, singular and plural nouns etc.),while the levelling of Morph-Suff above other morphemic constraints provides a protective mechanism against it. Notice, however, that such mergers, caused by affix’s deletion are found cross-linguistically (e.g. certain dialects of Spanish in which las casas ‘the houses’ merges with la casa ‘the house’, or some varieties of English which merge worked with work).9 In conclusion, it is perhaps worthwhile to emphasize the point already hinted at in §1, namely, that the distinction between the ‘regular’ and the ‘exceptional’ is based on statistical rather than principled criteria. There is no ‘special’ phonology affecting exceptions to the exclusion of ‘regular’ forms; nor is it that some ‘exceptions’ are particularly exempted from being affected by the regular phonology. On the contrary, all morphemes, includ- ing the ‘exceptional’ ones, are subject to the same general phonological
9 One could perhaps argue that there is a separate constraint against zero morphemes. Then McCarthy & Prince’s (1995) universal dominance (cf. fn. 8) could be still maintained and the English case would be accounted for by ranking of *Zero-Morpheme appropriately high. However, this solution seems rather artificial because the only role of such a constraint would be to prevent deletion in specific cases as English above; but this can be achieved by means of re-ranking of Morph constraints themselves, hence *Zero-Morpheme seems quite redundant.
130 Iwona Kraska-Szlenk constraints arranged in a hierarchy. If different morphemes are not uni- formly affected by phonological constraints, it is only because the ranking of particular Morph constraints in the hierarchy may vary. What we tend to call ‘exceptions’ are cases of unusually high or unusually low ranking of Morph constraints, assuming that ‘usual’ refers to the point where the condensa- tion of Morph constraints is greater than anywhere else. When the criterion of number happens not to be very significant, there is more room left for the difference of opinion as to what is ‘regular’ and what is ‘exceptional’. For example, Turkish disharmonic roots, mentioned in §2, are typically called ‘exceptions’, as in Lewis’s (1986 [1967]) grammar. However, Clements & Sezer (1982) argue that disharmonic roots are so numerous and psychologically ‘normal’ that there is no reason to think of them as ‘exceptional’. Under their interpretation, it is coincidental that some roots contain the harmo- nizing vowels and happen to be ‘harmonic’ by chance. For Clements & Sezer, this particular interpretation has a consequence for the formalism of their analysis; since it is ‘normal’ for the roots to contain any vowels, the vowels of the accidentally harmonic roots are all specified, contrary to the truly harmonizing vowels of most suffixes which are left underspecified. The present framework allows for a more flexible interpretation of the Turkish data. We can assume two classes of roots with their Morph constraints ranked respectively, above or below vowel harmony constraints as in (26). Note that in the latter class it is quite irrelevant whether Morph constraints require full specification or allow for partial underspecification of root vowels (cf. the discussion in §1). Because the two classes seem to be com- parable in size, it is not that obvious where the reference point for ‘normal- ity’ should be, as indicated by the arrows in (26). But such a reference point does not have any formal status in the analysis although it might reflect speakers’ feelings(which may vary, too).
(26) Morph-Roots ← ‘normal’ ? ↓ Vowel Harmony ↓ Morph-Roots ← ‘normal’ ?
Acknowledgements
I would like to thank three anonymous reviewers for their useful comments.
Syllable structure constraints in exceptions 131
References
Clements, G.N. & S.J. Keyser (1983). CV phonology: A generative theory of the syllable. Cambridge, Mass.: MIT Press. Clements, G.N. & E. Sezer (1982). Vowel and consonant disharmony in Turkish. The structure of phonological representations. Part 2, edited by H. van der Hulst & N. Smith (eds.), 213–255. Dordrecht: Foris. Cole, J. & C.W. Kisseberth (1994). An Optimal Domains Theory of Harmony. To appear in FLSM V, published by Studies in Linguistic Science, University of Illinois. Dogil, G., J. Mayer & J.C. Roux (this volume). Syllables and unencoded speech: Clicks and their accompaniments in Xhosa. 49–60. Itô, J. & A. Mester (1995). Japanese Phonology. The handbook of phonological theory, edited by J. Goldsmith, 817–838. Oxford: Blackwell. Itô, J., A. Mester & J. Padgett (1995). Licensing and underspecification in Optimality Theory. Linguistic Inquiry 26, 571–613. Kenstowicz, M. & C. Kisseberth (1979). Generative phonology. San Diego: Academic Press. Kirchner, R. (1993). Turkish vowel harmony and disharmony: An Optimality Theoretic account. Paper presented at Rutgers Optimality Workshop I, Rutgers. Kraska, I. (1992). From verb to clitic and nominal suffix: the Somali -e, -o nouns. Studies in the Linguistic Sciences 22, 89–106. Kraska-Szlenk I. (1995). The phonology of stress in Polish. University of Illinois Ph.D. dissertation. Lewis, G.L. (1986). Turkish Grammar. Oxford: Oxford University Press. McCarthy, J. & A. Prince (1993). Prosodic morphology I: constraint interaction and satisfaction. To appear Cambridge, Mass.: MIT Press. McCarthy, J. & A. Prince (1994). The emergence of the unmarked: Optimality in prosodic morphology. Proceedings of the North East Linguistic Society 24, edited by Mercé Gonzaléz, 333– 379. Amherst: Graduate Linguistic Student Association. McCarthy, J. & A. Prince (1995). Faithfulness and reduplicative identity. Papers in Optimality Theory(University of Massachusetts Papers in Linguistics 18), edited by J.N. Beckman, L. Walsh Dickey & S. Urbanczyk, 249–384. Amherst. Prince, A. & P. Smolensky (1993). Optimality Theory: Constraint interaction in generative grammar. (To appear Cambridge, Mass.: MIT Press.) Russell, K. (1995). Morphemes and candidates in Optimality Theory. Manuscript, University of Manitoba. Saeed, J.I. (1985). A concise reference grammar of Somali. Wheaton: Sunwoody Press.
On possible onsets
Michele Loporcaro
Abstract
This paper argues for a view of phonological analysis in which language- internal evidence is given priority over “metaphysical” assumptions. The specific theoretical question which is tackled here concerns the role of the sonority hierarchy in the description of syllable structure. The Sonority Sequencing Generalization (or comparable generalizations current in other frameworks), it is argued, should be understood as a preference statement rather than as an absolute constraint on the format of phono- logical representation. To substantiate this claim, empirical evidence from The Hague: Thesus. 133–151. Thesus.The Hague: 133–151. . Romance is discussed. 0. Syllables !? 1. Introduction
The syllable has again become a controversial notion over the past few years, as the question/exclamation mark in the title of this congress testi- Phonologica 1996: Phonologica 1996: fies. However, an important characteristic of today’s research in theoretical phonology is the fact that, in contrast with ‘no-syllable’ models of phonol- ogy current in the SPE era, everybody now agrees on the relevance of syllable constituents. Consequently, even those who currently reject the syllable as a theoretically significant structural entity acknowledge the relevance of such objects as e.g. onsets or rhymes (or, termed differently, of the binding relations among prenuclear segments or between the nucleus and segments in postnuclear position). One might even suspect, in some cases, that the same generalizations are being restated in only seemingly different terms. To put it very simply, what is an onset an onset of, if syl- lables do not exist? This question, however, will not be pursued any further in the present context. The problem to be discussed in this paper concerns the nature of the evidence which can and must be brought to bear in discussing syllable
John R. Rennison & Klaus Kühnhammer, eds. (1999). eds. R. Rennison & Klaus Kühnhammer, John (1999). structure (or, for those who attach importance to the distinction, the
134 Michele Loporcaro structure of syllable constituents). What must the linguist regard as relevant or crucial in this connection, and what are the appropriate procedures to be put to use in order to make decisions about how to parse segments into syllables and syllable constituents? In the following discussion I focus on the interplay between theory and evidence or, to put it more precisely, between theoretical expectations based on general principles, on the one hand, and empirical testing thereof, on the other. Although the argument to be developed here is confined to the specific case of syllable onsets, I would suggest that the conclusions apply to the study of syllable structure as a whole. The paper is organized as follows: §2 sketches the theoretical basics and surveys some of the criteria most often referred to in deciding whether a consonant cluster is syllabified as an onset. These criteria, it is argued, should be formulated in such a way that they allow us to empirically test theoretical claims, rather than be impervious to empirical verification/ falsification. This is far from stating the obvious, since the prevailing line of research nowadays seems to prefer the latter approach (i.e. a Platonic rather than a Galilean style of research). §3 puts to the test different criteria for onsethood, discussing an empirical case; in §3.2 theoretical conclusions are drawn, which are defended in §4 against possible counterarguments developed from a Platonic vantage point.1
2. Diagnostics for onsethood
Three different diagnostics for onsethood are summarized in (1a–c).
(1) A consonant cluster C1C2 is an onset if it a. is allowed to occur word-initially (Kuryłowicz, 1948: 196); b. patterns with onsets in the environment of rules of the language which are sensitive to syllable structure; c. complies with the Sonority Sequencing Generalization (Selkirk, 1984: 116: “In any syllable, there is a segment constituting a sonor- ity peak that is preceded and/or followed by a sequence of seg- ments with progressively decreasing sonority values”).
1 The following abbreviations are used throughout the paper: MPR = Morphophonological Rule, PR = Phonological Rule, PRom = Proto-Romance, SPE = The Sound Pattern of English (Chomsky & Halle, 1968), SSG = Sonority Sequencing Generalization, WRom = Western Romance.
On possible onsets 135
It has often been claimed that a VCV string is universally syllabified as V.CV (Maximal Onset Principle, see Blevins, 1995: 230 for discussion). This also follows from both criteria (1a–b). In fact, no language seems to bar word- initial CV-sequences, and in all languages, PRs affecting consonants in onset position usually affect prevocalic single initial consonants as well as inter- vocalic non-geminates. Consider for instance Gorgia Toscana (GT), which in Florentine and other Tuscan varieties turns velar stops into fricatives in syllable-initial post-vocalic position.
(2) GT applies (/k/ → [h]) GT does not apply (/k/ → [k]) ama[h]a ‘hammock’ mos[k]a ‘fly’ di[hj]aro ‘I declare’ man[k]a ‘fails’ emi[hr]ania ‘migraine’ bo[k:]a ‘mouth’ la [h]asa ‘the house’ la [hj]esa ‘the church’ la [hr]oce ‘the cross’
Criterion (1a), which is distributional in nature, is nowadays deemed by many to be unreliable. This rejection is exemplified very clearly by Kaye et al. (1990: 204): “the mere existence of words in various languages that begin with all kinds of exotic sequences of consonants is completely irrelevant to discussions of syllable structure”.2 In Kaye et al.’s (1990) perspective, to positively conclude that a word-initial cluster is indeed an onset, evidence must be provided to show that it actu- ally patterns with onsets in the environment of rules operating in the lan- 3 guage. This is criterion (1b), which I will call the RULE-BASED criterion. It is more restrictive than (1a), as it may be the case that a word-initial cluster behaves phonologically as if its first segment were not syllabified as part of the onset. A clear example of this is provided by the selection of the definite article allomorphs il/lo in standard Italian, shown in (3).
2 I will not take issue here with this rejection. As will become apparent directly, if constraints on syllabification are expressed (both universally and language-specifically) in the form of preference statements, (a revised version of) (1a) also proves useful in determining onset structure (cf. Vennemann’s, 1988: 33 Law of Initials, as well as the discussion of (14)–(15) below). Kaye et al.’s (1990) claim can be maintained in a weaker version, stating that criterion (1b) has to take precedence over (1a). 3 The rule metaphor has not been very much in fashion lately. By ‘rule’ I mean a synthetic statement of distributional regularities, be they phonological, morphophonological, mor- phological, etc. Empirically, I will call ‘rules’ observations about the (morpho)phonology of Italian like those summarized in (3)–(6) below. Although the same facts may be labelled dif- ferently in no-rule approaches or output theories of various kinds, this makes no difference, as far as I can see, for the present argument.
136 Michele Loporcaro
(3) a. [il 'ka:ne] b. [il 'pra:to] c. [lo 'spa:ro] d. [lo 'N:O:mo] ‘the dog’ ‘the lawn’ ‘the shot’ ‘the gnome’ /il/ is selected before onsets /lo/ is selected before non-onset initial Cs
According to a long-standing generalization (see e.g. Camilli, 1913: 170; 194 4 –45: 90; Davis, 1990; Kaye et al., 1990: 204f; Marotta, 1993), /il/-selection is evidence for onsethood in (3a–b) (/il$ka/, /il$pra/), whereas /lo/-selection in (3c) proves that /s/ + C clusters are not entirely syllabified as onsets under- lyingly.4 Rather, they are heterosyllabic (/lo#s$pa/) word-initially just as word- internally. Needless to say, the same goes for geminates: (3d). A further piece of evidence which has long been brought to bear for the same purpose (as early as Camilli, 1913: 170; see Loporcaro, 1997: ch.1, §1.2 for references) is that raddoppiamento fonosintattico (RF) affects the initial con- sonants in (3a–b) but not in (3c–d), as shown in (4).
(4) a. ['tre 'k:a:ni] b. ['tre 'p:ra:ti] c. ['tre 'spa:ri] d. ['tre 'N:O:mi] ‘three dogs’ ‘three lawns’ ‘three shots’ ‘three gnomes’ 1444 RF applies 4443 144 RF does not apply 443
However, here I will not dwell on this fact for reasons which will be made clear in §3.2 below. That /s/ + C clusters (again, like geminates) are heterosyllabic word- internally, can be shown in turn by considering vowel lengthening, in (5), and diphthongization, in (6), in stressed non-final open syllables.
(5) a. ['la:$na] b. ['la:$dro] c. ['las$ko] d. ['lat$te] ‘wool’ ‘thief’ ‘loose’ ‘milk’
$ $ $ $ (6) VE˘ NIT PE˘ TRAM VE˘S TEM SE˘P TEM Latin > a. ['vjE:$ne] b. ['pjE:$tra] c. ['vEs$te] d. ['sEt$te] Italian ‘comes’ ‘stone’ ‘dress’ ‘seven’
Prepausal open syllable lengthening does not apply before /s/ + C clus- ters or geminates (see (5c–d)). Open syllable diphthongization of open-mid vowels, which changed allophonically long PRom [E:], [O:] to [jE:], [wO:] respec- 5 tively, did not apply before /s/ + C clusters or geminates (see (6c–d)).
4 This is not to say that the two forms of the definite article can still be derived synchronically from one single underlying phonological form. Rather, the relation between il and lo is today an allomorphic one, as is convincingly and comprehensively argued by Dressler (1984). 5 Since diphthongization is no longer a synchronically active PR (rather, the generalization relating e.g. buono/bontà ‘good/goodness’, vuole/volete ‘(s/he) wants / you (PL.) want’. is best
On possible onsets 137
There is also a further, widely adopted diagnostic for onsethood (cri- terion (1c)): for a cluster to be an onset, it must comply with Selkirk’s (1984: 116 ) S ONORITY SEQUENCING GENERALIZATION (SSG), which states that the sonority of segments must decrease towards the edges of a syllable (a time- honoured observation in phonological research). Adopting this criterion, Rubach & Booij (1990: 122f) achieve a dramatic simplification of Polish onset structure. In analysing words such as those in (7), they argue that the “con- sonant at an edge of a word […] does not count from the point of view of the SSG” — especially if it violates it. Consequently, the initial consonant of the word-initial clusters in (7) is actually not part of an onset underlyingly.6
(7) rte˛c´ ‘mercury’, lwy ‘lions’, lnu ‘linen (GEN.SG)’, mnie ‘me’, szpieg ‘spy’, wtorek ‘Tuesday’, ptak ‘bird’, kpic´ ‘joke’
In the case of /s/ + obstruent clusters in Italian, thus, the application of criteria (1b) and (1c) gives an identical result: /s/ + obstruent clusters do not behave like onsets phonologically, and do not obey the SSG (they are SONORITY REVERSALS, in Clements’ (1990) terminology).
(8) criterion (1a): criterion (1b): criterion (1c): word-initial rules SSG /st/+ – – = non-onset /tr/ + + + = onset
What happens, however, if the rule-based and the SSG-based diagnostic give contradictory results for a given language? Suppose that we were to find a variety of Italian (which I will call “language x”; see (9b)) in which /il/ is con- sistently selected before some clusters violating the SSG, e.g. before clus- ters of equal sonority, such as /bd/, /dgˇ/, /mn/, whereas /lo/ is still selected before clear non-onsets (sonority reversals), as in standard Italian. If a variety like language x exists, and if, as I believe, the rule-based criterion (1b)
expressed in the form of an inverted MPR: cf. Dressler, 1985: 125), the evidence in (6) shows that /s/ + C clusters and geminates were heterosyllabic by the time diphthongization was completed (viz. in the mid 7th century: cf. Castellani, 1961: 95). However, the lack of vowel lengthening before the same clusters (see e.g. the measurements in Fava & Magno Caldognetto, 1976: 62) assures us that the generalization is still valid for Italian in its present stage (see however Bertinetto, 1996, who argues for the tautosyllabicity of /s/ + C clusters in Italian, both word-initially and word-medially). In fact, open syllable lengthening is an allo- phonic PR which was added to the phonology of Late Latin by the 6th century and was car- ried on unchanged into PRom and standard Italian (cf. Loporcaro, 1997: Ch. 3, §3.1.1). 6 In that framework such non-onset initial consonants are not linked to the syllable node but to the phonological word node (Rubach, 1996: 79).
138 Michele Loporcaro is valid, then the SSG-based criterion (or any other criterion making the same predictions) would have to be revised.
(9) Test /il/ if $C(C)V /lo/ if C$CV a. Italian: il kV- il prV- lo pkV- lo bdV- lo stV- lo ftV- b. language x: il kV- il prV- il pkV- il bdV- lo stV- lo ftV- good onsets sonority plateaus sonority reversals
In fact, criterion (1c) is not the only conceivable way in which the sonority hierarchy can be used to state a generalization about onset structure, and is reminiscent of a current procedure insightfully described by Dressler:
For decades theoretical linguistics has been plagued by the following scenario: 1) The most frequent/(intuitively) natural state of affairs (e.g. in American structuralist phonology the principle of biuniqueness […]) is elevated to the only state of affairs allowed by the given theory/ model. 2) This is encapsulated in a theoretical principle capturing valid generalizations about it, e.g. in the slogan “once a phoneme, always a phoneme”. Dressler (1988: 143)
Returning to our specific case, Dziubalska-Kołaczyk’s (1995: 78) criticism of Selkirk’s SSG also underscores the inadequacies of such an approach in determining what must count as a syllable onset: “The weakness of this approach lies in the fact that […] it does not differentiate between ‘better’ and ‘worse’ clusters among those which do observe the sequencing.” In addition, the SSG-based criterion, as it stands in (1c), is also unable to discriminate between worse and better violators of the sequencing. From an epistemological point of view, the approach to syllabification exemplified by criterion (1c) represents a procedure which I would call PLATONIC. First of all, an a priori principle is assumed, then empirical evidence is checked against the principle. If what is empirically observed happens not to conform to the latter, then it is not the principle which is revised but rather (our representa- tion of) the empirical data itself. (See above on Rubach & Booij’s (1990) analysis of the Polish data in (7).) Now I will explore the consequences of looking at language in a Galilean perspective, following Chomsky’s well-known definition. If the empirical evidence disconfirms the assumed principle, then the latter, not the former, has to be revised. It may turn out, for instance, that the putative principle is not actually a principle, but rather a parameter or a matter of preferences. As far as syllable onsets are concerned, we could replace criterion (1c) with a parametric statement of the conditions on onset formation like Vennemann’s Head Law (where head is equal to onset in the present context).
On possible onsets 139
(10) Head Law (Vennemann, 1988: 13) A syllable head is the more preferred: (a) the closer the number of speech sounds in the head is to one, (b) the greater the Consonantal Strength value of its [first segment], and (c) the more sharply the Con- sonantal Strength drops from the onset toward the Consonantal Strength of the following syllable nucleus.
It will then be the evidence provided by rules operating in each specific language (according to criterion (1b)) which will tell us exactly how the relevant parameter is set for that language. This approach is exemplified in §3 by considering a concrete empirical case.
3. Grizzanese
Grizzanese is an Italo-Romance variety spoken by approximately 2,000 people in Grizzana Morandi (southern Emilia, province of Bologna).7 Like most Emilian dialects, it is characterized by the occurrence of many (by Romance standards) unusual consonant clusters, both word-internally and at word edges, as in (11).
(11) word-initial word-internal a. ['klo:r] ‘colour’ ['dOndla] ‘weasel’ obstruent + lateral ['vlo] ‘wanted’ [ar'dlEt] ‘chosen’ ['tNi:] ‘to hold’ [pi'tne:] ‘to comb’ obstruent + nasal [dman'de:] ‘to ask’ [ar'kNOser] ‘to recognize’ b. ['bzON] ‘need’ [a'ksI-] ‘so’ stop + fricative c. ['dgˇe:va] ‘was saying’ ['ma:ndga] ‘sleeve’ stop + stop/affricate ['pkO˜:] ‘mouthful’ [as'pte:] ‘to wait’
In this dialect, much as in standard Italian, /s/ + obstruent clusters occur word-initially as well as word-internally.
(12) a. ['stmE:na] ‘week’, [zdiz'de:] ‘to wake up’, [zbdu'cˇe:] ‘to delouse‘, ['zdNa] ‘offended’ b. [ar'spOnder] ‘to answer’, ['vI-sta] ‘seen (FEM.)’, ['ka:ska] ‘(it) falls’, [jus'fO˜:] ‘Joseph’
7 The data discussed here stem from my own field notes (1990–96). They are presented in the phonetic form of the dialect of Veggio, a frazione (administrative subdivision) of Grizzana.
140 Michele Loporcaro
Let us now investigate whether there is evidence to decide on the syllabi- fication of these consonant strings. Consider the series of hypothetically conceivable word-initial clusters in (13), where the clusters are ordered left to right according to their degree of compliance with the SSG (better candi- dates for onsethood are on the right, worse candidates on the left).
(13) a. b. c. d. e. f. g. h. i. j. k. *JT- *RT- *LT- NT- ST- T1T2- TS- TN- TL- TR- TJ- worse better [with respect to the Head Law (10)] S = fricative, T = stop, N = nasal, L = lateral, R = rhotic, J = glide
The worst candidate onsets, (13a-c), do not exist word-initially, even though (13b and c) should have arisen in diachrony, as a product of a syn- cope process which deleted mid vowels before stress in Emilian dialects. As shown in (14), however, when a word-initial *rC- or *lC- cluster was about to be created, it was prevented through the application of /a/-prosthesis.
(14) Latin input Romance development Grizzanese output REMANERE > remanere > *rmanér > [arma'Ne:] ‘to remain’ RESPONDERE > respondere > *rspònder > [ars'pOnder] ‘to answer’ *LIGICARE > lekkare > *lker > [al'ke:] ‘to lick’ LAETAMEN > letame > ledame > *ldam > [al'da:m] ‘manure’ Wrom Emilian prosthesis voicing syncope
Nasal plus consonant clusters, (13d), have an in-between status. As illus- trated in (15), /n/C patterns with the clusters in (14), whereas /m/C, when it arose through syncope, was tolerated in initial position.
(15) Latin input Romance development Grizzanese output *NITIDARE > nettare > *nter > [an'te:] ‘to clean’ NEPOTEM > nepote > nevode > *nvod > [aM'vo:d] ‘nephew’ vs. MINESTRA(-RE) > minestra > mnestra > ['mnE:stra] ‘soup’ METEBAT > meteva > medeva > mdeva > ['mde:va] ‘was harvesting’
The very fact that a prosthetic /a/ was added in (14)–(15) shows that the clusters involved (viz. (13b–c) and partly (13d)) were not tolerated as onsets. As a result these clusters only occur word-internally.
On possible onsets 141
Note that the data in (14)–(15) confirm the validity of the distributional criterion (1a), provided that it is restated in a parametric form as discussed in fn. 2 above. After all, a default relationship between word-initial and syllable-onset position must be assumed, so that the word-initial occurrence of non-onsets is disfavoured and represents the marked option. Con- sequently, it is not merely coincidental that the non-onsets most often tolerated word-initially are those which depart least from the preference parameters (cf. /s/C- in Italian, as opposed to */l/C-, */r/C-, */j/C- etc.). The remaining clusters in (13e–k) all occur word-initially; we must ask whether there is evidence for discriminating among them with respect to syllabification. Comparison with Standard Italian (see (3)–(4)) suggests some legitimate expectations: we expect fricative + stop clusters (13e) not to behave as onsets; conversely, the best candidates (viz. (13i–k)) are expected to behave as onsets. In order to empirically substantiate these expectations, a test is needed, which parallels il/lo selection in Standard Italian. It would be especially interesting to check the behaviour of the in-between clusters in (13f-h). These clusters do not occur in the native vocabulary of Italian (all words displaying them are borrowed), whereas they are perfectly integrated in the native phonotactics of Grizzanese (and Emilian in general). Both points are dis- cussed in §3.1.
3.1. The test: clitic clusters and syllabification
Grizzanese has subject and object clitics, which all precede the verb in declarative utterances. The relevant paradigms are given in (16): (When two forms are given, the one following the slash is selected before an initial vowel.)
(16) a. subject 1sg 2sg 3msg 3fsg 1pl 2pl 3mpl 3fpl a t e/l la/l a a i/j al/aL b. direct ob. 1sg 2sg 3msg 3fsg 1pl 2pl 3mpl 3fpl refl m t (a)l la/l s v i li s
As can be seen, most subject clitics end in a vowel and most object clitics simply consist of a single consonant. A typical combination is given in (17).
(17) a t 'vEd/'krEd/'zdEzd 1SG.SUB. 2SG.OB. see/believe/wake (1SG.) ‘I see/believe/wake you’
142 Michele Loporcaro
Apart from some low level processes (optional voicing assimilation of obstruents and optional final devoicing), the phonological shape of clitics in (17) is not significantly altered, regardless of the number and type of initial consonants of the following verb. However, when the 3rd person feminine plural clitic /al/ (which ends in a consonant) precedes an object clitic (which consists of a consonant), cluster simplification takes place, as shown in (18).
(18) a. /al t 'vEden/ → [a t 'vEden] 3PL.FEM.SUB. 2SG.OB. see (3PL.) (by [l]-deletion) ‘they (FEM.) see you’ b. /al t 'krEden/ → [a t 'krEden] 3PL.FEM.SUB. 2SG.OB. believe (3PL.) (by [l]-deletion) ‘they (FEM.) believe you’ c. /al t 'zdEzden/ → [al ti 'zdEzden] 3PL.FEM.SUB. 2SG.OB. wake (3PL.) (by [i]-epenthesis) ‘they (FEM.) wake you up’
Two complementary strategies for cluster simplification are at work in (18): coda [l]-deletion applies before single consonants or C + rhotic clusters in (18a–b), whereas [i]-epenthesis (for some dialects and speakers, phoneti- cally [E]) applies before /s/ + C clusters. Indeed, epenthesis applies before all fricative + stop clusters, as well as before word-initial nasal + C clusters, as further exemplified in (19):8
(19) a. /al t 'stofen/ → [al ti 'stofen] ‘they (FEM.) are boring you’ b. /al s 'sbducˇe:ven/ → [al si 'zbducˇe:ven] ‘they (FEM.) were delousing us’ c. /al s 'vde:ven/ → [al si 'vde:ven] ‘they (FEM.) were seeing each other’ d. /al s 'mde:ven/ → [al si 'mde:ven] ‘they (FEM.) were harvested’
Both cluster simplification strategies in (18)–(19) are syllable-driven, as can easily be seen from the data in (20).
8 The 3rd person plural feminine clitic /al/ is the only one which provides a relevant testing ground for syllabification (by [l]-deletion or [i]-epenthesis). The only other clitic ending in a consonant, viz. 2nd person /t/, behaves in an entirely different manner, e.g. [tam 'vEd] ‘you see me’ from underlying /t m 'vEd/ via [a]-epenthesis.
On possible onsets 143
(20) a. [al t a'rE˜: 'vI-st] 3PL.FEM.SUB. 2SG.OB. have (3PL.FUT.) seen ‘they (FEM.) will have seen you’ b. [al t a've:ven 'pu˜:t] 3PL.FEM.SUB. 2SG.OB. have (3PL.IMPERF.) stung ‘they (FEM.) had stung you’
When the verb following the clitics begins with a vowel, both the subject and the object clitic surface without any changes. And if nothing happens before a vowel in (20), it is safe to conclude that what happens in (18) ([l]- deletion or [i]-epenthesis) depends on syllabification constraints: too many consonants are put together underlyingly for them to be properly syllabi- fied. If the processes in (18) are sensitive to syllable structure, a striking parallel with standard Italian il/lo selection emerges.
(21) Standard Italian Grizzanese a. [lo 'spa:ro] ‘the shot’ [al ti 'stofen] ‘they (FEM.) annoy you’ b. [il 'pra:to] ‘the lawn’ [a t 'krEden] ‘they (FEM.) believe you’
As underscored in (21), the occurrence of a word-initial /s/ + C cluster in both varieties requires the occurrence of a vowel-final form of the clitic, to allow the syllabification of the initial dental fricative. If this is correct, we now have a test for onsethood in Grizzanese which is just as reliable as Italian il/lo selection.
(22) Discovering language x. The test: [l]-deletion / __ # C # $ C (C)- (before onsets) vs. [i]-epenthesis / C # C __ # C $ C- (before non-onsets)
As apparent in (22), [l]-deletion applies before word-initial onsets, whereas [i]-epenthesis applies before word-initial non-onset consonants, which have to be resyllabified as coda consonants. And to do so, they need a preceding vowel to be inserted. Now let us try to apply test (22) to the remaining onset candidates in (13). Not unexpectedly (and as already partly illustrated in (18b) above), the three best candidate onset clusters in (13i–k) trigger [l]-deletion, just like single initial consonants.
14 4 Michele Loporcaro
(23) a. /al t 'pjE:zen/ → [a t 'pjE:zen] 3PL.FEM.SUB. 2SG.OB. please (3PL.) ‘you like them (FEM.)’ b. /al t 'brE:ven/ → [a t 'brE:ven] 3PL.FEM.SUB. 2SG.OB. scold (3PL.) ‘they (FEM.) scold you’ c. /al t klu're:ven la 'fa:sIa/ 3PL.FEM.SUB. 2SG.OB. paint (3PL.IMPERF.) the face → [a t klu're:ven la 'fa:sIa] ‘they (FEM.) painted your face’
This supports the hypothesis that clusters of types (13i–k) are syllabified as complex onsets. There remain the two onset candidates in (13f–g): so- nority plateaus (stop + stop/affricate, nasal + nasal), and clusters of increas- ing sonority with a small sonority distance between the two consonants (obstruent plus nasal, stop plus fricative). The behaviour of these clusters is exemplified in (24).
(24) a. [al ga'lI-n a t 'pke:ven] ‘the chickens pecked you’ b. [al 'me: 'sII- 5i a t 'dgˇe:ven] ‘my aunts told you’ c. [al 'fjo:li a t 'mne:ven] ‘the girls were beating you’ d. [al 'fjo:li a t dman'de:ven] ‘the girls were asking you’ e. [al 'fjo:li a t kman'de:ven] ‘the girls commanded you’ f. [al 'me: 'sII- 5i a t 'kNOsen 'bE˜:] ‘my aunts know you well’ g. [al su'rE:l a t 'pse:ven 'vEder] ‘the sisters could see you’ h. [aL ur'ti:g a t 'psIi:gen] ‘the nettle (F.PL.) makes you itch’
As apparent in (24), [l]-deletion takes place in this case.
3.2. Grizzanese vs. Italian onset structure: a parametric account
As we have seen, Grizzanese happens to match the description of language x, as hypothesized above in (9). In this language, sonority plateaus and all clusters of increasing sonority (including those with a small sonority dis- tance) qualify as true onsets according to test (22). (The other test for onset- hood mentioned in §2, viz. raddoppiamento fonosintattico, is not available for Grizzanese, since this variety has lost geminates altogether, like all of WRom.) All this is perfectly explicable without any ad hoc adjustment of the data (especially of the underlying representations). Given a universal sonority
On possible onsets 145 hierarchy, and universal preference laws for syllable structure like those proposed in Vennemann (1988), this dialect makes a more extensive choice than Standard Italian (and many other languages) in selecting both a) con- sonant clusters which are allowed word-initially; and b) word-initial clusters which are syllabified as onsets (where the latter are a subset of the former). (25) illustrates that this is far from fortuitous.
(25) [S = fricative, T = stop, N = nasal, R = rhotic, J = glide] a. b. c. d. e. f. g. h. i. j. k. *JT- *RT- *LT- NT- ST- T1T2- TS- TN- TL- TR- TJ- Grizzanese *JT- *RT- *LT- *NT- ST- (T1T2-) (TS-) (TN-) TL- TR- TJ- Italian non-occurring non-onsets onsets worse better
Thus, the contrast between Italian and Grizzanese reduces to a different choice along a scale defined by the parameters included in the Head Law (10). These parameters determine implicational preferences for the well- formedness of onsets, which are based on the consonantal strength relation between segments in prenuclear position. Such an account is possible, of course, only under a Galilean perspective. The rule-based criterion (1b) is what is really central to the discussion of syllable structure: this is the ex- periment which, in the Galilean style, gives us a chance to define operation- ally (viz. in an empirically measurable way, as opposed to metaphysically) what is an onset. Both the distributional criterion (1a) and the SSG-based criterion (1c), on the other hand, prove to be hierarchically subordinate: (1b) > (1c) > (1a). They have to be understood, in my opinion, as probabilistic preference statements, pointing to default options.
4. Cross-linguistic variation in syllable structure
The analysis presented in §3 provides a straightforward account of Grizzanese syllabification, as well as an explanation for two specific seg- mental processes which are sensitive to syllable structure. It also param- eterizes in a principled way the syllabification contrasts between Grizzanese and Italian. Of course, alternative accounts readily come to mind. Based on the discussion in §2 above, for example, a different solution such as that proposed by Rubach & Booij for Polish (cf. (7)) might be considered. Note, however, that such a move would not be of much use in the present case. If it were maintained that, for all C1C2 clusters not complying with the SSG (viz. at least (13e–f)), C1 is irrelevant for syllabification, then the systematic
146 Michele Loporcaro contrast observed in word-initial sonority reversals vs. sonority plateaus with respect to [i]-epenthesis and [l]-deletion would be left unexplained. One basic presupposition of the present account is that languages may differ parametrically with respect to important properties of phonological representation such as syllabification. After all, languages differ enormously in the number and type of segments they allow in their phonological inven- tories, ranging from only half a dozen voiceless obstruents plus a few sono- rants to several series of consonants produced with differentiated initiatory mechanisms. There is no reason to expect that this huge variation can be reduced to underlying uniformity; rather, it must be taken into account and it should form the object of phonological typology. The same is true of syllable structure, in my view: languages diverge empirically with respect to syllabification, and this has to be accounted for on a typological scale, as has in fact been done, through implicational generalizations (see e.g. Vennemann, 1988). To this ‘realistic’ conception of syllable structure, a radical alternative can be proposed, which is rather ‘idealistic’ in nature; according to this latter conception, cross-linguistic diversity boils down to mere surface effects, whereas underlying syllable (constituent) structure is assumed to obey universally invariable principles. Take Grizzanese and Italian. In spite of the fair amount of surface diver- gence exemplified above in §3, phonologists assuming an “‘enriched pho- netic representation’ which includes entities like empty positions” (Kaye et al., 1990: 229) would prefer to maintain that syllable structure is indeed the same underlyingly in the two languages (and in all languages, for that mat- ter). The question is now an empirical one: it must be decided which of the two alternative views allows for the most economical account of the facts described (in particular, of [i]-epenthesis and [l]-deletion in Grizzanese, as opposed to il/lo selection in Italian). Within the classical framework of Government Phonology (GP), as ex- pounded in Kaye et al. (1990), strings such as those in (25f–h) cannot possibly $ be analysed as complex onsets. By definition, a branching onset ( C1C2-) must consist of a negatively charmed segment plus a charmless segment: any other combination of consonants (e.g. stop + stop/affricate/fricative/ nasal) is ill-formed. Consequently, if word-initial strings like [dn], [pk], etc. occur, they cannot, on any account, be analysed as onsets.9
9 As for the exclusion of onsets such as [pk], the principle constraining the structure of branching onsets in GP makes predictions similar to those derivable from the SSG-based criterion (1c). The predictions are even stronger, in that clusters with small sonority distance such as [dn]-, [ps]- are not ruled out per se by (1c), whereas they are in GP.
On possible onsets 147
If (25f–h) are not onsets, the fact that they pattern with CV- initial strings as to the empirical properties described in §3 (viz. [i]-epenthesis and [l]-deletion) has to be explained some other way. This could be achieved, for instance, by assuming the representations in (26).
(26) a. O R O R O R O R b. O R O R O R N N N N N N N x x x x x x x x x x x x x x x p 0v0 k e: v e n v E d e n
c. O R O R O R N N N
x x x x x x x s t o f e n
According to (26a), initial [pk] in ['pke:ven] is not a branching onset, given that an empty nucleus intervenes between the two initial consonants. This empty nucleus is not spelled out phonetically because it is properly gov- erned. Consequently, given the representations (26a–b) we have no more reason to expect an epenthetic vowel to crop up before initial [pk] (in (26a)) than before an initial single consonant (in (26b)), since in both cases the word begins with a non-branching onset. The two segments within a word- initial [s]C cluster, on the other hand, are in a well-formed interconstituent relationship; thus, no empty nucleus has to be assumed to occur between them. Therefore, initial /s/ is not an onset consonant and [i]-epenthesis is needed for /s/ to be syllabified with the vowel inserted to its left. Let us now compare the two alternative accounts. I assume here that an explanatory hypothesis is to be preferred over an alternative one, firstly, if it makes empirically testable predictions (i.e. predictions with empirical content); and secondly, if it is corroborated by the empirical data available. Both conditions are satisfied by the account in §3. If syllabification obeys preference laws based on consonantal strength and empty constituents are not allowed in the underlying representation, then it is predicted, in our specific case, that if word-initial [s]C clusters behave as non-onsets, [m]C clusters will do the same a fortiori. The prediction is borne out, as shown in (27).
(27) a. [al 'spi:g al si 'mde:ven 'bE˜:] ‘the ears of corn were harvested easily’ b. /al s 'mde:ven/ → [al si 'mde:ven] ‘they (FEM.) were harvested’ c. /al s 'stofen/ → [al si 'stofen] ‘they (FEM.) get annoyed’
148 Michele Loporcaro
If, on the other hand, epenthesis took place in (27c) but not in (27a–b) (*[a s 'mde:ven]), this would constitute a serious problem for the present account: it would imply that [i]-epenthesis would no longer be conditioned by syllable structure. Conversely, given a representation including empty nuclei, there would be no cogent reason to exclude [m]C- from behaving like [pk]-, [dn]- but differently from [s]C-. Even worse, [i]-epenthesis could still be explained as syllable-driven, even in such a bizarre situation: it would suffice to assume that an empty nucleus occurs between /m/ and /d/ in ['mde:ven] in (27a–b), a possibility which, as far as I can see, cannot be ruled out on principled grounds.10 In short, one might suspect that a theory allowing for empty constituents could account for any arbitrary set of data, in a rather unconstrained way.11 But return finally to our empirical question. The above conclusions should, of course, be revised if compelling empirical evidence for the pos- tulated occurrence of an empty nucleus in (26a) could be found. And there actually seems to be such evidence. If empty nuclei are defined as phonetic zeroes which alternate with full vowels, then this is actually the case for the empty nucleus in (26a). As illustrated in (28a), the latter does alternate with a full vowel, which is found when the same lexical morpheme occurs under stress; the form in (26c), on the other hand, has no alternates with a vowel occurring between /s/ and the following stop (see (28b)).
10 It is true that an empty nucleus does not necessarily have to occur between /m/ and /d/, since the two segments are in a well-formed interconstituent relation. However, nothing can prevent [md]- from being analysed alternatively as a sequence of two onsets in a row. 11 To quote Scheer’s (this volume) humorous formulation, it could just as well account for worlds where peaches fall upwards. As far as I can see, the argument developed here ap- plies a fortiori to Lowenstamm’s (1996) strict CV-hypothesis. One anonymous referee, while agreeing that “standard [GP] encounters a problem in analysing the Grizzanese pattern”, objects that I “hastily conclude that the same criticism can be levelled against a strict CV account”. While details remain to be worked out, she argues, the starting empirical observa- tion on which a strict CV account could elaborate is the following: “the clusters not permit- ted word-initially are exactly those the mirror image of which would be possible complex onsets”. While this is true, it follows directly from the head law (10) under my approach: if C1C2 is a bad onset, due to sonority decrease, C2C1 has to be a good one. The crucial point here concerns the treatment of obstruent + fricative/nasal clusters (13g-h). They qualify as onsets by the rule-based criterion (1b) (see (24d-h)), but within GP they do not (see fn. 9). This leads to an incorrect prediction, in this case. On the other hand, revised versions of GP allowing for rightward governing within such clusters (which is tantamount to complex onset) would simply yield a prediction empirically indistinguishable from that derived via (10), a universal preference statement, combined with a universal sonority hierarchy.
On possible onsets 149
(28) Independent evidence for the empty nucleus (?) a. ['pke:ven] ‘(they) pecked (3PL.)’ ~ *['bE:ka] ‘pecks’ b. ['stofen] ‘(they) annoy (3PL.)’ ~ *['sVt-]
However, closer inspection shows that the existence of a C1C2-/C1VC2- alternation does not really correlate with the behaviour of initial consonant clusters with respect to [i]-epenthesis.
(29) a. /al v 'vde:ven/ → [al vi 'vde:ven] ‘they (FEM.) saw you (PL.)’ b. /al s 'mde:ven/ → [al si 'mde:ven] ‘they (FEM.) were being harvested’ (30) a. ['vde:ven] ‘saw (3PL.)’ ~ ['vEden] ‘see (3PL.)’ b. ['mde:ven] ‘harvested (3PL.)’ ~ ['me:den] ‘harvest (3PL.)’
As shown in (30), the verbs occurring in (29) have both C1C2- and C1VC2- alternates, as in (28a). Consequently, if such an alternation is considered evidence for an empty nucleus, the [C1C2-] sequences in (29) must be ana- lysed as underlying /C1v°C2-/. But this representation predicts the non- occurrence of epenthesis before [vd]- and [md]- in (29), a prediction which is not borne out by the data. Conversely, the analysis proposed in §3 accounts for the fact that epenthesis does take place here. Both /st/, on the one hand, and /vd/-, /md/-, on the other, qualify as non-onsets according to test (22). This is so for purely phonological reasons. They all show the same sonority shape; they are sonority reversals. The absence, in (28b), vs. presence, in
(30), of C1VC2- alternants plays no role in determining the behaviour of the initial clusters C1C2- with respect to syllabification and syllable-driven rules. This lack of correlation is finally confirmed by the fact that many clusters of categories (13f–h), although paralleling /pk-/ in that they do not trigger epenthesis, have no CVC- alternates either. The mutual independence of the two factors is graphically displayed in (31), where it is shown that all possible combinations occur. (In particular, the second column exemplifies the behaviour of verbs whose initial consonant clusters belong to class (13h).)
(31) sonority plateaus and sonority reversals small sonority distance ['pke:ven]['kNOsen] ['stofen]['vde:ven] a. alternates with C1VC2-: yes no no yes b. triggers [i]-epenthesis: no no yes yes
As is apparent from (31), empirical evidence for a putative role of empty nuclei cannot be found in the phenomena at hand. In the absence of such empirical evidence, the analysis proposed here, which does not require any
150 Michele Loporcaro ad hoc adjustment to the representation of the data involved, remains more economical and should be preferred.
Acknowledgements
I would like to thank W.U. Dressler, J. Kaye and J. Lowenstamm for com- ments on the oral presentation of this paper and, more generally, for stimu- lating discussion. I am also indebted to P.M. Bertinetto, with whom I have discussed at length the topics dealt with here, and to two anonymous referees. I am of course responsible for any errors and inadequacies which may have persisted beyond their remarks.
References
Bertinetto, P.M. (1996). La sillabazione dei nessi /sC/ in italiano: Un’eccezione alla regola generale?. Ms., Scuola Normale Superiore, Pisa. Blevins, J. (1995). The Syllable in Phonological Theory. The Handbook of Phonological Theory, edited by John A. Goldsmith, 206–244. Cambridge, Mass.: Blackwell. Camilli, A. (1913). I rafforzamenti iniziali in italiano. Archiv für das Studium der neueren Sprachen und Literaturen 131, 170–174. Camilli, A. (1944–45). Articoli, pronomi, preposizioni articolate. Lingua Nostra 6, 89–90. Castellani, A. (1961). Sulla formazione del tipo fonetico italiano. Fenomeni vocalici. Studi linguistici italiani 2, 24–45. Chomsky, N. & M. Halle (1968). The sound pattern of English. New York: Harper & Row. Clements, G.N. (1990). The role of the sonority cycle in core syllabification. Papers in Laboratory Phonology, edited by J. Kingston & M.E. Beckman, 283–333. Cambridge: Cambridge University Press. Davis, S. (1990). Italian onset structure and the distribution of il and lo. Linguistics 28, 43–55. Dressler, W.U. (1984). On the definite Austrian and Italian articles. Phono-Morphology, edited by E. Gussmann, 35–47. Lublin: Catholic University. Dressler, W.U. (1985). Morphonology: the dynamics of derivation. Ann Arbor: Karoma. Dressler, W.U. (1988). Preferences vs. strict universals in morphology: word-based rules. Theoretical morphology. Approaches in modern linguistics, edited by M. Hammond & M. Noonan, 143–154. San Diego: Academic Press. Dziubalska-Kołaczyk, K. (1995). Phonology without the syllable: A study in the natural framework. Poznan´: Motivex. Fava, E. & E. Magno Caldognetto (1976). Studio sperimentale delle caratteristiche elettroacustiche delle vocali toniche e atone in bisillabi italiani. Studi di fonetica e fonologia (SLI 9), edited by R. Simone, U. Vignuzzi & G. Ruggiero, 35–79. Rome: Bulzoni. Kaye, J., J. Lowenstamm & J.-R. Vergnaud (1990). Constituent structure and government in phonology. Phonology 7, 193–231. Kuryłowicz, J. (1948). Contribution à la théorie de la syllabe. Biuletyn Polskiego Towarzystwa Je˛zykoznawczego 8, 80–114 [Also in: Kuryłowicz, J.1973. Esquisses linguistiques I, 193–220. Munich: Fink]. Loporcaro, M. (1997). L’origine del raddoppiamento fonosintattico: Saggio di fonologia diacronica romanza. Basel: Francke.
On possible onsets 151
Lowenstamm, J. (1996). CV as the only syllable type. Current Trends in Phonology: models and methods, edited by J. Durand & B. Laks. Vol. 2, 419–441. Salford: European Studies Research Institute, University of Salford. Marotta, G. (1993). Selezione dell’articolo e sillaba in italiano: un’interazione totale?. Studi di grammatica italiana 15, 255–296. Rubach, J. (1996). Nonsyllabic analysis of voice assimilation in Polish. Linguistic Inquiry 27, 69–110. Rubach, J. & G. Booij (1990). Syllable structure assignment in Polish. Phonology 7, 121–158. Scheer, T. (this volume). A theory of consonantal interaction. 289–304. Selkirk, E.O. (1984). On the major class features and syllable theory. Language sound structure, edited by M. Aronoff & R.T. Oehrle, 107–136. Cambridge: MIT Press. Vennemann, T. (1988). Preference laws for syllable structure and the explanation of sound change. Berlin: Mouton de Gruyter.
The beginning of the word
Jean Lowenstamm
Abstract
In this paper an argument is made for the existence of an initial CV site to the left of every major category. In section 1, I will first discuss a noted asymmetry in the organization of word-initial consonant sequences. In section 2, the issue of word boundaries will be briefly revisited. Section 3 contains an answer to the questions raised in section 1. Finally, section 4 is a case study of definite article cliticization in French, and failure thereof in Biblical Hebrew.
The Hague: Thesus. 153–166. Thesus.The Hague: 153–166. 0. .
Syllables !? 1. Word-initial clusters
It has long been observed that some languages tolerate at most, one single consonant at the left margin of words, whereas other languages tolerate two, or more.1 Kusaiean, a language spoken in Kusaie, one of the eastern Phonologica 1996: Phonologica 1996: Caroline Islands (Lee, 1975; 1976), or Ticuna, a language spoken in Colombia, Peru and Brazil (Montes Rodriguez, 1995), belong to the first type. English, or Maghribi Arabic, belong to the second type. It has, moreover, been ob- served that an implicational universal obtains, whereby a language tolerat- ing a sequence of two word-initial consonants will tolerate single word-initial consonants (but, of course, not vice versa). The two types of languages just described appear to exhaust the attested cases. Indeed, to the best of my knowledge, no third type has been documented, whereby a language would tolerate two word-initial consonants while at the same time ruling out single consonants word-initially. In turn, two types can be recognized amongst languages tolerating word- initial consonant sequences. One type strictly enforces an arrangement,
1 sCC clusters will not be dealt with in this paper. See Harris (1994), Kaye (1992), Lowenstamm
John R. Rennison & Klaus Kühnhammer, eds. (1999). eds. R. Rennison & Klaus Kühnhammer, John (1999). (1981) for discussion of their special status.
154 Jean Lowenstamm where the least sonorous consonant must appear to the left of the more sonorous one. Thus, English has blue, and true but not *lbue, *rtue, etc. The other type allows both sequences of increasing sonority and sequences of decreasing sonority in word-initial position. Here too, an implicational universal has been recognized: if a language displays word-initial sequences of decreasing sonority, then it displays word-initial sequences of increasing sonority (of course, not vice-versa). To be sure, no language has been re- ported to display the mirror image of the type exemplified by English, that is a language with word-initial sequences restricted to the decreasing sonority type. The last two types are documented in (1), with English displaying familiar [obstr.][liquid] clusters, and Moroccan Arabic exhibiting similar sequences, but also their mirror images.
(1) English Moroccan Arabic obstr.–liquid obstr.–liquid liquid–obstr. brick brId ‘cool down’ rbIt. ‘bind’ dream d. rIb ‘hit’ rd. a ‘accept’ glow glI! ‘remove’ lga ‘find’
In fact, the range of word-initial sequences allowed in Moroccan Arabic extends well beyond the set defined in (1), including types of sequence that are not tolerated in English: sequences of stops, obstruent-nasal sequences, etc. each with their mirror image, e.g. bka ‘cry’ vs. kbIr ‘grow larger’; nzIl ‘descend’ vs. zna ‘commit adultery’; dna ‘come near’ vs. ndIm ‘regret’; bqa ‘stay’ vs. qbIl ‘accept’. The state of affairs in Semitic, only partially documented above, has led to the view put forward in Kaye, Lowenstamm & Vergnaud (1990), where templatic languages imposing no sequential restrictions on clustering are to be analysed as having only open syllables.2 I assume here the results of Lowenstamm (1996), an attempt to generalize this view to all languages.3 The basic mechanism involved is Proper Government (henceforth PG), as shown in (2) with the English example brick.
(2) C iVi C jVj C V
b 0 r I k 0
2 See Angoujard (1982) for pioneering work in this area. 3 See also Bendjaballah (1995), Boyé (1996), Goh (1996), Larsen (1994), Scheer (1996), Ségéral (1995).
The beginning of the word 155
In (2), Vi, the empty vocalic position between b and r, is licensed by PG from Vj, the phonetically identified vocalic position to its right and can therefore remain phonetically unidentified.4 This example illustrates how PG correctly handles all the English and Moroccan Arabic cases in (1). However, an entire area remains uncovered by this account. Clearly, the right predictions are made concerning the behaviour of the vocalic positions entering into the PG relation. The establishment of a PG relation also pro- vides the formal means for surface clustering to occur. However, since PG is defined only with reference to vocalic positions adjacent on the nuclear projection, it has nothing to say about the nature of the consonants flanking the governee. That is, under PG, no particular order of those consonants is expected. Thus, everything being equal, Moroccan Arabic with its free order is the expected case. Accordingly, the absence of the hypothetical class of languages alluded to above, where clusters of decreasing sonority would be the only type allowed, becomes a non-issue: why should such an outlandish restriction obtain? More generally, why should any restriction obtain at all? However, by the same token, the strict ordering actually evidenced by English and many other Germanic and Romance languages becomes myste- rious, indeed no less outlandish, than the hypothetical, unattested case just described.5
2. Word edges, word boundaries
An important notational convention of Chomsky & Halle (1968), henceforth SPE, is the assignment in certain places of the boundary symbol #. Thus, at some point in the derivational history of a syntactic string, an item, say the noun kit, becomes associated with occurrences of that symbol to its right and to its left as in (3).
(3) #kit#
4 I adopt the proposals of Kaye (1990) for the licensing of the final empty nucleus, a topic not further discussed in this paper. 5 It has been pointed out to me that the peculiar form of clustering displayed in Germanic and Romance follows, in straightforward fashion, from a theory of syllable structure coun- tenancing branching onsets. Thus, the inability of a strict CV framework to recognize branching onsets must, supposedly, be viewed as a serious, inherent flaw of such a pro- gram. Of course, the objection evaporates when it is realized that the only argument for branching onsets is precisely their being the seat of the restriction under discussion (see Blevins, 1995 for discussion).
156 Jean Lowenstamm
Kahn (1976), in his important dissertation, observed that the expression C occurs in the environment of many phonological rules. This can be # illustrated by a process of monophthongization in Yiddish. In most dialects of Yiddish, a diphthong will shorten to a monophthong before two con- sonants in items belonging to the Hebrew-Aramaic component of the lexi- con.
(4) soyxIr ‘merchant’ soxrIm (*soyxrIm) ‘merchants’ soyfIr ‘scribe’ sofrIm (*soyfrIm) ‘scribes’
The same phenomenon can be observed before a word-final consonant.
(5) dor ‘generation’ doyrIs (*doyr) ‘generations’ sod ‘secret’ soydIs (*soyd) ‘secrets’
In standard SPE notation, the phenomenon would be encoded as in (6).
C (6) Vy → V / ___ C #
Kahn also noted that # shares virtually no properties with the class of segments with which it so often co-occurs in such expressions. In an at- tempt to solve this paradox (viz. the popularity of such rules and the highly C unnatural character of their environment), Kahn submitted that is but a # notational artefact standing in reality for the syllable boundary. While Kahn's proposal led to the recognition of the syllable in post-SPE phonol- ogy, thereby significantly reducing the number of occurrences of # in the rule component of individual grammars, the need for # was not evacuated altogether. Indeed, a number of phenomena are characteristic of left word edges, although not of word-internal onsets. For instance, reference to word-initial position must crucially be allowed for the statement of syn- chronic allophonic selection. This point can be illustrated with an example from Chaha, an Ethio- Semitic language of the Gurage group. In Chaha, the segment /r/ is barred from word-initial position, where it is regularly replaced by its strong alter- nant /n/. Thus, all the Perfect forms of verb näkyäräm ‘win in a lawsuit’ are n-initial, whereas the Present and Jussive conjugations faithfully restore the original form of the root-initial consonant /r/ in their prefixed forms.
The beginning of the word 157
(7) Perfect Present Jussive y w y 1SG.COM. näk ärx Im äräk Ir nIräkIr y y 2SG.MASC. näk ärxäm tIräk Ir näkIr y y 2SG.FEM. näk ärçIm tIräk i näki y y 3SG.MASC. näk äräm yiräk Ir yäräkIr y y 3SG.FEM. näk äräcIm tIräk Ir tIräkIr y y w IMP. näk ärim yiräk ri yäräk ri y y 1PL.COM. näk ännäm nIräk Innä nIräkInnä y y 2PL.MASC. näk ärxum tIräk ro näkro y y 2PL.FEM. näk ärxImam tIräk räma näkräma y y 3PL.MASC. näk ärom yiräk ro yäräkro y y 3PL.FEM. näk ärämam yiräk räma yäräkräma
Another well-known example from the history of Biblical Hebrew which highlights the special status of the left word edge can be quoted from Gesenius (1910: 186): In consequence of a phonetic change which prevails also with few ex- ceptions in the noun, w in Hebrew and Aramaic always becomes y, at least when it is the initial consonant; but after preformatives it […] re- appears Gesenius (1910: 186) A third example of a phenomenon taking place only word-initially, is the palatoalveolarization of pre-consonantal /s/ in Standard German, e.g. streng, Stadt, spazieren vs. Astrid, erste, Wespe. Building on the representational format of Lowenstamm (1996), I offer in (8) an alternative to the traditional representation #kit#.
(8) C V C V C V k i t 0
Rather than being conventionally marked by the insertion of a # symbol to its left, the word is preceded by an empty CV span. The major difference between this proposal and the traditional view lies in the fact that the initial empty CV span is a true phonological site, over which a number of opera- tions will be shown to take place, or in terms of which a number of generali- zations will be shown to receive expression. In addition, I adopt the revised version of PG put forward in Scheer (1996). In the context of what must be the most comprehensive discussion of vowel/0 alternations in Czech, Scheer examines the behaviour of prefixes
158 Jean Lowenstamm such as od/ode, which can be loosely glossed as ‘away’. The selection of the long and short forms of the prefix is partially illustrated in (9).6
(9) a. ode-psat (*od-psat) b. od-pisovat (*ode-pisovat) c. od-blanit (*ode-blanit)
Cases (9a–b) are perfectly straightforward, as shown in (10a–b) respec- tively: an ungoverned prefix-final V position must be phonetically identified (10a), and a governed prefix-final V position remains silent (10b).
(10) a. C V C V - C V C V C V b. C V C V - C V C V C V C V o d e p 0 s a t o d 0 p i s o v a t
More problematic is (9c). Under the standard view, the vocalic position realized as a (bold in (11a)) governs the vocalic position to its left. Con- sequently, the final V position of the prefix remains ungoverned and should surface as [e]. Addressing this challenge, Scheer (1996) develops an in- dependently motivated account of consonant interaction such that a se- quence consisting of an obstruent followed by a liquid constitutes a closed domain.7 As a result, the position identified by a (11a) in od-blanit governs the final vocalic position of the prefix, as shown in (11b).8
(11) a. C V C V - C V C V C V C V b. C V C V - C V C V C V C V
o d ?0? b 0 l a n i t o d 0 [b[ 0 ]l] a n i t
The descriptive potential of this revised model will be illustrated in the form of a second pass at the issue raised in §1.
6 While the examples in (9) illustrate the point at hand in this paper, they hardly do justice to the complexity of the overall situation in Czech, or to the wealth of material dealt with in Scheer (1996). Readers with a closer interest in the analysis of Czech are urged to reserve judgement on these results until Scheer (1996) has been directly consulted. 7 This idea is reminiscent of a proposal put forward in Hirst (1985). 8 Certain aspects of Scheer's analysis are reminiscent of the proposals put forward by Gussmann & Kaye (1993) for Polish.
The beginning of the word 159
3. Word-initial clusters, again
In this section, we return to the challenge raised by the shape of word-initial strings of consonants, namely: why are the options limited to the strict increasing sonority type, or to the entirely unordered type? Let us first consider the case of a language of the first type, French. French word-initial configurations of the type discussed in this paper, i.e. with the exception of sC(C) clusters, are exhausted by the two cases exem- plified in (12), by a ‘single’ onset in (12a) and by a ‘branching’ onset in (12b).
(12) a. tapis [tapi] ‘rug’ b. plateau [plato] ‘tray, plateau’
The representations of these forms, in the terms advocated here, are given in (13).
(13) a. C V C V C V b. C V C V C V C V
t a p i s [p[ 0 ]l] a t o
In both cases, the first vowel (bold) governs the V position of the initial site. Consequently, the generalization in (14) obtains for French, and lan- guages of the same type.
(14) Initial site always licensed in French
Consider now the case of Biblical Hebrew, a language of the second type. A number of roots appear in the first column of (15), along with the singulars and plurals of the segholate nouns derived from those roots.
(15) Root Singular Plural a. √klb kéleß ‘dog’ k0laßim ‘dogs’ b. √lkd léxeð ‘capture’ l0xaðim ‘captures’ c. √qrb qéreß ‘midst’ q0raßim ‘midsts’ d. √rqh- réqah- ‘spice’ r0qah- im ‘spices’
The careful reader will have noticed that the first two consonants (bold) in the roots of (15a–b), and those of (15c–d), are mirror images of each other, viz. /kl/ vs. /lk/ and /qr/ vs. /rq/. The plurals in the segholate class are de- rived by syncope of the initial vowel, thus blindly giving rise, in this case, to surface sequences of obstruents and liquids of the strictly unordered type qr, rq, etc. Representations of those forms appear in (16).
16 0 Jean Lowenstamm
(16) a. obstr.–liquid b. liquid–obstr. [k[ 0 ]l] a b i m l 0 k a d i m
C V C V C V C V C V C V C V C V C V C V
[q[ 0 ]r] a b i m r 0 q a -h i m
In (16a), the obstruent and the liquid occur in such an order that the first vowel can govern the V position of the initial site. In (16b), on the other hand, the order of the first two consonants is such that government of the inter- vening empty vocalic position is required. As a consequence, in this case, the V position of the initial site remains ungoverned. Hence, the generaliza- tion in (17) can be said to obtain of Biblical Hebrew, and languages display- ing initial consonant sequences of the same ilk.
(17) Initial site “not always” licensed in Biblical Hebrew
The claim made here is that the government status of the initial site pro- vides a rationale for the mysterious state of affairs under scrutiny: lan- guages strictly abiding by the increasing sonority sequencing generalization will display a systematically licensed initial site, and languages oblivious to that generalization will not. It can readily be seen that no third option is available in terms of the pro- posed rationale, i.e. that the empirically correct dichotomy opposes “always licensed by government” and its logical negation “not always licensed by government”. The obvious alternative dichotomy might oppose languages in which the initial site is always licensed by government and those in which it is never licensed by government. Under what conditions would the initial site never be licensed by gov- ernment? Surely, a language that displayed only word-initial sequences of decreasing sonority, would meet such a condition. In such a language, sequences of liquids and obstruents would appear in just that order, say rdab, lbad, etc. to the exclusion of the reverse order, *drab, *blad. Words of that type would exhibit an unlicensed initial site as shown in (18), where /a/ governs the empty vocalic position between /l/ and /b/, leaving the V of the initial site ungoverned.
(18) Initial site status: unlicensed l 0 b a d
C V C V C V C V
The beginning of the word 161
It could then be said of such a language, that the initial site of its cluster- initial words is never licensed by government. However, for the initial site never to be licensed by government across the entire lexicon of that hypo- thetical language, an additional condition would have to be met, namely (19).
(19) No lexical item may display an initial sequence of a consonant fol- lowed by a vowel
Unless such a condition were met, items such as say dab, tap, etc. would exist, alongside *lbad, *rdab, etc., the former exhibiting an inevitably licensed initial site, as shown in (20).
(20) Initial site status: licensed C V C V C V
t a p
In actual fact, no language has been reported to enforce a requirement such as (19). I conclude that a state of affairs where the initial site would never be licensed is unsubstantiated, and that the empirically correct di- chotomy opposes languages in which the initial site is always licensed, and those in which it is not always licensed. Consequences of that dichotomy are examined in the next section.
4. Clitics
In this section, I intend to argue for the thesis in (21).
(21) a. The initial CV site is the site of cliticization. b. Cliticization can take place iff the site is licensed.
The discussion in this section will be limited to definite article cliticization in French and Biblical Hebrew. However, (21) includes in its scope pronominal cliticization to the left of verbal forms. Accordingly, several cautionary comments are in order before (21) can be illustrated. First, cliticization can be construed as a syntactic phenomenon or as affixation, two approaches to which the proposal put forward here is neutral. (For an overview and discussion, see e.g. Miller, 1992; Van Riemsdijk, in press, and references therein.) Indeed, (21) is to be viewed as an attempt at sketching out a subset of the factors conditioning cliticization, namely the morphophonological conditioning. Accordingly, it is to be expected that overriding syntactic or morpholexical considerations may block cliticization even under
162 Jean Lowenstamm morphophonologically favourable circumstances. Consequently, failure of cliticization to take place under such conditions will not be construed as necessarily problematic for (21). On the other hand, cliticization is not expected to take place under unfavourable morphophonological con- ditions, viz. when the site is unlicensed. Accordingly, cliticization occurring in spite of such unfavourable conditions would have to be construed as a possible source of refutation of (21). Second, the discussion below is re- stricted to clitics assuming a CV shape in the sense of the French data soon to be adduced. Clitics of a shape larger than CV will be dealt with elsewhere. French definite articles are reputed to cliticize onto the nouns they de- termine. Examples appear in (22).
(22) a. Before “single initial onset” words le + tapis ‘the rug’ → [l7tapi] la + tasse ‘the cup’ → [latas] les + tapis ‘the rugs’ → [letapi] b. Before “initial branching onset” words le + plateau ‘the tray’ → [l7plato] la + place ‘the place’ → [laplas] les + plateaux ‘the trays’ → [leplato] c. Before initial empty onset words le + ami ‘the male friend’ → [lami] la + amie ‘the female friend’ → [lami] les + amis ‘the friends’ → [lezami]
The respective positioning of the ingredients of cliticization just prior to its operation, I claim, are as shown below in (23). That is, the candidates to cliticization, le, la, les approach the item they are about to cliticize onto, from the left.
(23) a. C V + C V C V C V b. C V + C V C V C V C V
l 7 t a … l 7 [p[ 0 ]l] a … l a l a l e z l e z c. C V + C V C V C V
l 7 a m i l a l e z
The beginning of the word 163
The status of the initial site (licensed/unlicensed) undergoes checking. As the initial site is always licensed in French, cliticization successfully takes place in the form of a movement of the candidate item into the site, as in (24).
(24) a. C V + C V C V C V b. C V + C V C V C V C V
l 7 t a … l 7 [p[ 0 ]l] a … l a l a l e z l e z c. C V + C V C V C V
l 7 a m i l a l e z
The deserted CV, to which the article was originally attached, now stands ready to host a further candidate for cliticization if one is queuing up.9 In the absence of candidates, it will simply wither away. We now turn to the behaviour of Biblical Hebrew definite articles. Con- sider the ingredients involved in the combination of the definite article /ha/ and some of the plurals of the segholate class discussed earlier.
(25) ha + k0laßim ‘the dogs’ ha + r0qah- im ‘the spices’
Representationally, the configurations involved, shown in (26), are very similar to the French analogues in (23), prior to cliticization.
(26) a. h a [k[ 0 ]l] a b i m
C V + C V C V C V C V C V
b. h a r 0 q a h- i m
However, the status of the initial CV site in Biblical Hebrew differs from its French counterpart. Whereas the site is always licensed in French, it is not always licensed in Biblical Hebrew. I assume here the convention in (27) for the interpretation of a language such as Biblical Hebrew.
9 This is essentially the mechanism behind unbounded cliticization, e.g. je ne te le donnerai pas [I NEG you it give-will NEG] ‘I won't give it to you’.
16 4 Jean Lowenstamm
(27) The Uniformity Convention For any given language, cliticization operates in uniform fashion with respect to the licensing status of the host site
The intended meaning of (27) can be illustrated by a comparison of (26a) and (26b). In (26a), owing to the specific ordering of the obstruent and the liquid, the site is governed; by contrast, for the reverse reason, the site is ungoverned in (26b). However, according to (27), checking does not operate item by item.10 Rather, if the initial CV is not be licensed in all cases, then the site must remain unlicensed throughout the language. Accordingly, definite article cliticization in Biblical Hebrew is blocked across the board. A phonological consequence ensues: the manifestation of the initial site. Whereas an empty, uninterpreted initial CV normally remains silent, the initial CVs in the Biblical Hebrew cases under consideration are now stuck in the midst of a phonological word. As such, they have become part and parcel of that word, and they must undergo phonetic interpretation. Two strategies are available. The first, more popular strategy involves gemina- tion of the root initial consonant, /k/ in (28a); the second strategy, activated when the root initial segment is one of the five consonants resisting gemina- tion, involves compensatory lengthening of the definite article's vowel, (28b).11
(28) Gemination (=a.) vs. compensatory lengthening (=b.) a. h a [k[ 0 ]l] a b i m h a k 7 l a b i m
C V + C V C V C V C V C V → C V + C V C V C V C V C V
b. h a r 0 q a h- i m h a r 7 q a h- i m
5. Conclusion
In this paper, a number of arguments have been put forward to the effect that an empty CV site precedes every major category. The postulated site rationalizes the organization of word-initial consonant sequences. More- over, it accounts for the differential behaviour of definite articles in lan- guages such as French and Biblical Hebrew. Further work on Korean
10 Evidently, Biblical Hebrew enforces a version of the Uniformity Condition. While I am not familiar with languages in which checking would operate in an item-by-item fashion, such an option cannot be ruled out in principle. 11 This phenomenon is described in every grammar of Hebrew, see Bauer & Leander (1922), Ewald (1870), Gesenius (1881), Joüon (1923), and Lowenstamm & Kaye (1986) for discussion.
The beginning of the word 165 compounds, the Berber construct state, Fulani nominal classes, sC(C) clusters in French, German and Spanish, Romance clitics, etc., is expected to yield additional evidence for the hypothesized site.
Acknowledgements
I am grateful to Jonathan Kaye and Tobias Scheer for discussion of issues dealt with here, to David Odden and Curt Rice for detailed comments following the . oral presentation in Vienna, to Wiebke Brockhaus, Grazyna Rowicka, Peter Szigetvári and Shohei Yoshida for comments on a draft of this paper.
References
Angoujard, J.-P. (1982). Sur la représentation phonologique du verbe [ktib] (parler de Tunis), Analyses, Théorie 2–3, 1–17. Bauer, H. & P. Leander (1922). Historische Grammatik der hebräischen Sprache des Alten Testamentes. Halle a. S.: Verlag von Max Niemeyer. Bendjaballah, S. (1995). Aspects du système verbal du berbère (kabyle). DEA Thesis, Université Paris 7. Blevins, J. (1995). The syllable in phonological theory. The handbook of phonological theory, edited by J. Goldsmith, 206–245. Oxford: Blackwell. Boyé, G. (1996). Aspects du système verbal du népali. DEA Thesis, Université Paris 7. Chomsky, N. & M. Halle (1968). The sound pattern of English. New York: Harper & Row. Ewald, H. (1870). Ausführliches Lehrbuch der hebräischen Sprache des Alten Bundes. Göttingen: Verlag der Dieterichschen Buchhandlung. Gesenius, W. (1881). Hebräische Grammatik. Leipzig: Verlag von F.C.W. Vogel. Gesenius, W. (1910). Gesenius' Hebrew grammar. Second English edition revised in accordance with the twenty-eighth German edition (1909) by A.E. Cowley. Oxford: Clarendon Press. Goh, Y.S. (1996). The segmental phonology of Beijing Mandarin. Doctoral Dissertation, School of Oriental and African Studies, London. Gussmann, E. & J.D. Kaye (1993). Polish notes from a Dubrovnik Café: I. The yers. SOAS Working Papers in Phonetics and Linguistics 3, 427–462. Harris, J. (1994). English sound structure. Oxford: Blackwell. Hérault, G. (1989). Les rections syllabiques en soninké. Linguistique Africaine 3, 43–90. Hirst, D.J. (1985). Linearisation and the single-segment hypothesis. Grammatical representation, edited by J. Guéron, H. Obenauer & J.-Y. Pollock, 87–100. Dordrecht: Foris. Joüon, P. (1923). Grammaire de l'hébreu biblique. Institut Biblique Pontifical, Rome. Kahn, D. (1976). Syllable-based generalizations in English phonology. Doctoral Dissertation, MIT. New York: Garland, 1980. Kaye, J.D. (1990). Coda licensing. Phonology 7, 301–330. Kaye, J.D. (1992). Do you believe in magic? The story of s+C sequences. SOAS Working Papers in Linguistics and Phonetics 2, 293–312. Kaye, J.D. & J. Lowenstamm (1984). De la syllabicité. Forme sonore du langage, edited by F. Dell, D. Hirst & J.R. Vergnaud, 65–123. Paris: Herrmann. Kaye, J.D., J. Lowenstamm & J.R. Vergnaud (1990). Constituent structure and government in phonology. Phonology 7, 193–231.
16 6 Jean Lowenstamm
Larsen, U.B. (1994). Some aspects of vowel length and stöd in Modern Danish. M.A. Thesis, Université Paris 7. Lee, K. (1975). Kusaiean reference grammar. Pali Language Texts: Micronesia. Honolulu: The University Press of Hawaii. Lee, K. (1976). Kusaiean-English dictionary. Pali Language Texts: Micronesia. Honolulu: The Univer- sity Press of Hawaii. Lowenstamm, J. (1981). On the maximal cluster approach to syllable structure. Linguistic Inquiry 12, 575–604. Lowenstamm, J. (1996). CV as the only syllable type. Current trends in phonology, models and methods, edited by J. Durand & B. Laks, 419–443. Salford: European Studies Research Institute. Lowenstamm, J. & J.D. Kaye (1986). Compensatory lengthening in Tiberian Hebrew. Studies in compensatory lengthening, edited by L. Wetzels & E. Sezer, 97–133. Dordrecht: Foris. Lowenstamm, J. & T. Scheer (forthcoming). A note on initial and internal sC(C) clusters in French, German and Spanish. Miller, P. (1992). Clitics and constituents in phrase structure grammar. New York: Garland. Montes Rodriguez, M.E. (1995). Tonologia de la lengua ticuna. Bogota, Colombia: Centro Colombiano de Estudios de Lenguas Aborigines, Universidad de los Andes. Riemsdijk, H.C. van (in press). Clitics: a state of the art report. Clitics in the Languages of Europe. Language Typology, 8, edited by H.C. van Riemsdijk. Berlin: Mouton de Gruyter. Scheer, T. (1996). Une théorie de l'intéraction directe entre consonnes. Doctoral Dissertation, Université Paris 7. Ségéral, P. (1995). Une théorie généralisée de l'apophonie. Doctoral Dissertation, Université Paris 7.
Constraint ranking, Government Licensing and the fate of final empty nuclei
Krisztina Polgárdi
Abstract
In this paper, I propose to combine Government Phonology, a theory of phonological representation, with Optimality Theory, a theory of con- straint interaction. I use the “principle” of Government Licensing as the basis of the discussion. This principle is in conflict with Proper Govern- ment word-internally and with the faithfulness constraints word-finally. I show that an account using explicit constraint ranking is superior to a purely parametric approach. In addition, such an analysis makes it pos- The Hague: Thesus. 167–182. Thesus.The Hague: 167–182. . sible to dispense with parametrically licensed final empty nuclei. 0. Syllables !? 1. Introduction
Government Phonology (GP) is a principles-and-parameters approach, where principles are inviolable and language-specific facts are expressed by Phonologica 1996: Phonologica 1996: parameters (see Kaye, Lowenstamm & Vergnaud, 1985; 1990). Some analy- ses, however, have been proposed which involve a conflict between different principles that is resolved on a language-particular basis (see Charette, 1990; 1991; 1992; Cyran, 1994; 1996). In this paper, I use Government Licensing (GL) as an example, and I propose that such conflicts can be resolved by Opti- mality Theoretic (OT) ranking (cf. Prince & Smolensky, 1993), rather than ‘turning the principles off’ in a parametric style. This not only makes the grammar simpler, but it also allows the lower ranked constraints to have an effect in a non-conflict situation. In addition, I propose to get rid of the (controversial) parameter licensing final empty nuclei, and show how this move, apart from other advantages, simplifies the account of final con- sonant clusters. The project of combining GP with OT can also be of interest from a more general point of view, because both theories work with constraints, but they
John R. Rennison & Klaus Kühnhammer, eds. (1999). eds. R. Rennison & Klaus Kühnhammer, John (1999). concentrate on different aspects of the grammar. Namely, GP concentrates
16 8 Krisztina Polgárdi on issues of representation, while OT concentrates on issues of derivation. Or in other words, GP deals with the nature of constraints, while OT deals with their ranking. In such an investigation, several questions arise. For example, which of the principles of GP remain inviolable and which become violable in a combined approach, and is there any principled basis for such a distinction? Another question might be whether there are any parameters left at all. I will attempt to some answer these questions.
2. Government Licensing
According to Charette (1990; 1991; 1992), languages exhibiting vowel-zero alternations fall into three types on the basis of how they treat properly governable empty nuclei that follow a consonant cluster. These types are illustrated in (1).
(1) Typology of properly governable empty nuclei following a consonant cluster (Charette, 1990; 1991; 1992) a. French: phonetic realization of empty nucleus (i) suvv0nirv0 [suvnir] ‘to remember’ (ii) parvv0nirv0 [parv7nir] ‘to reach’ librv0ment [libr7ma˜] ‘freely’ b. Billiri: Proper Government + deletion of second consonant (i) sana + do [sando] ‘her food’ (ii) landa + zi [lanzi] ‘your (f.) dress’ c. Polish: Proper Government of empty nucleus (i) kopv0ra [kopra] ‘dill’ (ii) plastv0ra [plastra] ‘plaster’ ubrv0dacv0 [ubrdac] ‘imagine’
As can be seen from these examples, French chooses to phonetically realize the empty nucleus even though it could be properly governed; Billiri — after vowel syncope — deletes the second consonant from the cluster; and Polish treats empty nuclei following consonant clusters the same way as those following single consonants — that is, it leaves them phonetically null. Charette analyses the difference between these types of languages by the two principles in (2). The principle of GL expresses the fact that a con- sonant cluster requires stronger support from its following nucleus than a single consonant which has no governing tasks to fulfil, and that this stronger support can only be given by a full vowel.
Constraint ranking, Government Licensing and final empty nuclei 169
(2) a. An empty nucleus is properly governed by a head-adjacent un- licensed nucleus. (Proper Government)1 b. A governing consonant must be government-licensed by an unlicensed nucleus governing it. (Government Licensing)
According to Charette’s analysis then, Billiri opts to preserve (2a), Proper Government (PG) of empty nuclei, while French opts for (2b), GL. Polish on the other hand has a different parameter setting for GL, namely, in this language properly governed empty nuclei are government licensors as well. There are a number of problems with this analysis. One is that it is mis- leading to say that Billiri opts for PG over GL. The phonetic forms in Billiri in fact obey both of these principles — although at the expense of violating a third constraint, the one prohibiting deletion of segments. A more serious problem is that the derivation of the Billiri alternation violates the Projection Principle, as can be seen in (3). Here, the n is resyl- labified from a lexical rhymal position into a surface onset, and the origi- nally following onset is deleted, thus the coda-onset governing domain disappears.
(3) Billiri a. Lexical representation PG N O N N
O 1R1 O 2N2 O 3N3
x x x x x x x interconstituent government 0 l a n d ]v0] z i b. Surface representation PG N N N
O 1N1 O 2N2 O 3N3
x x x x x x x 0 l a n d ]v0] z i [lanzi]
1 The relation of Proper Government is defined as follows: A properly governs B iff (i) A governs B (adjacent on its projection) from right to left (ii) there is no intervening governing domain (iii) A is not properly governed
170 Krisztina Polgárdi c. Projection Principle (Kaye, Lowenstamm & Vergnaud, 1990) Governing relations are defined at the level of lexical representation and remain constant throughout a phonological derivation.
Therefore I would like to argue that the consonant~zero alternation in Billiri should not be analysed as phonologically conditioned, but rather as a frozen regularity in the lexicon, similarly to Closed Syllable Shortening in English in pairs like keep~kept (cf. Kaye 1990; Harris, 1994). Thus the Billiri alternation does not belong to the typology to be discussed here, and I will concentrate on French-type and Polish-type languages. In fact, I am going to argue that it is Polish, not Billiri, that violates GL.
3. French vs. Polish, and Optimality Theory
The two possibilities that remain are illustrated in (4) and (5).
(4) French N O N N N
O 1R1 2O2 <<2 N2 3O3 3N3 4O4 4N4 GL
x x x x x x x x x interconstituent gov’t p a r v 0v0 n i r 0v0 [parv7nir]
In (4), the full vowel under N3 is not allowed to properly govern the empty N2 so that this ungoverned nucleus can government license O2 (indicated by “<<”). Since the empty N2 is not properly governed, it receives phonetic content. The problem here is that there is no explicit means in the GP approach of Charette to express the fact that the principle of GL overrides the principle of PG in case of a conflict, as in (4). This is where principle ranking can provide a solution. This ranking, however, is not universal; it can be reversed, as evidenced by Polish, represented in (5).
Constraint ranking, Government Licensing and final empty nuclei 171
(5) Polish PG N N N
1O1 1R1 2O2 2N2 3O3 3N3
x x x x x x x x 0 p l a s t ]v0] r a const. & interconst. gov’t [plastra]
Here the empty N2 is properly governed by the full vowel in N3. N2 therefore remains silent and is thus not capable of giving the required license for O2 to govern. That is, here it is PG that overrides GL. Notice that O2 in (5), though not licensed, still governs the preceding rhymal complement. This is a legitimate state of affairs, if constraints are violable. This situation can be compared to driving a car without a license. Such a driver breaks a rule, but he is not deprived of his ability to drive. This is what happens in Polish. In French, however, the requirement for having a license is so strong that it is impossible to violate it. Then the only way out is to break some other rule.2 The contrast between (4) and (5) thus demonstrates that we have a genu- ine constraint conflict here, resolved in different ways in different languages. These facts invite an Optimality Theoretic analysis in terms of constraint ranking. The two (violable) constraints are given in (6).
(6) a. PROPER GOVERNMENT (PG) An ungoverned Nucleus properly governs a preceding empty Nucleus. b. GOVERNMENT LICENSING (GL) A governor Onset must be licensed by a Nucleus which is not properly governed.
The constraint PG works in conjunction with an inviolable representational principle, given in (7), the Empty Category Principle, or the Principle of P-Licensing (which is presumably part of Gen).3
2 In an OT-style analysis, in a “repair” situation (as in Billiri) the given constraint is satisfied vacuously. For instance if there is no governing relation, then there is no need for a license to govern. In a derivational approach, the implication is reversed, thus if there is no license, then no governing relation is possible. However, as we have seen in the Polish example in (5), this requirement is too strong, which shows that constraints can in fact be violated. 3 Although see Gussmann (1997) on Polish, where morpheme-internally we find examples of sequences of empty nuclei without phonetic content, in apparent violation of the ECP.
172 Krisztina Polgárdi
(7) Empty Category Principle (ECP — Kaye, 1990) An empty Nucleus is phonetically interpreted iff it is not properly gov- erned.
This principle is violated if (a) an empty nucleus remains empty even though it is not properly governed, or if (b) it is properly governed, nevertheless it is phonetically interpreted. The constraint PG is normally not a separate constraint of GP, but it is supposed to fall out from the ECP. However, I think that it is necessary to formulate it separately, since the two constraints perform different func- tions. The ECP tells us something about the governee, whereas PG refers to the governor. Moreover, it seems that the ECP is inviolable, while, as we have seen, PG is violated in French. The two possible systems resulting from the two different rankings of the constraints in (6) are given in (8a) and (8b).
(8) a. French: Empty N realized after C cluster parvv0nirv0 GL PG parv7nir * parvnir *! b. Polish: no “illicit” filling of empty nuclei plastv0ra PG GL plastera *! plastra *
In the first candidates, represented as in (4), PG is violated, since N2 is not properly governed by N3. As a result, N2 can government-license O2, and the ECP will make sure that N2 receives phonetic interpretation. In the second candidates, represented as in (5), on the other hand, N2 is properly governed (and thus, by the ECP, remains empty), and so GL is violated. As we have seen, French chooses to preserve GL, while Polish opts for PG. That PG is not ‘turned off’ in French is evidenced by the fact that before single consonants (that is, in the absence of a conflict), PG does have an effect, and the empty nucleus remains silent, as for example in [suvnir] in (1).4
4 In fact, in sequences of two consecutive empty nuclei, either nucleus can remain silent, and a word like devenir ‘to become’ can be pronounced either [d7vnir] or [dv7nir] (cf. Charette, 1991), with the second form violating PG, without any obvious reason. Although such forms as yet await explanation, the fact that they are considered more marked than the ones with the first nucleus interpreted indicates that the constraint PG still manifests itself here.
Constraint ranking, Government Licensing and final empty nuclei 173
4. Domain-final empty nuclei
Now let us turn to the behaviour of domain-final empty nuclei following a consonant cluster. In this respect, Polish and French behave the same way, so we should look at some other languages as well. The data in (9) illustrate the different types given by Charette. In French domain-final empty nuclei are government licensors, and words may end both in a coda-onset cluster and in a complex onset. In Wolof (cf. Ka, 1994) and Korean (cf. Sohn, 1994) they are not government licensors, thus a schwa is inserted after morpheme-final consonant clusters in Wolof, whereas in Korean, one of the consonants of a word-final cluster is deleted. And finally in English, domain- final empty nuclei government-license directly, but not indirectly, and words may end in a coda-onset cluster, but not in a complex onset.
(9) Word-final consonant clusters (Charette, 1991; 1992) a. French: Domain-final empty nucleus is government licensor carte [kart] ‘card’ table [tabl] ‘table’ castre [kastr] ‘castrate’ b. Wolof: Empty nucleus is phonetically realized nak ‘cow’ lEkk7 (*lEkk) ‘to eat’ rabb7 (*rabb) ‘to weave’ c. Korean: One of the consonants is deleted hulk [hul]/[huk] ‘earth’ d. English: Domain-final empty nuclei government-license only directly card post (*catr)
Charette analyses these facts by setting the parameter of potential gov- ernment licensors differently in these languages. One problem with this analysis is that this parameter setting in itself does not explain the differ- ence between (9b) and (9c), epenthesis vs. deletion. Moreover, I have reser- vations about positing domain-final empty nuclei, as will be seen in the next section.
Cyran (1996) also proposes (for Irish) to rank GL above PG instead of turning PG off com- pletely. However, he still follows Charette (1990; 1992) in claiming that the conflict only arises as a result of particular licensing properties of word-internal nuclei, and he further suggests that this ranking is universal — claims that I have been arguing against in this paper.
174 Krisztina Polgárdi
5. Syllable typology
As argued by Kaye (1990), word-final consonants behave differently from word-internal codas and can in fact be argued to pattern together with word-internal onsets. To account for this fact, Kaye proposes the principle of Coda Licensing, given in (10a). Onsets in turn must be followed by a nucleus, as a consequence of the principle of Onset Licensing, given in (10b) (cf. Harris, 1992). Consonant-final words thus in fact end in an onset followed by an empty nucleus. Since such empty nuclei cannot be properly governed, the only way to make sure they remain silent is to add to the ECP the extra clause in (10c), which parametrically licenses domain-final empty nuclei. Thus languages that have this parameter switched on allow words to end in a consonant (or rather, in an empty nucleus), while languages with this parameter in the ‘off’ setting oblige their words to end in a full vowel.
(10) a. Coda Licensing Principle (Kaye, 1990) Post-nuclear rhymal positions must be licensed by a following onset. b. Onset Licensing Principle (Harris, 1992) An onset head position must be licensed by a nuclear position. c. ECP Domain-final Parameter A domain-final empty nuclear position is licensed: YES/NO
One problem with such final empty nuclei is that they need to be deleted when they come to precede an empty onset as a result of suffixation. This operation was introduced under the name Reduction by Gussmann & Kaye (1993). It is defined in (11a) and illustrated in (11b). Reduction has often been used in the GP literature, without mentioning the fact that it violates the Projection Principle. This can be seen in (11b), where the governing domains contracted by the pairs O2 and N2 and O3 and N3 disappear, and a new governing domain is created between O2 and N3.
(11) a. Reduction (Gussmann & Kaye, 1993) An empty nucleus followed by a pointless onset are removed from any phonological representation in which they occur. b. Polish 1O1 1N1 2O2 2N2 3O3 3N3 4O4 4N4 → 1O1 1N1 2O2 3N3 4O4 4N4
x x x x x x x x x x x x x p e s k p e s k ‘dog DIMINUTIVE’
Constraint ranking, Government Licensing and final empty nuclei 175
An additional problem is encountered in languages such as Turkish (cf. Lees, 1961; Lewis, 1967), where unlicensed empty nuclei surface phonetically as the colourless vowel (/i-/, spelled ı). Turkish also allows words to end in a consonant and thus seems to license final empty nuclei. These facts are illustrated in (12a). Thus the colourless vowel should never appear in domain-final position, a prediction which turns out to be false, as shown in (12b).5
(12) Turkish (Lees, 1961; Lewis, 1967) a. qarv0nv0 → qari-n ‘belly (NOM.SG.)’ qarv0nv0+a → qarna ‘belly (DAT.SG.)’ b. kapi- ‘door’ kari- ‘wife’ aci- ‘grief’
In order to avoid these problems, I propose to turn the principle of Onset Licensing into a violable constraint called NUCLEUS, given in (13), and thus remove the extra clause from the ECP.
(13) NUCLEUS An onset is licensed by a following nucleus.
A language with underlyingly consonant-final morphemes has then the three possibilities listed in (14) for the surfacing of such forms. Notice that the underlying representation of these morphemes satisfies Coda Licensing and ends in an onset. Each possible surface representation, however, violates one of the universal (faithfulness) constraints defined in (15). Note that (14) is not a tableau, it is rather the summary of three tableaux, where each row represents the winning candidate in a possible language. The ranking of the constraints needs to be established in each language, as explained below. Square brackets stand for morpheme boundaries and angled brackets indicate underparsing.
5 Kaye (1990) argues that the possessive suffix -ı in Turkish is in fact -v0nv0 underlyingly, where the n is not realized phonetically for some reason when no other suffixes follow it. This, however, does not solve the problem completely, because there are plenty of lexical words (and suffixes, other than the possessive) in Turkish ending in the ominous ı, where there is no evidence of a following latent consonant.
176 Krisztina Polgárdi
(14) Syllable typology: Domain-final position N O
x ]x] PARSE FILL NUCLEUS " $ a. Samoan N O x ]x] * " <$> b. Zulu N O N x x ]x] * " $ c. English N O x ]x] * " $
(15) a. PARSE Segments are parsed. 6 b. FILL Empty positions are prohibited.
Depending on which of the constraints is ranked lower than the others, we get two types of languages. In one type, words can end in a consonant, as in English, illustrated in (14c). Here NUCLEUS is ranked below the faithful- ness constraints and therefore such morphemes have to surface identical to their underlying representation. In the other type of languages, words have to end in a vowel. This can be achieved in two ways: either by deletion of the word-final consonant, as in Samoan in (14a), i.e. by ranking NUCLEUS above PARSE; or by inserting an epenthetic vowel, as in Zulu in (14b), i.e. by ranking NUCLEUS above FILL. Thus under the present approach, languages differ in whether words have to end in a nucleus or not — instead of differing in whether they allow a
6 This constraint is slightly different from the one usually bearing this name in OT, since it is also violated by empty positions that actually remain empty as a consequence of being properly governed by a following full nucleus. Thus for instance the examples in (8a) and (8b) all violate FILL, containing underlying empty nuclei. But since every relevant candidate has the same number of violations, this constraint is irrelevant for the decision between the candidates.
Constraint ranking, Government Licensing and final empty nuclei 177 final empty nucleus or not. Consequently the extra clause of the ECP can be dispensed with. This way the problems mentioned above are avoided. Deletion of final empty nuclei, as in (11b), is not necessary, since they are not there in the first place. And in languages like Turkish, words ending in the colourless vowel, as in (12b), can now end in an empty nucleus underlyingly which consequently will have to receive phonetic interpretation, since it is always unlicensed. Moreover, NUCLEUS constitutes a perfect parallel to the already existing constraint ONSET, which expresses the difference between languages con- cerning the necessary presence (or not) of initial onsets. That is, the NUCLEUS and ONSET constraints together express the generalization that word edges like to coincide with syllable edges, a fact usually accounted for by a member of the ALIGN family. Notice that to preserve domain-internal empty nuclei in languages with word-final consonants (i.e. where NUCLEUS is ranked below FILL), and thus to achieve word-internal alternation of onsets and nuclei, we need an inviola- ble representational principle which prohibits sequences of consecutive constituents of the same type within a domain. This does not mean adding an extra constraint, however, since such a principle has always been needed to separate nuclei in hiatus by an empty onset. Also, the addition of empty nuclei now becomes necessary in certain configurations; but if we regard the Projection Principle as a monotonicity requirement, instead of forbid- ding both the deletion and addition of structure, this will cause no prob- lems. Notice too that by making the principle of Onset Licensing violable, the more general Licensing Principle of Kaye (1990), given in (16a), becomes violable as well. The most important aspect of this principle, namely that every domain must have a head, however, can be retained under a modified version of the Licensing Principle, given in (16b). (It is enough to say ‘nucleus’ in this principle, since in the absence of a proper governor, a sole nucleus will always receive phonetic interpretation.)7
7 There is another issue here, that of catalexis, that deserves attention. This phenomenon fitted very well in an account with final empty nuclei, where these nuclei could become “visible” in languages that have previously been analysed by means of a final dummy syl- lable. However, since the ECP might prove to be violable independently (cf. sequences of phonetically null empty nuclei in Polish, in Gussmann, 1997), it could be suggested that catalexis be a result of some higher ranked constraint as well. This issue requires further consideration.
178 Krisztina Polgárdi
(16) a. Licensing Principle (Kaye, 1990) All phonological positions save one must be licensed within a domain. The unlicensed position is the head of this domain. b. Licensing Principle (new version) Every domain must have a head, i.e. a nucleus.
6. Domain-final consonant clusters continued
Now we can return to the story of GL. The behaviour of word-final con- sonant clusters in (9) can now be accounted for in an analysis parallel to that of single final consonants, which in addition explains the difference between (9b) and (9c), and thus simplifies the overall analysis. Let us start with French, illustrated in (17). As we saw in (8a), in French GL is ranked above PG. And since there are word-final consonants in this language, we know that NUCLEUS is ranked below PARSE and FILL, as in (14c). The existence of word-final consonant clusters shows that GL is not inviola- ble either. Since no epenthesis or deletion occurs in these cases, this pro- vides evidence that GL, like NUCLEUS, is ranked below PARSE and FILL.
(17) French: Word-final consonant clusters allowed /tabl/ PARSE FILL GL NUCLEUS PG tabl * * tabl *! tab
As can be seen from this example, the ranking of GL with respect to PARSE and FILL is an independent issue from its ranking with respect to PG, just as in Charette’s analysis. That is, the behaviour of final clusters is independent of the behaviour of internal ones. Wolof and Korean, shown in (18) and (19), choose to preserve GL word- finally, although in different ways: Wolof by epenthesis, i.e. lower ranking of 8 FILL, and Korean by deletion, i.e. lower ranking of PARSE.
8 Notice that Korean, with underparsing, looks like Billiri at first glance, and so should also violate the Projection Principle. The difference, however, is that here we are still within one domain, and if underlyingly only nuclear nodes are associated to skeletal positions (cf. Charette, 1991), then the given (consonantal) skeletal positions will not get incorporated into any syllabic constituent by the time the form exits the lexicon, and thus no non-nuclear governing relation will be established (and we do not need to destroy such a relation).
Constraint ranking, Government Licensing and final empty nuclei 179
(18) Wolof: Epenthesis after word-final consonant cluster /rabb/ PARSE GL FILL NUCLEUS rabb *! * rabb * rab *! *
(19) Korean: Deletion of either consonant from word-final cluster /hulk/ FILL GL PARSE NUCLEUS hulk *! * hulk *! hul
Now let us turn to English. English looks like a case in between French and Wolof, since it allows word-final consonant clusters of the coda-onset type, but not of the complex onset type. In fact, if we look at words like cycle, fibre or centre more closely, we see that English treats these the same way as Wolof, at least if syllabic sonorants are analysed as the presence of an empty nucleus, i.e. as a violation of FILL. I suggest analysing this language by postulating a more specific GL con- straint, GL(O1,O2), requiring licensing of the head of a branching onset. Since complex onsets are subject to stricter phonotactic restrictions than coda-onset clusters, they are more likely to require more licensing. By Panini’s theorem, when a general constraint is ranked above a specific one, the effect of the latter becomes concealed. This way we account for the fact that there is no language that requires more licensing for coda-onset clus- ters than for complex onsets. The analysis of English is given in (20). Since FILL is sandwiched between GL(O1,O2) and GL, violating it will be fatal in the case of a coda-onset cluster, but optimal in the case of a complex onset.
(20) English: Epenthesis after word-final complex onset /kard, saikl/ PARSE GL(O1,O2)FILL GL NUCLEUS kard * * kard *! kar
The French dialect of Saint-Etienne mentioned by Charette (1992) pro- vides interesting extra evidence for the two GL constraints. But here the two are ranked differently with respect to another constraint: it is PG which is
180 Krisztina Polgárdi sandwiched in between; that is, this dialect represents a case in between standard French and Polish. The facts are illustrated in (21).
(21) Saint-Etienne French: Empty N realized after complex onset GL(O1,O2)PGGL parv7nir *! parvnir * librma˜ *! * libr7ma˜ *
Depending on the behaviour of word-final clusters, PARSE and FILL will be ranked either below, between or above the GL constraints. In summary, in this analysis, the ranking itself establishes the difference between all cases.
7. Predictions and summary
Now let us see how many possible (relevantly distinctive) permutations the constraints introduced in this paper give. The permutations are given in (22), where the symbol “>>” indicates ranking.
(22) a. Medial contexts GL >> PG (e.g. Polish) PG >> GL (e.g. French) GL(O1,O2) >> PG >> GL (e.g. Saint-Etienne French) b. Word-final contexts PARSE, FILL >> NUCLEUS: PARSE, FILL >> GL (e.g. French or Polish) PARSE, GL >> FILL (e.g. Wolof) FILL, GL >> PARSE (e.g. Korean) GL(O1,O2) >> FILL >> GL (e.g. English) GL(O1,O2) >> PARSE >> GL (e.g. French) NUCLEUS >> PARSE (e.g. Samoan) NUCLEUS >> FILL (e.g. Zulu) Total number of permutations: 3x7=21
In medial contexts, different rankings between PG and GL result in three different grammars. In word-final contexts, GL can only come into play if NUCLEUS can be violated (and thus word-final consonants are allowed). In this case, we have to look at the rankings between GL and the faithfulness constraints, PARSE and FILL. Either GL is at the bottom, or one of the faith- fulness constraints is in this position, or a faithfulness constraint is sand- wiched in between the two halves of GL. This leaves us with five possibilities,
Constraint ranking, Government Licensing and final empty nuclei 181 plus the two cases where NUCLEUS is inviolable, giving seven. Thus the total number of permutations, combining both contexts, is 3x7=21. The empirical testing of these predictions awaits further research. As a matter of fact, the parametric approach does not fare any better, because in its present formulation direct and indirect licensing are two separate parameters, predicting four possibilities word-internally, in con- trast to the present approach where GL and GL(O1,O2) predict only three. (Word-finally, the two approaches do not make different predictions.) How- ever, as we have seen, an analysis employing ranking has certain advantages over a parametric one. One such advantage is that it can account for cases of constraint conflict in an elegant way, a state of affairs that as yet has not received a principled solution in a parametric approach (where the only possibility is to turn constraints off completely, but they cannot be made violable). Another advantage of a ranking approach is that it is more explicit. For example, in the case of Wolof, the ranking of FILL under GL expresses both the fact that GL cannot be violated and the means to rescue the situation, namely epenthesis. In a parametric approach, the parameter about domain- final empty nuclei only says that these are unable to government-license, and we need a separate parameter to distinguish between Wolof and Korean, that is, between epenthesis and deletion. Thus the present ap- proach proves to be simpler than a parametric one. Finally, as I have shown, by turning the Principle of Onset Licensing into a violable constraint, the controversial stipulation of phonetically always null final empty nuclei can be dispensed with. Moreover, all the parameters mentioned in this paper can be expressed by different rankings of violable constraints which, in turn, are needed anyway (among them PG and GL). The principles that still seem to prove inviolable on the basis of this study are the Projection Principle (basically a monotonicity requirement), the (modi- fied version of the) Licensing Principle, Coda Licensing and the ECP. Further study will have to determine whether they are all really principles, or whether some of them can be violated under the pressure of other con- straints.
Acknowledgements
I would like to thank Monik Charette, John Harris, Harry van der Hulst and participants of the SOAS Phonology Workshop and of the 8th International Phonology Meeting in Vienna for their comments and discussion of earlier versions of this paper. All remaining errors are, of course, my own.
182 Krisztina Polgárdi
An earlier version of this paper appeared in P. Backley & J. Harris (eds.) (1996): UCL Working Papers in Linguistics 8, 595–616. The material discussed here has also become part of my dissertation (Polgárdi, 1998).
References
Charette, M. (1990). License to govern. Phonology 7, 233–53. Charette, M. (1991). Conditions on phonological government. Cambridge University Press. Charette, M. (1992). Mongolian and Polish meet Government Licensing. SOAS Working Papers 2, 275–291. Cyran, E. (1994). Vocalic elements in phonology: A study in Munster Irish. PhD dissertation, Catholic University of Lublin. Cyran, E. (1996). Licensing properties of nuclei and principle ranking in Irish. The Linguistic Review 13, 1–31. Gussmann, E. (1997). Govern or perish: Sequences of empty nuclei in Polish. A festschrift for Jacek Fisiak on his 60th birthday. [Trends in linguistics. Studies and monographs 101], edited by R. Hickey & S. Puppel, 1291–1300. Berlin: Mouton de Gruyter. Gussmann, E. & J. Kaye (1993). Polish notes from a Dubrovnik cafe: I. The yers. SOAS Working Papers in Linguistics and Phonetics 3, 427–462. Harris, J. (1992). Licensing Inheritance. UCL Working Papers in Linguistics 4, 359–406. Harris, J. (1994). English Sound Structure. Oxford: Blackwell. Ka, O. (1994). Wolof phonology and morphology. Lanham, Md.: University Press of America. Kaye, J. (1990). “Coda” Licensing. Phonology 7, 301–330. Kaye, J., J. Lowenstamm & J.-R. Vergnaud (1985). The internal structure of phonological ele- ments: A theory of Charm and Government. Phonology Yearbook 2, 305–328. Kaye, J., J. Lowenstamm & J.-R. Vergnaud (1990). Constituent structure and government in phonology. Phonology 7, 193–231. Lees, R.B. (1961). The phonology of modern standard Turkish. Volume 6 of the Uralic and Altaic Series. Bloomington Ind.: Indiana University Publications. Lewis, G.L. (1967). Turkish grammar. Oxford: Clarendon Press. Polgárdi, Krisztina (1998). Vowel harmony. An account in terms of Government and Optimality. Ph.D. dissertation, Leiden University. The Hague: Holland Academic Graphics. Prince, A. & P. Smolensky (1993). Optimality Theory: Constraint interaction in generative grammar. Manuscript, Rutgers University & University of Colorado at Boulder. Sohn, H. (1994). Korean. London: Routledge.
Can there be empty phonological elements? On empty heads and empty operators
John R. Rennison
Abstract
The theory of Government Phonology (GP) allows empty units in phono- logical representations, provided that these are licensed. This paper continues the line of research begun in Rennison (1996) and explores the possibility of maximally exploiting the theoretical device of empty positions by using them within the melody of segments, both as heads and as operators. This restricts the generative power of the theory, while still allowing a reasonable set of plausible segment inventories to be The Hague: Thesus. 183–193. Thesus.The Hague: 183–193. . generated. I propose that it is the syllable structure which determines the pho- netic realization of an empty head: in an Onset as a stop, and in a Nucleus
Syllables !? as the vowel [a] (barring modification by further elements). Every language allows empty heads (i.e. has stops and low vowels). Empty operators, on the other hand, are typical functors. They add a stronger posture (articulatorily) or clearer patterning (acoustically) to a segment: thus in an Onset an empty operator adds noise to a stop, pro- Phonologica 1996: Phonologica 1996: ducing a fricative; in a Nucleus it adds ATR. But in effect an empty opera- tor weakens rather than strengthens its phonological position. Empty operators are licensed parametrically (i.e. language-specifically). Thus a maximum of two elemental positions per segment (i.e. the head and one operator) are defined by the syllable constituent in which the segment is located. This partial integration of syllable structure into melodic structure has far-reaching and overwhelmingly positive con- sequences for the theory. 0.
1. Prerequisites
First of all I would like to state one theoretical assumption that lies behind
John R. Rennison & Klaus Kühnhammer, eds. (1999). eds. R. Rennison & Klaus Kühnhammer, John (1999). this paper, without which it would be fairly pointless. I assume that
184 John R. Rennison asymmetries of the head/operator type are omnipresent throughout gram- mar. In particular, if we have one or more elements associated to a skeletal point, one of those elements must be the head. In most respects, the view that there exist “headed” and “headless” expressions arrives at the same results, if we say that “headless” means “there is an empty head”. But I simply assume here that every elemental melody which is licensed to be realized phonetically, has a head, even if that head is itself empty. The theory of Government Phonology allows empty units in phonolog- ical representations, provided that these are licensed. In standard theory, empty heads have been proposed for two things: for velar consonants (this is fairly standard GP; cf. Kaye, Lowenstamm & Vergnaud, 1985) and for ATR vowels (Charette, 1991). As far as I know, no one has proposed that there are empty phonological elements as operators. This is hardly surprising, given the current interpretations of empty heads; it is difficult to imagine what an empty melodic operator would do.
2. The proposal
2.1. Empty heads in melodic expressions
I propose a different, and more unified interpretation of empty melodic heads: they are “transparent”, i.e. they transmit the properties of the Onset or Nucleus with which they are associated to the skeleton. This is formu- lated in (1). So an empty head in an Onset gives “prototypical onset-ness” — a good, solid consonant, preferably an obstruent, preferably a stop. And an empty head in a Nucleus is “prototypical nucleus-ness” a nice sonorous vowel, not high, preferably low. I deliberately leave open the question of the interpretation of empty melodic positions in Codas, since a) there have been proposals to eliminate rimal consonants from GP altogether (cf. Lowenstamm, this volume and Scheer, this volume), and b) the interpreta- tion could hardly differ from that of an empty melodic position in an Onset.1
(1) An empty head transmits the prototypical melodic properties of the syllabic constituent in which it is located (i.e. “Onset-ness” or “Nucleus- ness”) to its skeletal position.
1 The question also remains (posed by one reviewer) what property is transmitted to a Coda consonant. Since rhymal complements can be either vowels or consonants (see Ritter & Vago, this volume), this is clearly a problem; however, I no longer believe in Codas, or in branching constituents of any kind.
On empty heads and empty operators 185
The resulting system of elements is given in (2).
(2) The elements (from Rennison, 1996). (Shaded cells involve the empty me- lodic positions.) element articulatory correlates acoustic correlates old new I I palatal dIp* (formant structure) U U labial rUmp* (formant struc- ture) R R coronal (formant structure) H H high tone / voiceless high F0 L L low tone / voiced lower F0 N L (head of O) nasal stop2 nasal formant ? _ (head of O) stop (almost) silence h _ (operator of O) aspiration / friction noise A _ (head of N) non-high tongue mAss* (formant struc- body ture) ATR _ (operator of N) advanced tongue formant structure root
This move has eliminated two elements from the inventory, namely ? and A, which were a bit uncomfortable in a way. The stop element ? was always overworked in Onsets and almost unemployed in Nuclei. The A element was very frequent in nuclei, but almost unemployed in Onsets. So in a way they were just echoing syllable structure most of the time: if ever you saw a ? element, you could be sure it was in an Onset (or a Coda, of course), and if ever you saw an A element, you could be fairly sure it was in a Nucleus.3 What about uniform interpretation of elements? Well, this is no problem if the empty head only transmits the properties “Onset-ness” and “Nucleus- ness”, as stated in (1). Uniform interpretation is simply removed to a higher level of structure, and boils down to relative intensity. “Onset-ness” is quiet, “Nucleus-ness” is loud. In (3) I show some quite common onsets and nuclei in a fairly standard GP representation given as “old”, and my reanalysis of these under the new proposal, given as “new”.
2 The reanalysis of nasality is not relevant to the present discussion and will be dealt with elsewhere; the question of nasality in vowels and its relation to tone is particularly tricky. 3 Recently some researchers (p.c. and Williams, 1998) have considered economizing by collapsing the R element with the A element. Beyond economy, I see no particular motiva- tion for this move. Although R only appears in Onsets, it is frequently also realized in neigh- bouring Nuclei (e.g. in rhotacized vowels or “syllabic” r, n or l.)
186 John R. Rennison
(3) Some melodic expressions that have an empty head4 consonant old new vowel old new p or ph (h,H,?,U) (H,U,_) E (I,A,_) (I,_) t or th (h,H,?,R) (H,R,_) a (A) (_) k or kh (h,H,?,_) (H,_) O (A,U,_) (U,_) p or b (?,U) (U,_) t or d (?,R) (R,_) k or g (?,_) (_) b (L,h,?,U) (L,U,_) d (L,h,?,R) (L,R,_) g (L,h,?,_) (L,_)
Note that these consonants and vowels are traditionally regarded as be- ing phonologically strong. This contrasts with the much weaker, truly empty Onsets and Nuclei.
2.2. Truly empty Onsets and Nuclei
Truly empty Onsets and Nuclei are quite distinct from Onsets or Nuclei which contain an empty melodic head, in that they have no melodic content whatsoever (which includes empty elements).5 An empty Onset or Nucleus just has nothing. This is like the distinction between an empty set and no set at all. If empty positions receive phonetic interpretation in order to license themselves, they are extremely weak — typically a glottal stop for an onset and a schwa or some other central vowel for a nucleus. These are
4 Convention: The head of a melodic expression is the rightmost element. If an empty operator is present it is written in penultimate position. The “old” column is my own idea of what others might sensibly postulate, but should not be taken to reflect any particular person’s unshakeable belief. The transcriptions “p or ph”, etc. and “p or b” etc. refer to the “contrastively voiceless” and the “non-contrastively voiceless/voiced” stops respectively, and reflect variations of phonetic transcription, not imprecision of the phonological repre- sentation (which is quite unambiguous). A reviewer suggested using some different symbol for the empty melodic position, for the sake of legibility; I have not done so because a) the underscore has been used previously by others, and b) a more contentful symbol would suggest the presence of some melodic con- tent. 5 A reviewer points out that “empty element” is potentially a) a self-contradiction or b) nothing different from the old “cold vowel” (v0). Ad a) probably “empty melodic position” is a better term; ad b) there probably is no difference except for the phonetic interpretation which the “empty element” receives, and the fact that it can also occur in Onsets.
On empty heads and empty operators 187 given in (4), with the caveat that their phonetic realization can vary widely from language to language.
(4) Truly empty skeletal positions (typically unlicensed positions which need phonetic filling) consonant old new vowel old new ? (?) ( ) 7 (A,_), (_), ( ) ? ( )
3. Empty operators
If the interpretation of an empty head is “transparency” that allows the syllable structure to shine through, what could an empty operator be? And do we need one at all? Well, formally the theory allows for anything to be empty, provided that it is licensed and has some effect somewhere. If we look for semi-redundant elements in the standard version of GP theory, which might be derivable in some way from syllable structure, the choice is fairly simple. The noise element h is mainly restricted to noise in con- sonants, especially fricatives, and the ATR element is restricted exclusively to nuclei (and so has long been a candidate for elimination). What do noise in consonants and ATR in vowels have in common? In a way, they seem to make their host position more extreme versions of what they already were: in an onset position by adding more noise, and in a nucleus by slight changes of formant position. Articulatorily, both these effects are gained by more controlled movements of the musculature involved: a stop is a ballistic movement which during the closed phase involves no constriction of the muscles used to open the stop during the release phase. A fricative, on the other hand, involves a controlled posture, and both the closing and opening musculature are active (counterbalancing each other) throughout the whole segment. The ATR vowels probably involve no additional control of the tongue body muscles, but they have additional contraction of the tongue root muscles. All these muscular activities may amount to what is traditionally called “tenseness”; however, I would not like to use that term here because it is often used for other cases which definitely could not be ascribed to an empty operator (e.g. the lenis vs. fortis distinction). These effects are each the opposite of the syllable-structure constituent in which the elements reside: noise adds sonority / energy to an Onset (and thereby makes it more Nucleus-like), and ATR subtracts sonority / energy from a Nucleus (and thereby makes it more Onset-like). In both cases, the “opposite syllable-constituent effect” is quite minor, and is realized with melodies that are quite typical of the constituent in which they actually
188 John R. Rennison occur. In other words, an empty operator in an Onset does not produce a vowel-like quality, even though it adds sonority; an empty operator in a Nucleus does not reduce sonority by a drastic degree, but rather by a degree that is only just noticeable.6 I therefore propose that noise and ATR are empty operators which act as identity elements for the elemental head. This kind of identity element serves only to intensify or “sharpen” what is already melodically present in an expression. As an identity element, the empty operator can have no independent phonetic interpretation. Moreover, instead of adding to the strength of the position to which it is associated, the empty operator weak- ens it. This is stated in (5).
(5) An empty operator intensifies the melody, but weakens the skeletal position to which it is associated.
What (5) also tells us is that the empty operator is highly marked. This is especially true if there are no contentful melodic elements present to give a position some colour. An immediate objection which springs to mind is: what about the con- sonant h? Isn’t that the independent phonetic interpretation of noise? Well, I don’t think so. I think that the consonant h is the independent phonetic interpretation of voicelessness in an Onset (or Coda?), i.e. the element H on its own. Let us now consider the representation of the other fricatives. It seems that typologically a language has to have stops before it can have fricatives. So, for example, most Australian languages have no fricatives at all, but many stops. I know of no language in the world that has fricatives but no stops. In standard GP, this distribution cannot be easily be captured, even assuming that fricatives are no more complex than a minimal representation of a stop (i.e. without the h element). My proposal is that a stop is the struc- tural prerequisite for a fricative, i.e. an onset position needs both an empty head and an empty operator to be a fricative. My proposal for the representation of some fricatives is given in (6).
6 This is true especially for nuclear melodies which contain only one of the elements I, U or empty head (former A). Mid vowels, on the other hand, whose melodies contain more than one element, are perhaps more easily distinguishable.
On empty heads and empty operators 189
(6) Some melodic expressions in Onsets (here: fricatives) that contain an empty operator consonant old new consonant old new f (U,h) (U,_,_) S (R,I,h) (R,I,_,_) s (R,h) (R,_,_) ç (h,I) (I,_,_)
Here, in contrast with the stops, my proposal has a larger number of elements than a standard GP representation (if we also count the empty elements). Is this a disadvantage? If complexity is to replace charm, and is computed by the simple counting of elements, then yes. However, I propose that the strength of a position is not derived directly from its complexity. A metric for the “strength” of melodic expressions is given in (7).
(7) Strength metric 1. An expression with an empty head is stronger than any expres- sion without an empty head. 2. An expression with an empty operator is weaker than the corre- sponding expression without an empty operator. 3. Otherwise, strength corresponds to the number of elements in an expression.
Why is this strength metric so complex, and why is it the way it is? Part 1 of (7) reflects the intrinsic strength of the Onset or Nucleus to which the melodic expression is associated. It is akin to the relative strength of stressed vowels vis-à-vis unstressed vowels which is also acquired from a higher level of structure. Part 2 of (7) states in effect that an empty operator is a modifier only, with no special acoustic signature, and therefore contributes only “weak- ness” to an expression. And part 3 of (7) simply states the normal case when there is no differ- ence between two expressions as regards their empty elements. If we wanted to translate (7) into numbers, then (roughly) an empty head counts as plus 7 elements and an empty operator as minus half an element. This, of course, means that “strength”, in the sense of “what can govern or license what”, and “complexity”, by simple “non-weighted counting of ele- ments”, are now different things. Let me now complete the picture of my proposal by giving some nuclear expressions with empty operators in (8), without further comment.
19 0 John R. Rennison
(8) Melodic expressions in Nuclei that contain an empty operator (left column) and the corresponding expressions without an empty operator (right column) vowel old new vowel old new i (I) (_,I) I (I,_) (I) e (A,I) (I,_,_) E (I,A,_) (I,_) A (A) ?(_,_) a (A._) (_) o (A,U) (U,_,_) O (A,U,_) (U,_) u (U) (_,U) U (U,_) (U)
Now we have eliminated the elements A and ATR from the inventory of phonological elements, and are left with I, U, R, H, L, plus the empty head and the empty operator. Note that the effects achieved by this reduction could be couched in terms of constraints on the occurrence of the elements ?, h, A and ATR, as given in (9); but I see no motivation for these or any such constraints.
(9) An alternative way to express the distribution of ?, h, A and ATR in terms of standard GP a) ? occurs only in Onset or Coda (?) positions and when it occurs is always the head of the melodic expression b) h occurs only in Onset or Coda (?) positions and when it occurs is never the head of the melodic expression c) A occurs only in Nucleus positions and when it occurs is always the head of the melodic expression d) ATR occurs only in Nucleus positions and when it occurs is never the head of the melodic expression
However, what the formulation in (9) does not capture is that “stopness” in an Onset is the same animal as “A-ness” in a Nucleus. Now that we’ve got it, what does this new version of the theory buy us?
(10) What the new proposal buys us for segment7 inventories a) All languages allow empty heads. Therefore all languages have stops and non-high vowels. b) Empty operators are parameterized. Therefore not all languages have fricatives and ATR vowels.
7 A segment comprises a skeletal position and all the melodic elements that are associated with it.
On empty heads and empty operators 191
c) [k] / [g] is the least marked licensed consonant, i.e. (_), and [x] / [G] the most marked obstruent, i.e. (_,_).8 d) [a] is the least marked licensed vowel, i.e. (_), and [A] the most marked vowel, i.e. (_,_).
First of all, as mentioned above, it gives us very simple representations for the least marked consonant and vowel systems of languages, if we assume that all languages allow an empty head, both in Onsets and Nuclei (10a), but that empty operators are parameterized (10b). So some languages can get by without fricatives (e.g. most Australian languages), and some can get by without an ATR distinction in vowels (e.g. 3 and 5 vowel systems, which also includes most Australian languages). Moreover, the combination of an empty head with an empty operator is a marked option, especially if there are no “real” elements in the same melodic expression. So the velar fricative [x] or [G] should be the rarest obstruent in the languages of the world (10c) and tense low [A] the rarest vowel (10d). Why should this combi- nation be most marked? Because it contains no specific acoustic cues of the kind that we find say in elements like I or U, but only the general acoustic cue of relative overall energy, modified by what I have termed intensification or “sharpening”. The consequences of this proposal for syllable structure are given in (11).
(11) What it buys us for syllable structure a) Non-high vowels do not occur in Onsets, but high vowels do (e.g. [j] and [w]). b) Obstruents do not occur in nuclei, but nasals and liquids do. c) [ka] / [ga] is the least marked syllable, [xA] / [GA] the most marked.
There is no way to represent a non-high vowel in an Onset, (11a), because there is no A element. But we still have I and U, so the onsets [j] and [w] can be represented. There is no way to represent an obstruent in a nucleus, (11b), because we have no stop element ? or noise element h; but this does not affect the so-called “syllabic” (i.e. nuclear) sonorants, i.e. nasals, which I assume to have a low tone element L as their head, or liquids, whose head is the coronal element R. Finally, let me tell you the bad news. There is one segment type which looks as if it will be difficult to represent: Non-high nasal vowels in
8 A reviewer notes that there exist quite a number of languages with [x] as their only fricative. Although there are considerably more languages without [x], this fact (and the phonological representation of [x] in the languages concerned) should be explored in greater depth.
192 John R. Rennison languages with three tones. But do they exist?9 In Mòoré, which has a nasality distinction for [a] vs. [a%], but only two tones, we can just about scrape by, by assuming that low-tone vowels have no element for tone, but high tone vowels have an H element, as shown in (12).
(12) The low vowels of Mòoré (a Gur language of Burkina Faso, ca. 2 mill. native speakers) vowel old new vowel old new a´ (A,H,_) (H,_) a˜´ (A,H,N,_) (L,H,_) a` (A,L,_) (_) a˜` (A,L,N,_) (L,_)
This is the maximum number of distinctions that we can obtain using L and H; if there were one more distinctive tone, e.g. mid, we could not repre- sent it. (NB: Collapsing L and N was not my idea, but I find it attractive.) Finally, a quick look at the repercussions of this proposal for automatic speech recognition. It seems that this may provide the theoretical backing for work that has been done quite independently by Wiebke Brockhaus and co-workers on the recognition of vowels. If a mid vowel has an empty head, then the I or U element is an operator. But in a high vowel, the I or U must be the head. If we take headship seriously, then this suggests that the acous- tic pattern or signature of I or U will be stronger for high vowels than for mid. Of course, this is nothing new; the only point is that in the older theory headship was a matter that could be decided on one way or the other (because there were too many possible representations of mid vowels), but in my new approach there is no choice about what is the head.
Acknowledgements
I am grateful to many people for discussing issues raised here and for sharing their insights and data with me. Special thanks go to my reviewers, . Wiebke Brockhaus and Grazyna Rowicka, for many useful suggestions.
9 Stefan Ploch (p.c.) suggests that Yoruba is such a case. My own research on (one variety of) Yoruba, carried out several years before the question of “N = L” arose, came up with a two- tone system. Clearly, further investigation is required.
On empty heads and empty operators 193
References
Charette, M. (1991). Conditions on phonological government. Cambridge: Cambridge University Press. Harris, J. (1994). English Sound Structure. Oxford: Blackwell. Kaye, J.D., J. Lowenstamm & J-R. Vergnaud (1985). The internal structure of phonological elements: a theory of charm and government. Phonology Yearbook 2: 303–328. Lowenstamm, J. (this volume). The beginning of the word. 153–166. Rennison, J.R. (1996). Wann ist ein Merkmal kein Merkmal? Inhaltliche Aspekte phonologischer Strukturen in der Rektionsphonologie. Wiener Linguistische Gazette 57–59, 27-38. Rennison, J.R. (1998). Contour segments without subsegmental structures. Structure and Interpre- tation. Studies in Phonology and Syntax, edited by E. Cyran & P. Stalmaszczyk. Lublin: Wydawnictwo Folium. 227–245 Ritter, N.A. & R.M. Vago (this volume). Subsyllabic constituency in Hungarian: Implications for Moraic Phonology and Government Phonology. 289–304. Scheer, T. (this volume). On constraints vs. non-circular approaches to word-initial clusters. 289–304 Williams, G. (1998). The phonological basis of speech recognition. Unpublished Ph.D. dissertation, SOAS, University of London.
Reduplication as pure constituent copying In defence of the syllable against the phonemic melody
David Restle
Abstract
The aim of this paper is to show that contrary to the claims of Prosodic Morphology, when an appropriate syllable model is applied to reduplica- tion, then the reduplicated part represents a prosodic constituent in the base. As a result, it will not only be possible to reduce the complexity of existing theories of reduplication, but also to simplify the description of nonconcatenative processes, as will be exemplified by the formation of The Hague: Thesus. 195–217. Thesus.The Hague: 195–217.
. the Arabic broken plural. 0. Syllables !? 1. Introduction
Over the last 10 years, the theory of Prosodic Morphology (PM) has devel- oped into a relatively independent field of research, investigating the influ- Phonologica 1996: Phonologica 1996: ence of prosodic structure on morphological processes (cf. McCarthy & Prince, 1995: 318). Since the very beginning, this branch of research has concentrated on two main areas: a) concatenative phenomena, such as infixation and reduplication and b) nonconcatenative phenomena, such as the so-called broken plural in Arabic. Since the early work of Moravcsik (1978) and Marantz (1982), it has gener- ally been accepted that reduplication cannot be described as a process in which a suprasegmental constituent, such as the syllable, is simply doubled. In order not to lose sight of the obviously prosodic character of the phe- nomenon, on one hand, the base is broken down into (nonprosodic) single segments, and on the other hand, a prosodically specified template is used onto which the single segments are mapped.1
1 “A template that is affixed to a base will lead to copying or reduplication of the segments of
John R. Rennison & Klaus Kühnhammer, eds. (1999). eds. R. Rennison & Klaus Kühnhammer, John (1999). a base, which then satisfy the template. This is reduplication.” (McCarthy & Prince, 1995: 332)
196 David Restle
This complicated method is solely justified by absence of common pro- sodic constituents which correspond to the reduplicated parts of the base. This means that the usual structuring of bases is not compatible with the reduplication data, if one disregards the structure given by the phonemic melody, a structuring that results in excessive power due to its high resolu- tion. For this reason, considerably more processes are describable than occur in the data. Therefore, it seems reasonable to look for alternative phonological struc- tures that are compatible with the reduplication data.
2. Reduplication
2.1. What is reduplication?
Nothing is more natural than the prevalence of reduplication, in other words, the repetition of all or part of the radical element. (Sapir, 1921: 76)
This characterization of reduplication as a natural process is in accordance with two basic observations: a) early child language acquisition, and b) human intuition. According to Vihman (1996: 109), the child’s most striking milestone of language production in the first year is represented by the introduction of the reduplicated babbling stage. The term reduplication is only applicable to the form of the phenomenon,2 because the characteristic form-meaning- relation cannot be assumed at this “prelinguistic” stage of development. Nevertheless, we must start with the zero hypothesis that such patterns of child production do not fundamentally differ from the more elaborate reduplication patterns found in the languages of adults. The relevance of this “prelinguistic reduplication” for “linguistic reduplication” could only be challenged by the assumption of a categorical difference or a discontinuity between the babbling stage and the first-word-stage (following Jakobson, 1941). However, research has since proven that this discontinuity hypothesis is not tenable.3
2 According to Vihman (1996: 109f), drawing upon Holmgren et al. (1986), these form properties are: “Supraglottal closure, polysyllabicity, and temporal regularity of sequencing. Temporal regularity, or rhythmicity, is a product of the cyclic alternation of consonant- and vowel-like gestures which effect articulatory closure and opening, as in adult syllable production.” 3 “There is considerable evidence that there is a close connection between the specific features of babbling and of first words and that babbling and word use are not discrete stages but overlap in time as well as in substance. These facts show that there is no major
Reduplication as pure constituent copying 197
The acquisition of this basic production pattern seems to be a precondi- tion for further language acquisition — a pattern of which active use is made in the further course of development (cf. Ingram 1978: 144). The application of this basic production pattern to express semantic relations between related word forms is, in this sense, simply a further example of the eco- nomical use of linguistic resources. The second observation concerns human intuition. Most introductions to the phenomenon of reduplication point out that, according to our intui- tion, reduplication is a primitive process which copies primitive constitu- ents.4 It is, however, generally accepted as a fact that such intuition is in- compatible with the data and that reduplication is more complicated than may be first assumed. Why is there this discrepancy between facts and intuition? And how is it be explained? The currently prevailing linguistic theories of reduplication do not address this question. In this paper, I will propose an approach to re- duplication which inter alia does not suffer from such a discrepancy between facts and intuition.
2.2. What are the basic types of reduplication?
The most common reduplication patterns are listed below in (1)–(11). Exam- ples (1) and (2) pose no problems to most linguistic theories of reduplica- tion, because they indisputably show constituent copying processes (words or feet are copied). It is uncontroversial that a theory of reduplication must allow for such copying of whole words or feet. (Portions doubled by redupli- cation are shown here in bold type.)
(1) Reduplication of whole words: Walpiri (Nash, 1986) kurdu kurdukurdu ‘child SG./PL.’ kamina kaminakamina ‘girl SG./PL.’ mardukuja mardukujamardukuja ‘woman SG./PL.’
discontinuity between the so-called prelinguistic babbling period and the presumably phonological stages that begin in the second year.” (Macken, 1995: 675). Cf. also Vihman et al. (1985). 4 “The idea that reduplication involves affixing a template may seem surprising, since one might expect reduplicative operations to say something like ‘copy the first syllable.’” (McCarthy & Prince, 1995: 333) “At first sight, a case could be made for treating reduplication always as nothing more than constituent copying”. (Katamba, 1993: 182)
198 David Restle
(2) Prefixing foot reduplication: YidiN (McCarthy & Prince, 1990) mulari mulamulari ‘initiated man SG./PL.’ kintalpa kintalkintalpa ‘species of lizard SG./PL.’
Most of the remaining patterns can be described as affixing doublings of CV-, VC- and CVC-portions of the base, where C can also represent a cluster of consonants, as in (3)–(11).
(3) Prefixing CV reduplication: Ilokano (Hayes & Abad, 1989) li9?ét si-lili9?ét ‘perspiration / covered with perspiration’ buné9 si-bubuné9 ‘kind of knife / carrying a buneng’ (4) Prefixing CVC reduplication: Ilokano (Hayes & Abad, 1989) kaldí9 kalkaldí9 ‘goat SG./PL.’ púsa puspúsa ‘cat SG./PL.’ (5) Prefixing CVCV reduplication: Fox (Bloomfield, 1933) wa:pamE:wa wa:pawa:pamE:wa ‘he looks at him / keeps looking at him’ (6) Partially Infixing CV reduplication: Murut (McCarthy & Prince, 1995) bulud bubulud ‘hill / ridge’ ompodon ompopodon ‘flatter / always flatter’ (7) Infixing VC reduplication: Mangarayi (Merlan, 1989) yirag yirirag ‘father SG./PL.’ jimgan jimgimgan ‘knowledgeable person SG./PL.’ (NB: ‹y› = [j] and ‹j› = [dZÖ]) (8) Suffixing CV reduplication: Hausa (Sapir, 1921) suna sunanaki ‘name SG./PL.’ (9) Suffixing VC reduplication: Comox (Kroeber, 1988) pes peses ‘numb / get numb’ tih tihih ‘big / get big’ (10) Suffixing CVC reduplication: Mokilese (Harrison, 1976) pwirej pwirejrej ‘dirt / be dirty’ (11) Suffixing CVCV reduplication in Siriono (Moravcsik, 1978) erasi erasirasi ‘he is sick / continues being sick’
2.3. Is reduplication a process of copying constituents?
The question of whether or not reduplication is a process that copies con- stituents other than phonemes, can obviously be answered with both yes and no. Yes, in that words and stems do exist in total reduplication (see examples (1)–(2)). However, while the type of reduplication in (1) (“copy the
Reduplication as pure constituent copying 199 whole base”) occurs relatively frequently, syllable or foot based reduplica- tion, as exemplified in (2), is very rare. The remaining part and therefore, the majority of reduplications cannot be understood as a syllable copying process, because these reduplications are independent of one of the classical properties of syllables, namely their status of being open or closed. This can be seen from the data from Ilokano in (3), repeated here as (12), where only the initial part of the base up to the first vowel is copied. This results in the reduplication of the complete syl- lable in (b), but of only a part of the syllable in (a).
(12) a. li9?ét ‘perspiration’ si-lili9?ét ‘covered with perspiration’ b. buné9 ‘kind of knife’ si-bubuné9 ‘carrying a buneng’
In a second type of reduplication in Ilokano, illustrated in (4) and re- peated below as (13), the post-vocalic consonant takes part in the reduplica- tion process, independent of whether the consonant is a) an offset of the first syllable, or b) an onset of the second syllable.
(13) a. kaldí9 ‘goat’ kalkaldí9 ‘goats’ b. púsa ‘cat’ puspúsa ‘cats’
Whilst the source of reduplication, e.g. the chain of segments sa9 in the base sa´9it, apparently corresponds neither to a syllable (by the known phonological definitions) nor to any other suprasegmental constituent, the target is always a constituent, namely a heavy syllable, e.g. the first sa9 in sa9sa´9it. Therefore, the source and target would appear to share only the phonemic melody and their order. However, they obviously do not share their prosodic affiliation, because the 9 is part of the first (and only) syllable of the reduplicant, but the onset of the second syllable of the base.
2.4. Currently prevailing approaches
Two principal proposals have been made for tackling the aforementioned problems. The first proposal (see e.g. McCarthy & Prince, 1995) trims down all prosodic structure that appears to be incompatible with reduplication, and reconstructs the reduplicated form with the remaining phonemes, as illustrated in (14) with Ilokano pusa ‘cat’ and its plural form puspusa.
200 David Restle
(14) a. F F b. c.
µ µ
p u s a → pusa → pusa + pusa →
d. F e. F F F
µ µ µ µ µ µ
p u s a + pusa → p u s p u s a
In (14b), the base is trimmed down to its phonemic melody. The resulting string is then reduplicated in total (see (14c)), resulting in the string pusa + pusa. The first part, i.e. the reduplicant, is then mapped to a template, which, in this case, consists of a bimoraic syllable (see (14d)). The template is always maximally filled. Remaining material that fails to syllabify, e.g. the a of the reduplicant in (14d), undergoes stray erasure and is therefore suppressed. The final step is illustrated in (14e): the reduplicant and the base are con- catenated, and the missing prosodic structure of the base is rebuilt. The second proposal (see Steriade, 1988) differs in that the prosodic structure is not erased. Instead, resyllabification is used to produce the correct output.
(15) a. F F b. F F F F
µ µ µ µ µ µ p u s a → p u s a p u s a →
c. F F F F d. F F F F
µ µ µ µ µ µ µ µ µ µ p u s a p u s a → p u s a p u s a →
e. F F F
µ µ µ µ p u s p u s a
Diagram (15b) illustrates Steriade’s view that reduplication involves the copying of all of the prosodic structure of bases. In a second step, the pro- sodic affix (again a bimoraic syllable) crystallizes at the left edge of the reduplicated form, thereby introducing a second mora (see (15c)). This newly introduced mora maps to the following consonant (the first s in (15d) above).
Reduplication as pure constituent copying 201
Because the second syllable of the reduplicant is not licensed by the pro- sodic affix, the syllable is erased, yielding the output puspusa.
2.5. Weaknesses of the current theories
In assessing the first approach, we must be aware of its huge generative power: Single segments are copied or, conversely, templates are filled phoneme-by-phoneme. Although this generative capacity can be limited by a second step which limits the admitted templates and the method of tem- plate mapping, nevertheless the copying and the limiting of the copying are two independent modules. As a result of this independence, the plausibility of constraints must be derived from other areas, such as a general theory of affixation or general metrics, in order to at least partially eradicate the character of ad hoc con- straints (e.g. as in Marantz’s (1982: 446f) Conditions A–D). But even explicit reference to theories of affixation and metrics can hardly explain the choice of some constraints. There is no doubt that, for example, the so-called Prosodic Morphology Hypothesis of McCarthy & Prince (1995: 318), which claims that only authentic units of prosody may be used in the description of templates, as well as the Minimality Hypothesis (p.344), are permissible hypotheses. Nevertheless, they depend heavily on the chosen theoretical framework. Furthermore they are stipulative, since there is no secondary evidence (e.g. psycholinguistic evidence) for their relevance to reduplication. From this interplay between atomization of the phonological structure of the base and an independent module which assigns new structure to these atoms, problems result which are known under the cover term of prosodic transfer. There are many suggestions for solving this problem, yet a compre- hensive and convincing solution is still pending, even according to the main proponents of this theory (McCarthy & Prince, 1995: 338). Other flaws in the atomization of the phonological structure of the base can also be identified. When describing reduplication on the basis of pho- nemes, one wonders how reduplication in the early phases of language acquisition actually functions. Since the 1970’s we have known that the organization of phonology during the early stages of reduplicated babbling is based on units such as a syllable or a word, but certainly not phonemes (cf. Vihman, 1986: Ch. 9). This means that the first approach to reduplication, which depends on splitting bases into phonemes, is not capable of de- scribing the earliest stage of development involving reduplication. Because a strict division between prelinguistic and linguistic phonetics/phonology in general is no longer held, the first approach runs into difficulties.
202 David Restle
The second approach (Steriade, 1988) avoids the problems of prosodic transfer by leaving the prosodic structure intact during the copying process. It also appears capable of bypassing the problem of the organizing unit in early language acquisition; however, this approach is forced to make heavy use of resyllabification (from sa.9it-sa´.9it to sa9.it-sa´.9it, then sa9.-sa´.9it). This measure is obviously ad hoc; furthermore, it illustrates that even this approach heavily depends on the phoneme as an independent unit, since the very independence of the phoneme is a prerequisite for the phoneme to adjoin to another prosodic unit. But there are even greater problems. Both approaches can only deal with some types of reduplication by the introduction of further complexities, which lack motivation. These problems stem from the patterns that were introduced as “partially infixing CV-reduplication” and “VC-reduplication”. In both cases, PM is forced to fade out, or (to use the correct terminology) to negatively circumscribe, some portion of the base that is not a type of constituent desirable in that framework, namely, onsets and onsetless syllables. In moraic theory, neither onsets nor onsetless syllables can be circumscribed since onsets do not contribute to syllable weight. The data from Murut in (16) below shows that the first CV-sequence is doubled, as was the case with the light reduplication in Ilokano in (12). McCarthy & Prince cannot tackle this problem without weakening their theory. It is only by means of joining Optimality Theory (OT) and PM that their theory gains observational adequacy.
(16) (Partially) Infixing CV-reduplication in Murut bulud ‘hill’ bubulud ‘ridge’ ompodon ‘flatter’ ompopodon ‘always flatter’
Irrespective of the question of whether the constraints selected in the framework of OT are borne out by the reduplication data, and of the useful- ness of OT in other areas, another question arises: Why is it that PM can handle the reduplication type exemplified in (12) without OT, but not the type exemplified in (16)? There is no pre-theoretical reason to assume a higher degree of complexity in the latter case. The problem only arises when we try to describe this data within the framework of moraic theory. It results from the inability of moraic theory to make a moraic distinction between VC-syllables and CVC-syllables, and it evaporates when we change the theoretical framework. If we use the syllable model proposed by Vennemann (1988b), (16) can be described as a doubling of the BODY of the first COVERED SYLLABLE. Both (12a) and (16) can thus be seen as examples of body reduplication of the first covered syllable.
Reduplication as pure constituent copying 203
The data in (16) do not favour the application of OT in cases that cannot be described in moraic theory; rather, they show the inadequacy of moraic theory when applied to the area of reduplication. This is also pointed out by Brockhaus (1995: 216f) with reference to Sanskrit data. As a result, she notes advantages for a syllable model such as that of Government Phonology, in which onsets not only have independent status, but also internal structure. PM is faced with similar problems in the case of VC-reduplication. Exam- ples from Mangarayi are given in (17).
(17) Infixing VC-reduplication in Mangarayi singular plural yirag yirirag ‘father’ gabuji gababuji ‘old person’ jimgan jimgimgan ‘knowledgeable person’
The problem posed by this type of reduplication can be seen as a combina- tion of the problems introduced by the partially infixing type in (16) with those raised by CVC-reduplication in (13): On one hand onsets of initial syllables do not reduplicate. On the other hand the reduplicated part is scattered over two syllables of the base (e.g. ab in ga.buji, ga.ba.bu.ji). Whereas the first part of the problem could be solved by using a flat syllable struc- ture, most of the current models of syllable structure fail in describing the VC-portion internal to CVCV as a constituent. In sum, two problematic areas can be identified. The first results from the atomization of the phonological structure of the base, and the second from the selection of the syllable model in the framework of PM. What is desired is a theory of reduplication that is constrained only by the prosodic constituents supplied by phonological theory: all prosodic constituents, and not only a selection of them, should be reduplicated as a whole. As a result, the phonological constituents must tally with the por- tions of the base that are reduplicated. Most data show that CV-, VC- and CVC-portions of the base act like constituents under reduplication. The following section will introduce a phonological theory that provides for exactly those constituents in a simple or compound form.
3. The theory of syllabic oscillation
Obviously, there is no point in assembling a theory of the syllable on the basis of reduplication data alone. The model to be introduced in this article was developed independently of the phenomenon of reduplication. In fact, it stems from a study of the SYLLABLE CUT, a phenomenon shared by many
204 David Restle
Germanic languages. In German it manifests itself in minimal pairs like those given in (18).
(18) German Miete [mi:t7] ‘rent’ vs. Mitte [mIt;7] ‘middle’ Koma [ko:ma] ‘coma’ vs. Komma [kOm;a] ‘comma’ beten [be:tn@ ] ‘to pray’ vs. betten [bEt;n@ ] ‘to bed’.
This contrast of syllable cut shows a lively interaction with suprasegmen- tals like stress and syllable structure, as well as phenomena like syllable reduction (cf. Vennemann, 1991; 1994 and Becker, 1996), suggesting the prosodic character of this phenomenon. The phonological school founded by Theo Vennemann (but see also Reis, 1974; Ronneberger-Sibold, this volume) emphasizes the relevance of this prosodic concept, which can be traced back to the 16th century (cf. Restle, 1998). The phoneticians at the University of Munich have succeeded in finding a prosodic correlate of the syllable cut opposition from the articula- tory aspect. The investigation of Hoole et al. (1994) uses the method of electromagnetic articulography (EMA) which is based on a segmentation of the velocity signal into three phases. 1. a CV-segment, which will be called OPENING PHASE, 2. a VC-segment (for which the term CLOSING PHASE is used in this paper) and 3. the nucleus, which links opening and closing phase. What is measured is the time function of the velocity of the relevant ar- ticulator, e.g. of the lower lip in words like gepippe vs. gepiepe. This type of segmentation is illustrated in (19).
(19) Example of the segmentation procedure of Hoole et al. (1994) for an utterance with tense vowel (/pi:p/, left) and lax vowel (/pIp/, right). / 7 p i: p 7 / / 7 p I p 7 /
Reduplication as pure constituent copying 205
This investigation found that differences in syllable cut, which manifested themselves as differences in duration and quality of a vowel and a following consonant, did not influence the duration and the shape of opening or closing phases to any significant degree. Rather, Hoole et al. point out that:
the crucial durational difference between tense and lax vowels in- volved the nucleus segment. It should be emphasized that we do not see this segment as a kind of steady-state phase. Rather, the duration and the variability in the duration of the nucleus can more profitably be seen as a measure of the tightness of the coupling between the CV and VC movements: tight for lax vowels, loose for tense vowels. Hoole et al. (1994: 54)
What would be more natural than incorporating such a contrast in the degree of coupling into a phonological model of the syllable? The basic idea is as follows. We take two types of basic categories, namely the opening and the closing phase, as in (20), and use three degrees of coupling, i.e. relations between basic categories, shown in (21): LOOSE COUPLING to represent a smooth cut (a long vowel followed by a short lenis consonant) and TIGHT COUPLING for an abrupt cut (a short vowel followed by a longer, more fortis consonant). The third degree, namely a BLENDING COUPLING, is used to describe reduced syllables. The potential absence of any vowel (in the case of the so-called syllabic consonants) is interpreted as a reflex of a partial overlapping of opening and closing phase, i.e. the closing starts before the target of the opening, namely [7], is reached.
(20) Basic categories
opening phase
closing phase (21) Basic relations
loose coupling (longer V)
tight coupling (shorter V)
blending coupling (reduced V)
loose coupling (longer, more fortis C)
tight coupling (shorter, more lenis C)
In (22) three examples of representation within this theory are given. There are two levels: one for the syllabic oscillation consisting of the basic categories, which are opening and closing phases, and the second level which contains the respective phonological targets for each phase. It speci- fies at which articulatory configuration a movement starts and at which target configuration a movement stops.
206 David Restle
(22) Example representations Komma Koma kommen
ko om ma ko om ma ko om mV Vn
Two consecutive and different basic categories (i.e. opening + closing or closing + opening) build a syllabic oscillation. A sequence of an opening phase followed by a closing phase is called an oscillatory syllable. Each oscillatory syllable obligatorily consists of an opening and a closing phase. The outcome of this is the inability to represent closed syllables as having something additional when compared to open syllables. This can be seen below in (23), where both types of syllables, e.g. /ren/ in /ren$tV/ Rente vs. /rä/ in /rä$tV/ Räte, have a closing phase — at least in nonfinal position. The closing phase of the first oscillatory syllable in Rente starts from the articulatory configuration described by IPA [E]5 and ends with the configura- tion transcribed with IPA [n], while the closing phase in Räte starts from the configuration IPA [E] and leads to IPA [t]. In the present theory, the basic distinction between open and closed syllables is expressed as the difference in the degree of tightness of the coupling between the closing phase and a following opening phase. This is exemplified in (23).
(23) “Weight by position” as a difference in independence of a closing phase from the following opening phase Rente [rEn(:)th7] ‘pension’ rette [rEth(:)7] ‘save’ Räte [rE:th7] ‘councillors’
re en tV re et tV rE Et tV
Example (24) shows representations of reduced syllables and of hiatus configurations. The latter are seen as defective oscillatory syllables, a view resulting inter alia from typological considerations.
(24) Syllable reduction as overlapping of opening and closing phase; hiatus configuration retten [rEth(:)7n] ‘save’ Hiat [hia:t] ‘hiatus’
re et tV Vn hi ia at
As the theory of syllabic oscillation is not the focus of this paper, the inter- ested reader is referred to Restle (1998). I now turn to a discussion of the application of syllabic oscillation to reduplication.
5 Since the phonological opposition between /e/ [e] and /ä/ [E] is neutralized under abrupt cut, e can be used to represent short [E], as in (23).
Reduplication as pure constituent copying 207
4. The phenomena of reduplication as seen from the viewpoint of a theory of syllabic oscillation
In this section, our goal will be to show that the problems mentioned in §2.5 are avoidable by choosing an appropriate syllable model. The following illustration confines itself to three types of reduplication: first, prefixing CV and CVC reduplication, second, the problematic cases of (partially) infixing reduplication, and finally the suffixing VC reduplication type.
4.1. Ilokano
Ilokano exhibits two of the most common types of reduplication. The terms light vs. heavy reduplication, as well as the data itself, are taken from Hayes & Abad (1989).
4.1.1. Light (CV-) reduplication
Light reduplication “copies the initial consonant or cluster of a stem, plus the following vowel” (Hayes & Abad, 1989: 357). Examples are given in (25).
(25) Prefixing CV-reduplication in Ilokano li9?ét ‘perspiration’ si-lili9?ét ‘covered with perspiration’ pandilí9 ‘skirt’ si-papandilí9 ‘wearing a skirt’
The reduplicated portion of the base corresponds to a basic category of the syllable model described in §3, namely the opening phase. The light redupli- cation of Ilokano produces a copy of this opening phase, as in (27). The insertion of a corresponding default closing phase, whose targets are de- termined by the environment, happens automatically and obligatorily: it starts where the previous opening ends, and ends where the following opening starts. Such an indeterminate closing phase is necessary for rea- sons of production: We must first close the vocal tract before we can reopen it. The rule corresponding to light reduplication is formulated in (26).
(26) Light reduplication Copy the first opening phase and supplement a default closing phase. (27) li9?ét lili9?ét
li i9 ?e et li lii9 ?e et
208 David Restle
4.1.2. Heavy (CVC-) reduplication
The second type of reduplication in Ilokano also copies the closing phase, as illustrated in (28)–(31).
(28) Prefixing CVC-reduplication in Ilokano kaldí9 ‘goat’ kalkaldí9 ‘goats’ púsa ‘cat’ puspúsa ‘cats’ (29) Heavy reduplication Copy the first oscillation (opening and closing phase). (30) kaldí9 kalkaldí9
ka al di i9 ka al ka al di i9 (31) púsa puspúsa
pu us sa pu us pu us sa
4.2. Timugot Murut
The partially or pseudo-infixing type of reduplication of the Austronesian language Murut is handled in the same way as light reduplication in Ilokano. Again, it is the first opening phase that is copied; the moraic part of PM was unable to express this parallelism.
(32) Murut (data from McCarthy & Prince, 1995: 348) bulud ‘hill’ bubulud ‘ridge’ ompodon ‘flatter’ ompopodon ‘always flatter’ (33) Murut reduplication Copy the first opening phase and supplement a default closing phase. (34) ompodon ompopodon
om po od do on om po po od do on
We turn now to the suffixing and infixing VC(V) reduplication patterns, which can be handled without any additional devices.
4.3. Spokane
The suffixing VC reduplication pattern can be observed as one of four reduplication patterns in the Salish language Spokane. Comparing the examples in (35) below with those from (25) above, we see that this type of
Reduplication as pure constituent copying 209 reduplication is a mirror-image of the light reduplication of Ilokano in (26), i.e. that instead of the opening phase, the closing phase is copied.
(35) Spokane (Carlson, 1989) -VC (OUT OF CONTROL) kw’ul’ ‘make, do’ kw’ul’ul’ [kw’ul’l’] ‘it was born, created’ šil ‘chop’ šilil [šlil] ‘it accidentally got cut’ CV- (DIMINUTIVE) kw’ul’ ‘make, do’ kw’ukw’ul’ [kw’ukw’l’] ‘sth. small is created’ šil ‘chop’ šišil [ššil] ‘it accidentally got cut’ CVC- (PLURAL) kw’ul’ ‘make, do’ kw’ul’kw’ul’ [kw’l’kw’ul’] ‘a bunch of people are doing sth.’ šil ‘chop’ šilšil [šlšil] ‘a bunch of things are chopped’
(36) Spokane -VC (OUT OF CONTROL) reduplication Copy the first closing phase supplemented by a default opening phase. (37) šil šilil šlil
ši il ši il il šV Vl il šil šišil ššil
ši il ši il il ši ši il
The fact that both types, CV- and -VC reduplication, operate simultaneously, not only in Spokane, but in many other Salish languages (see e.g. Mattina, 1993; Carlson & Thompson, 1992), should be seen as strong evidence that the obvious symmetry between these two patterns should lead to a sym- metrical treatment within reduplication theory.
4.4. Washo
One of the most intriguing patterns of reduplication is found in the Hokan language, Washo. What looks at first like the well-known pattern of suffixing CV reduplication is in fact a type of prefixing and partially infixing doubling, as can be seen from the examples in Clearly, it is the last opening phase which is prefixed. Contrary to the patterns described so far, however, the copied opening phase is not realized in the position adjacent to its source, but in front of its preceding closing phase. As a result, this reduplicated opening phase materializes as a prefix when the base is vowel initial (i.e. starts with a closing phase), and as an infix when the base is consonant initial (i.e. starts with an opening phase). This pattern is illustrated in (39). (38).
210 David Restle
Clearly, it is the last opening phase which is prefixed. Contrary to the pat- terns described so far, however, the copied opening phase is not realized in the position adjacent to its source, but in front of its preceding closing phase. As a result, this reduplicated opening phase materializes as a prefix when the base is vowel initial (i.e. starts with a closing phase), and as an infix when the base is consonant initial (i.e. starts with an opening phase). This pattern is illustrated in (39).
(38) Washo (Winter, 1970) Singular Plural da?a da?á?a ‘mother’s brother’ ?élel ?elélel ‘mother’s father’ ?éwši? ?ešíwši? ‘father’s brother’ íši-š ši-ši- š ‘heavy’ íšp’i-š p’í-šp’i- š ‘narrow’ ín?y-ni ?yi-n?yi-n ‘varicoloured’ (39) ?éwši? ?ešíwši?
?e ew ši i? ?e ši (e)w ši i?
The resulting adjacency of the vowel configurations of the (copied) last opening phase of the base and the penultimate closing phase results in a coalescence of the two vowels. In most cases, it is the quality of the vowel of the copied opening phase that wins over the vowel of the closing phase (cf. ?ešíwši?, not ?ešéwši?), but there are some exceptions (see Winter, 1970).
4.5. Mangarayi
Our illustration of the application of syllabic oscillation to reduplication concludes with a description of the infixing VC(V) reduplication of Mangarayi, which shares this type, according to Evans (1995: 758), with other Australian languages such as Jingilu, MalakMalak, Warumungu and Kugu- Nganhcara. The relevant data are given in (40).
(40) Mangarayi (Merlan, 1989) yirag yirirag ‘father SG./PL.’ jimgan jimgimgan ‘knowledgeable person SG./PL.’ guryag guryuryag ‘lily/having a lot of lilies’
The last example above suggests that this pattern should not be ana- lysed as a type of infixing VC reduplication, i.e. doubling of the first closing
Reduplication as pure constituent copying 211 phase, but as an instance of VCV reduplication, i.e. of a complete oscillation. Although it would be unproblematic to regard img in jimgan as a closing phase (because there is little evidence in Mangarayi of loose vs. tight coup- ling between a closing and a following opening phase, e.g. there is no con- trast of syllable weight), it is impossible to analyse ury in guryak as a closing phase, because the vocal tract is more open for the production of the semi- vowel than for the rhotic. As a result, the y cannot be seen as a part of the closing phase. Therefore, Mangarayi is analysed here as an instance of a VCV reduplication pattern, which copies a complete oscillation.
(41) Mangarayi reduplication Copy both the first closing phase and the second opening phase. (42) jimgan jimgimgan
ji im ga an ji im g(a) img ga an
The [a] target of the second opening phase is overridden by the [i] of the copied first closing phase. This analysis is unproblematic in view of the fact that Mangarayi exhibits neither vowel length nor diphthongs. Support for this analysis comes from German reduplication (cf. Schindler, 1991), which exhibits the pattern exemplified in (43). (The semantic effect of these reduplications is discussed in Schindler, 1991).
(43) German (Schindler, 1991) klimpern klimpimpern ‘tinkle on the piano’ rumpeln rumpumpeln ‘rumble’ knutschen knutschutschen ‘snog’
The example knutschutschen shows that the opening of the second syllable is copied, too. The analysis proposed by McCarthy & Prince (1995) for the Mangarayi data cannot express the fact that both consonants in the onset of the second syllable are copied. The application of McCarthy & Prince’s Onset Rule would only require the inclusion of the first member of the onset, resulting in knututschen instead of knutschutschen. Therefore, only the VC(C)V analysis illustrated in (44) is possible for the data above.
(44) knutschen knutschutschen
knu ut tSV Vn knu ut tSV ut tSV Vn
This section has shown that the successful description of the major part of the problematic reduplication data hinges upon the selection of an ap- propriate model of the syllable. Therefore, it is astonishing that mainstream
212 David Restle
PM has not made much of an effort to find alternatives to moraic theory. The replacement of moraic theory is not only advantageous for reduplica- tion, but also for the description of other so-called nonconcatenative alter- nations such as the BROKEN PLURAL of Arabic.
5. The Arabic broken plural
In Arabic there are two ways of forming plurals, namely a) affixation (the so- called PLURALIS SANUS or EXTERNAL PLURAL, cf. Brockelmann, 1992: 84), and b) internal modification of the singular form (internal or broken plural). McCarthy & Prince (1990) showed that the pluralis sanus is restricted to a relatively marginal part of the nominal lexicon. The vast majority of nouns have broken plurals, and in particular IAMBIC BROKEN PLURALs, whose name stems from their initial iambic sequence CVCVV. Some examples, taken from McCarthy & Prince (1990: 217), are given in (45)–(49). (Pharyngealized t is shown as t!.)
(45) CVCV:C jaziir + at jazaa?ir ‘island SG./PL.’ kariim + at karaa?im ‘noble SG./PL.’ (46) CV:CVC xaatam xawaatim ‘signet-ring SG./PL.’ jaamuus jawaamiis ‘buffalo SG./PL.’ (47) CVCCV(:)C jundub janaadib ‘locust SG./PL.’ ! ! sult aan salaat iin ‘sultan SG./PL.’ (48) CVCC nafs nufuus ‘soul SG./PL.’ qidh- qidaah- ‘arrow SG./PL.’ (49) CVCVC ?asad ?usuud ‘lion SG./PL.’ ?inab /?anaab/ [?a!naab] ‘grape SG./PL.’
These plurals share the following properties: 1. In most cases the vocalism of the last three syllables is /a-a-i/; the excep- tions appear to be unpredictable and therefore lexical. 2. The vowel of the first syllable of the plural form is always short, that of the second syllable always long. 3. In most cases the length of the vowel of the third syllable correlates with that of the second-syllable vowel in the singular. 4. Both the /?/ in (45) and the /w/ in (46) are default consonants. Their position seems to depend on the position of the long vowel in the singu- lar. McCarthy & Prince (1990) first trim down the lexical singular forms to their phonemic melody. Long vowels are reproduced as a sequence of two
Reduplication as pure constituent copying 213 identical short vowels. This is surprising, coming from the inventor of root- and-pattern morphology and of V/C segregation. The remaining steps can be seen in (50).
(50) McCarthy & Prince’s (1990) procedure for deriving iambic broken plurals from singular forms 1. Trim down the lexical singular form to its phonemic melody. jundub jaziir jaamuus 2. Bracket off the first two moras, leaving behind some residue. jun * jazi * ir jaa * muus 3. Map this sequence of phonemes to the first iambic foot (F1) of the plural template (CµCµµ). junuu * dub jazii * ir jaCaa * muus 4. Link the first part of the templatic vowel melody /a_i/ to the vowel positions of the foot. janaa * dub jazaa * ir jaCaa * muus 5. Concatenate the first foot with the residue. janaadub jazaair jaCaamuus 6. Associate the second part if the templatic vowel melody /a_i/ with the remaining vowel position(s). janaadib jazaair jaCaamiis 7. Insert default consonant(s) into empty C-positions. janaadib jazaa?ir jawaamiis
This description produces the correct output, but it surely cannot be re- garded as a description of a productive, synchronic process that is spon- taneously repeatable. The portions of the phonemic melody that are copied are not intuitive prosodic constituents; instead, long vowels and consonants are represented as sequences of two short phonemes in a style of segmen- talism long considered forgotten. It is only such segmentalism that makes it possible to map one so-called “part” of the long vowel to the template and to keep the other “part” in reserve for reconcatenation. Again, this type of atomization is very problematic for cognitive reasons, and I doubt the descriptive adequacy of this solution. How could children acquire such a complex mechanism at an early developmental stage? If the formation of the broken plural is a synchronic process, we have to uncover a simpler mechanism. The following description of the phenome- non in the framework proposed in this paper shows, again, that the re- placement of moraic theory by the theory of syllabic oscillation makes the problems disappear. We take up one of McCarthy’s ideas: that the process involved here is infixation. In order to fulfil the criteria of descriptive and explanatory
214 David Restle adequacy, it is important to fulfil two basic conditions: a) the position at which infixation occurs is directly accessible to natural observation, and b) what is being infixed is a primitive phonological object. These conditions are fulfilled by the following analysis: a) The position at which infixation occurs is the first loose coupling. b) The infix consists of an additional oscillation. First, the position: The first loose coupling of the singular forms, in the left column of (52)–(56), may occur after the first opening phase, as in jaamus, after the first closing phase, as in jundub, or after the second opening phase, as in kariim+at. There is also the possibility that no loose coupling occurs at all, as in ?inab in (55); in this case, no infixation takes place. Next we turn to the infixed entity: an additional oscillation. This oscilla- tion starts with an opening phase if the first loose coupling occurs after a closing phase, as in (52), and with a closing phase if the first loose coupling occurs after a opening phase, as in (53) and (54). Since the infixed oscillation replaces the first loose coupling of the singular form, the loose coupling itself is lost. Next the vowels are replaced and the loose coupling of the second oscil- latory syllable is specified. When the trigger of infixation is absent, as in (55), no infixation takes place. However, the specification of vowels and of the coupling of the second syllable functions as expected. The steps of the new derivation are summarized in (51).
(51) Procedure for deriving broken plurals from singular forms 1. Insertion: Insert one additional oscillation (opening + closing or closing + opening) into the position of the first loose coupling (if none is available, do nothing). 2. Specification a) Make the coupling inside the second syllable (i.e. the coupling between the second opening and closing phase) loose. b) Change the target position of the first three opening phases to /a-a-i/. 3. Make it pronounceable: Insert default consonants.
This procedure works with entities which are transparent to naïve speak- ers. The basic idea can be formulated as: “Wait until the syllabic oscillation is delayed for the first time. Use this time to open and close the mouth once. Then proceed.” This can be done simultaneously, since no complex “cut & map” processes are involved.
Reduplication as pure constituent copying 215
(52) jundub janaadib
ju un du ub ja an a di ib (53) jaamuus jawaamiis
ja am mu us ja Ca am mi is (54) kariim(at) karaa?im
kaar ri im ka ar ra Ci im (55) ?inab /?anaab/
?i in na ab ?a an na ab (56) nafs nufuus
naaf s nu uf u s
6. Conclusion
The aim of this investigation was to apply the existing phonological model of syllabic oscillation, for which there is independent phonetic and phono- logical evidence, to the new domains of reduplication and the Arabic broken plural. The model avoided most of the problems presented under the frameworks of PM and led to simple, natural descriptions. I am hopeful that the results of this study will provide a new direction for further research on these fascinating phenomena.
Acknowledgements
For useful comments I am grateful to Theo Vennemann, Thomas Becker and Laura Catherine Smith.
References
Becker, T. (1996). Zur Repräsentation der Vokallänge in der deutschen Standardsprache. Zeitschrift für Sprachwissenschaft 15, 3–21. Bloomfield, L. (1933). Language. New York: George Allen & Unwin. Brockelmann, C. (1992). Arabische Grammatik. 24th edition. Leipzig: Langenscheidt. Brockhaus, W. (1995). Skeletal and suprasegmental structure within government phonology. Frontiers in phonology, edited by J. Durand & F. Katamba, 180–221. London: Longman. Carlson, B.F. (1989). Reduplication and stress in Spokane. IJAL 55, 204–213. Carlson, B.F. & L.C. Thompson (1982). Out of control in two (maybe more) Salish languages. Anthropological linguistics 24, 51–65. Clements, G.N. (1985). The problem of transfer in nonlinear phonology. Cornell working papers in linguistics 5, 38–73.
216 David Restle
Evans, N. (1995). Current issues in the phonology of Australian Languages. The handbook of phono- logical theory, edited by J.A. Goldsmith, 723–761. Cambridge, Mass.: Blackwell. Hall Dunn, M (1990). A phonetic study of syllable structure in Finnish and Italian. Papers from the 26th regional meeting of the Chicago Linguistic Society. Volume 2. The parasession on the syllable in pho- netics and phonology (=CLS 26), edited by M. Ziolkowsky, 97–111. Chicago: Chicago Linguistic Society. Hammond, M. (1988). Templatic transfer in Arabic broken plurals. Natural language and linguistic theory 6, 247–270. Harrison, S.P. (1973). Reduplication in Micronesian languages. Oceanic Linguistics 12, 407–454. Harrison, S.P. (1976). Mokilese reference grammar. Honolulu: University of Hawaii Press. Hayes, B. & M. Abad (1989). Reduplication and syllabification in Ilokano. Lingua 77, 331–374. Holmgren, K., B. Lindblom, G. Aurelius, B. Jalling & R. Zetterström (1986). On the phonetics of infant vocalization. Precursors of early speech, edited by B. Lindblom & R. Zetterström. Basingstoke: Macmillan. Hoole, P., C. Mooshammer & H.G. Tillmann (1994). Kinematic Analysis of Vowel Production in German. Proceedings of the ICSLP ’94, Yokohama. Vol. 1, 53–56. Ingram, D. (1978). The role of the syllable in phonological development. Syllables and segments, edited by A. Bell & J.B. Hooper, 143–155. Amsterdam: North-Holland. Jakobson, R. (1941). Kindersprache, Aphasie und allgemeine Lautgesetze. Uppsala [Reprinted in R. Jakobson (1962). Selected writings I: Phonological studies. Den Haag: Mouton, 328–401]. Katamba, F. (1993). Morphology. Basingstoke: Macmillan. Kroeber, P. (1988). Inceptive reduplication in Comox and interior Salishan. IJAL 54, 141–167. Kroos, C. (1996). Eingipflige und zweigipflige Vokale des Deutschen? Kinematische Analyse der Gespannt- heitsopposition im Standarddeutschen. M.A. thesis, University of Munich. Macken, M.A. (1995). Phonological acquisition. The handbook of phonological theory, edited by J.A. Goldsmith, 671–696. Cambridge, Mass: Blackwell. Mattina, A. (1993). (V)C2 Reduplication in Colville-Okanagan, with Historical Notes. American Indian linguistics and ethnography in honour of Laurence C. Thompson, edited by A. Mattina & T. Montler, 215–235. (=University of Montana Occasional Papers in Linguistics 10.) Missoula, Mont. Merlan, F. (1989). Mangarayi. London: Routledge. Marantz, A. (1982). Re reduplication. Linguistic Inquiry 13, 435–482. McCarthy, J.J. (1979). Formal problems in Semitic phonology and morphology. PhD dissertation, MIT. McCarthy, J.J. (1983). A prosodic analysis of Arabic broken plurals. Current trends in African Linguistics. Volume I, edited by I. Dihoff, 289–320. Dordrecht: Foris. McCarthy, J.J. & A.S. Prince (1990). Foot and word in prosodic morphology: The Arabic broken plural. Natural Language and Linguistic Theory 8, 209–283. McCarthy, J.J. & A.S. Prince (1995). Prosodic morphology. The handbook of phonological theory, edited by J.A. Goldsmith, 318–366. Cambridge, Mass: Blackwell. Moravcsik, E.A. (1978). Reduplicative constructions. Universals of human language. Volume 3: Word structure, edited by J.H. Greenberg, 297–334. Stanford: Stanford University Press. Nash, D. (1986). Topics in Walpiri grammar. New York: Garland. Reis, M. (1974). Lauttheorie und Lautgeschichte: Untersuchungen am Beispiel der Dehnungs- und Kürzungs- vorgänge im Deutschen. Munich: Fink. Restle, D. (1998). Silbenschnitt – Quantität – Kopplung: Zur Geschichte, Charakterisierung und Repräsenta- tion der Anschlußprosodie unter dem Blickwinkel einer Oszillationssilbentheorie. Doctoral Dissertation. University of Munich. Ronneberger-Sibold, E. (this volume). Ambisyllabic consonants in German: Evidence from dialectal pronunciation of lexical creations. 247–271. Sapir, E. (1921). Language: An introduction to the study of speech. New York: Harcourt, Brace & Co. Schindler, W. (1991). Reduplizierende Wortbildung im Deutschen. Zeitschrift für Phonetik, Sprach- wissenschaft und Kommunikationsforschung 44, 597–613 .
Reduplication as pure constituent copying 217
Sommer, B.A. (1970). An Australian language without CV syllables. IJAL 36, 57–58. Steriade, D. (1988). Reduplication and syllable transfer in Sanskrit and elsewhere. Phonology 5, 73–155. Vennemann, T. (1988a). The rule dependence of syllable structure. On language. Rhetorica phonologica syntactica. A festschrift for Robert P. Stockwell from his friends and colleagues, edited by C. Duncan-Rose & T. Vennemann, 257–283. London: Routledge. Vennemann, T. (1988b). Preference laws for syllable structure and the explanation of sound change. Berlin: Mouton de Gruyter. Vennemann, T. (1991). Syllable structure and syllable cut prosodies in modern Standard German. Certamen phonologicum II. Papers from the 1990 Cortona Phonology Meeting, edited by P.M. Bertinetto & M. Loporcaro, 211–243. Turin: Rosenberg & Sellier. Vennemann, T. (1994). Universelle Nuklearphonologie mit epiphänomenaler Silbenstruktur. Universale phonologische Strukturen und Prozesse, edited by K. Ramers, 7–54. Tübingen: Niemeyer. Vihman, M.M. (1996). Phonological development: The origins of language in the child. Cambridge, Mass.: Blackwell. Vihman, M.M., M.A. Macken, R. Miller, H. Simmons & J. Miller (1985). From babbling to speech: A re-assessment of the continuity issue. Language 61 , 397–445. Winter, W. (1970). Reduplication in Washo: A restatement. IJAL 36, 190–198.
Subsyllabic constituency in Hungarian Implications for Moraic Phonology and Government Phonology
Nancy A. Ritter & Robert M. Vago
Abstract
Compensatory lengthening and epenthesis processes of Hungarian are analysed in terms of structural elements appropriate for two drastically different theories of prosodic organization: Moraic Phonology and Gov- ernment Phonology. Theoretical conclusions are drawn for each frame- work. Within the Moraic Phonology approach, it is suggested that the The Hague: Thesus. 219–245. Thesus.The Hague: 219–245. . timing units of syllabic organization are better represented in terms of contentless x nodes, rather than c and v nodes containing featural specifi- cations, and that some coda consonants are not affiliated with the mora
Syllables !? (weight) units of syllabic organization. Within the Government Phonology approach, governing domains defined over the structural elements of the onset, nucleus, and rhyme are shown to obviate reference to the struc- tural units of the mora and the syllable. Notwithstanding the differences, the two approaches are unified in providing an explanation for variations Phonologica 1996: Phonologica 1996: of epenthesis in terms of differing lexical structures. 0.
1. Introduction
Hungarian is especially well-suited for the study of prosodic organization: it has long vowels, geminate consonants, shortening and lengthening phe- nomena, and a host of insertion and deletion processes that affect syllable structure.1 In this contribution we have a twofold purpose: (a) to motivate for Hungarian some prosodically relevant structural units, and (b) to draw
1 Detailed discussion and analysis of the Hungarian phonological system can be found in
John R. Rennison & Klaus Kühnhammer, eds. (1999). eds. R. Rennison & Klaus Kühnhammer, John (1999). Vago (1980) and Kiefer (1994), among others.
220 Nancy A. Ritter & Robert M. Vago inferences for two influential theories of prosodic structure, namely Moraic Phonology and Government Phonology.
2. Moraic Phonology
The framework of Moraic Phonology has significantly advanced the under- standing of the prosodic organization of languages. The principles of mora theory have been motivated by extensive investigations of both general phonological patterning and morphological/templatic structure; they figure prominently in the emerging framework of Optimality Theory (e.g. McCarthy & Prince, 1986; Prince & Smolensky, 1993). In particular, the process of com- pensatory lengthening (CL) has contributed its fair share in promoting the mora as an essential unit of prosodic representation (see especially Hock, 1986; Hayes, 1989). In the following subsections, we will examine some of the CL evidence for the mora as a structural unit in Hungarian, consider entail- ments for the representation of timing tier units, and, on the basis of both CL and epenthesis phenomena, take on the question of maximal syllabic weight.
2.1. Vowel length
The standard dialect of Hungarian contrasts 7 short and 7 long vowels:2
(1) a. Short vowels b. Long vowels i [i] ü [ü] u [u] í [i:] u´´ [ü:] ú [u:] ö [ ö] o [o] é [e:] o´´ [ö:] ó [o:] e [E] a [O] á [a:]
In (1), each vowel is given in its orthographic representation followed by its phonetic value. Unless noted otherwise, we will cite Hungarian forms and refer to individual segments in their orthographic representations. Over and above the phonetic facts, we can be certain that the two sets of vowels in (1) differ primarily in quantitative terms: as corollary to a number of phonological deletion processes, the vowels in (1a) are converted to their respective counterparts in (1b), i.e. they undergo what typological studies (e.g. Hock, 1986; Wetzels & Sezer, 1986; Hayes, 1989) unmistakably identify as CL. We may briefly refer to a well-established diachronic case. When
2 This dialect is identified by Nádasdy (1985) as Standard Literary Hungarian, distinguished in some respects from what he calls Educated Colloquial Hungarian.
Subsyllabic constituency in Hungarian 221
Hungarian lost root-final reduced vowels in open syllables, a process com- pleted by the end of the 13th century, the vowel of the preceding syllable was lengthened. This process produced vowels that belong to set (1b). Note the following examples (Kálmán, 1972).
(2) wizi > víz ‘water’ tüzü > tu´´z ‘fire’ utu > út ‘road’ nyaru > nyár ‘summer’ kezi > kéz ‘hand’
In sum, on the basis of articulatory duration as well as phonological pat- terning, we take it as incontrovertible that the vowels under (1b) are long. For detailed studies and discussion, see further Vago (1992), Nádasdy & Siptár (1994), and Kassai (1994). As concerns the representation of vowel length within the general frame- work of Moraic Phonology, short vowels and their respective long counter- parts have the same melodic (segmental) representation, except for the fact that long vowels have an extra element on the timing tier. Moraic Phonology, in particular, has advanced two theories of timing. Hayes (1989) argues for the interpretation of length in terms of prosodic frames: a segment is long if it is associated with two mora elements, as in the case of long vowels, or with a mora and a syllable node, as in the case of intervocalic geminate consonants. Proponents of the Moraic Phonology framework assume nearly universally that long segments have a single node on what Clements (1985) identifies as the segmental root tier. However, Selkirk (1990) proposes that long segments have double root nodes on which prosodic structures are erected. The following visual representations contrast the two theories of vocalic length in Moraic Phonology.
(3) Hayes (1989) Selkirk (1990) Syllable tier F F Mora tier µ µ µ µ Root tier v v v
The symbols c and v represent the root tier units of consonants and vowels, respectively. Earlier conceptions of prosodic structure recognized the skeleton as a central representational tier, containing either C/V elements (see especially McCarthy, 1985 and Clements & Keyser, 1983), or abstract x timing units (see most notably Kaye & Lowenstamm, 1984 and Levin, 1985). In general, Moraic Phonology does not recognize the skeleton as an in- dependent tier (in addition to the root tier); see in particular the arguments
222 Nancy A. Ritter & Robert M. Vago of Hyman (1985), McCarthy & Prince (1986), and Hayes (1989). We will present an argument in favour of contentless skeletal timing tiers in §2.3. As seen in (3), length is derived interpretively from double associations above the root tier in Hayes’s theory, and from double root representations (leading to double prosodic associations) in Selkirk’s theory. For present purposes, we will assume the correctness of Selkirk’s arguments for the two root representation of length; see further Vago (1992).
2.2. Weight by position: Compensatory lengthening
One of the basic insights of Moraic Phonology is that only mora bearing segments may induce CL upon deletion (cf. Hayes, 1989). The basic mecha- nism of CL is as follows: (a) some moraic segment is deleted below the mora tier, leaving its mora element behind; (b) this segmentally empty mora then reassociates to some neighbouring segment, usually a vowel on the left, bringing about the lengthening of that segment.3 On this basis, we may rely on CL phenomena to diagnose that codas bear weight in Hungarian. We will consider two separate cases. The first case entails compensatory lengthening in syllables closed by a single consonant. Over a dozen historically old verbal and nominal roots exhibit the following alternation in root-final position: long vowel before consonant-initial suffixes and word-finally, short vowel plus /v/ before vowel-initial suffixes. Representative paradigms of three such nouns are provided in (4).
(4) Nominative Inessive 1pl.poss. fu´´ fu´´-ben füv-ünk ‘grass’ ko´´ ko´´-ben köv-ünk ‘stone’ lé lé-ben lev-ünk ‘liquid’
We can immediately dispense with analysing the above alternation pat- tern in terms of an underlying final long vowel, augmented by vowel short- ening and /v/-insertion: the productive pattern, which exhibits no alterna- tion in length, would have to be treated as exceptional. Contrast for instance the paradigms in (5) with those in (4) above.
3 Mora elements are stable in that they escape the scope of deletion; as such, they undergo relinking.
Subsyllabic constituency in Hungarian 223
(5) Nominative Inessive 1pl.poss. tu´´ tu´´-ben tu´´-nk ‘needle’ no´´ no´´-ben no´´-nk ‘woman’ kávé kávé-ban kávé-nk ‘coffee’
The inessive forms evidence alternations due to vowel harmony. These alternations have no significance for the points raised herein; they will therefore be ignored. The last column shows the effects of vowel elision: a suffix-initial vowel drops out if the preceding morpheme ends in a vowel. Taking syllable structure into account, the alternations in (4) are under- standable if we assume that the roots end in a consonant underlyingly. This consonant is syllabified into onset position if a vowel follows; the fact that the root vowel is short under such circumstances argues for setting up the root vowel as short. However, if onset syllabification is not possible, as in the nominative (citation) form (no syllable follows) and before consonant- initial suffixes (onset position is already filled), then the root-final consonant disappears and the preceding vowel lengthens.4 The second case of compensatory lengthening, shown in (6), concerns the deletion of /l/ in doubly closed syllables in varieties of colloquial Hungarian, especially in casual and fast speech styles.
(6) Standard Colloquial küld ‘send’ [küld] [kü:d] bolt ‘shop’ [bolt] [bo:t] kulcs ‘key’ [kulc4] [ku:c4]
We assume that the standard forms represent the underlying representa- tions of the colloquial surface forms as well: multidialectal speakers can produce either output. Note in particular the correspondence of standard /Vl/ to colloquial /V:/.5 There no doubt are additional factors governing the process of /l/- deletion in regional dialects and in the various subdialects of the colloquial dialect; for discussion, see Nádasdy & Siptár (1994). What is of significance for present considerations is that the loss of /l/ always entails the lengthen- ing of a short vowel, and further, that CL may take place in doubly closed syllables. These facts strongly argue that in Hungarian postvocalic coda
4 Hungarian does not permit syllabifying a morpheme-final consonant into the syllable onset of a following CV-initial suffix (see Törkenczy, 1994). 5 According to Imre (1972), /l/-deletion cum CL, both in singly and doubly closed syllables, is characteristic of the majority of Hungarian dialects.
224 Nancy A. Ritter & Robert M. Vago consonants, or at least a subset of them, are moraic. We explore possible theoretical consequences in the next two sections.
2.3. Timing tier units
Previously we have motivated an analysis of the root alternations exempli- fied in (4) in terms of a final consonant which deletes in coda position and induces CL. It is observed that the root-final consonant shows up as /v/ if not subject to deletion. However, there are reasons to suppose that at the underlying level this consonant is not /v/. There exist roots ending in /v/ which, however, do not exhibit the effects of CL. In point of fact, these roots represent the productive paradigm, followed by the majority of /v/-final roots, e.g.
(7) Nominative Inessive 1pl.poss. öv öv-ben öv-ünk ‘belt’ sav sav-ban sav-unk ‘acid’
Clearly, the productive paradigm would have to be treated in exceptional terms if roots like those in (4) had a final /v/. Nor can an analysis positing a final long vowel plus /v/ for the root class in (4) be maintained in its full generality: cf. for example év ‘year,’ év-ben, év-ünk, among many other roots of this type, where length alternation does not take place. It should be obvious that any other melodically specified consonant in the final position of our representative roots in (4) can be assumed only at the expense of gratuitous abstractness. To bypass this undesirable con- sequence, and still maintain a general analysis in terms of CL, we could set up an empty consonantal root tier unit (unspecified for any other melodic feature) in final position (see Vago, 1989 for details). In this way, the CL process is effected within the double root theory of length in three steps (see Selkirk, 1990). First, the empty consonantal root node is deleted if it is moraified via weight by position (see Hayes, 1989), i.e. if it is syllabified as a coda. Second, the resultant empty mora unit is filled on the root tier with a vowel node. And third, the empty vowel root tier node is associated with the preceding vowel. The resultant structure, in which two vowel roots are associated with a single melody (and with two moras) is interpreted pho- netically as a long vowel.6 A sample derivation is provided in (8).
6 Equivalently, in Hayes’s (1989) single root theory of length, the insertion of a second vowel root node is obviated and the empty mora of the deleted coda consonant is associated directly with the single root node of the preceding vowel.
Subsyllabic constituency in Hungarian 225
(8) F F F
µ µ µ µ µ µ c v c → c v → c v v
f ü f ü f ü
If the empty consonant is syllabified into onset position, as is the case if a vowel-initial suffix follows, then weight by position is inapplicable. Instead of deletion, the weightless empty consonant receives default values, which in Hungarian equates with /v/ (see Vago, 1989). Now an examination of the account of CL in (8) immediately reveals the cumbersome nature of the process on the view that the timing tier is essen- tially defined as the segmental root tier, represented in terms of c and v units. More precisely, deleting a consonantal root and replacing it with a vocalic root is a stipulation that lacks explanatory force. If, instead, we represent the timing tier units in terms of abstract x slots, CL originating in coda deletion receives a simple and straight-forward explanation, in the spirit that Hayes (1989) originally proposed: it operates exclusively in the segmental domain, leaving prosodic structure intact, as in (9).
(9) F F
µ µ µ µ x x x → x x x
f ü f ü
We thus take CL phenomena of this kind to lend support to the view that the timing tier should be represented in terms of x slots. We will adjust our representations accordingly.
2.4. Trimoraic syllables?
The conventional view within the Moraic Phonology framework admits only two types of syllable weights: monomoraic and bimoraic. However, Hayes (1989) argues that a few phonological phenomena can be explained only if the upper limit of intrasyllabic moras is extended to three. Part of the evi- dence derives from CL in doubly closed syllables. The argument is as fol- lows. The quintessential component of CL is an unassociated mora unit, which gets incorporated into the preceding vowel. Now that mora is
226 Nancy A. Ritter & Robert M. Vago rendered free by the deletion of the first coda consonant. Clearly, the sec- ond coda consonant cannot belong with the first one to the same mora unit, for after deletion the mora would not be unattached and available to induce vowel lengthening. Thus, if doubly closed syllables are bimoraic, CL is not expected to follow, as illustrated in (10).
(10) F F
µ µ µ µ x x x x → x x x
k ü l d k ü d *[küd]
Trimoraic structures, on the other hand, derive empty moras, hence CL is an automatic consequence, as in (11).
(11) F F F
µ µ µ µ µ µ µ µ µ x x x x → x x x x → x x x x
k ü l d k ü d k ü d Following Hayes (1989), it has nearly universally been assumed that all coda consonants are gathered into mora units. On this view, trimoraic syllables are forced, at least in doubly closed syllables, as evidenced by CL phenomena. Yet this position flies in the face of overwhelming evidence, gathered from both Prosodic Phonology and Prosodic Morphology, that languages go out of their way to limit the number of weight units per syllable to two. Both this constraint and the requirements of CL can be respected under an alternative view that attaches the consonant(s) of the “third” mora directly to the syllable (or a higher level unit), on a par with onset con- sonants, as in Empirical evidence to choose between the trimoraic analysis and what we may call the appendix analysis is not easy to come by. In what follows, we wish to offer justification for the latter position.
(12).7 Empirical evidence to choose between the trimoraic analysis and what we may call the appendix analysis is not easy to come by. In what follows, we wish to offer justification for the latter position.
7 Csúri (1990) stakes out this position as an alternative to assuming trimoraic syllables in Hungarian (as in Vago, 1992).
Subsyllabic constituency in Hungarian 227
(12) F F F
µ µ µ µ µ µ x x x x → x x x x → x x x x
k ü l d k ü d k ü d
Hungarian contains a rule (see Vago, 1980; Nádasdy & Siptár, 1994) that inserts a low vowel between a verbal stem ending in a consonant cluster and a consonant-initial suffix; we will refer to it as Low Vowel Epenthesis (LVE). Morphemes with coda clusters, nearly all underived roots, may condition LVE optionally or obligatorily (rarely, not at all); we may assume lexical markings for the appropriate effects. Compare for example the paradigms of the /CC/-final verbs in (13b), where epenthesis obtains, with those of the /V(C)/-final verbs in (13a), where epenthesis is not triggered.
(13) 3pl.indef. Infinitive a. üt ‘hit’ üt-nek üt-ni vár ‘wait’ vár-nak vár-ni no´´ ‘grow’ no´´-nek no´´-ni b. áld ‘bless’ áld-anak áld-ani olt ‘extinguish’ olt-anak olt-ani zeng ‘resound’ zeng-enek zeng-eni
Consider now verb stems ending in /V:t/. On the face of it, these stems should not condition LVE, since a vowel precedes /t/. In point of fact, though, they do: they behave exactly as those stems in which a consonant precedes stem-final /t/ (e.g. the stem olt in (13)), as shown in (14).
(14) 3pl.indef. Infinitive fu´´t ‘heat’ fu´´t-enek fu´´t-eni mu´´t ‘operate’ mu´´t-enek mu´´t-eni
Three verbal roots ending in /V:t/ are exceptional in that they systematically fail to condition LVE. Of these, only lát ‘see’ is common: lát-nak, lát-ni. Let us now review, in (15), the patterning of /t/-final verb stems with re- spect to LVE.
228 Nancy A. Ritter & Robert M. Vago
(15) a. Stems ending in a short V + /t/ do not condition LVE: e.g. üt-ni. b. Stems ending in a C + /t/ condition LVE: e.g. olt-ani. c. Stems ending in a long V + /t/ condition LVE: e.g. fu´´t-eni. d. A few stems ending in a long V + /t/ do not condition LVE, e.g. lát-ni.
If mora theory permits trimoraic syllables, defining the triggering charac- teristics of LVE appears problematical. One cannot refer to the melodic level for applicability: with respect to segmental structure, üt and fu´´t are identical (bear in mind that long /u´´/ is represented featurally as short /ü/), yet they systematically differ as far as triggering LVE is concerned. Nor can a con- sonant cluster be pegged as the triggering agent for epenthesis: fu´´t ends in a single consonant yet allows LVE to apply. Finally, and crucially, it cannot be said that LVE is conditioned by a preceding trimoraic syllable. It is un- controversially the case that weight by position does not moraicize a coda consonant following a long vowel (cf. Hayes, 1989). Indeed, in Hungarian no CL process causes extra length on an already long vowel (Nádasdy & Siptár, 1994). Therefore, stems like áld and bánt ‘harm’ must be bimoraic and are not expected to induce LVE. But this prediction is false: cf. áld-ani, bánt-ani. In contrast, the appendix analysis encodes the triggering of LVE in struc- tural terms. Simply put, LVE applies if and only if the stem ends in a non- moraic consonant, i.e. an appendix. Representative stem structures are provided in (16).8
(16) a. Stems conditioning LVE F F µ µ µ µ
x x x x x x x
f ü t o l t
8 In imperative inflections, stem-final /t/ palatalizes to /s4/ if it is mora bearing (e.g. /üt+j/ → [üšš] ‘hit!’), to /c4/ if it is an appendix (e.g. /olt+j/ → [olc4c4] ‘extinguish!’). The palatalization facts of the lát and the fu´´t types fall out of their respective structures postulated here: cf. [lášš] ‘see!’ vs. [fü:c4c4] ‘heat!’ See Vago (1980; 1987; 1991) for discussion across different theo- retical models.
Subsyllabic constituency in Hungarian 229
b. Stems not conditioning LVE F F µ µ µ µ x x x x x x
ü t l a t
3. Government Phonology
In this section, we would like to offer an alternative approach to analysing a number of the phenomena discussed in §2, without assuming higher pro- sodic elements such as the mora or the syllable. In the model to be pro- pounded, relations between phonological objects attached to skeletal points are determined by structural governing domains which subsume the notions of locality and adjacency. By such an approach there is no longer a need to refer to supraskeletal organizing nodes, such as the mora or the syllable, in order to differentiate contextual relationships between skeletally adjacent points. §3.1 presents the fundamental precepts which underlie the notion of governing domains. In §3.2 arguments and evidence are given for general- izing the branching rhyme structure to include long vowels. This structure then is claimed to be a crucial component which conditions the possibility of epenthesis, and is claimed to be contained within lexical representations which show effects of epenthesis. §3.3 discusses the variations of epen- thesis that exist with respect to certain verbal stem types. It offers an ac- count which attempts to explain such variation based on differing lexical structures and the manner in which these structures are licensed.
3.1. Fundamentals
The approach proposed here is based on the Government Phonology (GP) model proposed by Kaye, Lowenstamm & Vergnaud (hereafter: KLV) (1990), Kaye (1990), and Charette (1991), among others. In this theory, the prosodic well-formedness of onset-rhyme sequences is determined by various types of governing relations between skeletal points. The three constituents recognized by this theory are the onset (O), nucleus (N), and rhyme (R). The rhymal constituent dominates the nucleus and a post-nuclear rhymal position, if present. Segments are linked to these constituents by means of an intermediary skeletal or timing tier, represented by x slots. Each
230 Nancy A. Ritter & Robert M. Vago constituent constitutes a governing domain which may be binary, unary, or null.9 A governing domain which is binary within a constituent is represented by structural branching, the head of which is initial, and which is subject to strict locality and strict directionality, as seen in (17).
(17) O N R N x x x x x x
A second governing relation occurs between skeletal points of different constituents. In this type of governing domain, the governing head is final and government operates from right to left, all the while adhering to con- straints on strict locality (linear adjacency) on the skeletal tier, and strict directionality (right to left). An example of this type of government is given in (18), where the structure represents the relation between an onset and its preceding rhymal complement position (traditionally known as the coda).
(18) R O N x x x
The fundamental phonological governing relations depicted in (17) and (18) are representations of canonical primitive relations which underlie Universal Grammar. As a result, such governing relations, established at the level of lexical representation, must be maintained throughout phonological processing. The Projection Principle (KLV, 1990) encodes this concept and ensures its inviolability. Such a principle guarantees that structure preserva- tion of a domain will ensue and that resyllabification cannot be invoked. Other principles of this theory will be discussed as they pertain to the analysis being presented.
3.2. Conditioning epenthesis
In considering the phenomenon of LVE, we have identified two classes of stems which trigger the possibility of epenthesis: one type ends with a long
9 Which constituents may branch in a language is determined by parametric setting. In Hungarian the only constituent not parametrically set for branching is the onset.
Subsyllabic constituency in Hungarian 231 vowel and the coronal stop /t/ (V:C), and the other ends in a short vowel followed by a (homorganic) consonant cluster (CC). Representations of these two verbal stem types, adhering to the GP model, are given in (19).
(19) a. V:C b. VCC R O R O N N x x x x x x V C V C C
The V:C structure in (19a) contains a branching nucleus, with the final consonant segment linked to a following onset. Due to the fundamental notion of binary branching of constituents in government relations, the final surface consonant underlyingly must be relegated to the onset position following the branching rhyme. The VCC structure in (19b) is represented by having the first half of the consonant cluster in the post-nuclear rhymal position and the second half of the cluster in the following onset. Again, the distribution of segments within constituents results from the binary theorem since the two positions of the branching rhymal constituent, i.e. the nucleus head and its comple- ment, will be filled respectively by the vowel and the first consonant of a cluster. While the rhymal constituent is a projection of its head (the nucleus), the rhymal constituent is a viable constituent itself and will be subject to the same universal concept of binary branching as the other constituents are. Therefore, structures containing branching nuclei within branching rhymes, as in (20) below, will always be disallowed, since the head position of the nucleus is not strictly adjacent to the post-rhymal constituent and so cannot govern it; a skeletal point (of the dependent nuclear position) intervenes between the governor head and the governee.
(20) *R*
N x x x X
Kaye (1990) also proposes that the presence of a post-nuclear rhymal position must be licensed by a following onset and formulates this in terms of the Coda Licensing Principle. In effect, this principle ensures that con- sonants are universally maximized into the onset position when they are
232 Nancy A. Ritter & Robert M. Vago intervocalic or domain-final. Additionally, under this principle, a segment that would previously have been considered to be an extraprosodic element will now have its place in a following onset position. In keeping with the notion that GP is a theory of prosodic well-formedness of onset-rhyme sequences, such word-final consonants belonging to an onset position, as in (19), must be followed by a rhymal constituent headed by a nucleus, albeit a phonetically empty one in such cases.10 Whereas extraprosodicity in the standard treatment has been a non-uniform language-specific property, in GP such segments are treated uniformly as lexically linked to the onset of a well-formed prosodic domain. Furthermore, the Coda Licensing Principle does away with the problem of determining at what level such extraprosodic segments should be syllabified in order to be prosodically licensed. Instead, prosodic licensing is an automatic consequence of the governing relations in a GP approach. At this point we will depart somewhat from traditional GP theory and propose a structure which captures a generalization about the representa- tions in (19) that trigger epenthesis, and which thus serves to unify these two representations. In each of the representations in (19), it can be seen that the rhymal constituent, a maximal projection of its nuclear head, evidences variations in the way it can branch in each of these. If we abstract from the specifics of each structure, we are left with a nuclear head position and some dependent position to the right of this head in both instances. Sup- pose we generalize the structure then to include both the (19a) and the (19b) possibilities. In other words, we would like to posit a general structure, as represented in (21), in which the rightmost position under the rhyme, im- mediately following the nucleus head, is interpreted as a DEPENDENT RHYMAL POSITION, which can be either vocalic or consonantal.
(21) R N x x
V V/C
In order to be licit, the dependent rhymal position will be followed by an onset position as per the Coda Licensing Principle. Together, this post- rhymal position and its following onset are bound in a transconstituent
10 Whether the presence of empty nuclei in word-final position is licensed or not is a para- metric choice of languages. Languages which parametrically select to license final empty nuclei will exhibit domain-final consonants, e.g. English, French, Hungarian.
Subsyllabic constituency in Hungarian 233 governing relation. This is just one of the structural conditions that we claim participates in the phenomenon of epenthesis. Encoded in this inter- constituent relation are also the notions of sonority sequencing and shared phonotactic properties. As segments are further removed from the nucleus head, they tend to have lower sonority. Thus, homorganic clusters of son- orant-obstruent and geminates commonly occur in this site. If now we extend this dependent rhymal position to include the second half of a long vowel, it would have to follow that certain phonotactic constraints or co- occurrence restrictions must also hold between the long vowel and its following consonant in onset position under this interconstituent domain. There is in fact evidence in Hungarian that illustrates that a phonotactic relationship does exist between a vowel in the dependent rhymal position and a following consonant in onset position. Such evidence then justifies positing one unifying structure as in (21). The first piece of evidence rests on the issue of how to structurally repre- sent a sequence of vowel and semivowel in Hungarian, as in the example hajt-ani ‘to drive (INF.)’. One approach would be to argue that the glide or semivowel is the dependent of a branching nuclear structure, in other words, the second component of a heavy diphthong. However, this argu- ment would not hold, since Hungarian lacks diphthongs in its inventory. The semivowel then must follow the nucleus as a consonantal object in what has traditionally been called the coda position, and, by being in this position, the semivowel is then subject to phonotactic restrictions with respect to its following onset. Thus, lexical items which are represented by the structure in (21) and which contain a high front glide in the dependent rhymal posi- tion, are generally followed by a coronal stop in the following onset, as in hajt. In such cases, both objects within the transconstituent governing domain share the property of coronality. By generalizing branching within the rhyme to consist of a nuclear head and a dependent rhymal position, the possible structural contradistinction of the position of the semivowel and its interpretation as either vocalic or consonantal is thus obviated. Further evidence to underscore that a phonotactic relationship exists between a vocalic object in the dependent rhymal position and a consonan- tal object in the following onset can be seen in the overall majority of verbs triggering epenthesis, in which a long vowel is followed by a consonant, generally a coronal stop. The long vowel is a front vowel, usually high, as in fu´´t ‘heat’. Since the articulatory production of a high front vowel is charac- terized by the raising of the tongue blade toward the hard palate, precisely one of the areas which is subsumed within the wider phonological concept of coronality, a high front vowel is considered to fall within the categoriza- tion of coronal. Once again, it seems that lexical items which have the
234 Nancy A. Ritter & Robert M. Vago structure of that in (21) manifest the sharing of coronality under the transconstituent governing domain. We should emphasize at this point that the structure in (21) does not sub- sume the representation of all long vowels in Hungarian. Rather, by extend- ing the context in (21) to include long vowels, we are proposing that this is a marked structure for only those long vowels which show evidence of some combinatorial restriction with their following consonant. We follow Ritter (1995) in assuming that the unmarked structure for long vowels consists of two separate syllabic nuclei which are skeletally adjacent. We will discuss the implications of these two structures for epenthesis in §3.3. Further evidence that the dependent rhymal position can contain either a vocalic or consonantal segment comes from the facts of CL, which can be seen in stems such as küld ‘send’, discussed in §2.2. In such cases, the /l/ in dependent rhymal position deletes and the vowel in the nuclear head lengthens to fill this position. Under a traditional GP approach, deletion of the segment attached to the rhymal coda position could never trigger a long vowel in a branching nuclear position since the binary theorem disallows the simultaneous presence of a second nuclear position and a coda posi- tion (as in (20) above).11 Moreover, a branching nuclear structure could not simply materialize at some point in the derivation due to the Projection Principle, which admonishes against altering underlying governing relations already established at the level of lexical representation. Consequently, it would be difficult to account for this phenomenon in such terms. However, by positing only one rhymal dependent position, which can contain both a vocalic as well as a consonantal object, CL is more readily explicable. The last piece of evidence that can be adduced in claiming that the sec- ond half of long vowels can reside in the dependent rhymal position comes from those dialects in which stems ending in the verbal derivational suffix -ít are produced with the short reflex of the vowel followed by a geminate /tt/ (Grétsy & Kovalovszky, 1983). While the structure for these different alter- nants is the same in both cases, variation in production is determined by which segment is construed to be in the dependent rhymal position. The evidence presented above has attempted to justify the validity of the unified structure proposed in (21), which extends the notion of a branching rhyme to also include long vowels. Essentially, it is the presence of the dependent rhymal position which is a crucial component in triggering the effects of epenthesis in certain verbal forms. By generalizing this structure then, we are able to unify two traditionally different groups of lexical items
11 In keeping with the spirit of the Projection Principle, the presumption is that the underlying structure would already have to contain a dependent nuclear position (albeit an empty one) as well as a dependent rhymal coda position.
Subsyllabic constituency in Hungarian 235 which participate in epenthesis (i.e. long vowels plus consonant, and short vowel plus consonant cluster) by positing a uniform underlying structural representation which subsumes both varieties. Conversely, stems whose lexical representations do not contain a dependent rhymal position are thus predicted to not exhibit effects of epenthesis. Stems which contain a short vowel and one final consonant, such as üt ‘hit’ in (22), should not trigger epenthesis since they do not have a post-rhymal position in their structure which requires governing by the following onset.12 This prediction is borne out by such paradigmatic forms as üt-nek and üt-ni, cited in (13a) above.
(22) R O R N N x x x
ü t
Since no dependent rhymal position is present, there is no interconstituent governing domain to trigger epenthesis. Similarly, stems with long vowels followed by a consonant which do not trigger epenthesis are claimed not to have a dependent rhymal position in their structures. Long vowels of this type are posited as being represented by the unmarked structure for long vowels in Hungarian, in which there are two separate syllabic nuclei which are adjacent at the skeletal level (see further Ritter, 1995). This is the case with lát ‘see’.
(23) O R O R O R
N N N x x x x x l a t
Since the long vowel is represented as separate nuclear constituents, no rhymal branching appears and thus no dependent rhymal position exists to condition epenthesis. Both the structures in (22) and (23) then show the absence of a dependent rhymal position. As a result, there is no structural
12 The stem-final consonant here could not be claimed to be generated in the post-rhymal position (as per the Coda Licensing Principle discussed above) since there is no following consonant that could govern and thus license it as a coda consonant. Other verbs ending in a short vowel and single consonant (e.g. ad ‘give’) will have a similar structural representa- tion and, as a consequence, will not trigger epenthesis.
236 Nancy A. Ritter & Robert M. Vago condition of an interconsituent governing relation which warrants the pos- sibility of epenthesis in either of these cases.
3.3. Optional vs. obligatory epenthesis
While the phenomenon of low vowel epenthesis has been analysed in dif- ferent models, little or no focus has been given to explaining the obligatory vs. optional status of epenthesis with certain verbal stems, other than to claim that such differences are marked as such in the lexicon. In this section, we look at the variations that arise with epenthesis and show that the vary- ing degrees of epenthesis reflect the varied makeup of lexical structures of such items, especially with respect to governing domains and the valid licensing of such domains. Thus far we have attributed the triggering of epenthesis to the presence of a final interconstituent governing domain which requires the presence of a dependent rhymal position and a following onset to license the dependent position. However, not all stems which appear to end in a consonant cluster on the surface have the structure in (21). In other words, a dependent rhymal position followed by an onset is not the only type of interconstituent gov- erning domain that this theory admits. Governing relations can also be established between successive onset constituents, as illustrated in (24) below. Therefore, some stems exhibiting final clusters on the surface have another possible underlying representation which also captures the close relationship of the two components of the cluster, yet in a different struc- tural context.
(24) 1O1 2O2 O R O N
x x x
In the structures of stems of the type represented in (24), the first mem- ber of the cluster is in an onset position (O1) and the second member of the cluster is in a following onset position (O2) such that the two onsets enter into an inter-onset governing relation, as seen with the verb hall ‘hear’ in (25).
Subsyllabic constituency in Hungarian 237
(25) 1O1 2O2 O R O R O R N N N x x x x x x h a l l
The empty nucleus positioned between these two onsets remains empty by virtue of its being inside this governing domain. In the two types of interconstituent government (cf. (18) and (24)), one position is the head or governing position and the other is the governed position. For example, in the relational domain of rhymal coda followed by onset, the onset is the head which governs the preceding post-rhymal position. Charette (1991) claims that the governing head of such a domain must be licensed or empowered to govern its dependent by some means. She proposes that government-licensing power is bestowed upon an onset head by the nucleus immediately following the governing head. She also argues that this nuclear licensor could either be phonetically overt or pho- netically null, depending upon the parametric choice a language makes with respect to the phonetic state of the nucleus as licensor. Therefore, a lan- guage would select whether a vowel need be realized or not in order to license a preceding interconstituent governing domain. In Hungarian, there is evidence which shows that nuclei that are potential licensors for preced- ing word-internal interconstituent governmental relations must be pho- netically realized in order to fulfil their licensing potential. In other words, word-internal empty nuclei are not government licensors in Hungarian. Evidence for this comes from degemination when a consonantal suffix follows a geminate-final stem, as in jobb-ra [yobrO] ‘to the right (SUBLATIVE)’
(26) X O R O R O R
N 1N1 2N2 x x x x x x x X j o b b r a
Here the empty nucleus N1, which becomes word-internal after affixation, is unable to license the onset head of the interconstituent geminate domain to govern its dependent rhymal position. The result of this rhymal position
238 Nancy A. Ritter & Robert M. Vago remaining ungoverned is phonetic non-realization, which translates into geminate reduction on the surface. However, this restriction against empty nuclei as government licensors only applies to word-internal nuclei and not to word-final empty nuclei, for word-final empty nuclei must be government licensors in this language. This is what allows for the appearance of word- final consonant clusters on the surface, as in gomb ‘button’. As with the post-rhymal/onset relations, inter-onset relations, as seen in (24), also require word-internal nuclear licensors to be phonetically content- ful in order to license the government within their preceding structural domains. Therefore, it appears that in order for the relevant licensing rela- tions in Hungarian to be satisfied, two conditions must hold true: (a) there must be some possible relationship between the segments attached to the positional components of the structural governing domain, and (b) there must be a locally adjacent phonetically overt nucleus which can act as licensor for the word-internal structural domain to be licit. The first condition can be satisfied in terms of either sonority sequenc- ing, element sharing, or perhaps even head alignment, depending upon the theory of elements being used. The requirements of the second condition are twofold: the given licensor for the word-internal interconstituent domain must be a full vowel, and the nuclear licensor must be locally adjacent to the onset/head governor. It is the interplay of these two components of con- dition (b) that we propose explains the varying types of epenthesis effects, namely: (a) some stems trigger epenthesis optionally (e.g. tanít ‘teach’, áld ‘bless’, mond ‘say’); (b) some stems trigger epenthesis obligatorily (e.g. hall ‘hear’, fest ‘paint’); and (c) some stems never trigger epenthesis in current usage, even though they look as though they might (e.g. varr ‘sew’, áll ‘stand’).13 Let us first consider the structure of those stems which exhibit an inter- constituent relation of rhymal dependent followed by an onset, as with tan-ít ‘teach’ (‘study-DERIV.’) in (27).
13 A subgroup of the verbs which never trigger epenthesis has been pointed out by Nádasdy (p.c.). While the verbs in this subgroup do not show signs of epenthesis in normal usage, in literary/archaic usage optional epenthesis may occur with some forms. Nádasdy also notes that there may be literary/archaic or obsolete historical forms of the verb áll ‘stand’ which display evidence of epenthesis. However, in current-day educated speech, epenthesis is rare and verbs such as áll are treated as ending in V:C, with a reduced geminate, as opposed to V:CC in the archaic pronunciation, where epenthesis is triggered. Consequently, it seems that the structural representation of these lexical items containing final geminates has changed in that there is no longer a governing domain present which could trigger epenthe- sis. Rather, the representation of these forms seems to pattern with items such as lát in (23). Thus, how a geminate is construed stem-finally manifests different lexical structural con- ditions that will either trigger epenthesis or not.
Subsyllabic constituency in Hungarian 239
(27) O R O R O R N N N x x x x x x x t a n i t
Once a following consonant-initial suffix is added, such as the infinitival -ni, the interconstituent domain and its following empty nucleus are no longer word-final, as in (28).
(28) O R O R O R O R
N N 1N1 2N2 x x x x x x x x x t a n i t n i
The empty nucleus N1 appears between the derived stem tan-ít and the following consonant-initial infinitival suffix -ni. According to the Empty Category Principle of UG, an empty position must be licensed in order to remain phonetically null on the surface. Proper government is the process which licenses empty nuclei and constrains the proliferation of such posi- tions. Only phonetically full vowels can be proper governors and license the lack of phonetic content of empty nuclei. When empty nuclei cannot be licensed to remain empty, they must become phonetically realized. As with all other operations, this operation of proper government is claimed to apply whenever structural conditions warrant. Since in example (28) an empty nuclear position is present, the structure warrants the application of proper government. Yet, if proper government, which operates from right to left, were claimed to apply in this structure, N2, being fully contentful, would license the emptiness of N1, leaving N1, a word-internal empty nucleus, to license its preceding interconstituent domain. Since word-internal empty nuclei in Hungarian cannot be government-licensors, this would predict that the governing domain preceding the empty nucleus could not be a licensed domain and that some sort of reduction or non-realization of one of the domain components should occur. Yet, no such reduction is prevalent in such stems, as opposed to the geminate case in (26). Rather, in cases like (28), where sonority sequencing of segments is a phonotactic constraint on the skeletally adjacent segments in the onset- rhyme relation, proper government does not apply at all, and government-
240 Nancy A. Ritter & Robert M. Vago licensing prevails.14 It seems that when structural conditions are such that these two processes can both apply, potential structural domains estab- lished at the level of lexical representation are obliged to be governed first before the determination of the segmental quality of a single position occurs. As Cyran & Gussmann (in press) have also pointed out with respect to governing domains for onset clusters, structural relations which encode some possible segmental relation seem to be perceived as fundamentally universal types of relations which require a licit government domain. As a consequence of recognizing the fundamental nature of such relations, the process of licensing and thus establishing structural domains supersedes the application of proper government when the operation of both mecha- nisms is possible. The establishment of structural relations thus needs to be addressed first. Since the role of proper government is to constrain the proliferation of empty positions and justify their existence, licensing gov- ernmental relations does not seem to obstruct this role in any way. In fact, the reverse is true. Proper government, as pointed out in the hypothetical case, can interfere with government-licensing relations and can prevent structurally sound relational domains from being realized. Therefore it seems that government-licensing is a more fundamental principle in this regard than proper government is. Returning to example (28), it is clear that the structural relation of onset and dependent rhymal complement has the potential of being government- licensing compatible since the segments attached to these positions com- ply with the sonority sequencing principle and share some phonetic proper- ties. Since the relevant licensing relation is word-internal, it requires a fully contentful vowel in order to be realized. N1, the word-internal nucleus, is empty underlyingly and therefore cannot license the government domain. Word-final N2 is a phonetically realized nucleus; however, it is too far re- moved from the domain in question and would violate the locality con- straint on the nuclear licensor. For the interconstituent governing relation’s potential to be met and to become legitimized, there are two options avail- able. Either the locality constraint on nuclear licensors must be violated in
14 Geminates which have the structure in (26) are a different case. Underlyingly they consist of a head onset position that is elementally full and a complement consonant coda position that is elementally void. As a result of this elemental makeup, this type of domain requires a stronger or stricter form of government, namely one in which both constraints on the nu- clear licensor are strictly upheld. Such a geminate relation can only be licensed by a locally adjacent fully contentful lexical vowel, i.e. one that is base-generated. In this way, such a geminate relation cannot succeed by forcing the phonetic realization of the empty nuclear position, as is optionally possible with other onset-rhyme clusters in which each of those positions is elementally contentful.
Subsyllabic constituency in Hungarian 241 selecting a more distant viable governor, such as N2, or phonetic realization of the empty nucleus N1 must be forced. If a speaker chooses the first option, as represented in (29) below, the lexical vowel in N2 will government- license the onset head of the domain and the N1 empty nucleus will remain empty, since it falls within the licensing domain.
(29) ← Government Licensing O R O R 1O1 R O R
N N 1N1 2N2 x x x x x x x x x t a n i t n i
Therefore, while locality is violated in this case, since N2 is not strictly ad- jacent to the onset head O1, this option complies with the parametric setting which requires that word-internal nuclear licensors be fully contentful in this language. If, however, a speaker chooses the second option, shown in (30), in an at- tempt to avoid violating the locality condition, the word-internal nuclear position N1 will be construed as the relevant position which can govern- ment-license the head (O1) of the interconstituent domain and, in accor- dance with the second condition’s requirement for phonetic content in government-licensing, will be forced to be phonetically realized, yielding the effect of an epenthetic vowel on the surface.
(30) ← Government Licensing O R O R 1O1 R O R
N N 1N1 2N2 x x x x x x x x x
t a n i t a n i ↑
In such cases, ultimately both conditions are satisfied; the word-internal nuclear licensor is both a fully contentful vowel and locally adjacent. Given the fact that structures such as (29) permit a violation of locality but struc- tures such as (30) exhibit no violations of any kind, it is not surprising that, though both forms in (29) and (30) are possible, the one in (30) is preferred.
242 Nancy A. Ritter & Robert M. Vago
Thus, in current educated speech, epenthesis is the norm; however, the other option in (29) may exist in dialectal variations.15 Thus in attempting to locate a valid licensor, a speaker must opt between violating locality or forcing phonetic realization of an empty position, both serious violations. Optional epenthesis, therefore, depends on which nu- clear position is construed by the speaker to involve the less offensive violation. In this way, government-licensing of the interconstituent domain is affected in the most cost-efficient and effortless manner available. In cases of obligatory epenthesis, we propose that stems which fall into this category have an inter-onset governing relation which needs to be licensed by a following nucleus. Stems of this class include lateral geminates (preceded by short vowel) and clusters whose first member is a coronal strident and second member a coronal stop. Example (25), repeated here as (31), illustrates this structure.
(31) 1O1 2O2 O R O R O R N N N
x x x x x x h a l l
Since there is some relationship between the segments attached to the successive skeletal onset positions O1 and O2 in this potential structural relation, this domain is government-licensing compatible and requires a licensor. The nature of inter-onset relations, however, does not allow for optionality as do onset-rhyme relations. In determining a possible licensor for this domain-type when word-internal, such inter-onset relations hold locality of the licensor to be of utmost importance. This can be claimed to result from the nature of the inter-onset relation itself: onsets cannot be skeletally adjacent; a nuclear position intervenes. As a consequence, the governing onset requires a stronger licensor (one that is strictly adjacent and not removed from the government domain). In order then for the struc- tural inter-onset relation to be licensed, it forces the realization of the nu- cleus N1 immediately following it, yielding the effect of epenthesis on the
15 Nádasdy (p.c.) has pointed out that the suffix -tok/tek/tök ‘2PL.INDEFINITE’ may be more acceptable without epenthesis in educated current-day speech (e.g. mond-tok (mond ‘say), tanít-tok).
Subsyllabic constituency in Hungarian 243 surface (e.g. hallani ‘hear’, INFINITIVE). Since the inter-onset structure in such cases requires locality of its nuclear licensor, epenthesis will always be seen to occur with these structures. Therefore, word-internal geminates repre- sented as inter-onset relations will obligatorily trigger epenthesis.
4. Conclusion
This paper has attempted to investigate a range of phenomena in the pho- nology of Hungarian with a view to drawing theoretically relevant con- sequences for both Moraic Phonology and Government Phonology. With respect to mora theory, two proposals have been advanced: the x-tier skele- ton might be a more appropriate alternative to the segmental root tier as far as the representation of timing units is concerned, and further, the evidence lends support to representations in which coda consonants may reside outside of the domain of syllabic weight. Such appendices obviate postu- lating trimoraic syllables, surely a desired consequence. It remains to be seen if other putative cases of trimoraicity are amenable to reanalysis. With respect to Government Phonology, the structure of a branching rhyme has been extended to also represent marked long vowels that are found in items which trigger epenthesis. Furthermore, this analysis seems to shed a new perspective on government-licensing in Government Phonology since we claim that in Hungarian if a cluster is government-licensing com- patible, i.e. a potential structural domain meets both structural and segmen- tal conditions, it must be licensed in some way. Moreover, if the locally adjacent nucleus of such a domain is not a valid licensor, then either locality can be violated or phonetic realization of the nucleus must be forced, de- pending upon the nature of the governing domain. It is this structural gov- ernment of skeletal points that causes the locality and adjacency conditions to obviate any reference to supraskeletal objects such as the syllable.
Acknowledgements
We are grateful to Ádám Nádasdy and Péter Siptár for comments and dis- cussion.
References
Benko´´, L. & S. Imre, eds. (1972). The Hungarian language. Budapest: Akadémiai Kiadó. Charette, M. (1991). Conditions on phonological government. Cambridge: Cambridge University Press. Clements, G.N. (1985). The geometry of phonological features. Phonology yearbook 2, 225–252.
244 Nancy A. Ritter & Robert M. Vago
Clements, G.J. & S.J. Keyser (1983). CV phonology: A generative theory of the syllable. Cambridge: MIT Press. Csúri, P. (1990). Moraic theory and the syllable inventory of Hungarian. Manuscript. Brandeis Univer- sity, Waltham, Mass. Cyran, E. & E. Gussmann (in press). Consonant clusters and governing relations: Polish initial consonant sequences. The syllable: Views and facts, edited by H.G. van der Hulst & N.A. Ritter. Grétsy, L. & M. Kovalovszky, eds. (1983). Nyelvmu´´velo´´ kézikönyv. Vols. I & II, Budapest: Akadémiai Kiadó. Hayes, B. (1989). Compensatory lengthening in moraic phonology. Linguistic Inquiry 20, 253–306. Hock, H. (1986). Compensatory lengthening: In defense of the concept ‘mora’. Folia Linguistica 20, 431–460. Hyman, L. (1985). A theory of phonological weight. Dordrecht: Foris. Imre, S. (1972). Hungarian dialects. The Hungarian language, edited by L. Benko´´ & S. Imre, 299– 326. Budapest: Akadémiai Kiadó. Kálmán, B. (1972). Hungarian historical phonology. The Hungarian language, edited by L. Benko´´ & S. Imre, 49–83 Budapest: Akadémiai Kiadó. Kassai, I. (1994). A fonetikai háttér. Strukturális magyar nyelvtan. 2, Fonológia, edited by F. Kiefer, 581–665. Budapest: Akadémiai Kiadó. Kaye, J. & J. Lowenstamm (1984). De la syllabicité. Forme sonore du langage, edited by F. Dell, D. Hirst & J-R. Vergnaud, 123–159. Paris: Hermann. Kaye, J., J. Lowenstamm & J-R. Vergnaud (1990). Constituent structure and government in phonology. Phonology 7, 193–231. Kaye, J. (1990). ‘Coda’ licensing. Phonology 7, 301–330. Kiefer, F., ed. (1994). Strukturális magyar nyelvtan. 2, Fonológia. Budapest: Akadémiai Kiadó. Levin, J. (1985). A metrical theory of syllabicity. Dissertation. Massachusetts Institute of Technology, Cambridge. McCarthy, J.J. (1985). Formal problems in Semitic phonology and morphology. New York: Garland Publishing. McCarthy, J.J. & A. Prince (1986). Prosodic morphology. Manuscript. University of Massachusetts, Amherst and Brandeis University, Waltham, Mass. Nádasdy, Á. (1985). Segmental phonology and morphophonology. Approaches to Hungarian: Volume one, edited by I. Kenesei, 225–246. Szeged: József Attila Tudományegyetem. Nádasdy, Á. & P. Siptár (1994). A magánhangzók. Strukturális magyar nyelvtan. 2, Fonológia, edited by F. Kiefer, 42–182. Budapest: Akadémiai Kiadó. Prince, A. & P. Smolensky (1993). Optimality theory: Constraint interaction in generative grammar. Manuscript. Rutgers University, New Brunswick, N.J. Ritter, N.A. (1995). The role of universal grammar in phonology: A government phonology approach to Hungarian. Dissertation, New York University. Selkirk, E.O. (1990). A two-root theory of length. UMass occasional papers: Papers in phonology 14 , 123–171 . Törkenczy, M. (1994). A szótag. Strukturális magyar nyelvtan. 2, Fonológia, edited by F. Kiefer, 273– 392. Budapest: Akadémiai Kiadó. Vago, R.M. (1980). The sound pattern of Hungarian. Washington, DC: Georgetown University Press. Vago, R.M. (1987). On the representation of length. Phonologica 1984, edited by W.U. Dressler, H.C. Luschützky, O.E. Pfeiffer & J.R. Rennison, 319–324. Cambridge: Cambridge University Press. Vago, R.M. (1989). Empty consonants in the moraic phonology of Hungarian. Acta Linguistica Hungarica 39, 293–316 . Vago, R.M. (1991). Elméleti szempontok a t-végu´´ igék elemzésében. Tanulmányok a magyar nyelvtudomány történetének témakörébo´´l, edited by J. Kiss & L. Szu´´ts, 682–690. Budapest: Akadémiai Kiadó.
Subsyllabic constituency in Hungarian 245
Vago, R.M. (1992). The root analysis of geminates in the moraic phonology of Hungarian. The structure of Hungarian (Approaches to Hungarian, 4.), edited by I. Kenesei & C. Pléh, 177–194. Szeged: József Attila Tudományegyetem. Wetzels, L. & E. Sezer, eds. (1986). Studies in compensatory lengthening. Dordrecht: Foris.
Ambisyllabic consonants in German Evidence from dialectal pronunciation of lexical creations
Elke Ronneberger-Sibold
Abstract
A notorious problem of German phonology concerns the syllabification of words such as schalle ['Sal7], an inflected form of the verb schallen ‘to re- sound’, as opposed to Schale ['Sä:l7] ‘shell, dish’. Whereas theoretical pho- nologists and naive language users unanimously place a syllable bound- ary after the long/tense vowel in Schale ['Sä:.l7], informants and linguists characteristically hesitate after the short/lax vowel of schalle between ['Sa.l7] with a boundary before [l] and ['Sal.7] with an ambisyllabic medial The Hague: Thesus. 247–271. Thesus.The Hague: 247–271. . consonant. This uncertainty, it is argued, results from dialectal variation: Whereas in most Bavarian dialects, ambisyllabic consonants result from heavy phonotactic constraints, these constraints do not obtain in Stan-
Syllables !? dard German. The pronunciation of certain artificially created neologisms shows characteristic differences between Bavarian and Standard German imputable to a difference in syllabification. This is interpreted as external evidence against ambisyllabic consonants in Standard German. 0. Phonologica 1996: Phonologica 1996:
1. Introduction
1.1. External evidence in Natural Phonology
One of the main fields in which Natural Phonology has obtained important results in the last decades is “the systematic use of external evidence on a par with internal evidence” (Dressler, 1996: 47). In a linguistic theory where language is viewed not only as a self-contained system, but also as a tool for communication, fundamental insights can be gained not only by investigat- ing the structure of the system by itself, but also by systematically observing the way in which users handle this tool, how they acquire and lose the fac- ulty to use it, how they (deliberately or nondeliberately) modify it, and even
John R. Rennison & Klaus Kühnhammer, eds. (1999). eds. R. Rennison & Klaus Kühnhammer, John (1999). how they play with it. Thus, besides first and second language acquisition,
248 Elke Ronneberger-Sibold foreign language perception, aphasia, sociophonology, and language change, conscious language creation by adults is also an important source of external evidence fruitfully used in Natural Phonology. Conscious lan- guage creation by adults presupposes a kind of competence that partly depends on classical linguistic competence, but transcends it. I call it “crea- tive competence” (Ronneberger-Sibold in press).1
1.2. The problem. Conceptual and notational framework
In this article, evidence from word creation will be applied to a notorious problem of syllabification in German words such as schalle ['Sal7], an inflected form of the verb schallen ‘to resound’, as opposed to Schale ['Sä:l7] ‘shell, dish’. Whereas theoretical phonologists and naive language users unanimously place a syllable boundary after the long/tense vowel in Schale ['Sä:.l7], infor- mants and linguists characteristically hesitate after the short/lax vowel of schalle between ['Sa.l7] with a boundary before [l] and ['Sal.7] with an ambi- syllabic medial consonant. In this paper, I will refer to the first type of syl- 2 labification as the PLAIN MODEL and the other as the AMBISYLLABIC MODEL. The uncertainty concerning the syllable boundary after a short/lax vowel on the part of linguists as well as of native speakers might be taken to con- firm the claim that syllables are a linguistic artefact anyway. This would be in line with the intentions of Dziubalska-Kołaczyk (1995; this volume). It may finally turn out that she is right, but in this paper, I would nevertheless like to make a more conservative claim: Native speakers of German do syllabify their words, but they do it in different ways, for there are important dialectal differences in the structural prerequisites from which syllable boundaries can be derived. These differences are related to the segmental phonemic interpretation of the opposition between words such as Schale and schalle. Whereas there is absolutely no doubt that these two words form a minimal
1 Pioneering work in this field within Natural Phonology and Morphology was performed by Kilani-Schoch (1988; 1996) and Dressler & Kilani-Schoch (1992) for French, Sobkowiak (1991) for English, Dressler & Merlini-Barbaresi (1994) for Italian, German and other languages, and to some extent by myself (1995; 1996; forthcoming) for German and French. 2 The problem by no means depends on the theory employed. As pointed out by Ramers (1992), all modern non-linear phonological theories except Government Phonology contain formal representations of the concept of ambisyllabic consonants, which have been (or can easily be) applied to German. (Early examples are Vennemann, 1982 for Syllabic Phonology, Clements & Keyser (1983) for CV Phonology, Giegerich (1985) for Metrical Phonology, Anderson & Durand (1986) for Dependency Phonology, and Hyman (1985) for Moraic Pho- nology.) What is unclear is whether the notion of ambisyllabicity itself is really applicable to the German words under discussion.
Ambisyllabic consonants in German 249 pair in Standard German (SG), linguists disagree with respect to the pho- netic features to be considered as relevant for this opposition. Adherents to the plain model normally consider either the quantity or the quality of the stressed vowel as relevant. (Quality can in turn refer to tense- ness or centralization.) This yields the phonemic transcriptions /'Sa:l7/ : /'Sal7/ or /'Säl7/ : /'Sal7/ respectively. In this paper, I will stick to the first possibility, because it is pertinent to prosody. Adherents to the ambisyllabic model can adopt one of these two possi- bilities as well, considering the segmental phonemic opposition as irrelevant for syllabification. However, the ambisyllabic model even allows for a third phonemic interpretation of the opposition between Schale and schalle, in which, the relevant feature is the SYLLABLE CUT or CONTACT. This supra- segmental feature concerns the way in which a vowel is “cut” by a following consonant or word boundary. In Modern German, as probably in most languages, a long vowel has a long, continuous decrescendo of its intensity, before a following consonant or word boundary sets in. This is called smooth/soft cut or loose contact. Examples are Schale ['Sä:°l7], schal ‘stale’ ['Sä:°l], (ver)schalte ‘became stale’ ['Sä:°lt7], or da ‘there’ [dä:°]. In contrast to this, a short vowel is not simply characterized by a shorter decrescendo — this seems to have been the case in earlier historical stages of the language — but by no decrescendo at all: The vowel is cut immediately after its intensity peak. This phonetic feature, which, of course, enhances the phonetic differ- ence between long and short vowels, is called abrupt/sharp cut or close contact. Examples are schalle ['Sa^l7], Schall [Sa^l] ‘sound’ or schalte ['Sa^lt7], an inflected form of the verb schalten ‘to switch’. Close contact / abrupt cut before a word boundary is possible in interjections only, e.g. in Bavarian (especially Austrian) geh [gE^] ‘come on, that’s not true’ < geh! ‘go away!’. As long and short vowels are thus automatically associated with smooth and abrupt cut (loose and close contact) respectively, it is theoretically possible to consider syllable cut/contact as the primary distinctive feature. This yields the phonemic notation /'Sa°.l7/ : /'Sa^l. 7/. Ambisyllabic consonants follow from this interpretation, for phonemically relevant abrupt cut / close contact is only possible within a syllable. Therefore, the medial consonant in a word such as schalle has to belong at least in part to the first syllable. This view, which goes back at least to Sievers (1901), although less ex- plicitly formulated, is quite common in German historical phonology and dialectology. In the recent theoretically oriented literature, it has been strongly defended by Vennemann (1991; 1994) for SG. In this paper, it will be argued that, whereas syllable cut/contact and hence ambisyllabic consonants are in fact structurally relevant in certain dialects, especially in Bavarian, this is not the case in SG because the struc- tural prerequisites are radically different in these two varieties of German.
250 Elke Ronneberger-Sibold
The variation resulting from this fact in the pronunciation of lexical creations will be used as external evidence in this paper. The term “syllable cut” will henceforth be reserved for the articulatory phenomenon in the strict sense, whereas the term “contact” will receive a more abstract structural definition derived from the perceptual impression associated with the articulatory feature (see §2.4). As far as notation is concerned, I choose the notational variant used in Reis (1974), where the symbols are placed between the vowel and the consonant or the boundary, i.e. a circle for smooth cut / loose contact, a circumflex for abrupt cut / close contact. This avoids a misinterpretation of syllable cut / contact as a seg- mental feature of the vowel alone, which is suggested by other notations (including Vennemann’s) where the diacritics for syllable cut/contact are placed above the vowel. Apart from these few peculiarities, my conceptual and notational framework is as theory-independent as possible. In fact, the most fundamental notions of Prosodic Phonology in their classic definitions are sufficient: There are long and short vowels, simple, geminate and ambi- syllabic consonants, open and closed syllables as well as minimal long and short syllables. The prosodic length of syllables is counted in moras, repre- sented as m in all the diagrams in this article.
2. Survey of the dialectal situation and diachronic evolution
2.1. Classical Middle High German
In disyllabic words without medial consonant clusters of Classical Middle High German, there was a three-way opposition (which had replaced an earlier four-way opposition) concerning the stressed syllables of the types schâle /'Sa:.l7/ ‘shell, dish’ with a long open syllable, schale /'Sa.l7/, an inflected form of schal ‘stale’, with a short open syllable, and schalle /'Sal.l7/ ‘to re- sound’ (inflected form) with a long syllable closed by a geminate,3 parallel to schalte /'Sal.t7/ ‘to switch’ (inflected form) with a heterosyllabic medial con- sonant cluster. As a consequence, there were phonemic oppositions both between long and short vowels and between geminate and simple con- sonants. It is generally assumed that the distinction between open and
3 This example is taken from Fourquet (1964), an article which in many ways inspired the present paper. There is a problem with the form schale, for, according to orthodox Classical Middle High German phonology, /-7/ should be truncated in this position (after /-l/ pre- ceded by a short stressed vowel). However, as this rule is better observed by modern edi- tors of Middle High German texts than by the authors themselves, I retain this example for its expository convenience in view of the development into New High German.
Ambisyllabic consonants in German 251 closed syllables was phonetically accompanied by a difference between loose and close contact respectively. Contrary to New High German, there could thus be loose contact after a short vowel in schale ['Sa°l7]. With respect to prosodic length, two short syllables (counting one mora each) were equivalent to one long syllable, in polysyllabic as well as in monosyllabic words. For instance, schale /'Sa.l7/ was equivalent with /Sa:-/ in Schale, with /Sal-/ in schalle, with dâ /da:/ ‘there’ and with uninflected schal /Sal/ ‘stale’. Thus, the minimal polysyllabic word counted the same number of moras as the minimal monosyllabic word, namely two. These equivalences were (with minor modifications) an important element of Old High German and Middle High German quantifying poetry. This constellation is sum- marized in (1).
(1) Basic prosodic contrasts and equivalences of Classical Middle High German schâle : schale : schalle ['Sa:°l7] ['Sa°l7] ['Sa^ll7] /'Sa:.l7/ : /'Sa.l7/ : /'Sal.l7/ mm m mm Prosodic equivalences /'Sa:.-/ mm = /'Sal.-/ /'Sa.l7/ mm = /'da:/ = m m mm = /'Sal/ mm minimal monosyllabic word = minimal polysyllabic word
In most parts of the German-speaking area, the Middle High German three- way opposition was reduced to two-way oppositions in different ways. For the purpose of this investigation, three different areas will be distinguished: One comprises Central and Low Bavarian, here simply called Bavarian, the second certain parts of Alemannic, here simply called Alemannic, and the third many North and Central German dialects which are close to SG with respect to the problem under discussion, though by no means in all other respects. Here I refer to these dialects in an abbreviated manner as “Stan- dard German” (SG).
252 Elke Ronneberger-Sibold
2.2. Alemannic
The reduction of oppositions was performed in the easiest way in Alemannic: The geminates of the schalle-type were degeminated: Schalle merged with schale. We are thus left with two types of open syllables: /'Sa:.-/ in Schale with a long vowel and /'Sa.-/ in schalle and schale with a short vowel. A fuller analysis of this system will be provided in connection with SG.
2.3. Non-Alemannic dialects (first step)
In the other two dialect areas, in a first step, it was the MHG type schale /'Sa.l7/ with a short open syllable which was given up. Thus, only /'Sa:.l7/ and /'Sal.l7/ remained. The quite numerous words of the medial type /'Sa.l7/ merged with either one or the other of the two remaining types. This merger with its prosodic consequences is represented in (2).4
(2) Development in the non-Alemannic dialects (first step) schâle schale schalle /'Sa:.l7/ /'Sa.l7/ /'Sal.l7/ mm schalte /'Sal.t7/ mm mm Isochrony Minimal monosyllabic word < minimal polysyllabic word /da:/ /'Sa:.l7/ /Sal/ /'Sal.l7/ mm mmm (< means “is prosodically shorter than”)
An obvious result of the loss of the type /'Sa.l7/ was syllabic isochrony: All minimal stressed syllables were now of the same length of two moras. By the same token, the isochrony of the minimal polysyllabic word with the
4 In the segmental phonology of this stage, the opposition between long and short vowels was maintained only before consonant clusters, e.g. in Trost /tro:st/ ‘consolation’ vs. Rost /rost/ ‘rust’. As for the geminates, it is not entirely clear whether an opposition in words such as schalten /'Sal.t7n/ ‘to switch’ : (sie) schallten /'Sall.t7n/ (?) ‘they resounded’ was in fact always realized phonetically.
Ambisyllabic consonants in German 253 minimal monosyllabic word was given up. The minimal disyllable now counted three moras, the minimal monosyllable two. This was the end of quantifying poetry in German, which had been based to a high degree on the metrical equivalence of one long syllable with two short ones. It is from this stage that present day standard orthography originated: If a letter which is to symbolize a short vowel is not followed by a consonantal cluster (in which case vocalic quantity is not systematically symbolized), the following consonantal sign is doubled to indicate shortness of the preced- ing vowel, e.g. in Schall ‘sound’ or schallen ‘to resound’. By way of redundancy, length of vowels is symbolized by different means in many words, but not in a systematic manner.
2.4. Bavarian
The stage just described, with its indisputable syllabic isochrony, has been retained up to the present in Central and North Bavarian with an additional constraint concerning obstruents: After a long vowel, which is always E smoothly cut, you can only have a short lenis obstruent as in Feder ['vo e: do 6] ‘feather’, whereas after a short vowel, which is always abruptly cut, only a ^ ^ long and/or fortis consonant is possible, as in Vetter ['vo e t6], ['vo e do do 6], or ^ ['vo e tt6] ‘cousin’. The two other possible sequences, namely short, abruptly cut vowel + short lenis consonant and long, smoothly cut vowel + long and/or fortis consonant are systematically excluded. This distribution had been developing from Old High German onwards, its ultimate diachronic cause being the Second Sound Shift (see Reiffenstein, 1965), but it was carried out in full only in Bavarian. Here, the entire system has been re- organized in such a way that it nowhere contains an opposition between short and long vowels or between lenis and fortis, short and long or simple and geminate consonants. In this situation, the choice of either vocalic or consonantal length or of the fortis-lenis-contrast as phonemically relevant would be completely arbitrary. To my mind, the theoretically most satisfying of the many solu- tions is the following (elaborated in full detail in Bannert, 1976): The entire sequence of vowel + consonant forms a structural unit, and two such units are in opposition to each other. The distinctive feature is the way in which these units are internally divided into vowel + consonant, i.e. the type of contact between the two: “loose”, “smooth” or “soft” vs. “close”, “abrupt” or “sharp”. It is irrelevant whether this distinction is in fact primarily realized by a difference in vowel intensity contour depending on the intensity and/or duration of the following consonant, by relative timing of the vowel and the
254 Elke Ronneberger-Sibold consonant (as suggested by Bannert), by consonantal strength instead of consonantal duration, or by combinations of these (and possibly other) features in various proportions (as suggested by Scheutz, 1984: 27). What is important is the fact that the different types of vowels and consonants available in the system cannot be freely combined with each other. Instead, the choice of a certain type of vowel necessarily entails a certain type of following consonant and vice versa. This heavy constraint on the possible sequences of vowel + consonant implies an extremely close structural rela- tion between the two. This relation is STRUCTURALLY RELEVANT CONTACT. In the opposition between close and loose contact, loose contact is the unmarked member, because it occurs in the position of neutralization, which is before a word boundary under stress. As pointed out in §1.2, close contact in this position is possible only in interjections, in which it marks their special status outside the core of the linguistic system. Also, abrupt cut is considered phonetically more marked than smooth cut (see §3.2). With respect to syllabification, structurally relevant close contact be- tween vowel + consonant means that this sequence should not be disrupted by a boundary. Thus, only in the case of loose contact do native speakers feel a boundary before the consonant, whereas they leave intact the close contact by placing the syllable boundary later, i.e. within the long con- sonant. This consonant thus becomes ambisyllabic. Consequently, the minimal stressed syllable has two moras, the minimal disyllabic word three. (See (3).)
(3) Prosodically minimal words in Bavarian Minimal disyllables Feder : Vetter ^ ^ ^ ['vo e:°d6] : ['vo e t6] ~ ['vo e tt6] ~ ['vo e do do 6] ^ /'vo e°.d6/ : /'vo e t.6/ mm mm Isochrony Minimal monosyllabic word < minimal disyllabic word
/dÄ°/ /'vo e°.do 7/ mm mm m ^ /'vo e t.6/ mmm
Thus, for Bavarian, the argument concerning structurally relevant contact and ambisyllabicity can be summarized in the chain of implications repre- sented in (4).
Ambisyllabic consonants in German 255
(4) Chain of implications in Bavarian Constraints on structurally isochrony of ambisyllabic sequences of → relevant → → minimal stressed consonants vowel + cons. contact syllables
In a nutshell: Ambisyllabicity and hence syllabic isochrony of stressed syl- lables in Bavarian has a structural base in the form of a particular segmental distribution.
2.5. Standard German
The situation is radically different in SG. Here, the diachronic evolution differed from the Bavarian one in several important ways. First, and most important, all geminate consonants were degeminated, as in Alemannic. The structural result is similar to Alemannic, but with a different lexical distribution. For instance, the word Vogel ‘bird’ is pro- nounced /'fog7l/ in Alemannic with the MHG short vowel preserved, whereas the standard pronunciation is /'fo:g7l/ with long /o:/ due to vowel lengthening in open syllables, which was one of the ways in which the MHG type schale /'Sal7/ had been eliminated in the non-Alemannic dialects (see §2.3). In Alemannic as well as in SG, close contact was transferred from the former geminates to all syllables with short stressed vowels (including the former short open ones). Thus, this feature, which before had been con- ditioned by a type of syllable (i.e. a closed one), is now conditioned by a type of vowel (i.e. a short one). Moreover, in SG, the contrast between long and short vowels has been enhanced by a co-occurring contrast of quality: The long vowels are system- atically tenser and less centralized than the corresponding short ones. (A purely normative exception is long lax /E:/, e.g. in Väter, the plural of Vater ‘father’.) There has been a long debate on whether quantity or quality is phonemically relevant in vowels; in the view of syllabification, it is certainly quantity. Finally, numerous loans from Low German such as Krabbe /'krab7/ ‘crab’ reintroduced the sequence of a short vowel + lenis stop, which had been systematically excluded by open syllable lengthening. (Contrary to the evolution in Bavarian, the opposite combination of long vowel + fortis stop had never been excluded in SG.) As a result, all four theoretically possible combinations of long and short vowels with fortis and lenis consonants are realized in SG: long vowel + lenis, e.g. in Rabe /'ra:b7/ ‘raven’ vs. long vowel + fortis in Kaper /'ka:p7r/ ‘caper’ vs.
256 Elke Ronneberger-Sibold short vowel + lenis in Krabbe /'krab7/ ‘crab’ vs. short vowel + fortis in Rappe /'rap7/ ‘black horse’. Consequently, quantity of vowels and the fortis-lenis-contrast in con- sonants are both phonemically relevant and independent of each other in SG. This is supported by the durational analysis of my standard pronuncia- tion of the stressed vowels and their following consonants of the words Gabi /'ga:bi/, a woman’s name, Kapern /'ka:p7rn/ ‘capers’, geschlabbert /g7'Slab7rt/ ‘slobbered’, Papa /'papa/ ‘daddy’ in (5).
(5) 295 ms. 208 87 [a:] [b] ['ga:bi] Gabi ‘woman’s name’
137 ms . 59 78 [a] [b] [g7'Slab6t] geschlabbert ‘slobbered’
344 ms. 182 91 (closure) 71 (VOT)
162 [a] [ph] ['kha:ph6n] Kapern ‘capers’
146 ms . 53 93 [a] [p] ['phapa] Papa ‘daddy’
As can be easily seen, the entire sequence is shorter with a short vowel than with a long vowel. This shows that consonantal length does not compensate for vowel length or vice versa; the two in no way depend on each other.
Contrary to this, the sequences [a:bo ] and [abo :] (i.e. the only combinations possible in Bavarian) pronounced by a native speaker of this dialect in (6) are approximately of the same length, due to compensation of quantity.5
5 I am grateful to Dr. Sylvia Moosmüller of the Acoustics Research Department, Austrian Academy of Sciences, for being the speaker and the analyst.
Ambisyllabic consonants in German 257
(6) 221 ms. 145 76
[a:] [bo ]
['ga:bo i] Gabi ‘woman’s name’
211 ms . 86 125
[a] [bo :]
['bo abo :a] Papa ‘daddy’
Thus, the close structural relation between a stressed vowel and its follow- ing consonant of Bavarian does not exist in SG. Therefore, the first and basic link of the chain of implications of (4) is missing. This situation leaves us with two possible conclusions: Either the other links of the chain also no longer exist, or some, or even all of them nevertheless exist, but for differ- ent reasons. The first possibility leads to the plain model, the second to the ambisyllabic model.
3. Plain vs. ambisyllabic model in the description of contemporary German
3.1. The plain model
The simpler and more straightforward view is certainly the plain model: As there is no cogent reason to assume structurally relevant close contact in SG, the assumption of ambisyllabic consonants is not necessary either, and hence there is no syllabic isochrony in SG. Instead, a single postvocalic consonant is always preceded by a syllable boundary. Closed syllables are possible only in word-final position or with a medial, heterosyllabic con- sonant cluster, as in (7). This view is to be found in many traditional hand- books, e.g. Mangold (1990: 57).
(7) Basic prosodic contrasts and equivalences in SG and Alemannic by the plain model Schale : schalle ['Sa:°l7] ['Sa^l7] /'Sa:.l7/ : /'Sa.l7/ mm m : schalte ['Sal.t7] /'Sal.t7/ m m
258 Elke Ronneberger-Sibold
Minimal monosyllabic word = minimal disyllabic word da schalle /da:/ = /'Sa.l7/
mm m m
Theoretically, according to this model, New High German could re-establish quantifying poetry (similar to Middle High German) based on the equiva- lences of long open and long closed syllables and of these two with a se- quence of two short open syllables, as described in §2.1. This possibility is not exploited, however. I see two possible reasons for this: For one, in most disyllables, the second syllable contains schwa as its nucleus. Contrary to Middle High German, the New High German reduced vowel differs from short full vowels with respect to poetic metre: It can never receive metrical stress. For another, it seems that a prerequisite for quantifying metre is the existence of geminates, such that the opposition between long and short vowels is matched by an opposition between long and short consonants. (See §4.3 for an interesting piece of external evidence.)
3.2. The ambisyllabic model
In the ambisyllabic model, the question of poetic metre does not arise, because, in this view, the prosodic equivalence of all stressed syllables of the preceding stage still prevails. Hence, the minimal monosyllabic word is prosodically shorter than the minimal disyllable (cf. §2.3 and (8)).
(8) Basic prosodic contrasts and equivalences in SG and Alemannic by the ambisyllabic model Schale : schalle ['Sa:°l7] ['Sa^l7] /'Sa:.l7/ : /'Sal.7/ mm mm or /'Sa°.l7/ : /'Sa^l.7/ mm mm schalte ['Sa^lt7] /'Sa^l.t7/ mm Isochrony
Ambisyllabic consonants in German 259
Minimal monosyllabic word < minimal disyllabic word da schalle /da:/ < /'Sal.7/ mm mmm or /da°/ or /'Sa^l.7/ mm mmm
This line of argumentation has a long tradition, too, and it is strongly fa- voured in modern phonological theory.6 For some authors, close contact is essential to the argument, for others not. In the following, a brief account of the most important arguments will be given. 1) Close contact is important, simply because it is phonetically there. This was the position of Sievers (1901), who introduced the whole notion of contact. The question of whether a given phonetic feature discovered by sound analysis is structurally relevant or not was not yet common in lin- guistics at his time. Therefore, in non-structuralist dialectological work, the relevance of contact is sometimes simply extended from dialects to SG without consideration of the different structural situation. 2) As mentioned in §1.2 and 2.4 above, in the opposition between long, smoothly cut and short, abruptly cut vowels, the long ones are structur- ally unmarked, because they appear in the position of neutralization, i.e. word-finally under stress. The inconvenience that the vowel which is structurally unmarked is at the same time phonetically marked by its length can be avoided by considering contact as distinctive instead of length, for, phonetically, cutting a vowel before its decrescendo is con- sidered more marked than simply letting it come to an end (as argued by Trubetzkoy, 1939: 196). It is, however, well-known that structural and pho- netic markedness need not necessarily coincide. 3) Under the assumption of ambisyllabic consonants, the double con- sonant letters of modern orthography with graphic syllabification be- tween them appear more iconic than with simple consonants belonging to the second syllable only. Of course, orthography need not be (and in most cases is not) a true copy of speech, but, as pointed out by Aronoff (1992) and (with experimental support) by Derwing (1992), there is a rela- tion of mutual evocation between the two in highly literate societies. This means that graphical syllabification may influence phonological syllabifi- cation in the competence of language users. For German, this argument
6 Ambisyllabic consonants are taken for granted in the description of a modern Alemannic dialect in Auer (1989). This assumption leads to a rather questionable explanation of certain vowel lengthenings in this dialect.
260 Elke Ronneberger-Sibold
has been developed in particular by Butt & Eisenberg (1990). Close con- tact is not essential for this argument, but easily compatible with it. 4) The same holds for the fourth argument, which is a purely structural one: As demonstrated above, in the ambisyllabic model, all minimal stressed syllables of SG and Alemannic are isochronic (see (8)). This is an impor- tant generalization which a linguist would not like to miss, especially in a generative framework. (Wiese, 1988 even extends it to unstressed syl- lables as well.) A strong proponent of this argument for stressed syllables is Vennemann, who has even formulated this as a universal preference law, the WEIGHT LAW, which claims that “in stress accent languages an ac- cented syllable is the more preferred, the closer its syllabic weight is to two moras” (Vennemann, 1988: 30). For SG, this analysis is supported by his assumption that in vowels neither length nor tenseness is distinctive, but only contact, (Vennemann, 1991; 1994). He starts, so to speak, one link earlier in the chain of implications in (4). His prosodic model, which entirely dispenses with oppositions of quantity, is quite impressive and, as far as I can see, consistent; but it is not necessitated by the segmental structure of SG in the same way as structurally relevant contact in Bavarian. 5) The logical structure of the arguments summarized in this section is akin to that of the preceding section. The arguments concern phonotactic or accentual regularities which remain isolated peculiarities of German if they are not explained by ambisyllabic consonants. For instance, the obligatory length of a vowel before medial -h- as in Uhu ['u:.hu:] ‘eagle owl’ is easily explained by the fact that /h/ can only occur syllable-initially. Be- cause /h/ cannot be ambisyllabic, the preceding vowel must be long. A similar argument can be constructed from the fact that in hiatus the first vowel has to be long (cf. Laeufer, 1985; Hall, 1992; Ramers 1992.)7 In a general way, all arguments based on the notion of “possible generali- zation” have to face the fundamental problem of descriptive adequacy. As we have learnt from the debate about the abstractness of underlying forms, not all generalizations that are theoretically possible (i.e. which are observa- tionally adequate) are in fact made by native speakers. It might well be the
7 Words in which medial -h- after a stressed vowel is actually pronounced are rare and periph- eral in the German lexicon. Thus, Uhu is onomatopoeic, huhu ‘yoo-hoo’ is an interjection, Johann ‘John’ is a personal name. In words which have followed the regular evolution of German phonology (such as nahe /'na:.7/ ‘nearby, close’ or sehe /'ze:.7/, an inflected form of ‘to see’) or are spelt in analogy to these (such as gehe /'ge:.7/ and stehe /'Ste:.7/, inflected forms of ‘to go’ and ‘to stand’) medial -h- is written but not pronounced except in hyper- correct spelling pronunciation. If in such words underlying /-h-/ is assumed, the obligatory length of the preceding vowel is due to this /-h-/, if not, it is due to hiatus.
Ambisyllabic consonants in German 261 case that native speakers of SG and Alemannic live quite happily with two different sorts of stressed syllables and also with the isolated fact that before medial -h-, a vowel is always long.
4. Pronunciation of lexical creations
4.1. Lexical creations as a source of external evidence
In sum, all the internal arguments discussed so far provide a certain plausi- bility for the existence of ambisyllabic consonants in the competence of native speakers of SG and Alemannic, but none of these arguments is as cogent as the distributional constraints imposed on sequences of vowel + consonant in Bavarian. In this situation, recourse to external evidence suggests itself: If there is a linguistic feature related to ambisyllabicity which is systematically handled in a different way by native speakers of Bavarian and non-Bavarian dialects, this would be strong evidence against ambisyllabicity in the non-Bavarian dialects. Fortunately, a source of such evidence does exist: a number of non- standard, but very common words created by different irregular techniques such as clipping, acronymic shortening, irregular resuffixation etc. Some examples from the corpus of Ronneberger-Sibold (1992)8 are given in (9).
(9) German lexical creations logo ‘sure, of course’ 8 This corpus contains, among others, all creations of the type discussed here which are contained in Duden (1986). 262 Elke Ronneberger-Sibold They are disyllabic with initial stress and contain one medial consonant or (rarely) medial clusters allowing for an ambisyllabic consonant. Secondly, they do not have a fixed standard pronunciation and thus allow for much more variation than normal German words, precisely the kind of variation we need. Thirdly, the sound shape of such words is not entirely determined by their source form and putative “shortening rules” as in the case of normal rule-governed phonological derivations. There are no productive “shorten- ing rules” in German. What is productive is rather a certain pattern or tem- plate of the output, and it is a matter of creative ingenuity to find a way leading from the source form to the desired result. Sometimes, simple clipping is sufficient, e.g. in Schokolade ‘chocolate’ > Schoko. But in other cases, quite complex operations are performed, e.g. in Kugelschreiber > Kuli ‘biro’. In particular, the source form does not determine the length of stressed vowels in the creations. Thus, a long vowel of a source form may be replaced by a short one as in Opa /'opa/ (besides /'o:pa/) < Großpapa /'gro:spapa/ ‘granddad’, or a short vowel by a long one, e.g. in Schupo /'Su:po/ (besides /'Supo/) < Schutzpolizist /'Sutspo:li:tsist/ ‘bobby, cop’. The case is a little more intricate when the long vowel of the source form is in pretonic position, e.g. in Schokolade. The standard norm here prescribes a short tense [o], which is interpreted as an allophone of long tense /o:/. Thus Schokolade /So:ko:'la:d7/ > Schoko /'Soko/ ['SOkO] (besides /'So:ko:/ ['So:ko]) seems to parallel Großpapa > Opa. A detailed discussion of German un- stressed full vowels is beyond the scope of this paper; suffice it to say that I adhere to the view that in this position the entire opposition between long tense and short lax vowels is neutralized. This interpretation is quite in line with the general tendency for reduced vowel inventories in unstressed position in languages with a great discrepancy between stressed and un- stressed syllables like the Germanic ones. Speakers currently vary between [o] and [O] without even being aware of it. But as soon as the vowel is stressed, a clear decision has to be made. Thus, with respect to the source form, the pronunciation /'Soko/ ['SOkO] is not really a substitution of a short vowel for a long one, but a choice in an undecided situation. Thus these words, once they are created, are new lexical entries, inde- pendent of their source form. This is true not only of their form, but also of their phonostylistic register and very often even of their meaning. Let us consider as an example the word logo, used in juvenile slang. Its source form is the adjective logisch ‘logical’, from which it has been created by an irregular substitution of final -o for the derivative -isch. But its meaning is no longer ‘logical’, but rather ‘sure, of course’. “Shall we go to the cinema tonight? — Logo.” The semantic distance can be shorter than with logo. For instance, Abi has the same denotation as it source form Abitur, namely Ambisyllabic consonants in German 263 ‘A-levels, high school finals’, but the stylistic register is still different: Abi is much less formal than Abitur. Obviously, the aim of these creations is a form for informal use which is shorter and easier to pronounce than the source form. In this perspective, we would expect them to be made as short as possible. With respect to the number of syllables, this is not the case, for few of them are monosyllabic. This has to do with alternating rhythm (cf. Ronneberger-Sibold, 1996). But within the disyllable, the shortest possible solution is targeted. In the plain model for SG, the shortest solution for a disyllable consists, of course, of two short open syllables. This would not only be the shortest possible disyllable but by the same token the prosodically minimal word, for it would count the same number of moras as a minimal monosyllable, namely 2 (cf. (7) above). In fact, in this model, words such as Opa /'opa/ would be very similar to the Japanese cases investigated by Itô (1992) under the heading of prosodic minimality. Thus, according to this model, the stressed vowel of our disyllables should be short. In the ambisyllabic model, on the other hand, the minimal length of a di- syllable is three moras, regardless of whether the first two are filled by a long vowel or a short vowel + ambisyllabic consonant. As pointed out above, the minimal word of two moras can only be realized in a monosyllable (cf. (8) above). Therefore, it does not make much difference whether the stressed vowel is long or short. If close contact is considered essential, a short vowel should even be disfavoured for being structurally and phonetically marked, as explained by Trubetzkoy (cf. §3.2 above) Thus, according to this model, there should be indifference with regard to vowel length or even a prefer- ence for long vowels. 4.2. Lexical creations in Standard German and in Alemannic In SG and even more so in Alemannic we find a clear preference for short vowels. In SG, only the small group of words listed in (10) is always pro- nounced with a long vowel. These are names of members of political or official organizations, the brand name Coca Cola which can be shortened either to Cola or to Coca, and words with medial lenis /z/, such as ASU, where length of the preceding vowel is phonotactically conditioned in SG. 264 Elke Ronneberger-Sibold (10) Lexical creations pronounced with long vowels only in SG • Designations of members of political or official organizations Sozi < Sozialist ‘socialist’ Nazi < Nationalsozialist [nats5ional-] ‘national socialist’ Judo < Jungdemokrat + o ‘young democrat’ Realo < Realpolitiker + o Vopo < Volkspolizist ‘member of the former GDR police’ • Brand names Coca and Cola < Coca Cola • Words with medial lenis /z/ (phonotactically conditioned) ASU (see above) Many very common words are pronounced with short vowels only, whereas others have two pronunciations (see (11) and (12) respectively). (11) Lexical creations pronounced with short vowels only in SG Abo < Abonnement ‘subscription’ Abi (see above) Trabi (see above) Akku < Akkumulator ‘accumulator’ Mathe < Mathematik ‘maths’ Tetra < Tetraeder in Tetrapack ‘cardboard container for liquids’ Rolli (see above) Pulli < Pullover + i Ami < Amerikaner + i ‘American’ Sani < Sanitäter ‘paramedic’ Uni < Universität ‘university’ (12) Lexical creations pronounced with long or short vowels in SG Oma (see above) Opa (see above) Schupo (see above) Kripo < Kriminalpolizei ‘criminal investigation force’ Schoko (see above) Öko- < ökologisch ‘ecological’ Foto < Fotografie ‘photo’ Limo (see above) Kuli (see above) logo (see above) It is important to note that in the cases with both pronunciations, this is not a matter of realization, but of underlying form. You cannot, for instance, shorten /'lo:go/ to /'logo/ in allegro style, but if you choose /'logo/ this is a Ambisyllabic consonants in German 265 decision for an underlying form with short /o/. This is one of the features which make those non-standardized words such a precious instrument for our investigation, for such a variation would be impossible in the normal vocabulary. You cannot choose to say /'fog7l/ instead of /'fo:g7l/ ‘bird’, because you like this pronunciation better for certain occasions. Where both forms are available to the same speaker, they are not used indiscriminately, but the forms with a short vowel belong to a less formal phonostylistic register than those with a long vowel. /'logo/ sounds much “cooler” than /'lo:go/; /'o:pa/ is tender and respectful, whereas /'opa/ is easily meant and understood in a pejorative sense, e.g. in ein alter Opa /'opa/ ‘an old gramp’. The higher phonostylistic value of long stressed vowels in such words is most probably due to their occurrence in spelling pronunciation. (As men- tioned in §2.3, a vowel followed by a simple consonant in New High German spelling should be pronounced long.) The influence of spelling is obvious in the words in (10) above, because of the importance of written texts (news- papers etc.) in their diffusion. But I suspect that even in creations with less affinity to the written language, many pronunciations with long vowels are (or at least originated as) spelling pronunciations used by people who are not in direct contact with the groups of speakers where the word is in oral use. For instance, I remember thinking as a child that Schupo was pro- nounced /'Su:po/, because I had only read it in novels for children, until I heard an adult pronounce it with a short vowel. An interesting case of transi- tion from oral to written use is logo. About fifteen to twenty years ago, logo was exclusively used in extremely colloquial speech by adolescents using SG. From that time, I only remember the pronunciation with a short stressed vowel. Meanwhile, the word has become quite fashionable in the media, a fact, which, of course, enhances the spelling pronunciation /'lo:go/. Gradu- ally, this pronunciation is winning out. The importance of spelling pronunciation is indirectly confirmed by speakers of those Alemannic dialects where Middle High German short vowels in open syllables were maintained. For these speakers, a syllable which is open in spelling is not necessarily pronounced with a long vowel. Thus the word written Vogel is pronounced /'fog7l/, not /'fo:g7l/. Therefore, writing should be less of an impediment to pronunciations with short vowels, such as /'logo/. Indeed, my informants from this region produced short vowels almost without exception, even in Judo or ASU, although, phonotactically, long vowels would be possible in these positions. Thus, in SG and in Alemannic, the prediction made for the plain model is borne out. 266 Elke Ronneberger-Sibold 4.3. Lexical creations in Bavarian As was to be expected, the ambisyllabic model is supported in Bavarian. Three cases have to be distinguished: vowel + lenis obstruent, vowel + fortis obstruent, and vowel + resonant. In words with medial lenis obstruents, i.e. of the logo-type, speakers not only prefer a long vowel with loose contact: ['lo:°go] /'lo°go/; they even cate- gorically refuse to produce such a thing as ['lo^k(k)o] or ['lo^ggo] /'lo^go/ with o o o a short, abruptly cut vowel and a long and/or fortis consonant. As men- tioned in §2.4, this would be highly marked for them. The only answer of this ^ ^ type I got was ['do ra bo bo i] /'do ra bo i/ for Trabi. The informant is a car mechanic, and his pronunciation is obviously an attempt to copy the pronunciation of some of his northern colleagues. (This words exists in SG with short vowel only.) Quite generally, professional contact with such words seems to en- hance standard or close-to-standard pronunciation. Other speakers, who are not in professional contact with the word Trabi either hedge into E E ['do ra: bo i] /'do ra bo i/ with smoothly cut long vowel or, if they are capable of producing it, standard ['tra^bi] /'trabi/ with a short, abruptly cut vowel with- out lengthened or fortized consonant. In words with medial fortis obstruents, close contact is of course the only possibility in full dialect. Thus, I frequently got ['So^ggo] or ['So^k(k)o] /'So^go/ o o o for Schoko. However, many speakers, especially in the big cities, have ac- quired from intense contact with SG the ability to produce loose contact before a fortis obstruent. This is frequently used in our words to avoid the heavily marked close contact, even in cases where this seems excluded by the orthography, e.g. in ['a:°ku] /'a°ku/ for Akku, SG /'aku/ < Akkumulator. Before resonants, there is a lot of variation due to a very diversified dia- lectal situation. Originally, only loose contact with long vowel was possible. An informant from Bad Ischl, for instance, had a smoothly cut long vowel followed by short [m] in Kammerl ['ka:°m6l] /'ka°m6l/ ‘small room’. Of course, he only produces forms such as ['li:°mo] /'li°mo/ Limo, ['a:°lu] /'a°lu/ Alu, ['u:°ni] /'u°ni/ Uni. In other regions, close contact with a lengthened resonant has been introduced in different ways. In Munich, you frequently hear ['li^mmo] /'li^mo/ Limo. Many speakers also have at their disposal the SG pronuncia- tion with short vowel and short consonant ['li^mo] /'limo/. This is particularly current with medial /l/. The result are pronunciations like ['ku^li] /'kuli/ as in SG. As a matter of fact, native speakers of Bavarian who have firmly inte- grated the ['ku^li]-type of SG into their competence on a par with the ['a:°lu] and the ['li^mmo]-types, have at their disposal the same three-way opposi- tion of stressed syllables in disyllabic words as the speakers of Middle High Ambisyllabic consonants in German 267 German. With the exception of smooth cut in all open syllables, the opposi- tion of schâle : schale : schalle in (1) is exactly matched by (13). (13) Kuli ‘coolie’ : Kuli ‘biro’ : Pulli ‘pullover’ ['ku:°li] : ['ku^li] : ['pu^lli] Obviously, long and short vowels as well as simple and geminate [l] are in opposition with each other. Therefore, the most straightforward phonemic and prosodic analysis is (14). (14) /'ku:.li/ : /'ku.li/ : /'pul.li/ mm m mm By this analysis, the prosodic equivalences represented in (15) obtain. (15) Prosodic equivalences in Bavarian-Standard dialect mixture /ku:.-/ mm = /kul.-/ /'ku.li/ = mm mm = /dÄ:/ ‘there’ mm Minimal monosyllabic word = minimal disyllabic word Very significantly, speakers of this variety do in fact attribute equal prosodic length to syllables such as /'ku:.-/ or /'kul.-/ and to words such as /'ku.li/. They obviously measure the stressed first syllable in /'kuli/ as short, in spite of the phonetically close contact between /u/ and /l/.9 This is clear evidence that close contact as a phonetic feature alone does not necessarily trigger ambisyllabicity of the following consonant. Only if it is phonemically rele- vant does it have this effect. 9 I was in no way expecting this behaviour, and discovered it by mere chance. Two of my best informants in Munich firmly insisted that the word Kuli consists of one syllable only. I noted this as one of those curiosities which might be significant in some way, but I was unable to guess what this significance might be. A few days later, I asked a student from Bad Reichenhall, who combined in his speech very marked Bavarian elements with completely standard ones, whether he was aware of the difference between his realizations of /l/ in Pulli and in Kuli. He answered: “Yes, of course [pul-li]” (with one knock on the table for [pul-] and another one for [-li]) “is of course longer than Kuli” (with one knock for the entire word). Of course, what he was knocking and what my other two informants had been counting in- stead of syllables were metrical beats. 268 Elke Ronneberger-Sibold In sum, what these speakers have developed without any academic training is a basic rule of quantifying metre quite similar to the one applied in the poetry of Classical Middle High German and, of course, in the Classi- cal languages. This metrical rule has spontaneously evolved in a structural constellation containing the necessary prerequisites for quantifying metre tentatively formulated in §3.1: Long and short vowels, geminate and simple consonants, and short open syllables with full vowels. The evidence con- cerning quantifying metre is still very scanty.10 But, quite independently of this feature, our findings concerning the preferred quantity of stressed vowels in lexical creations substantiate our predictions: They confirm the plain model for SG and for Alemannic, the ambisyllabic model for Bavarian. 5. Conclusion: The ambisyllabic model for Standard German — a reconstruction? In conclusion, let us return once more to the diachronic perspective. It is a well-known fact that shortly after the completion of a sound change it can be very easy for a linguist to reconstruct the system before this change. This reconstruction then comprises all the generalizations that were valid in the preceding system but may have been destroyed by the change. This can make the reconstructed state more attractive than the subsequent one in the eyes of a linguist looking for generalizations. Moreover, if the recon- struction is particularly easy to perform, it may be really difficult to distin- guish between reconstruction and synchronic reality. Even among the speakers themselves there may be some who use the old system as a syn- chronic underlying form in their competence, whereas others may be more surface-oriented. From this perspective, on might ask: Could it be the case that the ambi- syllabic model for SG is a reconstruction of the stage before degemination (described in §2.3) with its isochrony of all stressed syllables? In this recon- struction, the ambisyllabic consonants would play the part of the former geminates, although they are not real geminates either phonemically or phonetically: Phonemically, they are not in opposition to simple consonants in the same distribution,11 and phonetically, they are not longer than their 10 As suggested to me by Nikolaus Ritt (Vienna), metrically bound modern Bavarian pop songs could be a good source of such evidence. 11 If compounds are taken into account — which, from a strictly phonological viewpoint should be done, although it is often avoided in phonological analyses of German, because it makes them still more complicated — ambisyllabic consonants are even in contrast with true geminates in received pronunciation, for instance in Rücken /'ryk7n/ ‘back’ vs. Rückkehr /'ryk‚ke:r/ ‘return’. The lengthened [k] in Rückkehr may, however, be considered as con- Ambisyllabic consonants in German 269 alleged heterosyllabic counterparts. Close contact is considered a kind of substitute for the missing length by some authors. As pointed out above, it is not unthinkable that native speakers perform such a synchronic analysis, but in order to decide whether they really do so, external evidence has to be taken into account if internal evidence is not decisive. The evidence presented in this paper strongly suggests that the ambisyllabic model in SG and in Alemannic is in fact a reconstruction of a preceding stage made by linguists but not shared by the native speakers. Acknowledgements Many people have helped me to develop the ideas put forward in this paper by acting as informants or interviewers, by providing acoustic analyses and unpublished dialect maps, and, of course, by their discussion and criticism. I would like to thank all of them, especially Wolfgang U. Dressler and Sylvia Moosmüller (Vienna), Hannes Scheutz (Salzburg), and Damaris Nübling (Freiburg). The responsibility for all errors and possible misinterpretations is, of course, entirely mine. References Anderson, J. & J. Durand (1986). Dependency phonology. Dependency and non-linear phonology, edited by J. Durand, 1–54. London: Croom Helm. Aronoff, M. (1992). Segmentalism in linguistics: the alphabetic basis of phonological theory. The linguistics of literacy, edited by P. Downing, S.D. Lima & M. Noonan, 71–82. Amsterdam: Benjamins. Auer, P. (1989). Some ways to count morae: Prokosch’s Law, Streitberg’s Law, Pfalz’s Law, and other rhythmic regularities. Linguistics 27, 1071–1102 . Bannert, R. (1976). Mittelbairische Phonologie auf akustischer und perzeptorischer Grundlage. Lund: Gleerup. Butt, M. & P. Eisenberg (1990). Schreibsilbe und Sprechsilbe. Zu einer Theorie der Orthographie: Interdisziplinäre Aspekte gegenwärtiger Schrift- und Orthographieforschung, edited by Ch. Stetter, 34–64. Tübingen: Niemeyer. Clements, G.N. & S.J. Keyser (1983). CV phonology: a generative theory of the syllable. Cambridge MA: M.I.T. Press. Derwing, B.L. (1992). A ‘pause-break’ task for eliciting syllable boundary judgments from literate and illiterate speakers: preliminary results for five diverse languages. Language and Speech 35, 219–235. Dressler, W.U. & L. Merlini-Barbaresi (1994). Morphopragmatics: diminutives and intensifiers in Italian, German, and other languages. Berlin: Mouton de Gruyter. ditioned by the secondary accent on kehr and hence as subphonemic. Another problem is the substitution of simple consonants for the geminates in colloquial style. 270 Elke Ronneberger-Sibold Dressler, W.U. (1996). Principles of naturalness in phonology and across components. Natural Phonology: The State of the Art, edited by B. Hurch & R.A. Rhodes, 41–51. Berlin: Mouton de Gruyter. Duden (1986). Duden Rechtschreibung der deutschen Sprache und der Fremdwörter. 19th ed. Mannheim: Bibliographisches Institut. Dziubalska-Kołaczyk, K. (1995). Phonology without the syllable: a Study in the natural framework. Poznan´: Motivex. Fourquet, J. (1964). Zur Deutung der Isophonen der Quantität. Phonetica 11 , 155–163. Giegerich, H.J. (1985). Metrical phonology and phonological structure: German and English. Cambridge: Cambridge University Press. Hall, T.A. (1992). Syllable structure and syllable-related processes in German. Tübingen: Narr. Hyman, L.M. (1985). A theory of phonological weight. Dordrecht: Foris. Itô, J. (1992). Prosodic minimality in Japanese. Papers from the 26th regional meeting of the Chicago Linguistic Society. Vol. 2: The parasession on the syllable in phonetics and phonology, edited by M. Ziolkowski, M. Noske & K. Deaton, 213–239. Chicago, Ill.: Chicago Linguistic Society. Kilani-Schoch, M. (1988). Introduction à la morphologie naturelle. Bern: P. Lang. Kilani-Schoch, M. & W.U. Dressler (1992). Prol-o, intell-o, gauch-o et les autres: Propriétés formelles de deux opérations du français parlé. Romanistisches Jahrbuch 43, 65–86. Kilani-Schoch, M. (1996). Syllable and foot in French clipping. Natural Phonology: the State of Art, edited by B. Hurch & R.A. Rhodes, 135–152. Berlin: Mouton de Gruyter. Kranzmayer, E. (1956). Historische Lautgeographie des gesamtbairischen Dialektraumes. Wien: Böhlau. Laeufer, C. (1985). Some language-specific and universal aspects of syllable structure and syllabification: Evidence from French and German. PhD dissertation, Cornell University. Mangold, M. (1990). Duden Aussprachewörterbuch. Wörterbuch der deutschen Standardaussprache. 3rd. ed. Mannheim: Bibliographisches Institut. Ramers, K.H. (1992). Ambisilbische Konsonanten im Deutschen. Silbenphonologie des Deutschen, edited by P. Eisenberg, K.H. Ramers & H. Vater, 246–283. Tübingen: Narr. Reiffenstein, I. (1965). Geminaten und Fortes im Althochdeutschen. Münchener Studien zur Sprach- wissenschaft 18, 61–77. Reis, M. (1974). Lauttheorie und Lautgeschichte: Untersuchungen am Beispiel der Dehnungs- und Kürzungs- vorgänge im Deutschen. München: Fink. Ronneberger-Sibold, E. (1992). Die Lautgestalt neuer Wurzeln: Kürzungen und Kunstwörter im Deutschen und Französischen. Habilitationsschrift, Freiburg i. Br. Ronneberger-Sibold, E. (1995). On different ways of optimizing the sound shape of words. Historical Linguistics 1993, edited by H. Andersen, 421–432. Amsterdam: Benjamins. Ronneberger-Sibold, E. (1996). Preferred sound shapes of new roots: on some phonotactic and prosodic properties of shortenings in German and French. Natural Phonology: the State of Art, edited by B. Hurch & R.A. Rhodes, 261–292. Berlin: Mouton de Gruyter. Ronneberger-Sibold, E. (in press). Creative competence at work: the creation of partial motiva- tion in German trade names. Extragrammatical and marginal morphology, edited by U. Doleschal & A. Thornton. Unterschleißheim: LINCOM Europa. Scheutz, H. (1984). Quantität und Lenis/Fortis im Mittelbairischen. Beiträge zur bairischen und ostfränkischen Dialektologie, edited by P. Wiesinger, 13–33. Göppingen: Kümmerle. Sievers, E. (1901). Grundzüge der Phonetik zur Einführung in das Studium der Lautlehre der indogermani- schen Sprachen. Leipzig: Breitkopf und Härtel. Sobkowiak, W. (1991). Metaphonology of English paronomasic puns. Frankfurt: Lang. Trubetzkoy, N.S. (1939). Grundzüge der Phonologie. Göttingen: Vandenhoeck & Ruprecht. Vennemann, T. (1982). Zur Silbenstruktur der deutschen Standardsprache. Silben, Segmente, Akzente, edited by T. Vennemann, 261–305. Tübingen: Niemeyer. Vennemann, T. (1988). Preference laws for syllable structure and the explanation of sound change: with special reference to German, Germanic, Italian and Latin. Berlin, New York: Mouton de Gruyter. Ambisyllabic consonants in German 271 Vennemann, T. (1991). Syllable structure and syllable cut prosodies in Modern Standard German. Certamen phonologicum II: papers from the 1990 Cortona phonology meeting, edited by P.M. Bertinetto, M. Kenstowicz & M. Loporcaro, 211–243. Turin: Rosenberg & Sellier. Vennemann, T. (1994). Universelle Nuklearphonologie mit epiphänomenaler Silbenstruktur. Universale phonologische Strukturen und Prozesse, edited by K.H. Ramers, H. Vater & H. Wode, 7–54. Tübingen: Niemeyer. Wiese, R. (1988). Silbische und lexikalische Phonologie: Studien zum Chinesischen und Deutschen. Tübingen: Niemeyer. Wiesinger, P. (1983). Dehnung und Kürzung in den deutschen Dialekten. Dialektologie: Ein Hand- buch zur deutschen und allgemeinen Dialektforschung. Vol. 2, edited by W. Besch, U. Knoop, W. Putschke & H.E. Wiegand, 1088–1101. Berlin, New York: Mouton de Gruyter. On trochaic Proper Government . Grazyna Rowicka Abstract In standard Government Phonology the basic mechanism controlling the distribution of empty nuclei is Proper Government. An empty and in- audible nucleus is assumed to be dependent on the following contentful vowel. In this paper I argue that a more insightful analysis of vowel-zero alternations is possible under the assumption that an empty nucleus forms a Proper Government relation, resembling a trochaic metrical foot, with the preceding vowel. Trochaic Proper Government accounts for closed syllable shortening in Turkish and Yawelmani. Under a Government The Hague: Thesus. 273–288. Thesus.The Hague: 273–288. . Phonology analysis, the shortening occurs before an empty nucleus in the following syllable. However, the standard theory does not anticipate any direct connection between an empty nucleus and the preceding (long) Syllables !? vowel. Such a connection follows if trochaic Proper Government is adopted instead. Turkish and Yawelmani are analysed below as Strict CV languages. Long vowels consist of a contentful vowel and a following skeletally adjacent empty nucleus. The head of a long vowel properly governs its empty dependent and spreads its melody to it so that a long Phonologica 1996: Phonologica 1996: vowel surfaces. However, if the following syllable contains an empty nucleus the head cannot both govern its empty dependent and the empty nucleus in the next syllable. Shortening must occur, i.e. the deletion of the empty dependent of the long vowel. Such shortening is analogical to trochaic shortening known from metrical theory. 0. 1. Introduction The idea that phonological structure contains empty positions can be traced back at least to Anderson (1982) and Kaye & Lowenstamm (1984). It has been proposed that empty nuclei represent vowels alternating with zero. In standard Government Phonology (GP) the presence of empty nuclei John R. Rennison & Klaus Kühnhammer, eds. (1999). eds. R. Rennison & Klaus Kühnhammer, John (1999). in representations is basically controlled by means of the phonological . 274 Grazyna Rowicka Empty Category Principle, formulated as in (1) (see Kaye, Lowenstamm & Vergnaud (hereafter KLV), 1990: 219). (1) Empty Category Principle (ECP) A position may be uninterpreted phonetically if it is properly governed. Accordingly, a position which is not properly governed must be realized phonetically. A proper governor may not itself be empty or properly gov- erned. Although provision has been made for the parametric nature of headed- ness in Proper Government (Kaye, 1990: note 21), it has generally been assumed to be right-headed, i.e. the governor follows the governee.1 The standard application of Proper Government (PG) and the ECP is illustrated in (2). (2) Proper Government and the ECP (standard GP) X O 1N1 O 2N2 O 3N3 nuclear projection level x x x x x x skeletal tier C 0 C 0 C V melodic features ↓ V where represents (iambic) PG The contentful nucleus N3 properly governs the preceding empty N2, which may therefore remain empty. N1 lacks a potential proper governor to its right and must become audible. The prediction is that in a sequence of empty nuclei every other one must surface. In this paper I argue that in many cases previously analysed in terms of right-headed PG, a more insightful analysis is possible under the assump- tion that an empty nucleus is properly governed by the preceding, and not the following, contentful vowel. By analogy to the prominence patterns in stress feet, I will refer to the left-headed PG relations proposed below as TROCHAIC PG, as opposed to the IAMBIC PG of the standard theory. As far as rhythmic stress systems are concerned, a trochee is the most common type of foot, while an iamb is a more marked option (see Dogil, 1980). Van de 1 The possibility of left-headed PG has, in fact, been considered by Charette (1991) and argued for by van der Hulst (1991). Gibb (1992) applies left-headed PG in Finnish and Yoshida (this volume) in Cairo Arabic. However, none of these proposals involves the change of focus and the reformulation of the ECP which I propose here. On trochaic Proper Government 275 Vijver (1997) suggests that all alleged iambic patterns can be derived from special forms of trochaic footing. In this paper I claim that left-headedness is also the unmarked option in proper governing relations. Trochaic PG will be shown to make possible an insightful analysis of long vowel shortening in Yawelmani and Turkish. 2. Long vowel shortening 2.1. Facts Long vowels in Yawelmani and Turkish shorten before a word-final con- sonant and before a consonantal cluster. Some examples are given in transcription in (3), where the relevant vowels are underlined (see Kaye, 1990: 302). (3) a. Yawelmani sa:pit saphin sapnit ‘burn’ go:bit gobhin gobnit ‘take in’ panat pana:hin pana:nit ‘arrive’ ?ilet ?ile:hin ?ile:nit ‘fan’ ?amlal ?a:miltaw ?a:milka ‘help’ moxlol mo:xiltaw mo:xilka ‘grow old’ b. Turkish NOM. POSS. NOM. PL. merak mera:ki- meraklar ‘curiosity’ sevap seva:bi- sevaplar ‘good deed’ usuly usu:lyü usulylyer ‘method’ In standard generative phonology the phenomenon is generally accounted for by a rule of vowel shortening in closed syllables. 2.2. Kaye (1990) Kaye points out that, according to GP, in many cases the consonant before which shortening takes place should not be analysed as a coda. A con- sonant is allowed in the coda position only if it is followed by a less . 276 Grazyna Rowicka sonorous consonant in an onset.2 A word-final consonant, such as /k/ in Turkish [merak], is not followed by any other (less sonorous) consonant. It therefore cannot be included in the coda of the preceding vowel, but in the onset of a final empty nucleus. Word-internal clusters which trigger long vowel shortening often fail to meet the requirements on well-formed coda- onset sequences. For instance, the Turkish word [meraklar] cannot be phonologically syllabified as [me.rak.lar] because /l/ is more sonorous than /k/. The cluster /kl/ can only be analysed as two onsets separated by an empty nucleus.3 This means that long vowels are shortened not in closed syllables, but before an empty nucleus in the following syllable. This is illustrated in (4). In (4a) there is no shortening of N2 because the following nucleus, N3, is not empty. In (4b) and (4c), N2 is followed by empty N3 and it shortens. Observe that under this analysis there is no obvious reason for vowel shortening. In all cases the triggering empty nucleus is licensed, either word- finally, as in (4b), or by PG, as in (4c). (4) a. b. 1N1 2N2 3N3 1N1 2N2 3N3 m e r a a k i- m e r a a k 0 ↓ shortening c. 1N1 2N2 3N3 4N4 5N5 m e r a a k 0 l a r 0 ↓ shortening where = (iambic) PG = parametric licensing domain-finally 2 Sonority used to be expressed in terms of the charm value of segments (KLV, 1985). Nowa- days most GP advocates refer to the complexity of segments instead (e.g. Harris, 1990). Roughly speaking, less sonorous segments are more complex and may therefore govern segments which are less complex (and more sonorous). 3 The cluster /kl/ could in principle be analysed as a branching onset. But there is no obvious structural reason why long vowels should be disallowed before branching onsets. Moreover, such a representation is not possible for many other clusters before which the shortening takes place, e.g. for the geminate in [usulylyer] in (3b). Postulating various structures for con- sonantal clusters which all trigger the shortening would make it impossible to relate the process to the environment in which it occurs. On trochaic Proper Government 277 Yoshida (1993) and Lowenstamm (1996) seek a principled explanation for the shortening of long vowels before empty nuclei. They do not consider Yawelmani or Turkish specifically, but their analyses can also be applied to these languages. 2.3. Yoshida (1993) Yoshida argues that languages in which long vowels are sensitive to the presence of empty nuclei in the following syllable have neither branching onsets nor rhymes. He adopts a STRICT CV analysis of these languages (i.e. no branching constituents are recognized). All surface clusters are broken up by empty nuclei. All long vowels are interpreted as disyllabic. They con- sist of two nuclei separated by an empty onset, as represented in (5). (5) Disyllabic long vowel (Yoshida, 1993) N O N x x V To account for the fact that configurations of a long vowel and an empty nucleus in the next syllable are dispreferred, Yoshida extends the notion of government licensing (GL) (Charette, 1991: 241) to the heads of long vowels. This is illustrated in (6). For a nuclear head N1 to govern its complement N2, the head must be government-licensed by the following nucleus N3. (6) Internuclear government licensing <<<<<<<<<<<<<<<<<<<<<<< 1N1 → 2N2 3N3 O O O x x x x x x C V C V where → = internuclear governing relation within a long vowel <<<< = internuclear GL In most languages the government licensor must be a contentful vowel and not an empty nucleus. If N3 is empty, it cannot government-license N1 to be the head of a long vowel N1N2. Shortening must take place. . 278 Grazyna Rowicka However, according to Charette’s proposal, GL is based on a governing relation: the governor government-licenses its governee. Yoshida’s analysis does not obey this. There is no evidence that the government licensor N3 governs its licensee N1 at any representational level. Another problem with this analysis is the multiplication of representational levels. The standard theory recognizes a nuclear projection level where right-headed PG is contracted. The left-headed internuclear government within the long vowel must be at a different projection level. Finally, right-headed GL applies across the long vowel internuclear government, and hence must occur at yet another level. 2.4. Lowenstamm (1996) Lowenstamm, the main proponent of the Strict CV approach, adduces arguments that in disyllabic long vowels the first nucleus is contentful, whereas the second is underlyingly empty. Since Lowenstamm does not recognize a separate skeletal tier, long vowels are represented as in (7). (7) Disyllabic long vowel (Lowenstamm, 1996; also Larsen (1994)) 1N1 O 2N2 V The surface long vowel is the result of melodic spreading from the head of the long vowel N1 to the following empty nucleus N2. However, the spreading is allowed iff the target empty nucleus is licensed by PG from the following contentful nucleus. This is illustrated in (8a). If N3 is itself empty, it cannot properly govern the preceding N2. N2 is not licensed (to remain empty) and cannot be targeted by melodic spreading from the preceding N1. No long vowel can surface, as shown in (8b). (8) a. b. X O 1N1 O 2N2 O 3N3 O 1N1 O 2N2 O 3N3 X C V C V C V C 0 Lowenstamm’s analysis raises some questions. First, according to the ECP (1), an empty nucleus which is properly governed, such as N2 in (8a), is allowed to remain silent. However, in the configuration in (8a) N2 may not remain silent, but is targeted by melodic spreading from N1. It is phoneti- cally realized as part of the long vowel N1N2. Second, spreading and PG seem to be in conflict in (8a): one affects N2 from the left, the other from the On trochaic Proper Government 279 right; one provides N2 with melody, while due to the other N2 could remain silent. Why, then, is spreading impossible in (8b) if PG does not apply? Finally, in (8a) the spreading agent N1 targets empty N2, which is the proper governee of another nucleus, namely N3. Why can spreading interfere with a governing domain? PG, for instance, is not allowed to apply across a gov- erning domain, such as a branching onset (see KLV, 1990; Charette, 1991). Given the above criticism of analyses proposed so far, I develop below a new account of the shortening of long vowels before empty nuclei. 3. New analysis 3.1. Disyllabic long vowels I follow Yoshida and Lowenstamm in assuming a Strict CV analysis of lan- guages in which long vowels are not allowed before an empty nucleus in the next syllable. Yawelmani and Turkish both have tight restrictions on cluster- ing and disallow word-initial consonantal clusters (see Newman, 1944; 1946 on Yawelmani and e.g. Clements & Sezer, 1982 on Turkish). There are ad- ditional arguments in favour of a disyllabic interpretation of long vowels in these languages. First, a language in which the nucleus branching parameter is set to “on” can be expected to have both long monophthongs and diph- thongs in its inventory. Neither Yawelmani nor Turkish has diphthongs. Second, there are no native Turkish words with underlying long vowels, which suggests that the nucleus branching parameter is set to “off” for native vocabulary. Long monophthongs which are subject to shortening appear only in Arabic and Persian borrowings. Third, a disyllabic represen- tation of a long vowel containing an empty nucleus, as in (7), predicts that such a long vowel will be sensitive to the presence of a following empty nucleus because it ends itself in an empty nucleus. Sequences of empty nuclei are cross-linguistically dispreferred, as vowel-zero alternations in many languages show. For these reasons I adopt a disyllabic analysis of Yawelmani and Turkish long vowels. In Lowenstamm’s Strict CV framework no separate skeletal tier is recog- nized. In the present analysis I assume that the two nuclei of a disyllabic long vowel dominate skeletal positions, whereas the intervening onset does not. Hence the nuclear positions are skeletally adjacent, as shown in (9). . 280 Grazyna Rowicka (9) Disyllabic long vowel N 1 O 2N2 x x V Observe that the presence vs. absence of a skeletal position “x” associated to the onset can be relevant. Charette (1991: 91) uses an unspecified skeletal slot in an onset position to represent French h-aspiré. Although not realized phonetically, h-aspiré functions differently from an entirely empty onset. Words beginning with it behave phonologically like words beginning with a consonant, and not like vowel-initial ones. If there were a skeletal position attached to the intervening onset in (9) above it could be expected to pre- vent melodic spreading from N1 to N2, just like a fully specified consonant. Consequently, N2 would remain silent when properly governed by N1. When followed by another empty nucleus in the following syllable, on the other hand, it would be phonetically realized as the default vowel, giving rise to vowel hiatus on the surface. Further research must show to what extent cross-linguistic evidence justifies this kind of representation as distinct from the one in (9). (There is some evidence that it may be useful in the analysis of Turkish ‘soft g’; see fn. 7.) 3.2. Trochaic PG In all analyses of long vowels some left-headed governing relation has been recognized between the head of the long vowel and its complement. Given the representation in (9), this is a governing relation between contentful N1 and empty N2. I propose that this left-headed relation is one of PG. The proper governor N1 licenses the following empty N2 and, because no skele- tal position intervenes, it spreads its melody to the governee. This is illus- trated in (10a). Such spreading, I argue, can be seen as a manifestation of a PG relation between skeletally adjacent nuclei. Although N2 acquires phonetic content (by spreading), it remains properly governed and cannot possibly function as a governor for another empty nucleus. Phonetic realization itself is not enough for a nucleus to have the status of a contentful vowel and be a potential proper governor.4 4 This conclusion is further confirmed by Rowicka’s (1998; to appear) study of Mohawk. In Mohawk a properly governed nucleus may be phonetically realized in certain contexts. However, its governee status is manifested in the nucleus’ invisibility to stress, in contrast to proper governors which are phonetically realized and visible to stress. The relevant dis- On trochaic Proper Government 281 I suggest further that the same relation, trochaic PG, holds between any contentful vowel and the following empty nucleus. However, an intervening onset consonant will prevent melodic spreading, as shown in (10b). (10) Trochaic PG a. b. O 1N1 O 2N2 O 1N1 O 2N2 x x x x x x x C V C V C 0 [CV:] [CVC] Observe that in both cases PG enhances the prominence of the governing nuclear head, either by lengthening it or by suppressing the following nu- cleus (i.e. keeping it empty).5 The structure in (10b) also shows that trochaic PG makes largely super- fluous the parametric licensing of word-final empty nuclei, which was neces- sary under the standard iambic theory: final N2 is properly governed by N1 in the preceding syllable. This seems an advantage of the approach adopt- ing trochaic PG, given the fact that languages with word-medial empty nuclei generally also allow for such nuclei word-finally. Word-final licensing pa- rameter may still turn out useful, for instance, in the analysis of languages which allow for more than one empty nucleus at the end of the word (see Yoshida, this volume, on Arabic). An objection which may be raised against the present approach is that under a trochaic analysis one does not expect to find empty nuclei in word- initial syllables, where they are not preceded by a potential proper governor. Evidence for the presence of initial empty nuclei in some languages involves the occurrence of ‘ill-behaved’, or unusual, consonantal clusters word- initially. Initial empty nuclei are analogical to word-initial unstressed pre- tonic syllables. They are sometimes unavoidable, given the binarity re- quirement of feet and an odd number of syllables in the word. (For an analysis of empty nuclei word-initially in Polish, see Rowicka, to appear). There are, however, languages, including Turkish and Yawelmani, which do not allow empty nuclei word-initially. The special status of word-initial tinction between unlicensed and licensed nuclei is not simply one of phonetic presence, but one of phonological relevance. 5 Note that the PG domains in (10a) and (10b) correspond to heavy syllables in other frame- works. It is therefore possible within the present framework to express generalizations made in other approaches with reference to heavy syllables (e.g. concerning their stress attracting properties), while it is hardly possible to do so in Strict CV assuming iambic PG. . 282 Grazyna Rowicka syllables in such languages has been a problem for standard theory (see Kaye, 1990: 327). They apparently resist iambic PG of domain-initial nuclei even though all conditions seem to be met. Under trochaic PG this situation is not unexpected. It indicates that there is a rigid requirement on exhaus- tive licensing of empty nuclei and that trochaic PG is the only operative licensing mechanism in the languages. 3.3. Trochaic shortening Let us return to long vowel shortening in Yawelmani and Turkish. If a disylla- bic long vowel is followed by an empty nucleus in the next syllable, a se- quence of two empty nuclei in adjacent syllables arises. An example from Turkish is given in (11). A sequence of empty nuclei is ill-formed and avoided cross-linguistically. There is no way that both empty nuclei can be properly governed, either iambically or trochaically. Assuming trochaic PG, N2 can only properly govern N3. Melodic spreading within a PG domain occurs and N3 surfaces as part of the long vowel [a:]. However, N4 remains unlicensed because there is no potential proper governor to its left. The resulting form, represented in (11a), is therefore ill-formed. In standard GP governing relations are said to be established direction- ally, from right to left. Observe that the directionality of analysis and the headedness of PG relations are independent parameters. One can therefore assume trochaic PG and still analyse representations from right to left when establishing PG relations, as in the standard approach. In most cases the correct result is achieved. I argue, however, that the direction in which PG relations are formed is not really crucial and, in fact, difficult to established on independent grounds. Various well-formedness requirements are always of relevance and interfere in determining the final output form. In the case at hand, a strictly directional right-to-left analysis predicts wrongly that N3 should surface as the Turkish default vowel [i-], subject to 6 vowel harmony, in order to properly govern the following empty N4. This is shown in (11b). However, representation (11b) is ill-formed because it con- tains a vowel hiatus, i.e. two skeletally adjacent nuclei, each dominating its own melody. Vowel hiatus is cross-linguistically dispreferred.7 6 This actually happens elsewhere in Turkish and Yawelmani when a sequence of empty nuclei in adjacent syllables is not a long vowel configuration. For the analysis of Turkish and Yawelmani “epenthesis” by trochaic PG see van der Hulst & Rowicka (1997). 7 In fact, vowel hiatus may be tolerated in Turkish. An empty nucleus may surface as [i-] even though it is not separated from the preceding contentful vowel by an audible onset, as in agˇ ız [ai-z] ‘mouth’ (cf. agˇza ‘mouth-DAT’). However, dialect pronunciations as well as spelling indi- cate that phonologically the onset position in between [a] and [i-] is not entirely empty. Pre- On trochaic Proper Government 283 The actual output form of the word [meraklar] is represented in (11c). The empty sequence ON3 is deleted so that contentful N2 can properly govern empty N4. Deletion of underlying material is itself not unproblematic. In (11c), however, very little underlying information is lost (namely, no melodic elements). This is the reason, I conclude, why (11c) is preferred to the possi- bilities given in (11a) and (11b). (11) a. ? O 1N1 O 2N2 O 3N3 O 4N4 O 5N5 O 6N6 x x x x x x x x x x x m e r a 0 k 0 l a r 0 *[mara:klar] b. O 1N1 O 2N2 O 3N3 O 4N4 O 5N5 O 6N6 x x x x x x x x x x x m e r a 0 k 0 l a r 0 *[maraki- lar] ↓ i- c. O 1N1 O 2N2 O 4N4 O 5N5 O 6N6 x x x x x x x x x x m e r a k 0 l a r 0 [maraklar] Under the analysis proposed here, long vowel shortening in Turkish and Yawelmani is directly related to PG. It is also analogical to Trochaic Short- ening discussed by Hayes (1995: 145–149). In Fijian, for example, a long vowel is shortened when followed by a single light syllable, but not before another long vowel or two light syllables. Compare [nröé:] ‘pull’ vs. [nröéta] ‘pull-TRANS.’, but [nröè:nröé:] ‘difficult’ and [mböè:léti] ‘belt’. As argued by Hayes, Trochaic Shortening takes place in order to permit a more complete parse of the word into metrical feet. In the case of Turkish and Yawelmani considered here, a long vowel is shortened in order to permit an exhaustive parse of empty nuclei into PG relations. sumably intervocalic deletion of so-called ‘soft g’ delinks melody, but leaves intact the skeletal position. This position, although phonetically uninterpreted just like an entirely empty and skeletally pointless onset, counts phonologically as a non-empty onset and allows the default realization of the following empty nucleus. See 3.1 above. . 284 Grazyna Rowicka 3.4. The Empty Category Principle revised The change of headedness in the analysis of PG relations also necessitates a reformulation of the ECP. The standard ECP dictates that every ungov- erned empty nucleus must surface phonetically. This predicts that in repre- sentation (11a) the second empty nucleus N4 should surface as [i-] because it is not properly governed. This is shown in (12). (12) O 1N1 O 2N2 O 3N3 O 4N4 O 5N5 O 6N6 x x x x x x x x x x x m e r a 0 k 0 l a r 0 *[mara:ki- lar] ↓ i- The representation in (12) should in fact be preferred to that in (11c) because it is monotonic, i.e. involves no deletion of underlying information. This prediction is wrong. Cross-linguistically, the unmarked pattern of empty nucleus surfacing is the one where a surfacing empty nucleus is followed by a silent one. In terms of trochaic PG this means that a surfacing empty nucleus must be followed by a silent proper governee. To accommo- date this fact into a theory of trochaic PG, I propose the reformulation of the ECP given in (13). (13) The revised ECP An empty nucleus must be phonetically realized if it properly governs another empty nucleus. The revised ECP does not sanction the surfacing of the empty nucleus N4 in (12) because N4 is not followed by and does not properly govern any other empty nucleus. Therefore the representation in (12) is ruled out. Observe the change of focus in (13) with respect to (1): in contrast to the standard version, the revised ECP does not state when empty nuclei may remain silent, but when they must surface.8 Under the new approach, the surfacing of an empty nucleus is analogical to assigning (rhythmic) stress to a syllable. A nucleus surfaces if it has a proper governee. A syllable also 8 The new formulation of the ECP resembles the proposal of Gussmann (1997). According to both analyses an empty nucleus surfaces only in order to properly govern another one. Otherwise it remains silent. However, in Gussmann’s account of Polish proper governing relations are right-headed and established directionally from right to left. Moreover, Gussmann allows for an iterative application of PG while the present analysis does not. On trochaic Proper Government 285 becomes the head of a foot if it is accompanied by an unstressed syllable. The ill-formedness of (12), where a single empty nucleus surfaces, is analogi- cal to the ill-formedness of a non-branching metrical foot. In Optimality Theory terms, FOOTBINARITY is a high ranked constraint in the metrical systems of many languages which rules out degenerate feet. The analogical requirement of binarity of PG relations tends to be high ranked, too. The revised ECP leaves it open whether or not a properly governed posi- tion can be phonetically realized. Such a formulation is supported by the observation that a proper governee in a disyllabic long vowel configuration is phonetically audible (see also fn. 4 above). 3.5. NOLAPSE The surfacing of every other empty nucleus in a sequence, I propose, is triggered by a ‘rhythmic’ well-formedness condition. A sequence of empty nuclei constitutes a lapse in the representation, analogical to the stress lapse created by a sequence of unstressed syllables. The restriction on stress lapses in rhythmic languages has been formalized by van der Hulst (1994) as the constraint NOLAPSE, given in (14a). I propose that a parallel constraint, given in (14b), prohibits lapses at the nuclear projection level where PG relations are contracted.9 (14) a. NOLAPSE in stress b. NOLAPSE at nuclear projection * w w * N N σ σ 0 0 NOLAPSE is not always satisfied in stress: sequences of unstressed syllables do occur. In Optimality Theoretical terms (Prince & Smolensky, 1993; McCarthy & Prince, 1993), NOLAPSE is a ‘soft’, or violable, constraint. Similar NOLAPSE violations involving sequences of phonetically unrealized empty nuclei can also be found at the PG nuclear projection level (e.g. in Polish. See Rowicka, to appear). However, in Yawelmani and Turkish NOLAPSE is a high-ranked constraint and is satisfied by deleting an empty ON sequence. 9 Honoré Kamany has pointed out to me that postulating NOLAPSE makes it largely redun- dant to assume PG relations between a contentful nucleus and the following empty nucleus. All that matters would just be to avoid lapses. It is still possible, however, that other types of phonotactic considerations may provide additional evidence in favour of PG relations. One candidate is, for instance, the lenition of a consonant enclosed in a PG relation (see Charette, 1991: 138). The issue at hand is comparable to choosing between a grid-only ver- sion and a bracketed grid version of metrical theory. I will not pursue it further here. . 286 Grazyna Rowicka Note that a single empty nucleus does not constitute a violation of NOLAPSE, wherever it occurs in the word. Word-internally and finally a single empty nucleus will be properly governed by a contentful nucleus in the preceding syllable (by trochaic PG). A single empty word-initial nucleus cannot be properly governed (because no potential governor precedes it), nor does it violate NOLAPSE. There is no other empty nucleus for it to gov- ern, hence the revised ECP does not trigger its surfacing. Such an empty nucleus resembles an odd syllable which is left unparsed because it cannot constitute a well-formed binary foot on its own. NOLAPSE has nothing to say about such ‘unparsed’ empty nuclei. 4. Conclusion Based on the following assumptions, I have argued that trochaic PG facili- tates a straightforward analysis of long vowel shortening in Yawelmani and Turkish. 1) PG is a left-headed governing relation, i.e. the governor precedes the governee; 2) an empty nucleus surfaces in order to govern another empty nucleus (see the revised ECP (13)); 3) sequences of empty nuclei are ruled out by the well-formedness condi- tion NOLAPSE (14); 4) representations are not scanned directionally, but the best possible output form is selected (cf. Optimality Theory). The same assumptions can successfully be applied to the analysis of other types of vowel-zero alternations involving the surfacing of empty nuclei (see van der Hulst & Rowicka, 1997 and Rowicka, to appear). Let me point out some advantages of the model proposed in this paper. 1) A special parametric mechanism for licensing domain-final empty nuclei is largely superfluous. 2) There is no need for a directional analysis. Representations must be as well-formed as possible, irrespective of whether they are examined from left to right, or from right to left. On trochaic Proper Government 287 3) The surfacing of empty nuclei appears to be governed by largely the same principles as the stressing of certain syllables: a) left-headedness is the unmarked option in stress feet and in PG rela- tions; b) parsing syllables into feet and the surfacing of some empty nuclei are both due to a rhythmic constraint NOLAPSE (14); c) stress feet are (preferably) branching, and a surfacing empty nucleus must be accompanied by a silent one. Under this approach the parallel between internuclear PG relations and foot formation becomes even more obvious. This analogy lies at the origins of the analysis of vowel-zero alternations in terms of PG relations (see Kaye, 1989). An adequate formulation of the relevant principles can lead to more illuminating analyses of both stress and vowel-zero alternations. Acknowledgements I am grateful to Colin Ewen, Edmund Gussmann, Harry van der Hulst, Honoré Kamany, Jonathan Kaye, Iwona Kraska-Szlenk, Jean Lowenstamm, Roland Noske, Krisztina Polgárdi, Jeroen van de Weijer, Shohei Yoshida and the audience of the 8th International Phonology Meeting in Vienna for helpful comments and stimulating criticism. The usual disclaimers apply. References Anderson, S.R. (1982). The analysis of French schwa: Or, how to get something for nothing. Language 58, 534–574. Charette, M. (1991). Conditions on phonological government. Cambridge: Cambridge University Press. Clements, G.N. & E. Sezer (1982). Vowel and consonant disharmony in Turkish. The structure of phonological representations. Part 2, edited by H.G. van der Hulst & N. Smith, 213–256. Dordrecht: Foris Publications. Dogil, G. (1980). Elementary accent systems. Phonologica 1980, edited by W.U. Dressler, O.E. Pfeiffer & J.R. Rennison, 89–99. Innsbruck: Institut für Sprachwissenschaft. Gibb, L. (1992). Domains in phonology: With evidence from Icelandic, Finnish and Kikuyu. PhD disserta- tion, University of Edinburgh. Gussmann, E. (1997). Govern or perish: Sequences of empty nuclei in Polish. A festschrift for Jacek Fisiak on his 60th birthday. [Trends in linguistics. Studies and monographs 101], edited by R. Hickey & S. Puppel, 1291–1300. Berlin: Mouton de Gruyter. Harris, J. (1990). Segmental complexity and phonological government. Phonology 7, 255–300. van der Hulst, H.G. (1991). Notes on Government and Charm Phonology. Manuscript, University of Leiden/HIL. van der Hulst, H.G. (1994). Notes on foot typology and Optimality Theory. Manuscript, HIL, Leiden. . 288 Grazyna Rowicka van der Hulst, H.G. & G.J. Rowicka (1997). On some parallels between (un)realised empty nuclei and (un)stressed vowels. Phonology in progress – progress in phonology, edited by G. Booij & J.M. van de Weijer. The Hague: Holland Academic Graphics. Kaye, J.D. (1989). Government in phonology: The case of Moroccan Arabic. The Linguistic Review 6, 131–159. Kaye, J.D. (1990). Coda Licensing. Phonology 7, 301–330. Kaye, J.D. (1996). Do you believe in magic? The story of s+C sequences. A festschrift for Edmund Gussmann from his friends and colleagues, edited by H. Kardela & B. Szymanek, 155–176. Lublin: The University Press of the Catholic University of Lublin. Kaye, J.D. & J. Lowenstamm (1984). De la syllabicité. Forme sonore du langage, edited by F. Dell, D. Hirst & J.-R. Vergnaud, 123–160. Paris: Hermann. Kaye, J.D., J. Lowenstamm & J.-R. Vergnaud (1985). The internal structure of phonological elements: a theory of charm and government. Phonology Yearbook 2, 305–328. Kaye, J.D., J. Lowenstamm & J.-R. Vergnaud (1990). Constituent structure and government in phonology. Phonology 7, 193–231. Larsen, U.B. (1994). Some aspects of vowel length and stöd in Modern Danish. MA thesis, Université Paris 7. Lowenstamm, J. (1996). CV as the only syllable type. Current trends in phonology. Models and methods. Vol. 2, edited by J. Durand & B. Laks, 419–443. Paris: CNRS, ESRI. McCarthy, J. & A. Prince (1993). Prosodic morphology: Constraint interaction and satisfaction. Manuscript, University of Massachusetts, Amherst and Rutgers University. Newman, S. (1944). Yokuts language of California. New York: Viking Fund Publications in Anthropol- ogy 2. Newman, S. (1946). The Yawelmani dialect of Yokuts. Linguistic structures of native America, edited by H. Hoijer, 222–248. New York: Viking Fund Publications in Anthropology 4. Prince, A. & P. Smolensky (1993). Optimality Theory: Constraint interaction in Generative Grammar. Manuscript, Rutgers University and University of Colorado at Boulder. Rowicka, G.J. (1998) The audibility and visibility of Mohawk ghosts. Structure and interpretation. Studies in phonology, edited by E. Cyran. 247-260. Lublin: Wydawnictwo Folium. Rowicka, G.J. (in press). Prosodic optimality and prefixation in Polish. The prosody–morphology interface, edited by H.G. van der Hulst, R. Kager & W. Zonneveld. Cambridge: Cambridge University Press. Rowicka (to appear). On ghost vowels. A Strict CV approach. PhD dissertation, HIL/University of Leiden. The Hague: Holland Academic Graphics. Rowicka, G.J. & J.M. van de Weijer (1994). Prosodic constraints in the lexicon of Polish: The case of Derived Imperfectives. The Linguistic Review 11 , 49–76. van de Vijver, R. (1997). Accidental iamb. Dam phonology. The HIL Phonology papers 2, edited by M. Nespor & N. Smith, 191–224. The Hague: Holland Academic Graphics. Yoshida, S. (1993) The case of Palestinian vowel harmony. The Linguistic Review 10. 127–159. Yoshida, S. (this volume). Internuclear relations in Arabic. 335–354. On constraints vs. non-circular approaches to word-initial clusters Tobias Scheer Abstract Co-occurrence restrictions on word-initial consonant clusters are tra- ditionally viewed as a consequence of the relative sonority of both mem- bers of the cluster. In the first part of this paper, I aim to show that the reasoning underlying this approach is circular. The observation that sonority does increase in word-initial clusters is relabelled explanation in saying that sonority must increase. Since the crucial part of this cir- The Hague: Thesus. 289–304. Thesus.The Hague: 289–304. . cular argumentation is expressed by a constraint (“sonority must increase within word-initial clusters”), I address the more general issue of con- straints in linguistic theory. Apart from being circular, the approach based Syllables !? on sonority constraints is not explanatory: it could just as well account for the reverse phenomenology. Indeed, if word-initial clusters were always of decreasing sonority, the constraint would simply say "sonority must decrease within word-initial clusters". If it is assumed that the world is as it is for precise reasons and not simply by chance, scientific investigation Phonologica 1996: Phonologica 1996: can hope to discover those reasons and their effects. It follows that theo- ries should be unable to deal with the reverse phenomenology if it is never attested. Actually, an important feature of theories is to predict the non- existence of objects that do not occur. In the second part of the paper, I propose a constraint-free approach where restrictions on word-initial clusters follow from the interaction of more general principles, especially Government Licensing (Charette, 1990), segmental complexity (Harris, 1990) and a strict CVCV syllable- structure (Lowenstamm, 1996). A theory of consonantal interaction built on the mentioned principles is presented. This theory is shown to account for word-initial restrictions in a non-circular way. Unlike constraint-based approaches, it is unable to deal with the reverse phenomenology, predict- ing that word-initial sonorant-obstruent clusters are impossible in lan- guages of the Indo-European type. 0. John R. Rennison & Klaus Kühnhammer, eds. (1999). eds. R. Rennison & Klaus Kühnhammer, John (1999). 290 Tobias Scheer 1. Word-initial consonant clusters and circularity If a language exhibits word-initial consonant clusters, either there are no restrictions on possible combinations (like in certain Afro-Asiatic lan- guages), or only #TR-clusters (where T = obstruent, R = sonorant) exist, as in Indo-European (IE) languages. There is no language where only RT-clusters occur word-initially. The usual way of accounting for the exclusion of #RT clusters might be summarized as follows:1 1) Words cannot begin with a Coda. Thus, the context “word-initial” cor- responds to “Onset” on the syllabic level. 2) In languages of the IE type, consonant clusters are not free word-initially, but both …TR… and …RT… occur word-internally. This distribution matches that of syllabic constituents: “only Onsets in #__” vs. “both Onsets and Codas word-internally”. Thus, syllabic structure is responsible for the ob- served restrictions. 3) The sonority value for each segment can be established independently. Word-initially, i.e. within a branching Onset, sonority must increase. 4) #RT clusters do not exist because their sonority falls. Hence, they cannot occur within a branching Onset. They cannot be interpreted as a Coda- Onset sequence either because there are no word-initial Codas. Moreover, in typical syllabification algorithms of the Kahnian kind that rely on the maximal cluster approach, the set of existing initial consonant clusters de- fines the very set of possible branching Onsets for the whole language: “a possible branching Onset is all and only the consonant clusters found in the context # __” (cf. Kahn, 1976; Lowenstamm, 1981). This approach is circular: it puts the word “because” between two obser- vations. (1) a. observation: “sonority always increases within initial consonant clusters” b. syllabic interpretation: “TR = branching Onset” c. explanation: “there are no initial RT clusters because sonority must increase within branching Onsets” Circularity is introduced by the word “must”: the only thing the statement “sonority must increase” follows from is precisely the observation 1 For more discussion of this approach, see for example Clements (1990), Selkirk (1984) and references therein. On constraints vs. non-circular approaches to word-initial clusters 291 “sonority increases”. Thus, the whole approach simply says “X is the way it is because it is the way it is.” This kind of reasoning is unsatisfactory. The question “why do RT-clusters not occur word-initially in some languages” still requires an answer. Before turning to a different approach based on more general principles, I should like to discuss the concept of constraints that is crucially involved in the cir- cularity of the reasoning. 2. Constraints are observations, not explanations As has been illustrated above, circularity is induced by the word “must”. Constraints crucially rely on this word, turning an observation into a neces- sity. Consider the prototypical example in (2). (2) a. observation trees grow straight up b. the observed facts are not random trees always grow straight up c. they must be as they are there is a constraint: GR.UP “trees must grow straight up” d. WHY do we observe these facts? trees grow straight up be- Because there is a constraint. cause GR.UP forces them to do so Circularity is an intrinsic property of constraints. The observed facts are never viewed as the consequence of something independent of the observa- tion made, such as a more general principle. Uttering the “must” merely says “the observed facts are as they are because I observe that they are as they are”. Instead of observing “X is like that”, using the word “constraint” in order to refer to an observation suggests that “X is like that, and it cannot be any other way since it must be like that”. If X is constrained, something or somebody must be at the origin of its constrainedness. However, the quest for this origin is never undertaken. Still more oddly, constraints inhibit further investigation. Since the an- swer to the question “why?” is the constraint, there is no more need to look out for an explanation of the facts. In the case of the tree example, no one will have the idea of connecting the observations to a conditioning factor such as sunlight because the constraint has already explained why trees grow straight up. Or let us take the example of the peach falling down. If Isaac Newton had considered that the peach does not go up or zigzag around because there is a constraint saying that peaches must fall down, no relation would ever have been established between falling peaches, their mass and the mass of the earth. In the same way, the moon would be said to 292 Tobias Scheer turn around the earth because of a constraint that prevents it from drifting away. No relation between the peach, the moon and something like gravita- tion would ever have been proposed. Finally, one major goal of scientific theories is to predict what cannot possibly exist. For example, once physicists have observed the behaviour of falling peaches, the phenomenon of gravitation extends to anything that has a mass. Namely, a prediction is made to the effect that there is no possible world where masses repel each other. Nor is it possible for masses to re- main without effect on each other. Thus, the successive pieces of evidence adduced by Isaac Newton (masses attract each other), Yuri Gagarin (but not in space) and Neil Armstrong (they do on the moon, but not so much) showed different manifestations of the universally true principle “masses attract each other”. There was no way for physicists to elude a unified ac- count by positing three different constraints (1. On earth, masses attract each other strongly, 2. On the moon, attraction is poor, 3. In space, there is no attraction) or three different rankings thereof. Anyone can imagine the state of our understanding of physics if they had done so. Returning to initial consonant clusters, no prediction of any kind is made by the constraint “sonority must increase within a branching Onset”. If a planet were discovered where the reverse phenomenology was found, nothing would prohibit explaining the new data by a constraint “sonority must decrease within branching Onsets”. A theory that can cope with all possible data and their reverse is not a theory at all, but a notational arte- fact enumerating observations. The interpretation of the restrictions on word-initial clusters I develop in the remaining sections cannot possibly deal with the existence of such a planet. 3. Questions If word-initial restrictions on consonant clusters are viewed as a con- sequence of the constraint “sonority must increase within a branching On- set”, the problem has found an answer. The linguist is not bothered by any further questions. The questions I would like to ask arise on taking a closer look at the cor- rect observation “word-initial restrictions on consonant clusters are related to sonority”. As a matter of fact, impossible word-initial clusters belong to two different types: On constraints vs. non-circular approaches to word-initial clusters 293 (3) a. Syntagmatic restrictions Initial consonant clusters that do or do not occur depending on the syntagmatic order of their members: #TR is ok, but #RT out. In clusters of this type, the consonants always contrast in sonor- ity. b. Segmental/paradigmatic restrictions There are also consonant clusters of non-contrasting sonority that do not occur word-initially: e.g. *#lr, rl, nl, ln, tp. In these cases, the syntagmatic order of the members does not matter: they are unattested in any order. The property “consonant cluster of contrasting sonority” seems to be related to syntagmatic restrictions, whereas the non-occurrence of initial consonant clusters of non-contrasting sonority has nothing to do with syn- tagmatic ordering. Rather, it is the consequence of the cohabitation of two equally sonorant consonants. An explanation for the former is to be sought in lateral relations holding among segments, while the latter must be due to the genuine identity of the consonants involved. 4. Consonantal interaction: Paradigmatic aspect Let us first address the issue in (3b). In order to represent the situation ob- taining within a cluster of two consonants, some model of the internal struc- ture of consonants must be used. The phonological identity of consonants is a classical issue under debate. Various theories have put forward very different and partly incompatible models for consonantal representation. Representations diverge as to the phonological primitives they use. Feature Geometry (e.g. Clements, 1993; Sagey, 1986) assumes multi-valued features whereas Particle Phonology (Schane, 1984), Dependency Phonology (Anderson & Ewen, 1987) as well as Government Phonology (KLV, 1985) rely on bigger, monovalent objects. They contrast with respect to the relations that are supposed to hold between these primitives, i.e. arborescence (Fea- ture Geometry) vs. a dependency-type of relation (the other models men- tioned). But even within a given framework, the various proposals are far from being consensual. Space restrictions preclude a detailed exposition of the empirical evidence that grounds the consonantal representations used below. This evidence may be recovered from Scheer (in press, 1996, 1998c). I assume primitives of monovalent nature that reside on autosegmental lines. More specifically, the model of consonantal representation used here is couched within Government Phonology. It makes reference to work that has been carried out on consonants within this framework, including Harris 294 Tobias Scheer (1990; 1994), Harris & Lindsey (1995), Cyran (1994), Weijer (1994), Rennison (this volume). Consider the situation obtaining for the melodic lines hosting A, I and U when two consonants occur in a row. “ ” indicates the absence of any primi- tive. The juxtaposition of an empty and a filled position is noted by “⇐”.2 (4) a. p r t r k l f r I/U ⇐ I ⇐ I U I ⇐ I A ⇐ A ⇐ A ⇐ A A A b. n r s r ł r t p I/U I I I I U I A A A A A A A Typical branching Onsets as in (4a) oppose at least one empty and one filled position on a given line. By contrast, the consonant clusters in (4b), which do not occur word-initially for the paradigmatic reasons mentioned (they are *#__ in any order), never oppose an empty and a filled position on a given line. The representations in (4) of course are only a choice of pos- sible combinations of two consonants. Place restrictions preclude an ex- haustive survey. The reader may verify that the above statements have gen- eral value in Scheer (1996: 320ff). This distribution relating possible initial consonant clusters to the oppo- sition between filled and empty positions leads me to propose the following definition of consonantal interaction: (5) Infrasegmental Government (IG) Iff a phonological primitive faces an empty position on a given auto- segmental line, it may govern that position. According to (5), the consonant clusters in (4b) can contract no infrasegmental governing relation because either both positions on a given line are occupied ([nr], [sr], [łr]) or both are empty ([tp]). Note that the con- sonantal identities shown are established on the grounds of segmental alternation that make no reference to phonotactics at all. IG can be viewed as a development of Harris’s (1990) notion of segmental complexity. Harris argues that interconsonantal relations depend on the number of phonological primes the head and the dependent are made of. 2 Labiality/roundness is represented by an extra prime, B. U contributes velarity with no implication of lip rounding. On constraints vs. non-circular approaches to word-initial clusters 295 The more complex a consonant (i.e. the more primes it is made of), the bet- ter head of a consonantal domain it is, and vice versa. Obviously, IG relies on this notion: a consonant C1 may govern another consonant C2 iff C1 is more complex than C2. However, this approach would not be any better than the one based on sonority constraints if it could not say why possible initial consonant clus- ters should depend on consonantal interaction. The distributional relation between the possible interaction of consonant clusters occurring word- initially and the impossible interaction of consonant clusters that are banned from this context is nothing more than an observation. Therefore, the next section examines how IG interacts with lateral rela- tions holding between phonological categories. 5. The phonological ECP, CVCV and the beginning of the word In recent work, the hypothesis assuming a strict CVCV syllable structure has been evaluated for particular analyses in various languages.3 The CVCV model (Lowenstamm, 1996; this volume) views syllabic structure as a strict sequence of non-branching Onsets and non-branching Nuclei (i.e. no branching constituents, no Codas). For the sake of clarity, consider the rep- resentation of closed syllables, geminates, long vowels and the right edge of consonant-final words within this framework, exemplified in (6) with the consonant [t] and the vowel [a]. (6) closed syllable geminate long vowel […C#] O N O N O N O N O N O N O N t a t 0 t a t a … t 0 # All structural information contained in traditional syllabic approaches is preserved. For instance, the site of “closed-syllable” phenomena such as devoicing, lenition, shortening etc. that occur word-finally and before con- sonants usually receives the uniform description “Coda”. In a CVCV ap- proach, these phenomena are referred to as occurring “before an empty Nucleus”. The difference between these two descriptively equivalent state- ments is the causal relation obtaining between the relevant environment and the observed phenomena. Apart from the general observation that 3 See e.g. Lowenstamm (1988; 1996), Guerssel & Lowenstamm (in prep.), Bendjaballah (1995), Creissels (1989), Bonvino (1995), Ségéral (1995), Hérault (1989), Nikiema (1989), Ségéral & Scheer (in press), Larsen (1994; 1995), Heo (1994), Scheer (1996; 1997; 1998a,b). 296 Tobias Scheer Codas are “weak” because e.g. they admit only a subset of possible con- sonants, there is no reason why segments should devoice, deaspirate, lenite, in short decomplexify in this specific position. The correct cross- linguistic observation pointing to the weakness of Codas can only lead to a less surprised reaction when devoicing etc. occurs once more in a Coda- position. It can hardly explain this fact. By contrast, if the Onset is univer- sally viewed as a dependent of the Nucleus like e.g. in Government Phonol- ogy, then the fact that objects decomplexify before an empty Nucleus stands in a direct causal relation with the emptiness of the latter. That is, the licensing power of an empty category is less than that of a filled category. A CVCV structure multiplies the number of empty positions, namely of empty Nuclei. This situation raises the more general question of the status of empty categories in linguistics. It seems to be consensual that “you can- not get an empty position for free”. This idea is encoded within the Empty Category Principle, which states that an empty position may remain un- expressed if and only if precise conditions are met. These conditions are defined in terms of the relation the empty position contracts with a filled position that is laterally distant. In syntax, it was proposed that movement could only take place if the trace of the moved object in its now empty base- position is properly governed by this object in its new position. Proper Gov- ernment (PG) was defined by the structural relation which the filled and the empty position contracted (c-command, barriers). This example provides the kind of motivation that is typical for the existence of empty categories. If there were no structure preservation, i.e. if the position the object was moved from were deleted or even lexically absent, no explanation along the above lines would be available. Empty positions do burden the grammar because they require special care (defined e.g. as PG). Nevertheless, their existence is a necessary con- dition for an explanatory account. Hence, the burdening of the grammar with more empty categories should not be seen as an undesirable overload, but rather as a welcome source of explanation. If grammar is not free in its moves because it must create or maintain the conditions required for the existence of empty categories, a step towards more constrainedness is taken. The challenge, as for any other scientific theory, is to propose a model that is as constrained as possible while covering all relevant data. The same reasoning holds for phonology. KLV (1990: 219) proposed phonological PG based on the same kind of lateral long-distance phenom- ena involving an empty and a filled position that led to syntactic PG. In their view, empty categories are subject to the ECP in phonology as well as in syntax. An adapted version of their phonological ECP is given in (7). On constraints vs. non-circular approaches to word-initial clusters 297 (7) Empty Category Principle An empty Nucleus remains unexpressed iff it is properly governed. The long-distance phenomena mentioned are vowel-zero alternations that are typically sensitive to the object(s) occurring between the zero (empty Nucleus) and the vowel (filled Nucleus) to its right. Consider vowel- zero alternations from various genetically unrelated languages given in (8).4 (8) zero vowel vowel glosses CeC-V CeC-0 CeC-CV Moroccan k-it0b-u k0ti- b-0 k-itti- b-0 ‘write PERF.ACT.3PL./ Arabic 3SG./3SG.CAUSATIVE’ German inn0r-e inner-0 inner-lich ‘inner+INFL./inner/ (opt. elision) internal’ Tangale dob0-go dobe dobu-n-go ‘called/call/called (Chadic) me’ Somali nir0g-o nirig-0 nirig-ta ‘young female camel (Cushitic) PL./SG.INDEF./SG.DEF.’ Turkish dev0r-i devir-0 devir-den ‘transfer ACC./NOM./ ABL.’ Slavic lok0t-e loket-0 loket-ní ‘elbow GEN./NOM./ (e.g. Czech) ADJ.’ Hungarian maj0m-on majom-0 majom-ra ‘monkey SUPERESSIVE./NOM./ SUBLATIVE’ If the alternation site and the following vowel are separated by more than one consonant, as in the grey-shaded column, the expected zero surfaces as a vowel. The intervening consonant cluster is viewed as a barrier that does not allow the filled Nucleus to properly govern the empty Nucleus, which must therefore appear on the surface.5 However, the blocking effect of the barrier consonant cluster is a purely observational fact that does not follow from anything, see (9a). By contrast, the multiplication of empty Nuclei when assuming a CVCV structure offers an immediate answer to the question “why do intervening consonant clus- ters block Proper Government?”, namely (9b). 4 See e.g. Kaye (1989; 1990), Charette (1990), Scheer (1996; 1997; 1998b) for data and analyses concerning vowel-zero alternations. References for the data given in (8) can be found in Scheer (1997, 1998a). 5 See e.g. KLV (1990), Kaye (1990), Charette (1990), Scheer (1996; 1998a,b,c) on PG. 298 Tobias Scheer (9) a. Non-CVCV: Why do intervening consonant clusters block PG? PG | R R O N C O N C x x x x x x k i- t t -i b b. CVCV: why do intervening consonant clusters block PG? Because the [CC] encloses an empty Nucleus N☺, /CN☺C/, that seeks PG. PG is not blocked, it simply cannot reach the first [i- ]. PG O N O ☺N☺ O N O N k i- t t -i b However, intervening consonant clusters do not block PG in all circum- stances. In Czech prefixes, for instance, -e- alternates with zero: pode-brat ‘seize from below’ vs. pod0-bradek ‘double chin’. The alternation corresponds to a contrast in the lexical structure of the roots involved: /ber/ for [-br-at] where -e- is properly governable (cf. 1SG. [-ber-u]) vs. /brad/ for [-brad-ek] where -a- is not properly governable (see Scheer, 1996; 1997 for a complete demonstration). In the former case, -a- properly governs the -e- of /ber/. As a consequence, the prefixal -e- fails to undergo PG. It therefore appears on the surface. (Unassociated segments are inaudible.) (10) PG O N O N – O ☺N☺ O N O p o d e b e r a t By contrast, in the case of pod0-bradek, -a- properly governs the prefixal -0- although the consonant cluster [-br-] stands in between the governor and the governee. (11) PG O N O N – O ☺N☺ O N O N O N p o d e b r a d e k So far, only two phonological operations that are able to satisfy the ECP have been identified: 1) PG and, in extension of the above definition, 2) On constraints vs. non-circular approaches to word-initial clusters 299 Licensing of final empty Nuclei (cf. Kaye, 1990). Assuming a CVCV structure, the empty Nucleus N☺ in cases such as (11) is neither final nor involved in PG (the -a- governs the prefixal -e-). Nevertheless, it does not surface. Hence, the ECP must be satisfied by another phonological operation. I propose that consonantal interaction as described above can close its domain to the effect that the ECP is satisfied: (12) Phonological operations satisfying the ECP6 a. Proper Government b. Licensing of final empty Nuclei c. Infrasegmental Government: The empty Nucleus N☺ of a domain /CN☺C/ may remain un- expressed if a relation of Infrasegmental Government holds between its surrounding consonants. Under these provisos, PG can apply over /bN☺r/ (in pod0-bradek) because the cluster constitutes a domain of IG. By contrast, no domain of IG can be established within the cluster /ber/ (in pode-brat) since the properly govern- able -e- prohibits consonantal communication. In this case, PG applies to the nearest available target, which is the -e- under focus. As a consequence, this prefixal Nucleus fails to undergo PG and receives phonetic interpreta- tion. In cases like pod0-bradek, PG by -a- cannot be held responsible for the muteness of N☺ in /-bN☺r-/ because its effect can be seen on the prefix. Lowenstamm (this volume) argues that this situation in fact is general even if there is no prefix involved: the first vowel of a word governs what is gener- ally referred to as “#”. The phonological identity of the non-linguistic object “#” is an empty Onset followed by an empty Nucleus: (13) The beginning of the word: “#” is an empty CV PG O N O ☺N☺ O N … b r a … [#bra…] The initial Nucleus being subject to the ECP, it seeks PG from the first vowel of the word. Hence, the first vowel can never properly govern N☺. 6 Other proposals such as Interonset Government and magic licensing, which are discussed in Gussmann & Kaye (1993), Cyran & Gussmann (forthcoming) and Kaye (1992), are not rele- vant for the purpose of this paper. See Scheer (1998b) for discussion of Interonset Govern- ment. 300 Tobias Scheer We have now reached the point where an answer to the question raised at the end of the previous section can be provided. There we saw that impos- sible word-initial clusters such as #nl, #tp are precisely the ones within which no relation of Infrasegmental Government may hold. Hence, it was tempting to establish a causal relation between both facts saying “a word- initial consonant cluster can exist only if it constitutes a domain of Infrasegmental Government”. Nevertheless, there was no apparent reason why Infrasegmental Government should be a condition on possible word- initial consonant clusters. Assuming CVCV, and the phonological identity of “#” and CV, the question “why are consonant clusters within which no IG holds not possible word-initial clusters?” receives the answer “because the empty Nucleus they enclose is subject to the ECP and IG the only way to satisfy it”. As an example, consider the situation for possible #tr as opposed to impossible *#nr: (14) a. #tr b. *#nr PG IG PG O N O ☺N☺ O N … O N O ☺N☺ O N … ⇐ I a … I I a ⇐ A A A t r n r In both cases (14a–b), the ECP concerning the initial empty Nucleus is satisfied through PG by the first vowel of the word. In contrast, only the ECP applying to the N☺ of (14a) is satisfied: [tr] can interact and close their domain, whereas [nr] cannot. The cluster (14b) is ruled out because it con- tains an empty Nucleus, N☺, that is not licensed by any of the phonological operations that may satisfy the ECP. 6. Consonantal interaction: Syntagmatic aspect Up to this point, it has been argued that the set of consonant clusters that do not occur word-initially for paradigmatic reasons coincides with the set of consonant clusters within which no Infrasegmental Government may hold. Moreover, a causal relation between impossible IG and the non- occurrence of consonant clusters in word-initial position has been estab- lished. I now wish to address the syntagmatic aspect of the restrictions on word- initial clusters: why do consonant clusters of the #TR-kind occur word- On constraints vs. non-circular approaches to word-initial clusters 301 initially, but not their mirror-image *#RT? The particular identities of the consonants involved do not matter here because they are identical for the occurring as well as for the non-occurring consonant clusters. In theory, Infrasegmental Government is possible for both #TR (right-to-left) and *#RT (left-to-right): (15) Head-final IG Head-initial IG O ☺N☺ O N O ☺N☺ O N ⇐ I I ⇒ ⇐ A A ⇒ t r r t Based on independent evidence, Charette (1990) has noted that inter- consonantal relations depend on the availability of a vocalic support for the head of the consonantal domain. The generalisation she arrives at is given under (16). (16) Government Licensing (adapted from Charette, 1990) A consonant C1 can govern another consonant C2 iff C1 is licensed to do so by its Nucleus. Government Licensing was developed within a non-CVCV frame where interconsonantal relations are expressed by means other than Infrasegmental Government and the consonantal identities differ from the ones assumed here. However, the idea expressed by Government Licensing is theory-neutral: in order for a consonantal cluster to exist, its head needs vocalic support. Let us now see, in (17), what predictions are made by Gov- ernment Licensing as to word-initial clusters: (17) a. Well-formed structure b. Ill-formed structure PG PG Lic. |Lic. O N O ☺N☺ O N … O N O ☺N☺ O N … T ⇐ R V … R ⇒| Τ a IG IG Under the assumption of Infrasegmental Government, R is always the head of the domain of consonantal interaction, and T is the dependent. Accord- ing to Government Licensing, R needs to be licensed by its Nucleus in order to be able to govern T. In (17a), the Nucleus following R hosts the first vowel 302 Tobias Scheer of the word. This vowel can license R so that R is able to establish IG over T. In (17b), however, the Nucleus N☺ following R is empty in any event. Hence, it can never license R, which, in turn, is unable to govern T. Although the two members of #RT fulfil segmental requirements for an Infrasegmental Gov- ernment relation, no such relation can be established for syntagmatic rea- sons. As a consequence, N☺ does not satisfy the ECP, and (17b) is ill-formed. Charette’s Government Licensing thus correctly predicts the non- occurrence of initial *#RT clusters.7 7. Conclusion In this article, I have tried to develop a non-circular alternative to the stan- dard way of handling the distributional restrictions on word-initial con- sonant clusters. At no point of the argumentation does a constraint inter- vene. Rather, the set of observations expressed by commonly used con- straints such as “within a branching Onset, sonority must increase” follow from more general principles. According to the view advocated in this arti- cle, restrictions on word-initial consonant clusters follow from the interplay of the factors in (18). (18) a. Government Licensing (Charette, 1990), b. segmental complexity (Harris, 1990), c. the phonological ECP (KLV, 1990), d. CVCV and “#”=CV (Lowenstamm, 1996; this volume), e. the consonantal identities developed in Scheer (1996; 1998c; in press). All of these devices are assumed to be generally operative in Phonology. None of them makes special reference to the particular issue discussed, viz. word-initial consonant clusters. For this reason, the approach presented is not circular. 7 Domains of IG are defined in the lexicon. Accordingly, in languages where consonants are lexically unrelated to syllabic constituents, IG does not occur. It is interesting to note that languages allowing for both #TR and #RT clusters are precisely representatives of Afro- Asiatic. These languages have a templatic structure, i.e. syllabic constituents and segmental information do not co-habit in the lexicon. Lowenstamm (this volume) argues for a different status of the initial CV in this kind of language. According to his analysis, the initial empty Nucleus is not always subject to PG in templatic systems. If this view is correct, then the possible occurrence of both #TR and #RT clusters in templatic languages follows: in #CN☺CV, V properly governs N☺, not the initial CV. Consequently, the surrounding con- sonants are subject to no co-occurrence restrictions. On constraints vs. non-circular approaches to word-initial clusters 303 Finally, the theory of consonantal interaction presented, unlike constraint-based models, makes the prediction that a world where only word-initial clusters of decreasing sonority occur could not possibly exist. Acknowledgements This article is an abridged version of Scheer (in press). I am grateful to Krisztina Polgárdi, Honoré Kamany, Patricia Cabredo-Hofherr, Ferdinand Zˇblabunˇka, Joaquim Branda˜o de Carvalho, John Rennison and an anony- mous reviewer for valuable comments. References Anderson, J.M. & C.J. Ewen (1987). Principles of Dependency Phonology. Cambridge: Cambridge Uni- versity Press. Bendjaballah, S. (1995). Aspects apophoniques de la vocalisation du verbe berbère (kabyle). Langues et Grammaire II-III, Phonologie, edited by P. Sauzet, 5–24. Paris: Université Paris 8. Bonvino, E. (1995). Aspects de l’article défini en romanesco. MA thesis, Université Paris 7. Charette, M. (1990). Licence to govern. Phonology Yearbook 7, 233–253. Clements, G.N. (1990). The role of the sonority cycle in core syllabification. Papers in Laboratory Phonology, I: Between the Grammar and Physics of Speech, edited by J. Kingston & M. Beckman, 283–333. Cambridge: CUP. Clements, G.N. (1993). Lieu d’articulation des consonnes et des voyelles: une théorie unifiée. L’architecture des représentations phonologiques, edited by B. Laks & A. Rialland, 101–145. Paris: CNRS Editions. Creissels, D. (1989). Structures syllabiques du dialonké de Faléya. Mandekan 18, 1–30. Cyran, E. (1994). Vocalic Elements in Phonology: A study in Munster Irish. Doctoral dissertation, Uni- versity of Lublin. Cyran, E. & E. Gussmann (forthcoming). Consonant clusters and governing relations: Polish initial consonant sequences. The syllable: views and facts, edited by H. van der Hulst & N. Ritter. Berlin, New York: de Gruyter. Guerssel, M. & J. Lowenstamm (in prep.). The derivational morphology of the Classical Arabic Verb. Ms., Université Paris 7 & UQAM. Gussmann, E. & J. Kaye (1993). Polish Notes from a Dubrovnik Café. SOAS Working Papers in Lin- guistics and Phonetics 3, 427–462. Harris, J. (1990). Segmental complexity and phonological government. Phonology Yearbook 7, 255–300. Harris, J. (1994). English sound structure. Oxford: Blackwell. Harris, J. & G. Lindsey (1995). The elements of phonological representation. Frontiers of Phonology, edited by J. Durand & F. Katamba, 34–79. London: Longman. Heo, Y. (1994). Empty categories and Korean Phonology. Doctoral dissertation, SOAS, London. Hérault, G. (1989). Les rections syllabiques en soninké. Linguistique Africaine 3, 43–90. Kahn, D. (1976). Syllable based generalizations in English Phonology. Doctoral dissertation, MIT. Kaye, J. (1989). Government in Phonology: The case of Moroccan Arabic. The Linguistic Review 6, 131–159. Kaye, J. (1990). ‘Coda’ licensing. Phonology Yearbook 7, 301–330. 304 Tobias Scheer Kaye, J. (1992). Do you believe in magic? The story of s+C sequences. SOAS Working Papers in Linguistics and Phonetics 2, 293–313. Kaye, J., J. Lowenstamm & J-R. Vergnaud (1985). The internal structure of phonological represen- tations: A theory of Charm and Government. Phonology Yearbook 2, 305–328. Kaye, J., J. Lowenstamm & J-R. Vergnaud (1990). Constituent structure and government in pho- nology. Phonology Yearbook 7, 193–231. Larsen, B.U. (1994). Some aspects of vowel length and stød in modern Danish. MA thesis, Université Paris 7. Larsen, B.U. (1995). Italian vowel length, Raddoppiamento Sintattico and the selection of the definite article in modern Italian. Langues et Grammaire II-III, Phonologie, edited by P. Sauzet, 87–102. Paris: Université Paris 8. Lowenstamm, J. (1981). On the maximal cluster approach to syllable structure. Linguistic In- quiry 12, 575-604. Lowenstamm, J. (1988). Another look at Ngizim syllable structure. Talk given at the 19ème con- férence de linguistique africaniste, Boston University, April 1988. Lowenstamm, J. (1996). CV as the only syllable type: Full interpretability in phonology. Current trends in Phonology: Models and Methods, edited by J. Durand & B. Laks, 419–441. Salford: ESRI. Lowenstamm, J. (this volume). The beginning of the word. 153–166. Nikiema, E. (1989). Gouvernement Propre et licenciement en phonologie: le cas du Tangale. Langues Orientales Anciennes, Philologie et Linguistique 2, 225–251. Rennison, J.R. (this volume). Can there be empty phonological elements? On empty heads and empty operators. 183–193. Sagey, E. (1986). The representation of features and relations in non-linear phonology. PhD dissertation, MIT. Schane, S.S. (1984). The fundamentals of Particle Phonology. Phonology Yearbook 1, 129–155. Scheer, T. (1996). Une théorie de l’interaction directe entre consonnes. Doctoral dissertation, Université Paris 7. Scheer, T. (1997). Vowel-zero alternations and their support for a theory of consonantal inter- action. Certamen Phonologicum III, edited by P.M. Bertinetto, L. Gaeta, G. Jetchev & D. Michaels, 67-88. Torino: Rosenberg & Sellier. Scheer, T. (1998a). A unified model of Proper Government. The Linguistic Review 15.1, 41-67. Scheer, T. (1998b). Governing domains are head-final. Structure and Interpretation. Studies in Phonol- ogy, edited by E. Cyran, 261-285. Lublin: Folium. Scheer, T. (1998c). La structure interne des consonnes. Langues et Grammaire II-III, Phonologie, edited by P. Sauzet, 140–172. Paris: Université Paris 8. Scheer, T. (in press). A theory of consonantal interaction. Folia Linguistica. Ségéral, P. (1995). Une théorie généralisée de l’apophonie. Doctoral dissertation, Université Paris 7. Ségéral, P. & T. Scheer (in press). A generalized theory of Ablaut: The case of Modern German Strong Verbs. Models of Inflection, edited by A. Ortmann, R. Fabri & T. Parodi. Tübingen: Niemeyer. Selkirk, E. (1984). On the major class features and syllable theory. Language Sound Structure, edited by M. Aronoff & R.T. Oehrle, 107–136. Cambridge, Mass.: MIT Press. van de Weijer, J.M. (1994). Segmental structure and complex segments. Doctoral dissertation, Rijks- universiteit te Leiden. The syllable as a cognitive unit in reading and spelling Margit Tornéus & Björn Andersson Abstract The purpose of this theoretical paper is to present an alternative to pre- vailing theories of reading/spelling and dyslexia. It is argued that reading and spelling skills are dependent on the use of appropriate cognitive units, and that syllables are the appropriate units in the context of reading and spelling. Syllables preserve the prosodic and articulatory parameters governing the “translation” of spoken words into sequences of letters (i.e. written words) and vice versa. It is argued that dyslexia is the behavioural The Hague: Thesus. 307–313. Thesus.The Hague: 307–313. . consequence of using phonemes as cognitive units in reading and spell- ing. This strategy will cause short term memory problems, problems in detecting orthographic regularities in spelling as well as in reading, Syllables !? problems in constructing prosody in reading and serious problems in reading comprehension. It is also argued that the use of phonemes as cognitive units is caused by the fact that some pupils do exactly what they are taught to do: they use phonemes. 0. Phonologica 1996: Phonologica 1996: 1. Introduction The common feature of mainstream theories of reading/spelling and of dyslexia is that the phoneme is considered to be of critical importance, being conceptualized as the link between the spoken and the written word (e.g. Hulme & Snowling, 1991; Lundberg, 1994). “Cracking the code”, i.e. learning to segment the spoken word into speech sounds, is supposed to be the main obstacle in learning to read and spell. According to the most popular explanation of dyslexia, dyslectic persons suffer from some kind of linguistic (phonemic or phonological) deficit (“metaphonological disabilities”, “disturbances”, etc.) which causes problems in segmenting words into phonemes and blending sounds into John R. Rennison & Klaus Kühnhammer, eds. (1999). eds. R. Rennison & Klaus Kühnhammer, John (1999). words. The common explanation of the large body of empirical research 308 Margit Tornéus & Björn Andersson demonstrating that dyslectics encounter great problems in phoneme seg- mentation tasks is that they suffer from this deficit, which is often assumed to have neurological and sometimes genetic correlates (e.g. Lundberg, 1994). The main purpose of this paper is to give an alternative approach to reading/spelling and to dyslexia, in which the syllable is assumed to be the working unit in spelling as well as in reading. However, we want to empha- size that we are not making any claims about the status of the syllable in the context of oral language or in any task other than in reading and spelling. 2. The role of the syllable in phoneme segmentation Segmenting words into phonemes demands some kind of cognitive organi- zation unit as soon as the number of sounds approaches the limit of the memory span (we are referring to one of the most established empirical facts in psychological research). This means that it is impossible to remem- ber all the items and their serial positions if the number of elements ex- ceeds about seven. Performance is characterized by omissions of elements as well as position reversals. However, if the elements are organized into larger units (e.g. single numbers organized into pairs forming tens) the same number of units are remembered, i.e. about seven, but each unit will also now consist of a number of original elements. Thus, the total number of original elements remembered will depend on cognitive organization. Studies of phoneme segmentation have demonstrated significant effects of word length and significant interactions between groups and length. The latter effect means that the difference between dyslectics and normal readers/spellers is significant for longer words but less pronounced or absent for shorter words. We suggest that the explanation of the dyslectics’ low performance is that although they do in fact segment the words into phonemes they do not use the syllable as an organizing unit; when the words get longer their performance will be impaired. This is the entrance into a vicious circle: when “phoneme segmentation problems” are detected, additional training in phoneme segmentation is introduced, which confirms and accentuates the phoneme even more as the exclusive unit to use. Normal readers/spellers, however, invent the use of the syllable as an organizing unit despite the lack of teaching. In other words, the trouble for the dyslectics is that they do exactly what they are taught to do — they use the phoneme. The syllable as a cognitive unit in reading and spelling 309 3. Empirical evidence Results from one of our ongoing studies of Swedish dyslectic persons demonstrated that 18 of the 20 dyslectics in one group (age 12–27 years) encountered problems on a syllable segmentation task (mean correct performance .77 ).1 As a contrast, the same task was easily performed (mean performance .91) without any training by a group of 32 randomly selected normal preliterate children (5–6 years old). Fifteen of the 18 dyslectics who encountered problems on the syllable segmentation task scored 100% correct on the first part of a phoneme segmentation test; this part included all ONC-structures in Swedish (the most complicated structure includes 3 onset and 2 coda members, e.g. struts, or vice versa, e.g. trilsk). However, none of the 20 dyslectics succeeded on the second part of the phoneme segmentation test, which included polysyllabic words composed of three to five O1N syllables (e.g. visare, föreläsare). Those 18 who did not succeed in the syllable segmentation task were then trained to segment words into syl- lables. After 1–3 training sessions, (each session lasting 20–40 min.) they all scored 100% correct on a similar syllable segmentation task. All 20 were then trained to apply syllable segmentation as a first step in segmenting polysyllabic words into phonemes. After one such training session all 20 achieved 100% correct answers on both parts of the phoneme segmentation test mentioned above. We consider these results as counter-evidence for any deficit hypothesis, since it is possible to teach dyslectics syllable as well as phoneme segmen- tation in a very short period (results from studies of training phoneme segmentation have been reported by a number of authors, e.g. Tornéus, 1984; Byrne & Fielding-Barnsley, 1993). The reason for not using the syllable might be that most teaching practices, at least in Sweden, do not include instructions on how to use syllables as working units in reading and spelling. As a contrast, teaching, which starts in pre-school, is very extensively directed at separate speech sounds. Our interpretation is that, as a con- sequence of instructions, dyslectics are using the phoneme to disorganize the word, instead of using the syllable as a unit for organizing the word. This disorganization results in elements which are too many to handle as well as too distant from the spoken word. To sum up, we suggest that the syllable is an appropriate organizing unit in the context of spelling and reading, since it does not destroy the word — all prosodic and articulatory characteristics are still left. 1 Syllable boundaries were not considered; any segment consisting of one and only one vowel was scored as a syllable. 310 Margit Tornéus & Björn Andersson 4. The role of syllables in spelling If the pupil is trying to spell by using the phoneme as the working unit, spelling is dependent on phoneme segmentation. Omissions of letters, as well as position reversals of letters in longer words, can be predicted from the failure to segment the words into speech sounds. Preliminary data from our ongoing studies show that such errors disappear when the student is taught to use syllables rather than phonemes as working units. However, the majority of spelling errors cannot be explained by incom- plete phoneme segmentation of the word. We will call these errors “trans- lational errors” (e.g. tri instead of try, refered instead of referred). In Swedish orthography many sounds are spelled in different ways depending on a 2 number of parameters. The LAWFULNESS of the spelling of these sounds is about 80 percent (varying from 63 percent for the /S/ to 97 percent for [9]) which means that on average 80 percent of Swedish words containing these 3 MULTISPELLED sounds can be described by a spelling rule. Learning to spell includes implicit learning of these orthographical rules (we want to empha- size parenthetically that we are referring to implicit learning, as opposed to explicit knowledge of spelling rules). The most common parameters gov- erning these translations are the syllable’s position within the word (e.g. the /O:/ is translated by the letter Å in the first syllable and otherwise by the letter O), stressed/unstressed syllable, position within the syllable, i.e. onset/coda position. All these parameters include aspects of the syllable. Thus, in order to learn how to spell a word containing these multispelled sounds, the student must make use of the word’s syllabic structure; he/she must use the syllable as a cognitive unit in spelling. We argue that the spelling errors made by dyslectics are caused by their use of the phoneme as the exclusive unit in spelling, since this strategy makes it impossible to discover much of the lawfulness in spelling. Preliminary data from our ongoing research show that dyslectics learn to spell even the most difficult sounds (/S/, which has 6 alternative spellings according to the spelling rule) after a very short training period (3–5 training sessions, each lasting 10–40 minutes) when they are taught to use the syllable as a cognitive unit. 2 The term “lawfulness” here means “regularity” in the sense of susceptibility to the formula- tion of a spelling rule that requires no idiosyncratic, non-phonological information. 3 This is the result from analyses of a data-based corpus which includes the transcriptions of pronunciations and spellings of 75.000 words. The database was originally developed by Per Hedelin, Chalmers, Gothenburg. The syllable as a cognitive unit in reading and spelling 311 5. The role of syllables in reading Needless to say, reading means translating the text into words, sentences, paragraphs and so on, i.e. extracting meaning. There is, however a gap between the text and the spoken language; oral language is characterized by a number of prosodic features, most of which are not represented in text. Thus, reading includes adding such features as stress, accent and duration within single words as well as within sentences, since they are not to be found in the text. Most of these features are relative and belong to the syllable; one syllable is more stressed than others, one syllable is length- ened more than others and so on. Using the phoneme as a cognitive unit in reading will not bridge this gap since most prosodic features are not realized as relations between separate sounds. Instead, the use of the phoneme as a cognitive unit will strongly limit the possibility of adding intonation and therefore of ending up with language in the proper sense. The consequence for the reader is always lack of reading comprehension. At a more superficial level, “sound reading” might lead to monotonous reading, rereading of parts of words (e.g. the word dog is read as [d], [dO] and finally [dOg]), or unsynthesized sounds / sound groups (e.g. dog is read as [d], [O], [g]) Thus, learning to read can be conceptualized as learning to create words and sentences by adding relational properties to appropriate units. We argue that the syllable is the appropriate unit in this context and that adding the relational features of stress, syllable position and onset/coda-position is learned by manipulating syllables; thus, the main part of learning to read does not involve letters. Learning how to generate the syllables from letters is the final step in learning to read. According to this view, reading problems are to be expected if the teaching is directed more or less exclusively at the final level in the learning process, at the letters. Furthermore, if this teaching is aimed at decoding the letters into separate speech sounds, learning problems are even more probable. In our ongoing research we are teaching dyslectics how to use the syl- lable as a unit of organizing the word. They are taught to manipulate sylla- bles in those aspects which have to be added in the reading task (stress, accent, duration, vowel quality). Preliminary results show an increase in reading comprehension as well as a substantial decrease in all kinds of reading behaviour indicating the use of separate sounds. 312 Margit Tornéus & Björn Andersson 6. Principles for teaching Our model includes one assumption about initial as well as remedial learn- ing/teaching in the context of reading and spelling, viz. that all teaching should be aimed at developing implicit knowledge which enables the pupil to perform the task we are giving him/her. This is realized by the means of “contrastive good examples” and the pupil’s learning by doing. Let us illus- trate this by an example: The spelling rule for the sound [E] in Swedish is that it is translated with an Ä in monosyllabic words and in the first syllable if it is stressed; with an E in initial unstressed syllable and with E in the following syllables regardless of stress. Suppose that a pupil translates the [E] with an Ä in the first syllable and in monosyllabic words and with an E in other syllables. This means that he/she has to learn to take stress into account in the word’s initial syllable. In order to teach him/her, we present pairs of words where the first word in the pair is a word with an initial stressed syllable (the performance on these words is already correct) and the second word has an unstressed initial syllable. The pupil’s task is to repeat the pairs and then to write each word. Misspellings of the [E] are corrected (no other spelling errors are considered in this context) but no explanations are given. The teacher says, ”This is the way I would spell this word” and then she writes the word. This continues until the pupil has reached a criterion of correct spellings of about 10 suc- cessive pairs. The next step is to give the same kind of words one by one in random order. If necessary, corrections are given and the teaching starts from step one again. We want to highlight some features in this example. First, the teacher has to know exactly what implicit knowledge the pupil has to learn (in our ex- ample the pupil has to learn to implicitly take stress into account when spelling the [E] in the first syllable). Knowing that, she chooses word-pairs to accentuate that parameter. No rule is given, since giving rules would break the principle of implicit learning and thereby change the task into a con- sciously controlled action. The reason for the principle of implicit learning is quite simple: we should never teach a student anything which is obviously not true for a skilled speller (or reader if the teaching concerns reading). Skilled spellers do not usually begin the spelling of a word by thinking of spelling rules. As a matter of fact, most skilled spellers do not even know the spelling rules. They have extensive implicit knowledge which is realized in the action of spelling. The syllable as a cognitive unit in reading and spelling 313 7. Conclusions We have suggested that the syllable is the appropriate cognitive unit in reading and spelling, and that this is the unit used by skilled readers/ spellers. Using the phoneme as the cognitive unit in reading/spelling will lead to reading/spelling behaviour summarized as dyslexia. This theoretical approach has been formalized into a comprehensive theory of the processes included in reading and spelling which is being tested in a number of ongoing studies. We claim that any theory of reading/ spelling and of dyslexia should reach the criterion of being applicable to any individual person regardless of age, level of success, and art and degree of problems. In other words: the theory should have explicit implications for diagnosis and teaching which should result in reading and spelling perform- ance equal to or above the average performance level (as compared to the appropriate age group). All data from our studies analysed so far are in accordance with these claims. Finally, we again repeat that we are not making any claims about the status of the syllable in the context of oral language or in any task other than reading and spelling. References Byrne, B. & R. Fielding-Barnsley (1990). Acquiring the alphabetic principle: A case for teaching recognition of phoneme identity. Journal of Educational Psychology 82, 805–812. Hulme, C. & M. Snowling (1991). Phonological deficits in dyslexia: A ‘sound’ reappraisal of the verbal deficit hypothesis? Progress in Learning Disabilities, edited by N. Singh & I. Beale, 260–283. New York: Springer-Verlag. Lundberg, I. (1994). Reading difficulties can be predicted and prevented: a Scandinavian perspective on phonological awareness and reading. Reading Development and Dyslexia, edited by C. Hulme & M. Snowling. London: Whurr Publishers. Tornéus, M. (1984). Phonological awareness and reading: A chicken and egg problem? Journal of Educational Psychology 76, 1346–1358. The syntax of syllables Why syllables are not different Michael Völtz Abstract The present article attempts to develop a reasonably different interpreta- tion of the relation of the phonological unit “syllable” and the overall organization of Universal Grammar. I will give an abbreviated introduction to an approach (cf. Völtz, 1996) that outlines non-accidental similarities between the phonological unit “syllable” and the syntactic unit “sentence”, The Hague: Thesus. 315–321. Thesus.The Hague: 315–321. . i.e. it will be shown that the syllable can be described and explained by an inventory of descriptive and explanatory tools, viz. the Theory of Prin- ciples and Parameters, that is already available as a fully developed Syllables !? (syntactic) theory. The assumptions proposed here have further consequences for Uni- versal Grammar: there is no modularity in such a way that modules like syntax or phonology are separated within the language system itself, i.e. certain subtheories (maybe all) as provided by Universal Grammar (for Phonologica 1996: Phonologica 1996: instance X-bar Theory, Binding Theory, etc.) do not exclusively apply to syntax, but also within other (maybe all) areas of grammar. A large part of the Principles and Parameters Theory (that is believed to form a part of Universal Grammar) can take over tasks of traditional descriptions and explanations of syllable structure and handle more adequately various processes that take place within the syllable. 0. 1. Introduction Bromberger & Halle (1989: 69) conclude their article by the statement that “syntax and phonology are essentially different”. In the following a reversal of this claim will be attempted, to show that John R. Rennison & Klaus Kühnhammer, eds. (1999). eds. R. Rennison & Klaus Kühnhammer, John (1999). syntax and phonology are in fact very much alike with respect to 316 Michael Völtz non-accidental similarities concerning the phonological unit “syllable” and the syntactic units “sentence” or “phrase”.1 2. The model This paper does not argue in favour of an absolute parallel between the internal organization and properties of syntax and phonology. Instead, it is assumed that syntactic and phonological units can be described and ex- plained by just one set of descriptive and explanatory tools as provided by 2 the PRINCIPLES AND PARAMETERS THEORY (PPT). The main intention of this paper is to outline possible strategies to dispose of redundant structures, principles, or rules occurring as identical structures, principles, or rules in both syntax and phonology as predicted within the framework of Govern- ment Phonology for instance by Kaye et al. (1990: 193): “[…] some of the same principles at work in syntax will be seen to be operative in phonology”. Simplified structures of a clause (parameters are set for a head-first lan- guage such as English) and a syllable in (1) show obvious structural parallels. (1) Clause in English vs. hierarchical structure of the syllable3 IP syllable NP ’I’ onset rhyme I VP peak coda These parallels were probably first observed by Levin (1985), and is even earlier reflected in Selkirk (1982: 337): “The notion of the syllable […] is […] one of a hierarchical unit, an internally structured tree quite analogous to a tree representing syntactic structure”. The approach developed below goes beyond a mere observation of identical structures by claiming that a univer- sal syllable structure can be explained by means of a mature linguistic theory, the PPT. The first aspect of the application of the PPT to the syllable concerns a module that is quite central in syntactic analysis, namely X’-Theory. Apart 1 This paper is extracted from a larger study (Völtz, 1996) where various details of the ap- proach sketched below are discussed and motivated at greater length. 2 For a detailed outline of the PPT and its internal organization into modules, see Haegeman (1994) or Webelhuth (1995). 3 Evidence in favour of this structure (and against a ‘flat’, ‘CV’, etc. structure) comes e.g. from Fudge (1969), Selkirk (1982) and the experimental results provided by Treiman (1989). 317 from plain structural aspects as sketched in (1) the integration of further standard assumptions of X’-Theory such as • The Endocentricity Constraint, • The Modifier Maximality Constraint, • a constraint that requires heads to project maximally, etc. may lead to a highly general (universal) structure of the syllable as shown below in (2). This structure emerges solely as a consequence of the introduc- tion of various aspects of X’-Theory to the syllable, i.e. the internal structure is a mere reflection of theory-internal properties of X’-Theory.4 (2) Universal structure of the syllable ”P” ”O” ’P’ … ’O’ P ”C” O … … ’C’ C … The structure in (2) provides specific domains within a syllable: • P” = maximal projection, i.e. complete syllable • P’ = traditional rhyme • P = head of P” • O” = complete onset • C” = complete coda The latter domains O” and C” are realized in each syllable with a head and projections; this does not necessarily imply that the heads are realized overtly. The structure of the affix -ing as shown in (3) displays a head posi- tion in O” that is overtly empty, but covertly filled by eO.5 Apart from the primary constituents of the syllable (P, O, C, and their projections) a secondary constituent6 Add, as shown in the word send in (3), 4 It is recognized, however, that some aspects of X’-Theory do not apply to the syllable; syntactic analysis of subordinate clauses for instance shows that the complement position to the head might contain another maximal projection, i.e. another CP. Clearly, in syllabic analysis C” cannot contain a full P”. 5 For a more detailed outline of the procedure that identifies the head of the overall syllable and the heads in O” and C” clusters see Völtz (1996). 6 See Völtz (1996) for further comments on secondary constituents such as Deriv and Infl that are designed to contain morphological material. 318 Michael Völtz is identified. Add is the domain for segments occupying a non-head position within O” or C” clusters. (3) Syllable structures for a) send and b) -ing a. ”P” b. ”P” ”O” ’P’ ”O” ’P’ s P ”C” Spec ’O’ P ”C” e ’C’ O I 9 C Add” eO n d Certain other domains are empty in base-generated forms, but can be overtly filled if a segment is moved (the original segment moves into the target position leaving behind a co-indexed trace) or copied (the original segment/feature moves into the target position leaving behind a precise co- indexed copy). The overt filling of empty positions directly leads to the introduction of another module of the PPT, i.e. Bounding Theory. Movement or copying processes are predominantly triggered by mor- phological processes; the traditional notion of resyllabification will be replaced by the more precise (and non-redundant) processes of movement and copying; (4) provides an example of a typical disyllabic form that results from an inflectional process: it comprises a root morpheme plus an affix, the latter receiving a segment for its domain O” through movement. (4) Surface structure of sending after movement (ignoring headedness on foot level) ”P” ”P” ”O” ’P’ ”O” ’P’ s P ”C” Spec ’O’ P ”C” e ’C’ idi O I 9 C Add” eO n iTi 319 Motivation for movement and copying processes is provided through a specific interpretation of another module of the PPT, the EXTENDED PRO- JECTION PRINCIPLE (EPP). This principle will be applied to the syllable in order to replace traditional syllabic principles such as the MAXIMAL ONSET PRINCIPLE (MOP) or the ONSET FIRST PRINCIPLE (OFP). The EPP applied to syntax demands a SpecIP position, or, in other words “every sentence has a subject”. The EPP applied to the syllable demands a SpecP”, or, “every syllable has an onset”. The parallel is obvious: in both cases a Specifier position is required for the maximal projection. This leads to a redefinition of the EPP in a very general fashion: “maximal projections have (pre-modifying) specifiers”. 3. Some examples This clear analogy is also apparent in the effects of the EPP on both syntac- tic and syllabic units: In syntax the EPP attempts to provide an element for the subject position, usually by overt realization of an NP, or, depending on the clause or language type, covertly by pro or PRO. With regard to the syllable the EPP initially provides the domain O”; in (3) the head of the affix syllable is covertly filled by eO. Furthermore, the EPP also attempts to supply O” with an overt element, in (4) this task is performed by movement into SpecO”. More evidence for the effect of the EPP can be drawn from the examples in (5). (5) a. send#ing [sen$dI9] b. idea of [aIdI7$r7v] c. score [skO7] scoring [skO7$rI9] d. It worries me that … The EPP not only provides an overt segment in O” by movement (5a), but also, lexically and postlexically, by an intrusive r (5b) or a linking r (5c). It is quite important to note that these intrusive or linking segments are nothing but dummies that are exclusively motivated by the EPP. Interestingly, there is also an apparent analogy in syntactic analysis: in English the dummy it is introduced into subjectless clauses, as in (5d). In general the EPP is applied in order to explain movement between syl- lables. This application of a UG principle seems more plausible than refer- ence to an MOP which is merely a reflection of observations. Furthermore, any traditional treatment (MOP, OFP, Onset Licensing Principle, etc.) of the fact that onset positions seem to attract overtly realized segments is 320 Michael Völtz restricted to a phonological unit and should therefore be discarded by Occam’s razor. Finally, movement or copying do not seem to apply unrestrictedly; both processes appear to be restricted by morphological boundaries.7 The examples in (6) provide a sketch of a rather heterogeneous picture: inflec- tional boundaries in (6a) are transparent for movement, compounding in (6b) blocks movement,8 derivational boundaries in (6c) do show no uniform behaviour. (6) a. paint#ed [peIn$tId] weak#er [wi:$k7] b. field#apple [fi:ld$æpl], *[fi:l$dæpl] c. duck#ling [dAk$lI9] nation#al [næ$S7$nl] The discussion of a possible application of the syntactic EPP to the syl- lable is only one aspect of the more general quest for ways to reduce the amount of information contained in Universal Grammar (UG). 4. Conclusions This paper claims that this can be performed by the application of the PPT, i.e. it seems to be the case that some modules of the PPT (e.g. X’-Theory, Bounding Theory, Binding Theory) can successfully explain structural and procedural aspects of the syllable, and therefore allow UG to be relieved of redundancies: X’-structures seem to be relevant to both syntactic phrases and syllables, i.e. there is no need to assume separate structure-building algorithms for syntactic and syllabic structures. Furthermore, this paper identifies another case of redundancy with regard to the effects of the (syntactic) EPP and the (syllabic) MOP: both demand a maximal projection to have a specifier. A more general reading of the EPP allows both units to be explained by one principle. Finally, the general application of the PPT does not imply an obligatory application of all aspects of its modules, but this approach provides a basis for the PPT as a whole to be used in the analysis of linguistic units. 7 This paper does not consider phonological phenomena such as stress that might influence the range of movement. 8 Note that the power of the EPP is severely constrained in compounding processes, i.e. in (6b) the EPP is too weak to supply the domain O” in [æpl] with an overt segment moved out of C” of the preceding syllable. 321 References Bromberger, S. & M. Halle. (1989). Why Phonology is different. Linguistic inquiry 20, 51–70. Fudge, E. (1969). Syllables. Journal of linguistics 5, 253–286. Haegeman, L. (1994, 1991). Introduction to government and binding theory. 2nd ed. Oxford: Blackwell. Kaye, J. & J. Lowenstamm & J.-R. Vergnaud. (1990). Constituent structure and government in Phonology. Phonology 7, 193–231. Levin, J. (1985). A metrical theory of syllabicity. Dissertation, MIT. Selkirk, E. (1982). The syllable. The structure of phonological representations II, edited by H. van der Hulst & N. Smith, 337–383. Dordrecht: Foris. Treiman, R. (1989). The internal structure of the syllable. Linguistic structure in language processing, edited by G. Carlson & M. Tanenhaus, 27–52. Dordrecht: Kluwer. Völtz, M. (1996). The syntax of syllables: A principles and parameters approach to the structure of the syllable. Dissertation, Philipps-Universität Marburg. Webelhuth, G. (1995). Government and binding theory and the minimalist program. Oxford: Blackwell. Syllable count and Tzeltal segmental allomorphy Laura Walsh Dickey Abstract Tzeltal, a Mayan language spoken in southern Mexico, exhibits allo- morphy of an unusual type. The vowel quality of the perfective suffix is determined by the number of syllables in the stem to which it is attaching. This paper presents previously unpublished data of this allomorphy and demonstrates that a syllable-count analysis of the phenomenon is the proper one. This finding is put in a more general context of segment- The Hague: Thesus. 323–334. Thesus.The Hague: 323–334. . prosody interaction in allomorphy. 0. Syllables !? 1. Introduction Affixes usually have a constant shape. This default case can be seen in the English prefix /pVíj-/, meaning ‘before’, which does not vary in size or seg- Phonologica 1996: Phonologica 1996: ments. (1) pre-consonantal pre-vocalic pre-pausal Nothing about the phonological size or shape of the stem affects the form of the prefix; this morpheme has no allomorphs. In contrast, the English negating prefix /In-/ does arise in different forms depending on the shape of the stem. In the case in (2), the prefixal nasal assimilates in place of articulation to the stem-initial consonant. (2) im-perfect in-tolerable i9-convenient This variability in affix shape, known as allomorphy, occurs when the shape of a morpheme depends on the shape of what it attaches to. Most often in John R. Rennison & Klaus Kühnhammer, eds. (1999). eds. R. Rennison & Klaus Kühnhammer, John (1999). affixal allomorphy, as in (2), the segments of the affix change as a result of 324 Laura Walsh Dickey the segmental make-up of the stem. But segment-segment interaction is only one possibility. This paper examines the range of possibilities in the typology of allo- morphic relations between stems and affixes. In doing so, we will see that stem prosody can determine the segmental structure of an affix (CROSS DOMAIN ALLOMORPHY). In particular, I will demonstrate that syllables can be referred to in segmental allomorphy. For the purposes of this study, I divide phonological triggers and con- sequences into two primary categories: segmental and prosodic. “Segmen- tal” is defined as pertaining to vocalic and consonantal features. “Prosodic” refers to syllable count, to weight, or to size in general.1 Given this, there are four logical possibilities of allomorphic interaction: the stem segments can either affect affixal segments or prosody; and stem prosody can either affect affixal segments or prosody. This can be seen in the table in (3). (3) Possibilities of allomorphic interaction Stem Segments Stem PROSODY Affix Segments: Influence within a domain! influence across domains Affix PROSODY: Influence across domains influence within a domain The more darkly shaded boxes are the cases of influence within a domain: segments influence segments and prosody influences prosody. This WITHIN DOMAIN ALLOMORPHY is what is customarily discussed in the literature. What has not been examined in the literature, to the best of my knowledge, are cases where the allomorphic influence is across domains. I call this Cross Domain Allomorphy. It is precisely a case of this type that I will be most concerned with in this paper: the case of Tzeltal, a Mayan language spoken in Mexico, which shows an affixal vowel quality distinction dependent on the syllable count of the stem. 1 For cases of Cross Domain Allomorphy, I leave aside the well-known influence of stress on segments (Chomsky & Halle, 1968; Halle & Vergnaud, 1979; Selkirk, 1984; Kager, 1995, among many others). The syllable weight and count of an affix can indeed influence the stress pat- tern of the whole concatenated word. When word stress is determined, segmental changes can reflect the stress pattern of the word (such as with vowel reduction). It is not the stress of the affix per se which influences the segments of the stem (that would be Cross Domain Allomorphy), but rather, segmental changes are part of a more general property of stress on words and phrases. I do, however, consider affixal syllable count and weight in this study; stress is also analysed as a possible explanation for the Tzeltal allomorphy examined here. Syllable count and Tzeltal segmental allomorphy 325 2. Filling in the chart Within-domain segmental effects in allomorphy are uncontroversial. This type of allomorphy can be seen when the affected segments are adjacent (as in local assimilation) as well as non-adjacent (often in dissimilation). In the dissimilation case below, the segments of the stem determine the segments of the affix, though at a distance. The ethnonymic suffix in Georgian (Fallon, 1993) takes the basic shape of /-uri/ There is also an allomorph [uli], which arises when the stem contains the phoneme /r/. (4) Georgian (Fallon, 1993) Underlying Surface Gloss /dan-uri/ [danuri] ‘Danish’ /p’olon-uri/ [p’olonuri] ‘Polish’ /somX-uri/ [somXuri] ‘Armenian’ /asur-uri/ [asuruli] ‘Assyrian’ /ungr-uri/ [ungruli] ‘Hungarian’ /aprik’-uri/ [aprik’uli] ‘African’ /p’rusi-uri/ [p’rusiuli] ‘Prussian’ /tSerk’ez-uri/ [tSerk’ezuli] ‘Cherkessian’ In this case, whether the suffix contains an [1] or an [r] directly depends on the segmental shape of the stem. Cases of the size of the stem determining the size of the affix seem to be less common than pure segmental allomorphy, but this type of interaction has been described in the literature. Tagalog reduplication provides an illustration of prosody-prosody interaction in allomorphy. In Tagalog root reduplication (Carrier-Duncan, 1984; Aronoff et al., 1987; McCarthy & Prince, 1990), disyllabic roots reduplicate completely, but larger roots reduplicate only the first two syllables and lengthen the last vowel of the reduplicant. (5) Tagalog (Carrier-Duncan, 1984) Disyllabic root Reduplicated Form Gloss pantay pantay + pantay ‘level/quite level’ mag-li:nis mag-li:nis + li:nis ‘clean/clean a little’ mag-walis mag-walis + walis ‘sweep/sweep a little’ Larger root Reduplicated Form Gloss tahi:mik tahi: + tahi:mik ‘quiet/rather quiet’ baluktot balu: + baluktot ‘bent/variously bent’ kalansi9 kala: + kalansi9 ‘jingle of coins/id.’ 326 Laura Walsh Dickey In this case, the syllable count of the root determines whether the affix will be a complete copy or a disyllabic copy with vowel lengthening. This has the effect of a sort of size compensation. The smaller roots reduplicate totally and are therefore augmented maximally; the larger roots are kept to a reasonable size by limiting the amount of material which can be redupli- cated. We can tell that there is not a simple disyllabic reduplication for all roots because the vowel lengthening distinguishes the disyllabic roots from the longer ones. Tagalog indeed varies the size of the reduplicant based on the size of the base. With the cases of Georgian and Tagalog, we have seen the chart fill up halfway; they provide us with the Within Domain Allomorphy cases. This can be seen in (6), which is a revised version of the original table.2 (6) Within Domain Allomorphy cases Stem Segments Stem PROSODY Affix Segments: Georgian Affix PROSODY: Tagalog We turn next to the case study of this paper, Tzeltal, to explore the pos- sibility of Cross Domain Allomorphy. 3. Cross-domain allomorphy: The case of Tzeltal What has not been discussed in the literature, to the best of my knowledge, are cases of allomorphy being determined across domains. This paper provides a study of previously unpublished data from the fieldwork of Penelope Brown (Brown, 1996) which show a case of affixal segment identity being determined by stem syllable count. 2 One of the difficulties in the change in size cases is that when you change the size of a morpheme, the content also must shrink or expand. For example, in Dyirbal (Dixon, 1972), the ergative suffix changes its size depending on the syllable count, and not shape, of the stem. The ergative marker is /-9gu/ with disyllabic stems, but /-gu/ with longer ones. Root Ergative Gloss yara yara-9gu ‘man’ yamani yamani-gu ‘rainbow’ The syllable count of the stem determines the size of the ergative affix in this language, making it a good candidate for prosody-prosody interaction. But it is difficult to maintain that the suffix effect, of /-gu/ to /-9gu/ is purely a size effect, since there is also an addition of a nasal consonant. Syllable count and Tzeltal segmental allomorphy 327 Tzeltal (Slocum, 1948; Kaufman, 1971; Brown, 1996) is a Mayan language spoken in southern Mexico, mostly in the state of Chiapas. Tzeltal currently has about 150,000 speakers; the data reported here are from the community of Tenejapa, which has about 10,000 speakers. 3.1. The basics of Tzeltal phonology The phonemes of Tzeltal are given in (7) and (8), taken from Brown (1996).3 (7) The vowels of Tzeltal Front Central Back High i u Mid E o Low a (8) The consonants of Tzeltal Bilabial Apico-dental Postalveolar Palatal Velar Glottal Voiceless Stops p t 1 k ? Voiced Stops 4b4 Affricates Cö töS Glottalized Stops ’p’ ’t’ 1 ’k’ Glottalized ’Cö’ ’töS’ Affricates Fricatives s 1 S h Nasal Stops m n 1 1 Laterals l 1 Tap r Glides w j Tzeltal stress is always word-final, except for Spanish borrowings. The canonical root shape is CVC. 3.2. The pattern of Tzeltal allomorphy There is very little allophony or allomorphy in Tzeltal. However, there is one suffix which has two different allomorphs: the perfective suffix /-Vh/. 3 I have converted Kaufman’s transcription, as well as the practical Tzeltal orthography used by Brown (1996) and others, to the IPA as revised in 1993. I have used the dental symbol [ ]1 in the phoneme chart, but I have omitted this symbol from the transcriptions. 4 The bilabial stop is implosive when not word-initial. 328 Laura Walsh Dickey Transitive verbs can take this extremely productive suffix to form the perfec- tive. When the stem is monosyllabic, the suffix vowel is [o]. This can be seen in (9) below. All unsuffixed forms in the left column are monosyllabic (al- though bimorphemic because of the requirement of person marking). [ja] is the incompletive aspect particle. (9) Tzeltal transitive monosyllabic verbs Incompletive Perfective ja s-mah ‘he hits s.t.’ s-mah-oh ‘he has hit s.t.’ ja j-il ‘he sees s.t.’ j-il-oh ‘he has seen s.t.’ ja s-pas ‘he makes s.t.’ s-pas-oh ‘he has made s.t.’ ja s-nEt’ ‘he squashes s.t.’ s-nEt’-oh ‘he has squashed s.t.’ ja s-nuCö ‘he chases s.t.’ s-nuCö-oh ‘he has chased s.t.’ ja j-al ‘he tells s.t.’ j-al-oh ‘he has told s.t.’ ja s-Cöak ‘he takes s.t.’ s-Cöak-oh ‘he has taken s.t.’ ja s-jom ‘he gathers it’ s-jom-oh ‘he has gathered it’ Other suffixes can be attached to this perfective form; the perfective suffix remains [-oh]. (10) s-pas-oh ‘he has made something’ s-pas-oh-ik ‘they have made it’ s-pas-oh-bE ‘he has made something for someone’ s-pas-oh-iS ‘he has already made something’ When the stem itself has more than one syllable, the suffix vowel of the perfective is [E]. The examples in (11) are formed from disyllabic verb roots. (11) Disyllabic verb roots with perfective suffix Incompletive Perfective ja s-majlij ‘he waits for s.o.’ s-majlij-Eh ‘he has waited for s.o.’ ja s-mak’lin ‘he feeds s.o.’ s-mak’lin-Eh ‘he has fed s.o.’ ja s-tikun ‘he sends s.t.’ s-tikun-Eh ‘he has sent s.t.’ ja s-maklij ‘he listens to s.t.’ s-maklij-Eh ‘he has listened to s.t.’ The suffix remains /-Eh/ even when other suffixes come between it and the root, as long as the whole stem is more than one syllable. This is shown in (12). Syllable count and Tzeltal segmental allomorphy 329 (12) The suffix /-Eh/ after root and other suffixes Incompletive Perfective ja s-hol-intaj s-hol-intaj-Eh ‘he thinks about it’ ‘he has thought about it’ ja h-pak’-antaj h-pak’-antaj-Eh ‘I patch it’ ‘I have patched it’ ja s-kutöS-laj s-kutöS-laj-Eh ‘she carries it repeatedly’ ‘she was carrying it repeatedly’ cf. s-kutöS-oh ‘she has carried it’ The perfective suffix in Tzeltal is a VC suffix whose vowel quality depends on the syllable count of the stem to which it attaches. In order to maintain that this is the correct characterization, however, we must examine alterna- tive explanations. In the following section, I take up a number of hypotheses about Tzeltal allomorphy. We will see that the alternatives do not accurately characterize the pattern. 3.3. Non-explanations for this allomorphy 3.3.1. Not morpheme count It is possible that the allomorph is conditioned by the number of mor- phemes to which it attaches. But we can see that this is not so because both allomorphs [-Eh] and [-oh] can attach to forms with two morphemes (the hyphens mark morpheme boundaries). (13) j-il-oh s-tikun-Eh In example (13), we have a minimal pair of sorts; we see that both the /-oh/ and the [-Eh] allomorphs can be expressed in identical morpheme count conditions. 3.3.2. Not triggered by stem vowel A good candidate explanation for the vowel alternation is vowel harmony. The hypothesis is that the vowel of the suffix is determined by the vowel of the stem. This explanation is of the common segment-segment interaction type. However, in (14) I have laid out both allomorphs arising after all five vowels of the language. 330 Laura Walsh Dickey (14) Stem vowels and both allomorphs -Eh -oh a h-pak’-antaj-Eh s-mah-oh E s-sol-Es-Eh s-nEt’-oh i s-majlij-Eh j-il-oh o s-k’opon-Eh s-yom-oh u s-tikun-Eh s-nuCö-oh The vowels of the stem have no effect on the vowel quality of the suffix. 3.3.3. Not triggered by stem consonant Another segment-segment interaction would be the suffix vowel being determined by a stem consonant. Unfortunately, the morphology of the language does not allow a comparison of all consonants. Most verb roots are CVC and most consonants can end a root. For these cases, we will always get the /-oh/ variant no matter the stem-final consonant. But stems longer than one syllable are restricted in their final consonant simply be- cause there are only a few suffixes which can be added to the root before the perfective suffix. Most of these suffixes end in [j] or [s] But there are indeed minimal pairs: both [-Eh] and [-oh] can attach to both [j] and [s]. (15) Stem consonants and both allomorphs -Eh -oh j s-jaklij-Eh j-a?j-oh s s-sol-Es-Eh s-pas-oh-iS The examples in (15) show us that the consonant preceding the suffix does not determine the vowel of the suffix. 3.3.4. Not semantics Another conjecture about the cause of this vowel change could come from the realm of semantics. It is possible that these two forms are not really allomorphs at all, but are grammatically related suffixes with slight semantic variation. However, field workers report no semantic distinction made between these suffixes by Tzeltal speakers. The allomorphs do not have any distinct Syllable count and Tzeltal segmental allomorphy 331 semantic connotations. The suffix is also 100% productive, regardless of the meaning of the stem. We can see, additionally, that these allomorphs are not restricted by the semantics of the root because the same root can take both suffixes. (16) s-kutöS-oh s-kutöS-laj-Eh Moreover, semantics cannot account for the complementary distribution of the allomorphs. The alternation observed in the perfective suffix there- fore seems to be phonological, not semantic. 3.3.5. Not sound symbolism Another place to check for any possible semantic connotations of the allomorphs is in sound symbolism. There are some sound-meaning corre- spondences in Mayan languages. For example, in the Tzeltal deictic system, [i] is the vowel in proximal words, while [E] is the vowel in distal words. Consonantal sound symbolism is more controversial, but glottalized vs. plain consonants seem to have an antonymic function. (17) uCö’in ‘to annoy, pester’ uCöin ‘to love, take care of’ One problem with this explanation is that the vowels of the perfective /-Vh/ suffix ([E] and [o]) are not known to be a symbolic pair or to carry any distinct semantic weight in Mayan. The other difficulty is that sound sym- bolism necessarily brings a difference in meaning, which we just saw is not the case with the perfective suffix. There is no Mayan sound symbolism associated with this morpheme, according to all Tzeltal sources. 3.3.6. Not stress or footing Stress in Tzeltal is always word-final (excluding Spanish borrowings). Both allomorphs [-Eh] and [-oh] can be stressed, as in (18), and both can be un- stressed, as in (19). (18) j-uCö’in-E´h s-jom-o´h (19) h-no-h-p-tEs-Eh-íS s-pas-oh-bE´ Assuming that the footing in Tzeltal is iambic, the examples in (18) and (19) show that both allomorphs of the suffix can appear in the weak and strong 332 Laura Walsh Dickey branches of the iamb. The vowel quality does not seem dependent on the prosodic structure of the whole word. 3.3.7. Conclusion from the Tzeltal data Ruling out accounts based on vowel-vowel interaction, consonant-vowel interaction, semantics including sound symbolism, morpheme count, stress, and footing, the vowel alternation seen in the Tzeltal perfective suffix really does seem to be due to syllable count. The one confounding factor is phoneme count; the /-oh/ affixes onto words of three or four phonemes, while /-Eh/ affixes onto words with five or more phonemes (5, 6, 7, 8, or 9). This circumstance arises because of both the phonotactics and the mor- phology of the language. This suffix only attaches to verbs. Almost all verbs have a CVC shape. The same phoneme count cannot have two separate syllable counts. All of the verb stems with four phonemes, for example, are of the shape CCVC (a consonantal person marker prefix plus a CVC root). A disyllabic stem with four phonemes, such as CV.CV simply does not exist as a verb stem in Tzeltal. The syllable count analysis and the phoneme count analysis cannot be distinguished from each other (certainly not with the available data, but I believe they are actually indistinguishable in principle). The syllable count analysis does have the usual linguistic characteristic of a “one vs. many” effect. Counting phonemes presents the unusual thorn of how to formally characterize “4 or less vs. 5 or more”. Under either analysis, the segmental identity of the perfective suffix in Tzeltal is determined by the size of the stem. This means that Tzeltal pro- vides us with a case of Cross Domain Allomorphy. (20) Tzeltal’s place in allomorphic possibilities Stem Segments Stem PROSODY Affix Segments: Georgian Tzeltal Affix PROSODY: Tagalog 4. Other cross-domain candidates The English adjectival comparative suffix can also be viewed as a case of the segments of the affix being determined by the prosody of the stem. With monosyllabic stems or disyllabic stems with light final syllables, the form is /-7V/. Syllable count and Tzeltal segmental allomorphy 333 (21) hip hipper swarthy swarthier hot hotter cranky crankier black blacker dreamy dreamier blue bluer dainty daintier With larger stems and stems of a different syllable shape, the allomorph is ‘more’ + stem. (22) blasé more blasé *blaséer boring more boring *boringer hypocritical more hypocritical *hypocriticaler anxious more anxious *anxiouser talented more talented *talenteder Thus, the shape of the English comparative morpheme depends on the syllable count and moraic structure of stem. This is akin to the Tzeltal case. However, I know of no candidates for the other cell in the chart, in which stem segments determine affix prosody. It is possible that this direction of interaction is rare (if not non-existent) not in principle, but simply because affixes rarely have distinctive prosody. 5. Implications and conclusion Tzeltal -oh/-Eh allomorphy of the perfective suffix provides a new view on the possible types of allomorphic interactions. The data show that there can indeed be segment-prosody interaction in allomorphy (Cross Domain Allomorphy). The vowel of the perfective suffix in this language is entirely determined by the number of syllables in the stem to which it is attaching. The fact that Tzeltal is sensitive to one vs. many syllables also supports the reality of the syllable in morpho-phonological processes. Acknowledgements I would like to thank the following people for their help with this work: Mike Dickey, David Odden, Doug Pulleyblank, John McCarthy, the audience at the 1996 Phonology Meeting in Vienna, an anonymous reviewer, and especially Penny Brown. All errors are my own. 334 Laura Walsh Dickey References Aronoff, M., A. Arsyad, H. Basri. & E. Broselow (1987) Tier configuration in Malassarese redupli- cation. CLS 23 Part Two. Parasession on Autosegmental and Metrical Phonology, edited by A. Bosch, B. Need, & E. Schiller. Chicago Linguistic Society. Brown, P. (1996) Tzeltal field notes March 1996: Tenejapa, Mexico. Ms., Max Planck Institute. Carrier-Duncan, 1. (1984) Some problems with accounts of reduplication. Language sound structure, edited by M. Aronoff & R. Oehrle. Cambridge: MIT Press. Chomsky, N. & M. Halle (1968) The sound pattern of English. New York: Harper & Row. Halle, M. & J-R. Vergnaud (1979) Metrical structures in phonology. Ms., MIT.. Kager, R. (1995) The metrical theory of word stress. The handbook of phonological theory, edited by J. Goldsmith. Cambridge: Blackwell. Kaufman, T. (1971) Tzeltal phonology and morphology. Berkeley: University of California Press. McCarthy, J. & A. Prince (1990) Prosodic morphology. NLLT 8:2, 209–283. Selkirk, E. (1984) Phonology and syntax: the relation between sound and structure. Cambridge: MIT Press. Slocum, M. (1948) Tzeltal (Mayan) noun and verb morphology. IJAL 14 , 77–86. Wolf, J. (1966) Beginning Cebuano: Part 1. New Haven: Yale University Press. Inter-nuclear relations in Arabic Shohei Yoshida Abstract In Generative Phonological literature, the recognition of the ternary contrast of syllable types in Arabic (i.e. light, heavy and superheavy) plays a crucial role in analysing such phenomena as syncope, closed syllable shortening and stress assignment. In contrast, Arabic is viewed as a language containing open syllables only in Government Phonology (GP). Apparent consonant clusters and final consonants are analysed as two onsets separated by a phonetically unrealized nucleus and onsets fol- lowed by a phonetically unrealized final empty nucleus, respectively. In The Hague: Thesus. 335–354. Thesus.The Hague: 335–354. . GP, it is assumed that internal empty nuclei are unrealized when properly governed and final empty nuclei are unrealized if parametrically licensed. However, it is yet to be investigated whether proper governors and their Syllables !? governees form constituents and whether such constituents are further constituentized on some higher level of prosodic licensing hierarchy. This paper argues that inter-nuclear relations, which constitute a subset of prosodic licensing (Harris, 1994), define the sites for Proper Government (syncope), closed syllable shortening, and stressed nuclei. In so doing, it Phonologica 1996: Phonologica 1996: is possible to account for the aforementioned phenomena with no refer- ence to syllable weight. 0. 1. Introduction In Government Phonology (GP), vowel-zero alternations do not involve the insertion or deletion of nuclear positions. Instead, it is assumed that the relevant nuclear positions are always present, but may remain phonetically uninterpreted under certain conditions. Specifically, a nuclear position is unrealized if it is properly governed. (Charette, 1991; Kaye, 1989; 1992; 1995). In Kaye (1989), Guerssel & Lowenstamm (1988) and Yoshida (1993), vowel- zero alternations in modern Arabic dialects are claimed to be manifesta- John R. Rennison & Klaus Kühnhammer, eds. (1999). eds. R. Rennison & Klaus Kühnhammer, John (1999). tions of right-headed PROPER GOVERNMENT (PG) holding between two nuclear 336 Shohei Yoshida heads. However, this directionality contradicts the left-headed foot in Arabic, forcing one to recognize two different types of inter-nuclear rela- tions in Arabic, i.e. PG and metrical relations. Pursuing such an analysis further requires an understanding of the nature of two different relations and their interaction. In regard to this, Yoshida (1993) notes the case of an Arabic dialect where one nucleus unexpectedly fails to properly govern a preceding nucleus if the preceding nucleus is stressed even though all the other conditions are met. Does this mean that the metrical structure is erected prior to the application of PG? This does not appear to be the case since the metrical constituents cannot be built without referring to the melodic quantity of nuclear positions, i.e. whether or not they are empty. In what follows, I shall take up Cairene Arabic, as a basis for discussion, and put forward the view that it is possible to unify PG and the metrical structure under the general notion of PROSODIC LICENSING (Harris, 1994). I shall show that the relation between the strong position and the weak position within a metrical constituent on the one hand, and the relation between a proper governor and its governee on the other are in fact identi- cal, requiring only a single prosodic structure. 2. Generative analysis Let us recall how three classical phonological problems of Cairene Arabic have been perceived in traditional Generative Phonology. It should be noted that the recognition of the three types of syllable weight is crucial to the understanding of the problems in this framework. First, (1) shows syncope of high vowels. (1) a. nízil ‘he descended’ šírib ‘he drank’ b. nízl-u ‘they descended’ šírb-u ‘he drank it’ The standard Generative Phonological (Kenstowicz, 1980; 1981) solution to this problem is to specify a syllabic context in which syncope takes place by stating that an unstressed high vowel is syncopated in an open syllable as in (1b). An additional condition must be proposed. For a high vowel to be syncopated, it must not be at the left edge of a word, as shown in (2). (2) a. nizíl-na ‘we descended’ kitaab ‘a book’ b. ma-nzil-náa-š ‘we didn’t descend’ wi ktáab ‘and a book’ Inter-nuclear relations in Arabic 337 However, the fundamental question of why a high vowel should be synco- pated if it is unaccented, non-initial and in an open syllable remains un- answered in such a purely descriptive account. Second, vowel shortening in (3) is viewed as ‘closed syllable’ shortening. (3) a. šúuf ‘look!’ fanagíin ‘cups’ b. šúf-na ‘look at us!’ fanagín-ha ‘her cups’ The standard explanation such as Kenstowicz (1994) is to claim that the rhyme is maximally bimoraic. A circular nature of this analysis should be noted. The pre-consonantal string šúf- in šúf-na is analysed as a closed syllable containing two moras because it is permitted by the bimoraic constraint, and the pre-consonantal *šúuf- is analysed as trimoraic because it is not permitted in this position, given that extrasyllabicity applies only to word-final segments. The conclusion that a syllable permits up to two moras only is drawn from the analysis that the permitted pre-consonantal šúf- is bimoraic and the impermissible *šúuf- is trimoraic. Third, in previous generative work (e.g. Welden, 1980; Kenstowicz, 1980; Halle & Vergnaud, 1987) stress assignment is analysed by referring to the different types of syllable weight. (4) a. katábt 1SG.PAST √ktb ‘to write’ katbíin /kaatibiin/ PARTICIPLE PL. ma katabnáaš 1PL.PAST NEG. b. katábti 2SG.FEM.PAST tíktib 2SG.MASC.NONPAST káatib PARTICIPLE MASC.SG. c. kátab 3SG.MASC.PAST kátabit 3SG.FEM.PAST katabítu(h) 3SG.FEM.PAST + ‘it’ katabítak 3SG.FEM.PAST + ‘you (SG.MASC.)’ tiktíbu 2PL.NONPAST It is claimed that stress is assigned to the final syllable if it is superheavy (or heavy if extrasyllabicity applies) as in 4a), and to the penult if it is heavy as in 4b). According to Halle & Vergnaud, the patterns in 4c) can be accounted for as follows: Otherwise, stress is on the antepenult or penult whichever is sepa- rated by an even number of syllables from the immediately preceding heavy syllable if there is one, or from the beginning of the word if there is none, where zero is counted as even. (Halle & Vergnaud, 1987: 61) 338 Shohei Yoshida This point is more clearly illustrated by the Cairene accentuation of longer Classical Arabic words in (5). (5) mukaatíbah ‘reporter (FEM.PAUSAL)’ mukaatíbatun ‘reporter (FEM.NONPAUSAL)’ mukaatibatúhum ‘their (PL.) reporter (FEM.)’ mukaatibatúhumaa ‘their (DUAL) reporter (FEM.)’ báqarah ‘a cow (PAUSAL)’ baqarátun ‘a cow (NONPAUSAL)’ baqarátuhum ‘their (PL.) cow’ baqaratuhúmaa ‘their (DUAL) cow’ 3. The syllable structure of Arabic in GP In marked contrast with the Generative Phonological treatment of the syllable structure of Arabic, Guerssel & Lowenstamm (1988), Yoshida (1993) and Lowenstamm (1996) treat Arabic as an O[nset]-N[ucleus] language lacking codas, as in (6). (6) a. O N O N O N b. O N O N O N x x x x x x x x x x x x d a r a s a m a r a [darasa] ‘he wrote’ [marra] ‘he passed’ c. O N O N O N d. O N O N x x x x x x x x x x q a l a q u l [qaala] ‘he said’ [qul] ‘Say! (MASC.SG.)’ In this analysis the Classical Arabic verbs darasa, marra and qaala share the identical canonical form ONONON, requiring only one syllable template for all types of Binyan I verbs regardless of whether they are triliteral, biliteral or hollow.1 Example (6d) shows the treatment of a ‘final’ consonant in GP, where it is not a coda, as in illicit (7), but an onset followed by an empty nucleus. 1 Hollow verbs are verbs whose underlying consonantal roots contain a medial /y/ or /w/. In the phonetic form, the medial glide disappears and a preceding vowel is lengthened. Inter-nuclear relations in Arabic 339 (7) *O* R N x x x q U l The representation of a final consonant in (6d) is derived from the Coda Licensing Principle (Kaye, 1990: 311). (8) The Coda Licensing Principle A post-nuclear rhymal position must be licensed by a following onset. According to this principle, if a position were truly a coda then we would expect the presence of a licensing onset (and therefore the presence of a following syllable). A significant implication of this principle is that a branching rhyme can occur only word-internally and words in all languages end with a nucleus I shall now give two pieces of evidence in support of the treatment of sur- face biconsonantal clusters as two onsets rather than as coda-onset se- quences. First, there is a theory-internal argument in GP against the view that Arabic has coda-onset sequences. The Coda Licensing Principle in (8) ensures that the presence of a coda is possible if and only if it is licensed by a following onset which governs it from right to left. It follows that if two phonological positions such as those containing /k/ and /t/ in yaktub ‘he writes’ truly form a coda-onset sequence, then severe phonotactics must be observed in these positions. In a true coda-onset sequence, the coda is occupied by a segment that can be governed (i.e. nasals, liquids, a small number of obstruents) and the onset by a segment that can govern (i.e. obstruents). But absolutely no phonotactics are observed in the putative coda-onset sequences in Arabic: if the cluster C1C2 is found, then its mirror image C2C1 also exists. Thus the representation of yaktub appears as in (9a). Structure (9b) is unacceptable in GP, as the Coda Licensing Principle is doubly violated, i.e. there is no attested governing relation between the first coda and the following onset, and there is no onset that licenses the final coda. (9) a. b. *O* R O R O N O N O N O N N N x x x x x x x x x x x x x x I A k t U b I A k t U b 340 Shohei Yoshida Second, Berg & Abd-El-Jawad (1992) note that onsets and apparent codas in Arabic are freely replaceable in speech errors, unlike in Germanic lan- guages, where consonants almost always replace consonants in like syllabic positions. This is expected in GP since the two members of a cluster in Arabic occupy two onsets as in (9a) and are therefore of like syllabic affilia- tion. Let us turn to the treatment of long vowels as binuclear sequences as in (6c). Supporting evidence in Classical Arabic is that the second half of a long vowel involves re-initiation of pulmonic pressure before a geminated con- sonant, converting a long vowel into a sequence of two short vowels, as in (10). (10) Daalla [Dáalla ~ Da:la] *[Da:lla] ‘straying’ šaadda [šáadda ~ ša:da] *[ša:dda] ‘he spoke violently’ (where the uppercase D is a pharyngealized/velarized voiced alveolar stop, and VV and V: indicate two vowels vs. a single long vowel, respectively) In fact, the re-initiation of the second half of a long vowel may even be accompanied by stress (e.g. [Dáálla]). A speaker’s attempt to avoid the rearticulation results in degemination of the following consonant (see Yoshida, 1996: 59–60 for detailed discussion).2 With this said, however, long vowels in Cairene Arabic are more mono- lithic in the sense that the second half of a long vowel is never rearticulated or stressed. Where Classical Arabic has the orthographic pattern -VVCjCjV Cairene Arabic has -V:CjV- only. This suggests that the two positions of a long vowel are in a steady governor-governee relation. To take account of the difference between the separability of long vowels in Classical Arabic and their inseparability in Cairene Arabic, I shall deviate from the treatment of long vowels as two nuclei, and represent long vowels in Cairene Arabic as branching nuclei which always contract a left-headed governing relation. 4. Phonological licensing and proper government It is a basic assumption of GP that licensing operates in phonology as well as in syntax. Kaye defines phonological licensing as follows. (11) The Licensing Principle (Kaye, 1990: 306) All phonological positions save one must be licensed within a domain. The unlicensed position is the head of this domain. 2 See Yoshida (1996: 59–60) for a detailed discussion. Inter-nuclear relations in Arabic 341 The basic insight of this principle is that the occurrence of a phonological position within a domain presupposes and depends on the presence of another position. The head of such a domain is unlicensed within the domain but licensed on some higher level of the prosodic hierarchy. Below the level of the onset-rhyme sequence, intra-constituent licensing relations hold between the two positions of each of the three maximally binary branching constituents, i.e. the Onset, the Nucleus and the Rhyme. Inter- constituent licensing relations hold between an onset and a following rhyme, and between a post-nuclear rhymal position (coda) and a following onset. Beyond the level of the onset-rhyme sequence, licensing holds be- tween two nuclear heads at nuclear projection. The directionality of intra-constituent licensing and that of inter- constituent licensing are not subject to parametric variation (Kaye, 1990; Harris, 1994). The former is universally left-headed whereas the latter is universally right-headed. The directionality of inter-nuclear licensing is parameterized. PG is one case of inter-nuclear licensing relations, and all the previous work on Arabic within the framework of GP agrees that PG in the language is right-headed. The recognition of empty nuclei in phonology raises the question of what principles regulate their distribution. In GP the Phonological EMPTY CATE- GORY PRINCIPLE (ECP) controls the distribution of empty nuclei. The following relevant definitions are taken from Kaye (1995: 295). (12) The Phonological ECP A p-licensed (empty) category receives no phonetic interpretation. P-licensing: 1. Domain-final (empty) categories are p-licensed. (parameterized) 2. Properly governed (empty) nuclei are p-licensed. 3. A nucleus within an inter-onset domain [is p-licensed]. Proper government: α properly governs β if: 1. α and β are adjacent on the relevant projection, 2. α is not itself licensed, and 3. no governing domain separates α from β. It is claimed that a phonetically empty nucleus is licensed when it is properly governed by a following nucleus which is not itself licensed. Consider the vowel-zero alternation in Moroccan Arabic taken from Kaye (1989). (13) a. tan kti-b ‘I am writing’ b. tan ki-tb-u ‘we are writing’ 342 Shohei Yoshida Kaye argues that the underlying representations of the verbs are 5. Problems with the previous GP analysis Let us apply right-headed PG to syncope in Cairene Arabic. The data (1) and (2) are reproduced below as (14). Syllable boundaries are indicated by periods, and licensed empty nuclei are indicated by 0’s. (14) a. ní.zi.l0. ‘he descended’ ší.ri.b0. ‘he drank’ b. ní.z0.lu. ‘they descended’ ší.r0.bu. ‘he drank it’ c. ni.zí.l0.na. ‘we descended’ ki.taa.b0. ‘a book’ d. ma. n0.zi.l0.náa.š0. ‘we didn’t descend’ wi. k0.táa.b0. ‘and a book’ It seems that the proposed parametric setting of final empty nuclei and right-headed PG can account for the vowel-zero alternations in the suffixed forms in (14b). However, despite its appeal, there are at least five reasons for seeking alternatives to right-headed PG. Inter-nuclear relations in Arabic 343 The first problem is that the alternating high vowels in Cairene Arabic are underlying, unlike in Moroccan Arabic, where, according to Kaye (1989), all non-branching nuclei are lexically empty (surfacing as [i-] if unlicensed) and underlying vowels can only occupy branching nuclei. There are two reasons for concluding that the high vowels in Cairene Arabic are not manifestations of empty nuclei. The first reason is that both [i] and [u] alternate with zero as in (15). (15) a. kútub ‘books’ yáaxud ‘he takes’ b. kútb-u(h) ‘his books’ yáxd-u(h) ‘he takes it’ The second reason is that if the alternating vowels in (14) and (15) were underlyingly empty nuclei, one could not explain why penultimate empty nuclei are phonetically realized in (14a) and (15a), but not in (16). (16) nahr The problem is that the ECP in (12) is supposed to apply to underlyingly empty nuclei only, giving rise to a need to modify the ECP by stating that it applies to nuclei occupied by high vowels as well as to empty nuclei. The second problem is that the GP account on the basis of right-headed PG cannot explain why word-initial nuclei cannot be properly governed at the left edge of a word, as illustrated by a comparison between (14c) and (14d). The third problem is that the forms in (16) themselves are unanalysable in terms of right-headed PG. These forms contain two successive underlying empty nuclei. The final empty nuclei cannot properly govern the penulti- mate nuclei since the final nuclei are parametrically licensed. Thus in a language in which PG is right-headed, final onset clusters are expected to be disallowed. The penultimate empty nuclei in forms like [nahr] in (17) cannot be properly governed. (17) n a h 0 r 0 ← the final empty nucleus is licensed ↑ cannot be licensed by the licensed final nucleus We need to understand how the penultimate nucleus is licensed. Final consonant clusters also occur as a result of attaching the suffix /t/ ‘I/you (MASC.SG.)’ to verbs, e.g. katab+t ‘I/you (MASC.SG.) wrote’, fihim+t ‘I/you (MASC.SG.) understood. It is possible to treat the two successive empty nuclei in these forms as being parametrically licensed because they are both final, 344 Shohei Yoshida i.e. [[ka.ta.b0.]t0.], [[fi.hi.m0.]t0.]. However, calling for such a morphological solution for the nouns in (16) is untenable, since they are monomorphemic and have no internal structure. The fourth problem is the previous treatment of so-called closed syllable shortening in GP — a theory that does not recognize closed syllables in Arabic. The relevant data from (3) is reproduced here as (18). (18) a. šúuf ‘look!’ fanagíin ‘cups’ b. šúf-na ‘look at us!’ fanagín-ha ‘her cups’ It should be noted that in GP terms the shortening occurs before a word- internal empty nucleus and not before a final empty nucleus, e.g. <šúu.f0>, <šú.f0.na.>. Yoshida (1992; 1993, also quoted in Kaye, 1995: 299–300) sug- gested that a long vowel is shortened before an internal empty nucleus because the head of a long vowel is not government-licensed. That is, a governing relation holds between the two positions of a long vowel, and its head is licensed to govern by a following nucleus which is not licensed, as shown below. (19) a. <<<<<<<<<<<<< ← government licensing >>> ← government š u u f u h 0 [šuufu(h)] ‘see him!’ 1 2 3 4 ← nuclear positions (N) b. <<<< ← government š u u f 0 n a [šufna] ‘see us!’ 1 2 3 4 nuclear positions (N) This analysis is based on the assumption that a governing relation holding between the head and the complement of a long vowel is licensed by a following nucleus which is not properly governed. In (19a) the underlying long vowel is not shortened since its head (N1) is government-licensed to govern its complement (N2) by the following nuclear position (N3) containing the lexical melody [u]. On the other hand, the initial vowel in (19b) cannot be long because the head (N1) cannot be government-licensed by N3, which is properly governed by N4. This analysis appears to be untenable because it requires an arbitrary stipulation that parametrically p-licensed final empty nuclei are potential government-licensors but those empty nuclei p-licensed by PG are not. As the fifth problem, it remains to be answered why stressed nuclei can- not be properly governed. It is possible to invoke an arbitrary ordering and Inter-nuclear relations in Arabic 345 state that PG applies after the metrical structure is erected, by recognizing two different prosodic licensing relations, the PG relation and the metrical relation. Needless to say, such an analysis complicates the grammar of Arabic. In fact, a solution like this ignores other aspects of phonological structure. However, like PG, the metrical structure is also thought to be a manifestation of inter-nuclear licensing. It follows that a study of PG alone without considering stress assignment or vice versa is local and inadequate. 6. Analysis The five problems sketched above can be solved by adopting a holistic system of inter-nuclear p[rosodic]-licensing. I define inter-nuclear p-licensing as a binary asymmetrical relation holding between two nuclear positions at some level of the prosodic hierarchy. In the following analysis, inter-nuclear p-licensing constituents are indicated by grids for typographical conven- ience. Binary constituents and degenerate constituents are represented as in (20), where asterisks are nuclear positions. (20) a. (** left-headed binary constituent b. **) right-headed binary constituent c. (* or *) degenerate constituent The principles operative in inter-nuclear p-licensing in Cairene Arabic are as follows. (21) Inter-nuclear p-licensing in Cairene Arabic (the low variety of the di- glossia) a. Two successive nuclear positions A and B are constituentized at Level 0 nuclear projection if: i. B is an internal empty nucleus, where A licenses B, overriding any other licensing duty of A, or ii. B is a parametrically licensed final empty nucleus unless A is the head of a long vowel, or iii. A and B are nuclear heads, B being a non-branching nucleus, or iv. A is the head and B its complement in a long vowel (unless an internal empty nucleus follows, as per (a.i). b. Licensing constituents at Level 0 nuclear projection are left-headed and constructed from left to ride. c. Licensing constituents at Level 1 nuclear projection are left-headed and constructed from right to left. 346 Shohei Yoshida d. Licensing constituents at level 3 nuclear projection are right-headed and constructed from right to left. I assume that the principles in (21a) are parametrically ordered and their different orderings are responsible for dialectal variation. Before solving the problems, the lexical representations of some mor- phemes in Cairene Arabic should be clarified, as it is important to have correct inputs for the following analysis. I believe that the syllable structures of these morphemes have not been considered with sufficient care in previous work. First, let us consider the final vowels of the representative words in (22). (22) fihím-na ‘we understood’ fihím-ha ‘he understood her’ fíhm-u ‘they understood’ tifhám-i ‘you (SING, FEM.) understand’ wára ‘behind’ nísi ‘he forgot’ These final vowels are usually analysed as short. However, they are length- ened when stressed as a result of suffixation. (23) ma fhim-náa-š ‘we didn’t understand’ fihim-náa-k ‘we understood you (MASC.SG.) ma fhim-háa-š ‘he understood her’ ma fhim-úu-š ‘they understood’ fihim-úu-ha ‘they understood her’ tifham-íi-h ‘you (FEM.SG.) understand him’ tifham-íi-ni ‘you (FEM.SG.) understand me’ waráa-ya ‘behind me’ ma nsíi-š ‘he didn’t forget’ Therefore, underlyingly, these vowels should be branching nuclei. This analysis is supported by the fact that their counterparts in Classical Arabic are always long. I propose that melodic material in the head of a branching nucleus spreads to its complement only if the head is stressed. Second, the representation of the feminine nominal suffix deserves spe- cial attention. This suffix involves an intriguing three-way alternation, [a(h)], [t] and [it]. (24) –suur-a(h) ‘picture’ – sur-t-i ‘my picture’ – sur-it-na ‘our picture’ (The [s–] is a pharyngealized/velarized voiceless alveolar fricative) In Cairene Arabic, the suffix appears as [a], optionally followed by [h], if no further suffixes are attached. Earlier studies of stress assignment treat this Inter-nuclear relations in Arabic 347 suffix as /a/ alone. It should be pointed out that the correct Classical pro- nunciation of this suffix before a pause, by the norm of Quranic recitation, requires the final [h]. As Classical Arabic is the high variety in the diglossic situation of the Cairene Arabic community, the users of Cairene Arabic are familiar with Classical Arabic in varying degrees. Therefore, I assume that the feminine suffix contains an onset position to which [h]/[t] is attached along with a following empty nucleus. I shall provide an analysis of this suffix in the next section. For our present purposes, suffice it to say that the feminine suffix has the following disyllabic structure with floating melodic material. (25) O N O N x x x A t Let us now apply the principles in (21) to the five problems, i.e. syncope, the immunity of the initial syllable to syncope, final consonantal clusters, vowel shortening and stress assignment. First, syncope of high vowels is accounted for by the following simple statement. (26) Proper government P-licensed nuclei at Level 0 nuclear projection cannot autosegmentally license (a-license) the elements I and U.3 In other words, a p-licensed nucleus cannot license the attachment of a high vowel. The representation of [nizil] is given in (27). Constituents at Level 0 nuclear projection alone are shown here as constituents at higher levels of nuclear projection are irrelevant to the present discussion. (27) a. b. (*) (*) * O N O N O N O N O N O N x x x x x x x x x x x x n I z I l n I z I l [nizil] Inter-nuclear p-licensing constituents are then constructed in conformity with the principles in (21). Principle (21a.ii) ensures that the parametrically licensed final empty nucleus and preceding nuclear head are assembled into a constituent. Principle (21b) then applies to erect left-headed Level 0 3 In GP, the high vowels [i] and [u] are represented as the elements I and U. 348 Shohei Yoshida licensing constituents. But since only the initial nucleus is available, it forms a degenerate constituent as in (27b). The initial and second high front vowels are not syncopated as the nuclei containing them are not licensees, and so PG as defined in (26) does not apply. Next, let us consider the representa- tion of [nizlu] in (28). (28) a. b. (*) * (*) * O N O N O N O N O N O N x x x x x x x x x x x x x x ╧ n I z I l U n I z I l U [nizlu] The head of the final long vowel and its complement form a constituent as per (21a.iv). The initial and second nuclei are assembled into a left-headed constituent as per (21a.iii). PG applies to a nucleus which is licensed at Level 0 nuclear projection, as defined in (26). As seen in (28b), the attachment of the second high vowel to its position is not a-licensed as the second nucleus is licensed by the initial nucleus. It should be clear that syncope does not apply to the initial nuclei in (27b) and (28b) because these positions are never licensed at Level 0 nuclear projection. Second, let us examine why a word-initial high vowel can be syncopated when it is demoted to a second vowel. Consider the representation of [kitaab] and [wi ktaab] in (29). (29) a. (*) (*) * (*) b. (*) * (*) * (*) O N O N O N O N O N O N O N x x x x x x x x x x x x x x x x ╧ k I t A>>> b w I k I t A>>> b [kitaab] [wi ktaab] The parametrically licensed final empty nucleus is not incorporated into a licensing constituent since the preceding nuclear head, which is the head of a long vowel, must license its complement as per (21a.iv). In (29a) only the initial nucleus is left, forming a degenerate constituent by itself. In (29b), on the other hand, two nuclei are available and so they form a left-headed licensing constituent. Consequently, the second nucleus in (29b) loses its ability to a-license a high vowel, resulting in syncope. Third, the existence of final consonantal clusters in Cairene Arabic is easily accounted for in the present analysis. Inter-nuclear relations in Arabic 349 (30) (*) * (*) O N O N O N x x x x x x n A h r [nahr] The internal empty nucleus must be licensed by the initial nucleus as per (21a,i). The final empty nucleus need not be licensed because final empty nuclei are parametrically licensed in Arabic Fourth, shortening of long vowels before a word-internal empty nucleus is accounted for by (21a,i) and (21a.iii). The representations of [šuuf] and [šufna] are given below. (31) a. (*) * (*) b. (*) * (*) * O N O N O N O N O N x x x x x x x x x x x x x š U>>> f š U>>> f n A [šuuf] [šufna] In (31a) the head of the long vowel licenses its complement because the parametrically licensed final empty nucleus does not necessarily have to be incorporated into a constituent, producing the form with a long vowel. On the other hand, the internal empty nucleus in (31b) must be licensed by the head of the long vowel. Licensing the following empty nucleus, the head cannot license its complement at the same time, generating the form with a short vowel. Finally, stress assignment is handled by constructing licensing constitu- ents at higher levels of nuclear projection. Following Kaye’s Licensing Prin- ciple given in (11), I propose (32). (32) The stressed nucleus in a word is the head of this domain. This means that the stressed nucleus is the head of a constituent at the highest level of nuclear projection in a given phonological domain. The complete inter-nuclear p-licensing structures of the words in 4) are as fol- lows. (33) a. * (*) b. * (*) (*) (*) * (*) (*) * (*) (*) * (*) (*) * (*) * (*) k a t a b 0 t 0 k a a t i b i i n 0 [katábt] [katbíin] 350 Shohei Yoshida c. * (*) d. * (*) (*) (*) * (*) (*) (*) * (*) * (*) * (*) * (*) (*) (*) * (*) * m a k a t a b 0 n a a š 0 k a t a b 0 t i i [ma katabnáaš] [katábti] e. f. (*) * (*) * (*) * (*) * (*) * (*) * t i k 0 t i b 0 k a a t i b 0 [tíktib] [káatib] g. h. (*) * (*) * (*) (*) * (*) * (*) * k a t a b 0 k a t a b i t 0 [kátab] [kátabit] i. * (*) j. * (*) (*) (*) (*) (*) (*) * (*) * (*) (*) * (*) * (*) (*) * k a t a b i t u h 0 k a t a b i t a k 0 [katabítuh] [katabítak] k. * (*) l. * (*) (*) (*) * (*) * (*) * (*) * (*) (*) * (*) (*) * (*) (*) * t i k 0 t i b u u m u k a a t i b a h 0 [tiktíbu] [mukaatíbah] m. * (*) n. * (*) (*) * (*) * (*) (*) * (*) * (*) (*) * (*) * (*) * (*) (*) * (*) * (*) (*) * m u k a a t i b a t u n 0 m u k a a t i b a t u h u m 0 [mukaatíbatun] [mukaatibatúhum] o. * (*) p. (*) (*) * (*) * (*) * (*) (*) * (*) * (*) * (*) * (*) * (*) * m u k a a t i b a t u h u m a a b a q a r a h 0 [mukaatibatúhumaa] [báqarah] q. * (*) r. (*) * (*) (*) * (*) (*) * (*) * (*) (*) * (*) * (*) * (*) * b a g a r a t u n 0 b a q a r a t u h u m 0 [baqarátun] [baqarátuhum] s. * (*) (*) * (*) * (*) * (*) * (*) (*) * b a q a r a t u h u m a a [baqaratuhúmaa] It should be noted that the forms (33l-s) represent the pronunciation of the high variety of the and cannot be generated if (21a.iii) applies before (21a.iv). Such an ordering results in the pronunciation of the words in the low Inter-nuclear relations in Arabic 351 variety. These forms without syncope are generated by applying (21a.iv) before (21a.iii). 7. The feminine suffix In this section, I show that the present analysis can provide us with an explanatory account of another phonological phenomenon in Cairene Arabic. Let us recall the intriguing three-way contrast of the feminine nomi- nal suffix. The data in (24) is reproduced below as (34). (34) –suur-a(h) ‘picture’ – sur-t-i ‘my picture’ – sur-it-na ‘our picture’ The underlying representation of [s–úurah] is (35a), and (35b) shows its inter- nuclear p-licensing structure. (35) a. b. (*) * (*) * (*) * O N O N O N O N O N O N x x x x x x x x x x x x x x –s U r A t – s U>>> r A t [s–úurah] The final empty nucleus is parametrically licensed. The floating segment A then is linked to its position so as to allow this position to the head of a constituent. The suffixal consonant /t/ remains floating and its position is phonetically interpreted as a glottal fricative. (36) (*) * (*) * (*) * O N O N O N x x x x x x x x –s U r A t I [s–úrti] Next, let us consider the representation of [s–úrti], given in (36). In this form the feminine suffix is followed by a personal pronominal suffix /ii/. Here, the second nucleus is licensed by the preceding nuclear head, shortening the initial long vowel. Being licensed, the second nucleus can remain empty and A remains detached. The suffixal /t/ is linked to its position. 352 Shohei Yoshida (37) * (*) (*) (*) * (*) * (*) * (*) * O N O N O N O N x x x x x x x x x x –s U r A t n A [s–urítna] The representation of [s–urítna] is given in (37). Here, the feminine suffix is followed by a consonant-initial suffix /naa/. The final empty nucleus of the feminine suffix, which is the penultimate nucleus of the string, must be licensed by the preceding nucleus containing the floating A. What makes this form challenging is the fact that the licensor of the empty nucleus is pho- netically realized as [i] and not as [a]. Why can A be linked to its point in (35b) and not in this form? I propose the constraint in (38). (38) A nucleus that p-licenses a following empty nucleus cannot a-license a floating segment A. A floating segment A remains detached when its nucleus p-licenses a fol- lowing empty nucleus. Such a nuclear position enters into a paradoxical situation. It needs to obtain melodic material to p-license the following empty nucleus, but the floating A cannot be linked to its skeletal point. Instead, I claim, this empty position is phonetically realized as [i]. Further evidence for the appropriateness of the present treatment of the feminine suffix is provided by the apparently unrelated process of vowel raising in the language. The underlyingly long mid vowels [e] (A.I) and [o] (A.U) are raised when shortened before an internal empty nucleus in the speech of many Cairene speakers, as in (39). (39) beet ‘house’ gooz ‘husband’ bit-na ‘our house’ guz-ha ‘her husband’ Once we assume the elemental representations of [e] and [o] in GP where they are represented as the compounds (A.I) and (A.U), respectively, the rest follows. Inter-nuclear relations in Arabic 353 (40) a. (*) * b. (*) * (*) * (*) (*) * (*) * O N O N O N O N O N x x x x x x x x x x x x x b I >>> t b I t n A ╧ A>>> A [béet] [bítna] In (40a) the initial nuclear head does not license an empty nucleus. It only licenses its complement within its syllable constituent. Thus the lexically floating element A is linked to its skeletal position and fuses with the ele- ment I, yielding the complex segment [e]. In (40b), on the other hand, the initial nuclear head must license the following internal empty nucleus, preventing the element A from being linked to its position. Acknowledgements . I am indebted to Grazyna Rowicka, John Rennison and an anonymous Phonologica reviewer for extremely helpful and valuable comments. The remaining errors and infelicities are of course mine. References Berg, T. & H. Abd-El-Jawad (1992). The structure of Arabic syllables and words. Paper presented at Seventh International Phonology Meeting, Krems. Brockhaus, W. (1995). Skeletal and suprasegmental structure within Government Phonology. Frontiers of Phonology, edited by J. Durand & F. Katamba, 180–221. Essex: Longman. Charette, M. (1991). Conditions on Phonological Government. Cambridge University Press. Guerssel, M. & J. Lowenstamm (1988). Classical Arabic ‘metathesis’. Paper presented at the Genera- tive Phonology Workshop, University of Leiden. Halle, M. & J.-R. Vergnaud (1987). An essay on stress. MIT Press. Harris, J. (1994). English Sound Structure. Oxford: Blackwell. Kaye, J. (1989). Government in phonology: the case of Moroccan Arabic. The Linguistic Review 6, 131–159. Kaye, J. (1990). ‘Coda’ licensing. Phonology 7, 301–330. Kaye, J. (1992). On the interaction of theories of Lexical Phonology and theories of phonological phenomena. Phonologica 1988, edited by W.U. Dressler, H.C. Luschützky, O. Pfeiffer & J.R. Rennison, 141–155. Cambridge: Cambridge University Press. Kaye, J. (1995). Derivations and interfaces. Frontiers of Phonology, edited by J. Durand & F. Katamba, 289–332. Essex: Longman. Kaye, J., J. Lowenstamm & J.-R. Vergnaud (1990). Constituent structure and government in phonology. Phonology 7, 193–231. 354 Shohei Yoshida Kenstowicz, M. (1980). Notes on Cairene Arabic syncope. Studies in the Linguistic Sciences 10, 39-53. Department of Linguistics, University of Illinois. Kenstowicz, M. (1981). Vowel harmony in Palestinian Arabic: a suprasegmental analysis. Linguis- tics 19 , 449–465. Kenstowicz, M. (1994). Phonology in generative grammar. Cambridge: Blackwell Lowenstamm, J. (1996). CV as the only syllable type. Current Trends in Phonology: models and methods, edited by J. Durand & B. Laks. Vol. 2, 419–441. Salford: European Studies Research Institute, University of Salford. Yoshida, S. (1992). Licensing of governing nuclear heads. Paper presented at Government Phonology Workshop, Seventh International Phonology Meeting, Krems. Yoshida, S. (1993). Licensing of empty nuclei: the case of Palestinian vowel harmony. The Linguistic Review 10, 127–159. Yoshida, S. (1996). Phonological Government in Japanese. Faculty of Asian Studies Mono- graphs, The Australian National University. Welden, A. (1980). Stress in Cairo Arabic. Studies in the Linguistic Sciences 10, 99–120. Department of Linguistics, University of Illinois. Phonological domains and binary inter-nuclear licensing for pitch accent Yuko Yoshida Abstract This paper highlights the benefits of a theoretical approach to accent assignment in Standard Japanese that rejects the idea of an inter-syllabic foot construction in favour of the phonological notions of domain and inter-nuclear licensing. This paper also aims to provide a clearer defini- tion of a licensing domain — from its minimal form, a binary licensing domain within an Onset or Nucleus (Rhyme) constituent, as proposed in Kaye, Lowenstamm & Vergnaud (1990), to its maximal form, a phrase The Hague: Thesus. 355–371. Thesus.The Hague: 355–371. . (Yoshida, 1995). At all projection levels of phonological representation, including the skeletal level (projection zero), the government/licensing relation is maximally binary (Kaye, Lowenstamm & Vergnaud, 1990), and Syllables !? accent assignment in Standard Japanese supports this binarity of inter- nuclear licensing. A closer consideration of the process in question from various angles — including innovative forms, and also in comparison with accent phenomena observed in other pitch-accent systems such as Basque — reveals crucial differences between the nature of a phonolog- Phonologica 1996: Phonologica 1996: ical domain and that of a lexical domain at the word level. 0. 1. Introduction The goal of this paper is to exemplify maximal binarity of licensing relations and to explicate the status of PHONOLOGICAL DOMAINS, based on the analysis of pitch accent in Standard Japanese (SJ). Here I aim to demonstrate that various pitch accent phenomena, including lexical accent assignment, can be accounted for in terms of binary INTER-NUCLEAR LICENSING at nuclear projection levels, rather than as inter-syllabic relations. In other words, my claim endorses the viewpoint of Government Phonology (GP): while the syllable constituent is not recognized as an independent category, the three John R. Rennison & Klaus Kühnhammer, eds. (1999). eds. R. Rennison & Klaus Kühnhammer, John (1999). constituents O(nset), N(ucleus) and R(hyme) enter into governing/licensing 356 Yuko Yoshida relations and thereby motivate a range of phonological phenomena (Kaye, Lowenstamm & Vergnaud (henceforth KLV), 1990). First, the proposal of Yoshida (1995) on the algorithm of lexical accent assignment recapitulates how the maximally binary licensing relations at P0 (skeletal level) (KLV, 1990) also serve to define larger prosodic domains such as foot, word and phrase. The binarity of governing relations at P0 extends to the constituent projection level for nuclei — the level at which metrical structure is defined. The following discussion is devoted to an explication of the status of lexi- cal and phonological domains, and of their capacity to accommodate certain linguistic innovations evident in the speech of younger generations — innovations which appear prima facie as counterexamples to the assump- tion of binarity in licensing relations at the nuclear projection levels. In this discussion, the arbitrariness that results from employing the syllable as an active phonological unit is highlighted. 2. Antepenultimate Accent in Standard Japanese (SJ) Prior to defining the role of binary licensing between nuclei, I present a set of data to illustrate how pitch accent functions. I share the viewpoint held by Haraguchi (1991), who claims that SJ has lexical accent assignment to the antepenultimate vowel of a given word — of course, with the exception of some words classed as lexically marked and accentless (Haraguchi, 1991; Yoshida, 1995). Presented in (1) below is a list of some Yamato (native), morphologically simplex words, classified according to their accentual property. An accent is marked by a star (*) above a nuclear segment, and the bar over segments shows that the relevant part is high-pitched; the other parts have no contrastive pitch (Yoshida, 1995). The location of lexical accent in the examples in (1a) is the antepenultimate nuclear position. It may be noted that the examples in (1a) are mostly long items, containing four nuclei or more, except for those with inherent lexical marking (1b), and those without a lexical accent (1c). The general pitch pattern is as follows: in any phonological phrase, regardless of grammatical category, the accented nucleus (the final nucleus of the string, if no accent is found) and all the nuclei to its left are high-pitched, except for the initial nucleus of the string unless the initial nucleus itself is accented (or is string-final). Inter-nuclear binary licensing for accent assignment 357 (1) a. Antepenultimate b. Lexically c. Accentless marked accent (no lexical accent) _∗_ _∗______u gu i su (ga) te (ga) ha / ha ga ‘bush warbler’ ‘hand’ ‘leaf’ _∗_ _∗______a za ra Si (ga) sa ke (ga) sa ke (ga) ‘seal’ ‘salmon’ ‘rice wine’ _∗_ _∗______mu ra sa ki (ga) ha Si (ga) ha Si (ga) ‘purple’ ‘bridge’ ‘edge’ _∗_ __∗______ka ge ro o (ga) ta ma go (ga) u sa gi (ga) ‘ephemera’ ‘egg’ ‘rabbit’ ___∗_ ____∗______ho to to gi su (ga) a ta ma (ga) ka e ru (ga) ‘cuckoo’ ‘head’ ‘frog’ NB: (ga) is a nominative case marker One class of lexical items has been omitted from the list in (1): those items consisting of three Onset-Rhyme (OR) pairs whose accent is on the initial (i.e. antepenultimate) nucleus. Examples are shown in (2). _∗_ __∗_ (2) na mi da (ga) ma ku ra (ga) These words carry inherent lexical marking, like the forms in (1b), rather than having lexical stress assigned via the process responsible for (1a). Distributional facts (Yoshida, 1996) and diachronic change (Yoshida, 1995) provide motivation for this claim, which will be justified below. (3) Distribution of Yamato words based on accent property 645 Yamato words 532 Yamato words (2 OR pairs) (3 OR pairs) No. % No. % Initial Accent 252 39 89 17 Medial Accent — — 13 2 Final Accent 208 32 52 10 Accentless 185 29 378 71 358 Yuko Yoshida (4) Diachronic change 158 Ya m a t o w o rd s 113 Ya m a t o w o r d s (2 OR pairs) (3 OR pairs) No. (1957) No. (1994) No. (1957) No. (1994) Initial Accent 49 49 24 23 Medial Accent — — 1 1 Final Accent 59 59 26 15 Accentless 50 50 62 74 The distribution facts in (3) imply that the proportion of each accent type is fairly even for words containing two nuclei, but is noticeably uneven for those containing three nuclei. For the former, the reason for the even distri- bution is one of shortness: the words are too short to be analysed with reference to antepenultimate position. So accent properties (i.e. presence/ absence of accent, and if present, location of the accent) amount to a mere arbitrary lexical stipulation. Stability in diachronic change, as observed in (4), also confirms this: the distribution of words does not change from 1957 to 1994. As for the words consisting of three OR pairs, the majority (70%) of the total are accentless; this is rather surprising, because the expected landing site of the accent is the antepenultimate nucleus, the initial nuclear position of words containing three nuclei. Initially-accented items only count for 20% of the group. Being antepenultimate, the initial accent of the words containing three nuclei is comparatively stable in diachronic change; however, one less is observed in 1994. Medial position seems to be the most marked of all types, accounting for only 2% of the group. Most vulnerable to historical change are the finally-accented terms. They demonstrate that the lexically marked terms tend to lose their lexical marking and become ac- centless. This fact alone can lead to two incorrect interpretations: i) lexical items are all becoming accentless, and ii) the reason for the loss of the final accent is confusion because in isolation (i.e. without the nominative marker) the pitch patterns of a finally accented word and an accentless word are identical. Both interpretations should be denied on the basis of the words containing two nuclei; lexical marking is not lost in that group of words over the same period of time. Based on these observations, Yoshida (1995) proposes an algorithm for accent assignment in SJ which motivates the loss of accent in finally- accented words consisting of three OR pairs, and accounts for the fact that the majority of terms containing three nuclei are accentless. Inter-nuclear binary licensing for accent assignment 359 3. Licensing at nuclear projection levels: Parameter settings for Standard Japanese 3.1. Algorithm Although three logical possibilities arise for antepenultimate accent as- signment, the head-final binary metrical structure at (NP1) proposed in Yoshida (1995), viz. (5a), invalidates the two other possibilities — namely, either to assume a head-initial ternary foot in (5b), or to stipulate extra- metricality of the word-final position in the metrical system of a head-initial binary foot in (5c) (Haraguchi, 1991). (5) a. Binary licensing b. Head-initial binary feet c. Ternary foot …C V C V C V C V …C V C V (C V) …C V C V C V Two advantages of the accent assignment algorithm in Yoshida (1995) are that the system accounts for i) the fact that the majority of lexical items containing 3 nuclei are accentless rather than initially accented, and ii) the leftward direction of so-called accent shift. The two other possibilities at least fail to account for such issues of pitch accent phenomena. Both struc- tures predict that the majority of words containing three nuclei will be initially accented. In contrast, the licensing structure proposed in Yoshida (1995) predicts not only that the lexical items in question will be accentless, but also the landing site of an accent which in the past analyses was be- lieved to be shifted. SJ has head-final licensing relations at the first nuclear projection (NP1) and head-initial licensing at the higher projection NP2, where only heads of the NP1 are projected; this determines the antepenulti- 1 mate nuclear position as the head position of that LEXICAL DOMAIN. Nuclear positions at NP1 and NP2 are shown as projected skeletal positions, x; for convenience, I distinguish a licensor position from a licensee position by marking the former as 8. The structure in (6a) shows that the antepenulti- mate nuclear position is established as the head of the lexical domain, following the algorithm illustrated above. The ill-formed structure (marked by *) in (6b) demonstrates that the initial position of an item consisting of three OR pairs is not the landing site of its accent. 1 The term “lexical domain” will be discussed in detail in §4.1. Here I assume that a word constitutes a lexical domain, and that the existence of a head nucleus (accented nuclear position) defines that domainhood. 360 Yuko Yoshida (6) a. domain b. * domain ...... 8 x NP2 8 x ...... x 8 x 8 NP1 x x 8 ...... O R O R O R O R O R O R O R N N N N N N N x x x x x x x x x x x x x a z a r a S i s a k a n a _∗______a za ra Si (ga) ‘seal (NOM)’ sa ka na (ga) ‘fish (NOM)’ Concentrating on the first nuclear projection level NP1, we observe that the antepenultimate nucleus is not a licensor: no licensee nuclear position is found in the string further to the left. Only licensor positions are projected from NP1 to the higher nuclear projection NP2. It follows that the antepenul- timate nuclear position cannot be projected to NP2 to license the ultimate nuclear position. Thus, the initial nuclear position of a term containing three OR pairs cannot undertake the headship of the lexical domain. Here we see that the accent on the initial nucleus of a word consisting of three nuclei is the inherent lexical one. 3.2. The natural landing site dispenses with accent shift Another important prediction which is possible only via the algorithm proposed above is the assignment of preantepenultimate accent as the natural landing site (and not as accent shift, as was believed to be the case), when the antepenultimate vowel is not available for accenting. From the numerous examples available, here I outline one involving so-called high vowel devoicing, which is an optional phonological process. The following data involve a class of compounds that conform to morphologically simplex native words in terms of antepenultimate accent assignment (Yoshida, 1995), as evident in the left column in (7), where the forms without the high vowel devoicing are presented. Here I only show the compounded forms and omit all morphological details. Inter-nuclear binary licensing for accent assignment 361 (7) _____∗_ __∗_ i) a. n a n a o| u S i g i b. n a n a o| u S i g i ‘seven wonders’ ______∗_ ___∗_ ii) a. o| u k k a C u s a i b. o| u k k a C u s a i ‘Easter’ _____∗_ __∗_ iii) a. b o o e k i S o o b. b o o e k i S o o ‘international trader’ To account for the way the accent is assigned on one nuclear position to the left, rather than to the right, makes the head-final structure in (5a) the only viable option. To assume either a head-initial ternary foot (5b), or the metri- cal system of a head-initial binary foot with extrametricality in the word-final position (5c), we need to stipulate an accent shift rule to slide the accent one nucleus to the left. For both configurations in (5b) and in (5c), the left- adjacent vowel, the preantepenultimate one, does not occupy the comple- ment position of the expected head (antepenultimate) nucleus to inherit headship. Also, even the preantepenultimate nucleus is not the head of the neighbouring foot (and therefore not a candidate to dominate the foot containing the antepenultimate nucleus at a higher projection). To demonstrate the potential of algorithm (5a) to predict the landing site of an accent in all cases, the government-based analysis of this high vowel devoicing must first be outlined. Since the analysis involves PHONOLOGICAL EXPRESSIONS (PE’s), composed of PHONOLOGICAL ELEMENTS in combination, the relevant elements should briefly be introduced. KLV (1985) propose a set of phonological elements2 which include I and U, whose phonetic values are [i] and [u] respectively, and the h element for stiff vocal fold state (voicelessness). S. Yoshida’s (1991) GP account of devoiced high vowels, involves a nuclear position dominating a phonological element I or U that fails to be phoneti- cally interpreted when sandwiched by onsets both dominating voiceless consonants3 (i.e. onsets dominating an h element). I will take this analysis of the non-interpretation of I and U (S. Yoshida, 1991), and provide further support from pitch accent assignment (Yoshida, 1995). The nuclear position trapped within the inter-onset governing domain is said to be p-licensed. P- licensing is defined below; part 2 is particularly relevant to the present discussion. 2 Since then the elements have gone through revisions, the number of elements has been reduced and the characteristics of each element have been revised. Such revisions are not central to the present discussion of pitch. 3 Devoicing of high vowels occurs also word-finally, when preceded by a voiceless consonant. 362 Yuko Yoshida (8) Phonological Empty Category Principle (Kaye, 1995) (relevant parts only) A p-licensed (empty) category receives no phonetic interpretation. P-licensing: 1. Domain-final (empty) categories are p-licensed (parameterized) 2. A nucleus within an inter-onset domain [is p-licensed]. A nuclear position within an inter-onset domain is p-licensed, and receives no phonetic interpretation. Note that in some languages nuclear positions with phonetic content are subject to non-interpretation when in a p-licensing environment (see Yoshida, 1995 for the case of Tonkawa, and Charette, 1991 for the cases of Tangale and Khalkha Mongolian), and I assume that in the case of Japanese high-vowel devoicing, the I and U elements in question receive no phonetic interpretation. In the representations in (9), the environment is illustrated where this type of p-licensing occurs. N3 is surrounded by onsets dominating voiceless PE’s, o| and S, which contain an h element. The diacritic o under a high vowel denotes that the vowel in question is inaudible. (9) a. U is interpreted O R O R O R O R O R 5 4 3 2 1 N5 N4 N3 N2 N1 x x x x x x x x x x n a n a o| U S i g i nana|uo Sigi ‘seven wonders’ b. U is not interpreted inter-onset domain O R O R O R O R O R 5 4 3 2 1 N5 N4 N3 N2 N1 x x x x x x x x x x n a n a o| U S i g i nana|uo o Sigi ‘seven wonders’ ↑ p-licensed N3, being in an inter-onset domain, is p-licensed, and thus inaudible. In (10a), we observe an orthodox case of accent assignment as in §3.1. The ante- penultimate nucleus is the head of the domain, following the regular algo- rithm. In the case shown in (10b), N3 is the expected landing site of an ac- cent; however, that position is p-licensed within the inter-onset governing domain. Inter-nuclear binary licensing for accent assignment 363 (10) a. domain ...... 8 x NP2 ...... x 8 x 8 NP1 ...... O R O R O R O R O R N N N N N x x x x x x x x x x n a n a o| U S i g i ____∗_ na na o|u Si gi b. * domain ...... 8 x NP2 ...... x 8 x 8 NP1 ...... O R O R O R O R O R 1N5 1N4 1N3 1N2 1N1 x x x x x x x x x x n a n a o| U S i g i ____∗_ na na o|uo Si gi As explained above in §3.1, a position which is not a licensor at NP1 can- not be the head of an accent domain. N3 is p-licensed at the skeletal level, and the nuclear point cannot be projected to the NP1 as a licensor. Under these circumstances, N3 cannot take on the headship of i) the binary licensing relation at NP1 ii) ultimately, the lexical domain. So the expected 1 licensee position at NP dominates the p-licensed position (N3). Thus the preantepenultimate nuclear position (N4), which is the expected licensee in the usual cases, has to take over the headship of the binary licensing at NP1. This way, the non-interpretation of the I/U element dominated at the antepenultimate position causes the accent to be assigned to the preante- penultimate nuclear position. In conclusion, the licensing configuration in §3.1 predicts the accent locations, and the phenomena outlined above are no longer treated as an arbitrary shift process. So far, I have illustrated how the accentual behaviour of SJ can be de- duced from binary licensing at all projection levels. In order to focus on 364 Yuko Yoshida what an accentless word means in a pitch accent language, I will turn to another innovative aspect of linguistic behaviour in SJ. 4. Domain Edges and Domainhood In this section, lexical domains and phonological domains are defined, in order to deal with their behaviour as apparent counterexamples to the algorithm presented above. The prediction the configuration of binary licensing in SJ makes is that Yamato morphologically simplex words con- sisting of 3 OR pairs will be accentless (§3.1). The data collected recently consist of accentless (though not Yamato) words, which contain 4 or more OR pairs. This section solves the question of how and why a word contain- ing more than 3 OR pairs turns out to be accentless. 4.1. Accent loss in loan words and the status of native (Yamato) lexical items Among language-conscious people in Japan, including professional lin- guists, one of the most prevalent issues of recent times is that of de- accenting phenomena. That is, a word that has antepenult accent in the speech of older adults (say, over the age of 25), has no accent at all in the speech of the younger generation, roughly between 15~24 years of age. However, indisputable is the fact that, for all the speakers to be discussed in this section, antepenultimate accent location of Yamato morphologically simplex words consisting of 4 OR pairs is stable. Thus the change in (11) never occurs. (11) * _∗______u gu i su (ga) → u gu i su (ga) ‘bush warbler (NOM)’ This fact confirms that the algorithm discussed above must be active for all SJ speakers. However, once the data extends to loan words and various derived nouns,4 a careful study of the pitch patterns used by different generations and social groups reveals an interesting fact. Inoue, in his article in a major Japanese newspaper, reports on a lady who uses an accentless form of o|akkusu ‘fax’ instead of the normal antepenultimate accent pattern, 4 These are Sino-compounds, which consist of more than one Sino-morpheme (Sino- morpheme = one Chinese character that contains one or two OR pairs in its Japanese reading). These terms conform to the native pattern in terms of pitch accent assignment. Inter-nuclear binary licensing for accent assignment 365 after being in charge of a fax machine for a long time in an office (Inoue, 1996). Many native speakers find the unaccented form odd, and may even try to correct the speaker’s form. Be that as it may, an accentless pronuncia- tion de facto exists, and consequently has to be explained. The list in (12) is from the results of Inoue’s test on university students. I categorize these speakers as generally falling into the under-25 age group. The two columns (12a) and (12b) show the contrast between the two generations, where the dividing line is age 25. (12) a. Generation 1 b. Generation 2 (age: c.25 and above) (age: up to c.24) _∗______to So ka W˜ (ga) to So ka W˜ (ga) toSo ‘books’ + kaW ˜ ‘building’ = toSokaW ˜ ‘library’ _∗______se ka i Si (ga) se ka i Si (ga) sekai ‘world’ + Si ‘history’ = sekaiSi ‘world history’ __∗______ni ho W˜ Si (ga) ni ho W˜ Si (ga) nihoW ˜ ‘Japan’ + Si ‘history’ = nihoW ˜Si ‘Japanese history’ The pitch pattern observed in (12b) is as striking to any older speaker as an exotic vernacular of Japanese would be.5 This pitch pattern might be a temporary phenomenon; however, this is the grammar of the generation in question for de-accented words such as those above. Here the grammar of those speakers who adopt de-accenting is consid- ered, and we will see that the unit syllable plays no role. This time, we resort to the notion of domain, defined in terms of licensing relations. First the status of domains for such accentless terms is examined, and then the speakers’ phonological system (grammar) is discussed. In §3 I showed that a domain with three nuclear positions fails to be assigned an accent; of course, the absence of a lexical accent means the absence of a domain head. The question is whether a domain can exist without any head at all, to which the Licensing Principle (Kaye, 1990) provides an unequivocal answer. (13) The Licensing Principle (Kaye, 1990) All phonological positions save one must be licensed within a domain. The unlicensed position is the head of the domain. 5 Diversity in Japanese local dialects depends mostly on differences in pitch pattern, as well as some lexical variation. 366 Yuko Yoshida To wit, the presence of a head defines a domain. Compare an accented term and an accentless term. The presence of a lexical accent, inherently marked or lexically assigned, defines a lexical domain: the accented nucleus is the head of that lexical domain. Then, the status of an accentless item is questionable in terms of its domainhood. As has been discussed, an ac- centless term consisting of three OR pairs is the outcome where it fails to obtain a head via lexical accent assignment in that lexical domain. Then the domain cannot assume the status of a lexical domain. The resulting phono- logical string would be treated by the phonology as a phonological domain, which also requires a head by virtue of being a domain. (The dispute where the location of the head should be is confirmed in §4.2.) To conclude, it seems that the speakers who prefer the de-accented words in (12) employ this system: the words are not recognized as lexical domains, but merely as phonological domains. 4.2. Motivating phonological domains (in relation to the domain-final nuclear position) 4.2.1. The head position of a phonological domain The location of a phonological domain head is the domain edge, and in many cases, the domain-final one. Resorting to the interpretation of a pitch accent, we can identify the head location of a phonological domain. (14) A parameter6 for pitch accent languages and stress accent languages (Yoshida, 1995) The interpretation of the domain head is that: i) in a stress language, the head nuclear position is fortified (pri- mary stress) and is stronger than other nuclei, and ii) in a pitch accent language, the head of the domain and the nuclei to the left (except the domain-initial one) are all high-pitched. The parameter setting cited in (14) is based on the pitch patterns of SJ and also on some dialects of Croatian (see Yoshida, 1995 for a detailed discus- sion). Recall the pitch pattern in (1): in any phonological string, regardless of grammatical category, the accented nucleus (the domain final nucleus, if no accent is found) and all the nuclei to its left are high-pitched, except for the initial nucleus of the string unless the initial nucleus itself is accented (or is 6 This parameter also replaces any stipulation of another metrical structure for the mapping of pitch e.g. a metrical tree to percolate desired pitch to relevant segments (e.g. in Abe, 1987). Inter-nuclear binary licensing for accent assignment 367 domain-final). For items without lexical accent, the domain-final nucleus takes on headship, i.e. the head nucleus is the stem-final nucleus when in isolation, and the string-final nucleus when suffixed by a case-marker. In other words, when in isolation, the stem forms the phonological domain, and when suffixed by a case-marker, the stem plus the marker together form a phonological domain. The phonological domain nominates the head nucleus, whose interpretation will be identical to that of the head of a lexical domain: the head nucleus and the nuclei to the left except the initial nucleus are interpreted as high-pitched. The claim that the domain-final nucleus is the head nucleus of a phono- logical domain gains support from similar analyses of other pitch accent languages and other Japanese dialects. For example, the Kyoto dialect has a class of nouns containing only one high-pitched nucleus. (15) One type_ of accentless noun in the Kyoto Dialect__ a. ki __i ‘tree’ ki i ga__ ‘tree (NOM)’ b. a na __ ‘hole’ a na ga__ ‘hole (NOM)’ c. u sa gi ‘rabbit’ u sa gi ga ‘rabbit (NOM)’ Evidently, the domain-final nucleus is the head of the phonological domain in such cases. That nuclear position is the only high-pitched one in the relevant domain, regardless of the presence/absence of the nominative marker -ga. In fact, besides this parameter, there are also other phonological prin- ciples which make crucial reference to domain-final position (e.g. h-licensing — see Kaye, 1994). The Phonological EMPTY CATEGORY PRINCIPLE (ECP), which was cited in (8) above, is one example. The relevant part is repeated in (16). (16) Phonological Empty Category Principle (part) A p-licensed (empty) category receives no phonetic interpretation. P-licensing: 1. Domain-final (empty) categories are p-licensed (parameterized) The domain-final nuclear position is licensed, by virtue of being domain- final, to license the preceding onset. In systems like the Gernika dialect of Basque, even the domain-final posi- tion which is subject to ECP has to take on the headship of the phonological domain, by virtue of being domain-final. Gernika is a system whose parame- ter for the ECP, as in (17) is set to “on”, thus the domain-final position is licensed to be empty. The following data set is taken from Hualde (1991), but adapted to the GP context here; 0 in (17) stands for an empty nuclear posi- tion. 368 Yuko Yoshida (17) Gernika (absolutive forms) indefinite singular plural ______gi xo_ n0______gi xo_ na______gi xo______na k0 ‘man’ gun __tzu rru n0 gun _tzu__ rru ne _ gun___ tzu rru ne k0 ‘kidney’ ‘girl’ _nes___ ka nes___ kie _ nes ka k0 a r0 a rra a rra k0 ‘worm’ As mentioned above, a licensed nuclear position cannot be accented in SJ; in other words, the licensed nuclear position cannot take on the head- ship of a lexical domain. The idea that the licensed domain-final nuclear position should take on headship is indicative of the lack of any lexical domain for such phonological strings. Instead, the presence of a phonolog- ical domain is confirmed. 4.2.2. Remarks on syllable and mora The status of domain-final positions sheds light on the analysis of some segments that can never be accented in SJ, such as N, the so-called moraic nasal [W˜ ], and the second member of an apparent diphthong and a long vowel, which appears with high pitch when in domain-final position. I focus on the example of a word-final N. From the fact that N can be recognized with high pitch, S. Yoshida (1991) proposed that N should involve an empty nuclear position, to give N an independent OR pair status, rather than treating it as the coda position of a (C)VC syllable. Yoshida (1995) proposed filling the nuclear position with an unassociated U element,7 so that a word- final N does not violate the ECP. (Japanese selects “off” for that parameter, i.e. the word-final position has to be filled with a PE). It should be mentioned that this nuclear position is not associated to the point, as was the case with the non-interpretation of I and U discussed in §3.2, and as such, cannot take the headship of a lexical domain. Therefore an accent cannot be assigned to the N. However, the nucleus in the phonological domain-final position should take the headship by virtue of being domain-final. This is where a system without an independently motivated syllable con- stituent has an advantage over a standard analysis, where another constitu- ent, the MORA has to be introduced to account for phonological processes in SJ. Some Japanese phonologists employ both syllables and morae, 7 The reason for choosing the element U is that N demonstrates complementary distribution with nu, i.e. a nuclear position dominating a U element, preceded by an onset dominating a nasal element (Yoshida, 1995). Inter-nuclear binary licensing for accent assignment 369 alternating through a derivation as active phonological units (e.g. McCawley, 1968; Poser, 1984; 1990). Japanese has been classified as a mora-counting syllable language. Heavy syllables such as (C)VV or (C)VC are said to consist of two morae, whereas a short syllable (C)V consists of one (McCawley, 1968). In SJ, if an accent falls on a heavy syllable, such as a VN sequence, the accent is always on the initial mora. To accommodate this fact, an accent never falls on the second mora within the heavy syllable, even if that mora is located on the expected accent landing site (Poser, 1984; 1990). In other words, the head of the word (accent) has to be on the head portion (initial mora) of the heavy syllable. Turning to the case of accentless words, a stipulation is necessary: from the fact that in accented words no high pitch is observed on any segments to the right of a pitch accented one, a high pitch is assigned to the word-final mora. In other words, the reason for this high pitch insertion in a specific position stems merely from observation, rather than from any particular theoretical motivation. It should also be noted that the unit relevant to high pitch insertion is the mora (and not syllable, which plays no role at this stage). This is to accommodate examples with high-pitched word-final Ns (the second mora of the heavy syllable), where the pitch spreads to the relevant portions within the given word (Poser, 1984). I consider the proposal of syllable and mora shift during derivation to be arbitrary. From the GP point of view, the headship of the phonological domain is identified with the domain-final nuclear position; in this way, it becomes possible to predict and account for the fact that a sound which can never bear an accent can nevertheless be recognized as high-pitched. 4.3. Domain-initial nuclear position and phonological domain There is another significant point to discuss, which highlights the impor- tance of the notion of a phonological domain. So far, the focus of discussion has been the domain-final nuclear position; now, the other domain edge is examined. As Charette (1989: 91) reports for Quebec French and Tangale, Yoshida (1990) for Tonkawa, and Yoshida (1995) for Japanese, in many lan- guages, the domain-initial nuclear position exhibits unique behaviour in contrast to other nuclear positions within the domain. In the case of SJ, the domain-initial position fails to participate in high pitch sharing. Unless the initial nucleus itself has a lexical accent, the head nucleus and the nuclei to its left except the initial nucleus are interpreted as high-pitched. This is due to the inaccessibility of the initial nuclear position (Yoshida, 1995). Phrasal interaction of high pitch in connected speech provides further evidence for the analysis (Yoshida, 1995). Here, generally speaking, two 370 Yuko Yoshida neighbouring phrases (i.e. two adjacent phonological domains merge) in a sentence merge (to form a single domain) for pitch assignment (Poser, 1984; Yoshida, 1995, etc.). Consider a sentence consisting of two phrases without an overt subject, as in (18). (In Japanese, subjects are often not explicit.) ______(18) a. [sa ka na -o] [mo ra t ta]8 b. [sa ka na -o] [mo ra t ta] fish (ACC) be given (PAST) ‘(SUBJECT) was given fish’ In their citation forms, the two phrases show separate pitch assignment; each phrase nominates its head nucleus (the domain-final nucleus, because the phrases are both accentless), and all the nuclei to its left are high- pitched except the domain-initial nucleus. In contrast, in connected speech, the initial nucleus of the second phrase also has high pitch; in other words, the position is not treated as domain-initial. When preceded by another phrase, a phrase merges into the other to form one single domain of pitch. Here, the domain-initial nuclear position of the second phrase is no longer domain-initial, and thus the position is subject to high pitch sharing from the head nucleus (i.e. the domain-final nucleus), as in (18b). Here, the definition of the domain and its corresponding head explain the pitch pattern. More- over, this shows that domain edges have a unique status, and that the notion of domain is the significant factor in the discussion of accent. In a sentence in connected speech, the third (from last) phrase is unlikely to be joined to the other two, except when it is accentless (Poser, 1984; Yoshida, 1995). This can be explained by the notion of lexical domain and phonological domain. Consider the following connected speech forms. The third phrase (in this case, a subject) is added before the two phrases of (18). The subject phrases of the sentences in (19a–b) are accented, while those in (19c–d) are accentless. _∗______(19) a. [ne ko -ga] [sa ka na -o] [mo ra t ta] ‘A cat was given fish’ _∗_ b. * [ne ko -ga] [sa ka na -o] [mo ra t ta] ‘A cat______was given fish’______c. [ne zu mi -ga] [sa ka na -o] [mo ra t ta] ‘A mouse______was given fish’ d. [ne zu mi -ga] [sa ka na -o] [mo ra t ta] ‘A mouse was given fish’ 8 Note that I employ an accentless verb here for the current discussion of accentless phrases and domains. However, the behaviour of the domain-initial nuclei in question is the same for accented verbs (see Yoshida, 1995 for detailed discussion). Inter-nuclear binary licensing for accent assignment 371 The only impossible case among the sentences (19a–d) is the one in (19b), which shows the addition of a third merging accented phrase. We see that the accentless subject phrase is easier to merge with the other phrases in connected speech. Accentless phrases have no lexical domain, and are recognized as phono- logical domains with their enclitics. Perhaps a phrase without a lexical domain can more easily merge than one with a lexical domain, since the phonological domain is only ultimately recognized when the size of the phonological sequence is defined, i.e. when the whole sequence is pro- cessed as connected speech. This also endorses the analysis of a pitch accent language as actively involving the notion of phonological/lexical domains. References Abe, Y. (1987). Metrical structure and compounds in Japanese. Issues in Japanese Linguistics, edited by T. Imai & M. Saito, 5–52. Dordrecht: Foris.. Charette, M. (1988). Some constraints on governing relations in phonology. PhD dissertation, McGill University. Charette, M. (1991). Conditions on phonological government. Cambridge: Cambridge University Press. Haraguchi, S. (1991). A theory of stress and accent. Dordrecht: Foris. Hualde, J.I. (1991). Basque phonology. London: Routledge. Inoue, F. (1996). Nihongo Ichiba 20: Shitashisano shoomei. Article in Asahi Shimbun (Feb. 25). Kaye, J.D. (1990). ‘Coda’ licensing. Phonology 7, 301–330. Kaye, J.D. (1994). Current issues in phonology. Lectures held at SOAS, Term 2. Kaye, J.D. (1995). Derivations and interfaces. Frontiers of phonology: atoms, structures, derivations, edited by J. Durand & F. Katamba, 289–332. Harlow: Longman. Kaye, J.D., J. Lowenstamm & J-R. Vergnaud (1990). Constituent structure and government in phonology. Phonology 7, 193–231. Poser, W. (1984). The phonetics and phonology of tone and intonation in Japanese. PhD dissertation, MIT. Poser, W. (1990). Evidence for foot structure in Japanese. Language 66, 78–105. Yoshida, S. (1991). Some aspects of governing relations in Japanese phonology. PhD dissertation, SOAS, University of London. Yoshida, Y. (1995). On pitch accent phenomena in Standard Japanese. PhD dissertation, SOAS, Univer- sity of London. Yoshida, Y. (1996). Pitch accent phenomena in Standard Japanese. Kansai Linguistics Society 16 (Proceedings of the 20th meeting of the Kansai Linguistics Society), edited by T. Sugimoto, 143–152. The proceedings of the 20th meeting of the Kansai Linguistics Society, Osaka.