96. Metrical Patterns 1347

96. Metrical patterns 1347

Ruhlen, Merrit. 1987. A guide to the languages of Vennemann, Theo. 1971. „The phonology of Gothic the world. Stanford: Stanford University Press. vowels“. Language 47: 90Ϫ132. Schane, Sanford A. 1973. Generative phonology. Vennemann, Theo. 1972. „Phonetic detail in assim- Englewood Cliffs. ilation: Problems in Germanic phonology“. Lan- Skalicˇka, Vladimir. 1979. Typologische Studien. guage 48: 863Ϫ892. Braunschweig/Wiesbaden: Vieweg. Vennemann, Theo. 1986. Neuere Entwicklungen in Sommerstein, Alan H. 1977. Modern phonology. der Phonologie, Berlin etc.: de Gruyter. London: Arnold. Vennemann, Theo. 1988. „The rule dependence of Stampe, David. 1979. A dissertation on Natural syllable structure“. In: Duncan-Rose, Caroline & Phonology. New York & London: Garland. Fisiak, Jacek & Vennemann, Theo (eds.). Rheto- Trubetzkoy, Nikolaj S. 31958. Grundzüge der Pho- rica, Phonologica, Syntactica: A Festschrift for nologie. Göttingen: Vandenhoek & Rupprecht. Robert P. Stockwell, London & New York: Rout- ledge, 257Ϫ283. Ultan, Russel. 1978. „A typological view of me- tathesis“. In: Greenberg 1978, 367Ϫ402. Vihman, Marilyn M. 1978. „Consonant harmony: its scope and function in child language“. In: van der Hulst, Harry & Smith, Norval. 1982. „ An Greenberg 1978, 281Ϫ335. overview of accent and metrical phonology“. In: van der Hulst & Smith (eds.) 1982, 1Ϫ45. Vogel, Irene. 1982. La sillaba come unita` fonolo- gica. Bologna: Zanichelli. van der Hulst, Harry & Smith, Norval (eds.). 1982. The structure of phonological representation. Dor- drecht: Foris Publ. Thomas Krefeld, München (Deutschland)

96. Metrical patterns

1. Introduction linguistic variables is fundamental to lan- 2. The syllable as a constituent guage typology (cf. Plank 1997), our goal is 3. The foot not merely to list the observed metrical pat- 4. Typology and metrical structures terns, but also to examine possible relation- 5. Universals, implications and correlations 6. Conclusion ships between the different patterns. 7. References To this end we will focus on ‘metrical co- herence’ from two perspectives and address the following questions. First, we ask whether 1. Introduction a given metrical constituent varies in its properties within a single language. For in- Metrical patterns in languages are obtained stance, the metrical constituent ‘foot’ is gen- by combining various elements of prosodic erally used to account for word stress. How- structure: syllables and their constituents, ever, there may be other processes which are feet, and other higher level organisational sensitive to foot structure. If so, one would units like prosodic words, phrases and so on. like to know if foot types vary for different Within a given metrical organisation, a par- processes within a given language, or whether ticular constituent may be the most promi- with respect to a given metrical constituent, nent. This relative prominence is marked by the system is coherent (Dresher & Lahiri stress, which is the central theme of this arti- 1991). The second issue is whether the type cle. Stress, under this conception, is not mer- of stress a language has can predict the prop- ely a phonetic feature, but is the means of erties of its metrical constituents. This par- marking relative prominence within various ticular perspective has not been an issue in organisational groupings of metrical units the phonological descriptions of metrical pat- (cf. Liberman 1975; see the articles by Ka- terns, but is extensively discussed in typologi- ger, and Halle & Idsardi in Goldsmith 1995 cal literature on the covariation of stress with for surveys of different metrical theories of the nature of syllables, headedness of phrasal stress). In order to establish stress patterns, stress and such (cf. Donegan & Stampe 1983; we first discuss how different metrical con- Gil 1986). Thus, we begin by motivating syl- stituents are relevant for the phonological lables and feet as necessary metrical constitu- systems as a whole. Since the covariation of ents in the description of phonological sys- 1348 XII. Phonology-based typology tems. For each of the constituents we provide (1) Syncope and syllable final devoicing evidence from segmental processes as well as in German for stress, and then move on to issues on met- Standard German rical coherence and covariation of metrical ‘flirt’ ‘sail’ ‘go by bicycle’ units, which are rather crucial for typologi- Infinitive li:belϩn ze:gelϩn ra:delϩn cal research. 1sg.ind.pres li:blϩe ze:glϩe ra:dlϩe Standard German, Northern pronunciation ϩ ϩ ϩ 2. The Syllable as a constituent li:bl e ze:gl e ra:tl e After syncopation, the consonant clusters In this section, we first review the syllable’s status in phonology before discussing the role that are created are not equally accepted as of the syllable in the assignment of stress. onsets in the Standard German as compared The syllable has traditionally been assumed to the Northern pronunciation. In Standard to consist of an onset followed by a rhyme German, the sequence [dl] is accepted as a which is divided into a nucleus and a coda. syllable onset, and the maximisation of onset The nucleus is the obligatory and most im- prevents the [d] being in the coda. Hence, portant part of the syllable, while the onset coda-devoicing does not apply. In contrast, and coda are optional. The most frequent syl- the Northern pronunciation which allows lable inventory in natural language consists [bl] and [gl] clusters, permits maximisation of of the following: V, CV, VC, CVC (see Blev- consonants in these cases, but prevents [dl] ins 1995 for a survey). The more complex syl- from being part of an onset. As a result, lable inventories arise from including more coda-devoicing applies and the surface form segmental material in the onset and the coda, is [ra:t.le] rather than *[ra:.dle]. and even the nucleus can be branching. Com- Maximisation of the onset is closely re- plex onsets and codas are generally governed lated to the notion of a core syllable, or a CV by the Sonority Scale which states that onset syllable. There is a general tendency to avoid consonants increase in sonority and codas onsetless syllables such that in most if not decrease in their sonority (cf. Clements & all languages, a VCV string is syllabified as Hume 1995). The accepted sonority scale in [V.CV]. Resyllabification to prevent onsetless terms of rising sonority is obstruents Ͻ na- syllables is central to the analysis of German sals Ͻ liquids Ͻ glides Ͻ vowels. devoicing as well (cf. Rubach 1990, Giegerich The notions ‘closed’ and ‘open’ syllables 1992). The following alternations are rele- play an important role in phonology. Closed vant.

(2) Resyllabification in German glaub [p] glaub-en [b] glaub-lich [p/b] ‘believe’ 2sg imp./inf./adj. Tag [k] Tag-e [g] täg-lich [k/g] ‘day’ sg./pl./adv. syllables are those which are closed by a As we have seen before, coda devoicing makes coda consonant, while open syllables end in the word final consonants in the first column a vowel (long or short) or a diphthong. To voiceless. A suffix vowel is added to the decide whether medial consonants are part words in the second column. Here the medial of onsets or codas, the principle of maximi- sequence VCV is syllabified as [V.CV] forcing sation of the onset is often invoked. That is, the medial consonant to be an onset, thereby when there is more than one intervocalic blocking coda devoicing. Oddly enough, when consonant, whether all of them are part of the suffix begins with a sonorant consonant, the onset of the second syllable, depends on and although the obstruent ϩ liquid is a pos- whether the language permits ‘maximising sible onset (as we saw in the previous exam- the onset’ based on sonority principles. ple), resyllabification can be blocked for Phonological processes can help determine certain speakers and coda devoicing applies. whether consonants fall in the coda or not. Obviously, for those speakers who devoice This is illustrated with an example from Ger- the obstruents, resyllabification is sensitive to man which has a process of syllable final certain morphemes even if allowable onsets devoicing. The data are from Vennemann may arise. However, the crucial point is that (1972). German has a rule of syncopation when a suffix with an initial vowel follows, which follows for the following types of al- resyllabification is obligatory since German ternations. always requires a syllable with an onset. 96. Metrical patterns 1349

Words or syllables without a surface conso- (4) Bengali degemination as cluster sim- nant are always preceded by a glottal stop: plification cf. Atmen [?atmen] ‘breathing’, abteilen [?ap- (a) su-lam su-tstshi ‘sleep 1past/ tai len] ‘to separate’, mitarbeiten [mıt?arbai - 1present’ ten] ‘to cooperate’ etc. For some speakers the (b) bos-lambos-tshi *bos-tstshi ‘sit 1past/ glottal stop insertion is restricted to stressed 1present’ syllables; hence, Theater [the?a´6tB]) ‘theatre’, but Bebauung [beba´ wwn] ‘building develop- Bengali does not allow coda clusters. Since ment’, and not [beba´ w?wn]. a geminate consonant belongs to the coda of one syllable and the onset of the following 2.1. Preferred syllable structure syllable, if the preceding syllable ends in a Once we accept the fact that languages have consonant, the geminate introduces a coda preferred syllable structures, any deviation cluster and is degeminated to fit the syllable from these preferences are repaired. Strate- template of the language. gies for repairing them can differ. For in- Thus, both deletions and insertions are stance, if affixation leads to unacceptable frequently found in languages, and almost al- syllables, either epenthesis or syncope are in- ways in the context of repairing an unaccept- voked to maintain the preferred structures. able syllable. Preference for syllable types, In a language like Koryak (a Paleosiberian and hence repairs, is usually restricted to the language spoken in Kamchatka; Spencer 1996: lexical level. In the postlexical level, there is 63Ϫ64), the most complex syllable structure more variation. The last example of degemi- permitted is CVC. Hence any affixation which nation can also be viewed as shortening, and leads to complex structures is resolved by as we will see in § 2.3., lengthening and short- schwa epenthesis. ening phenomena are also linked to syllable structure. However, in these cases it is the (3) Koryak schwa epenthesis weight of the syllable which plays a crucial Verb root /pnlo/ ‘ask’ role. Prefixes: t- 1sg.subj. mt- 1pl.subj., pl subj na- 3 . . 2.2. Syllable quantity and weight (a) t-pnlo-n tep.ne.lon ‘I asked him’ (b) mt-pnlo-n met.pen.lon ‘we asked him’ One view of representing syllable weight is by (c) na-pnlo-n nap.ne.lon ‘they asked him’ using moras. The moraic theory of represen- tation views moras as phonological positions If the segments are syllabified from left to which come between prosody and segments right obeying the preferred CVC syllabic tem- (rooted in the feature tree). Long and short plate, then the introduction of the schwa is vowels, and long and short consonants (i. e. entirely predictable. If we did not assume that geminate and single consonants) are dif- epenthesis was syllable based, it would not ferentiated by their moraic representation. be possible to account for the difference be- Moraic representations in (5) are based on tween the schwa insertions in the verb root Hayes (1989). in (3a) and (3b): pnel vs. penl. Epenthesis is one of the most frequent (5) Moraic representations ways to resolve unwanted clusters and to mmm m obtain a preferred syllable template. Related t ͉ ͉ tt languages often exhibit a difference in the [a] [a:] [p] [p:] acceptance of initial and final clusters. A striking example comes from certain final li- Short vowels have one mora, long vowels quid ϩ obstruent clusters in Germanic lan- have two moras, a single consonant has no guages. English and German allow [l ϩ ob- moras, and a geminate consonant comes with struent] clusters in words like milk or Milch, one mora. A single consonant is not assigned but Dutch disallows such clusters and intro- a mora in the lexical representation. It may duces a schwa as in melek. or may not be assigned a mora depending on Along with epenthesis, deletion is another whether it is in the coda and whether the lan- means for cluster simplification. In Bengali, guage treats closed syllables as heavy. If the the present indicative ending begins with a coda is counted as heavy, then weight-by-po- geminate affricate -tstsh which is degeminated sition assigns a mora to the coda consonant. when added to a verb root ending in a con- Geminates, on the other hand, are part of the sonant (Fitzpatrick-Cole 1994, 1996; Lahiri onset of a syllable, but must close the pre- 2000). ceding syllable as well, automatically adding 1350 XII. Phonology-based typology weight to this syllable. (A problem arises in (7) Bengali total assimilation languages where geminates do not contribute pcrda pcd6a ‘curtain’ to weight but long vowels do; see Lahiri & por-tam pot6am ‘wear-1sg.past Koreman (1988), Hayes (1989), Kager (1989) habitual’ for further discussion.) Hypothetical syllabi- ghcrdzama i ghcdz6ama i ‘house son-in-law; fications are given below. son-in-law who lives in the house (6) Syllable structure assignment of his in-laws’ Other common instances of compensatory lengthening involve the loss of a coda conso- nant which leads to the lengthening of the preceding vowel. We find this in Old English with the loss of a coda nasal. If we compare the words for five and tooth in Old High German, Old English and their modern de- scendants, we find the pattern in (8). Since ungrammatical forms are marked elsewhere with an asterisk, the Proto-Germanic recon- structed forms will be indicated with the sign †. (8) Compensatory lengthening in Ger- manic German OHG English OE Proto- Just as languages often try to preserve pre- Germanic ferred syllable structures, we often find pro- fünf fimf, five fı¯f †fimfi cesses which attempt to maintain the weight fumf of a syllable. Bimoraic syllables are heavy, ir- Gans gans goose go¯s †gans respective of whether they are closed syllables (the coda consonant adding weight to the The Proto-Germanic words had a short vowel syllable), or whether they have a long vowel. followed by a nasal consonant. The nasal has However, not all languages necessarily con- been retained in German and the vowels are sider closed syllables to be heavy. Languages still short. The loss of the nasal in English, tend to avoid trimoraic syllables although however, has led to long vowels (which they do exist. Further consequences of sylla- were later sometimes diphthongised) Ϫ an ble weight will be discussed when we con- instance of compensatory lengthening which sider stress. we can represent in a nonlinear fashion. In (9a), V and N represent any vowel or a nasal. 2.3. Compensatory lengthening Since long vowels are bimoraic, delinking Similar to deletions and insertions, shorten- after the loss of the nasal and reassociation, ing and lengthening processes are closely re- gives us the desired result. In (9b), the same lated to the syllable. A frequent process of effect is realised for the [r] deletion and con- lengthening is compensatory lengthening, comitant gemination in Bengali, except that where the loss of a segment is compensated only a single mora is involved. Here the mora, by lengthening an adjacent segment. This can which was originally linked to the [r] in the be accomplished by total assimilation or by coda, is then linked to onset consonant (rep- vowel lengthening. For instance, in Bengali resented by C) in the next syllable, thus creat- an [r] followed by a coronal consonant is op- ing a geminate. tionally deleted and the consonant becomes a geminate (Hayes & Lahiri 1991). The as- (9) Compensatory Lengthening as spread- similation can apply within words, across ing morphemes, as well as across words, the con- (a) Germanic straint being that the [rC] sequence must be- mm mm long to a single phonological phrase. Some ttJ͉ ͉ examples are given in (7). VN V¯ 96. Metrical patterns 1351

(b) Bengali The usual onset and coda constraints of the language apply. This procedure leading to ambisyllabicity is labelled as Extended Right Capture in Gussenhoven (1986: 130), the for- mulation of which is based on two different processes in Kahn (1976). (11) Ambisyllabicity Extended Right Capture It is worth noting that when the loss of the consonant in such circumstances is closely linked with vowel lengthening, it is invariably confined to a particular syllable position, and a similar loss elsewhere in the phonology of a language will not show any concomitant lengthening. For instance, in Old English the [n] sometimes disappeared between conso- nants: OE elboga beside elnboga ‘elbow’; OE Ambisyllabicity accounts for a number of sAterdAg beside sAterndAg ‘saturday’. postlexical phonological rules of American English like flapping, aspiration, glottaliza- (10) Loss of nasal in Old English not lead- tion etc. Flapping weakens coronal stops ing to compensatory lengthening [t, d] to a flap [J] when they are ambisyllabic. Compensatory lengthening as spread- This accounts for why the coronal stops in ing onto a free mora later, shouting, matter are subject to flapping, while those in latex, bait, tail are not. In the latter set of words, the stops are either fol- lowed by a stressed syllable (cf. la´te`x), are only in the coda (cf. bait), or only in the onset (cf. tail), and hence none of them are ambi- syllabic. Similarly, aspiration is also subject to ambisyllabicity. Aspiration of voiceless stops in American English occurs when in absolute syllable onset position, and ambi- In (10) the deleted [n] is not immediately pre- syllabic consonants cannot be aspirated. This ceded by a vowel. It is in a branching coda, is different in British English where absolute sharing the mora with another consonant. onset position is not required for aspiration. The loss of the nasal does not free the mora Thus, words like happy, where the medial of the coda and therefore there is no spread- consonant is ambisyllabic, may be aspirated ing and no lengthening. Thus, compensatory in British English, but never in American lengthening can be viewed as maintaining the English. However, British English also re- weight of a syllable. quires ambisyllabicity as a structural possi- bility, since rules like weakening (which 2.4. Ambisyllabicity ‘weaken the oral closure of obstruents’ in fast The notion of ambisyllabicity has been used informal speech, Gussenhoven 1986: 125Ϫ6) in two ways: as an environment for syllable- can operate on the output of aspirated conso- based processes and as a means of providing nants. However, weakening only operates on a coda to add weight to a syllable. The most ambisyllabic aspirated consonants, and those frequently discussed phenomena where am- that are in absolute onset position are ex- bisyllabicity plays a role are aspiration and empt. flapping in English (cf. Kahn 1976; Gussen- As we mentioned above, ambisyllabicity hoven 1986). Both processes are governed has also been argued to play a role in assign- by surface syllable structure, and hence are ing syllable weight (cf. van der Hulst 1985 stress-sensitive. The ambisyllabicity results and Lahiri & Koreman 1988 for Dutch; from the attraction of the first consonantal Ghini 2001 for Miogliola, a northern Italian onset of an unstressed syllable to form a coda dialect). Under these analyses, ambisyllabi- of the preceding syllable. This consonant then city not only allows a consonant to be part becomes ambisyllabic, since it belongs both of an onset in one syllable and a coda in the to the onset and the coda of two syllables. other, the coda consonant also projects a 1352 XII. Phonology-based typology mora. The representation would look as fol- bles are considered to be grouped into metri- lows: cal feet consisting of strong and weak sylla- bles. The feet differ in terms of whether the (12) Ambisyllabicity and weight head of the foot, i. e. the stressed position, occurs at the left or right edge. A left headed foot is known as a trochee, and a right headed foot is an iamb. We will not however, begin with the assumption that syllable weight and feet are crucial elements in stress assignment. Instead, with examples from two languages, we will trace step-by-step the After ambisyllabicity, the phonological repre- motivations for assuming (a) syllables are a sentation is identical to that of a geminate. necessary constituent for assigning stress, (b) However, in languages in which ambisyllabi- syllables are not enough to account for uni- city is invoked for syllable weight, there are versal stress systems, (c) a fixed inventory of no contrastive geminates (cf. also Borowsky, foot types built on syllables can delimit all Itoˆ & Mester 1984). Whether one could then stress patterns and (d) the weight of syllables assume that all such ambisyllabic consonants play a role in building feet. could be treated as geminates is a much de- bated topic. 3.1. Are syllables necessary for stress? The notion of ambisyllabicity has not Chomsky & Halle (1968) accounted for Eng- found favour with many researchers particu- lish stress using only a linear sequence of larly because of the dual linking of a single consonants and vowels, ignoring any hier- consonant to two syllables (Kiparsky 1979; archical constituent like the syllable. How- see also Blevins 1995 for a discussion). How- ever, referring only to a linear sequence is not ever, the arguments in Gussenhoven (1986) enough. Let us consider the facts of the well are very persuasive and since dual linking has known Latin stress rule, which has been dis- to be permitted for geminate consonants that cussed in metrical terms for a long time. The in itself is not a sufficient argument against length of a vowel is indicated with a macron. ambisyllabicity. So far we have focused on two different (13) Latin stress aspects of syllable structure: syllable as a 3rd vowel 2nd vowel context for phonological rules and strategies from end from end to maintain preferred syllable structure and mı´nimus refe´ctus incu˘dis syllable weight. We now move on to discuss mu´rmuris volu´ptas rela˘tus the role of syllable weight and its interaction exı´stimo de¯le´ctat inimı˘cus with metrical stress. adsimı´liter exce´rpsit refe˘cit If asked which vowels are stressed without 3. The foot taking recourse to syllables, one would come up with the rule in (14): A fundamental insight in metrical theory is that syllable weight plays a crucial role in (14) Latin stress rule based on vowels stress assignment. As we mentioned above, (i) If the penultimate vowel is long, it is the weight of a syllable usually depends on stressed. whether it has a long bimoraic vowel or (ii) If the penultimate vowel is followed whether the coda of a closed syllable con- by two consonants, it is stressed. tributes a mora to the syllable. Vowel quality (iii) Else, the antepenultimate vowel is is never taken into account where syllable stressed. weight is concerned. It would be very odd in- Such a rule, however, would cause problems deed, if for instance, all front vowels were for the following words. treated as heavy while other vowels including long vowels were light. However, syllables are (15) Problem cases not sufficient to account for stress assign- 3rd from left Expected 2nd from left ment in languages of the world. Recent theo- te´nebras *tene´bras ries of metrical stress argue that the foot, vo´lucres *volu´cres which is a constituent built on groups of syl- ma´niplis *manı´plis lables, accounts for stress. In general, sylla- la´tebras *late´bras 96. Metrical patterns 1353

If we take the second part of the stress rule acolakı´ ‘old timers, elders’ which says that if the second to last vowel osa´hwa ‘crow’ is followed by two consonants it should be ahicita´ ‘one to look after’ stressed, we have incorrect results. The solu- ca´:lo ‘trout’ tion lies in the evidence from syllabification so´kca ‘sack’ following the sonority hierarchy. The above famı´:ca ‘canteloupe’ words are syllabified as [te.ne.bras], [vo.lu. cres], [ma.ni.plis] and [la.te.bras], as against If we assume that all consonant clusters are [re.fec.tus], [vo.lup.tas] etc. Thus, the stress broken up into onset and coda, and that rule can be simplified as follows: If the pe- closed syllables are heavy, Creek stress can be nultimate syllable has a short vowel with no described as follows: coda then the antepenultimate syllable bears (18) Creek stress Ϫ first approximation the main stress; otherwise the penultimate (i) If the penultimate syllable is heavy, it syllable is stressed. is stressed. This description is very characteristic of (ii) Else, the last syllable is stressed. stress rules. Syllables with long vowels or with a coda consonant pattern together. And So far, the Creek data look quite similar to now we come to syllable weight. As we have Latin, the only difference being that in Creek seen before, this division is descriptively if the penult is not heavy the last syllable is characterised as heavy syllables versus light stressed. Now let us consider a few more syllables. The Latin stress rule can then be words. stated as: (19) Further data in Creek (16) Latin stress rule in terms of syllable ifo´ci ‘puppy’ weight imahicı´ta ‘one to look after for (i) If the penultimate syllable is heavy, it (someone)’ is stressed. itiwanayipı´ta ‘to tie each other’ (ii) Else, the antepenultimate syllable is acahankatı´ta ‘one to count me’ stressed. In the above words, the penultimate syllable 3.2. Are syllables enough for stress? is light, but nevertheless it is stressed. Ac- So far we have seen that a linear string of cording to our preceding assumption, the last consonants and vowels is not enough to syllable should have borne stress. Perhaps account for stress. Instead we require the we can salvage the analysis by the following notion of syllables, and particularly syllable statement: weight. But is syllable weight enough to cap- (20) Creek stress Ϫ second approximation ture stress facts of various languages? That (i) If the penultimate syllable is heavy, it is, is it always the case that stress assignment is stressed. can be characterised in terms of heavy and (ii) If the penultimate syllable is light, light syllables? The answer is no. Let us look Ϫ stress the final or penultimate which- at a more complicated case Creek, a ever is even-numbered, counting left Muskogean language. The data in this paper to right. comes from Haas (1977). Haas describes Creek as having tonal accent, which falls on Unfortunately this does not solve the prob- a ‘key syllable’ (p. 195). There can be more lem. Consider the following words: than one ‘key’ syllable, each one being ‘one step lower than the preceding’ one (p. 196). (21) More data from Creek The tones themselves can be level, falling or aktopa´ ‘bridge’ rising. Assuming that the key syllables are the wa:kocı´ ‘calf’ prominent syllables indicating main and sec- hoktakı´ ‘women’ ondary stresses, Creek provides us with a rich inkosapita´ ‘one to implore’ source of data. Further data is given in Hayes Clearly, our previous rules will not suffice. In (1995: 64Ϫ65). Consider the following facts: all the words, the penult is light, but the syl- (17) Creek data (length is indicated with lable that is stressed is not even-numbered the diacritic [:]) counting left-to-right. What we need to do is poco´swa ‘axe’ not to start counting from the beginning of cofı´ ‘rabbit’ the word, but from the rightmost heavy sylla- 1354 XII. Phonology-based typology ble. The stress rule could then be described (24) Trochees and Iambs as follows: (a) Syllabic trochee (weight insensitive) Ϫ (x .) (22) Creek stress third approximation ss (i) If the penultimate syllable is heavy, it (b) Moraic trochee: left headed (con- is stressed. structed over two light syllables or (ii) If not, examine the maximum string one heavy syllable) of light syllables at the end of the (x .) (x) word. ss s (iii) Within this string, stress the right- tt ͉ ͉ most even-numbered syllable count- mm mm ing left-to-right. (c) Iamb: right headed (construed over We go through two examples following the two light syllables, a light plus a steps elaborated above. heavy syllable, or one heavy syllable) (. x) (. x) (x) (23) Deriving Creek stress Ϫ third ap- ss s s s proximation tt t͉ ͉ ͉ ͉ inkosapita´ acahankatı´ta mm mmmmm (i) ϪϪ (ii) in (ko sa pi ta) acahan (ka ti ta) Although the moraic trochee and the iamb (iii) * * are both weight sensitive (i. e. the weak 1234 123 branch cannot be heavier than the strong in (ko sa pi ta´) acahan (ka tı´ ta) branch), under Hayes’ analysis these two feet are asymmetric. Under this system, an iamb This description is not particularly illuminat- may have a [L(ight) H(eavy)] sequence, but a ing. Clearly we are missing a generalisation. trochee is not permitted to have a branching If the penult is not heavy, it is not syllable head. The way stress assignment works is as counting that gives us the right answer, but follows. A string of syllables are parsed into some constituent which groups syllables to- feet going from left-to-right or right-to-left. gether. Such a constituent in poetic meter is The last foot on the left or the right is assigned known as a foot, and in the next section we main stress: End Rule (left/right). Thus, main discuss the universal inventory of feet that stress is always at an edge of a word, edge have been suggested for natural language. being defined by foot structure and not by syl- lables or vowels. To assign stress, we therefore 3.3. Inventory of feet require the following parameters: Hayes (1995) argues that there are three basic (25) Stress assignment foot types used in linguistic systems univer- (a) Foot type sally: a syllabic trochee, a moraic trochee, and (b) Direction of parsing an iamb. The syllabic trochee groups any two (c) End Rule syllables together regardless of their weight. A moraic trochee and an iamb are weight We illustrate this first with the most straight- sensitive. These three foot types are given forward foot type, namely the syllabic tro- below. The foot is demarcated between pa- chee, which is weight insensitive. The syllabic rentheses and the strong and weak branches trochee groups syllables together regardless are indicated by a [x] and a dot [.] respec- of their internal structure. The analysis is tively. from Hayes (1995: 62Ϫ63).

(26) Syllabic trochee (Pintupi, a Pama-Nyungan language of Australia) Foot construction: Left to Right Main stress: End Rule Left (indicated with X) (X ) (X ) (X ) (X ) (x .)(x .) ( x .)(x .) (x .)(x .)(x .) (x .)(x .)(x .) ssss sssss sss sss sssss ss ma´lawa`na pu´linka`latjutja´mulı`mpatju`nku t´ılirı`nula´mpatju ‘through from behind’ ‘we (sat) on the hill’ ‘our relation’ ‘the fire for our benefit flared up’ 96. Metrical patterns 1355

Let us now turn back to Creek and investi- not exist for footing and, therefore, not for gate which of the foot types would be appro- stress. Thus, there are four parameters to be priate to account for the entire set of data. taken into account: foot type, extrametri- Clearly syllable weight plays a role since the cality, direction of foot parsing and the end penultimate syllable is stressed only when it rule. In the two following examples taken is heavy. The End Rule appears to be on the from Hayes (1995) we see instantiations of right, since stress falls always towards the the moraic trochee with and without extra- right edge of the word. Now we need to de- metricality (cf. (28)). termine the direction of parsing and whether This brings us back to Latin stress. Recall the foot type is a moraic trochee or an iamb. that in Latin, stress fell on the penultimate The syllabic trochee cannot be considered syllable if it was heavy. Otherwise the ante- since it is quantity insensitive. The decision is penultimate syllable bore stress regardless of not a difficult one since in our third approx- weight. We can now analyse Latin in the imation we saw that when a sequence of following way (cf. (29)). light syllables occur at the end of a word, the The moraic trochee along with the extra- rightmost even numbered syllable can get metricality does away with the oddity of the stressed. As a result final light syllables may syllable based description which required that bear stress and this is not possible for a tro- syllable weight was responsible for attracting chee. Thus, if the foot inventory is indeed suf- stress on to the penultimate syllable but not ficient, then the foot type must be an iamb. for the antepenultimate syllable. The antepe- The final decision regarding stress assign- nult could be stressed regardless of syllable ment must be the direction of parsing. Again weight; it depended on the lack of weight of in the last approximation, if the penult was the penult. In the foot based analysis, the ex- not stressed, the grouping of syllables into a planation rests on the fact that the antepenult larger constituent began after the last heavy and the penult together can make up a single syllable. Hence, the parsing must be from foot if both are light. left-to-right. Following Hayes (1995), apply- The inventory given above excludes the ing these parameters to Creek we obtain the possibility of asymmetric moraic trochees following structures: which are the mirror images of iambs. How-

(27) Stress assignment in Creek: final version Syllable weight: Long vowels and closed syllables are heavy Foot type Iamb Foot construction: Left to Right Main stress: End Rule Right (X)(X) ( X) ( X) (X ) (. x) (. x) (. x)(. x) (. x)(. x) (. x) m m mmm mmmm mmmmm mmmm co fı´ ifo´ ci acolakı´ i mahi cı´ ta fa mı´:ca ( X) ( X) ( X ) (x)(. x) (x)(. x)(. x) (. x)(x) (. x) mmmm mmmmm m mmmmmmm wa: ko cı´ in ko sa pi ta´ acahan katı´ ta

Thus, although syllables may provide an ade- ever, others like Dresher & Lahiri (1991), quate description for stress patterns in some Lahiri & Dresher (1999, for Germanic), Ja- languages, they are not sufficient to account cobs (1989, 2000, for Latin), Kager (1989, for for the complicated systems like Creek. Once English) have claimed that asymmetric we introduce a larger constituent grouping trochees incorporating [H L] sequences are syllables together into feet, the analysis of the necessary as well. For instance in Latin, using stress pattern becomes very simple. an asymmetric trochee would mean that In addition to the three basic foot tem- words like murmuris would be parsed into plates, it is necessary to invoke the notion feet as ([mur. mu] *ris+). The traditional tro- of extrametricality to understand some other chee is, in fact, asymmetric (cf. Hayes 1981). stress patterns. Syllables or segments (usually Although we are not in a position here to ex- consonants) at right edges are often extra- haustively compare these various proposals, metrical; that is, they behave as if they do while discussing the foot based phonological 1356 XII. Phonology-based typology

(28) Moraic trochees with and without extrametricality Moraic trochee (with extrametricality) Cairene Arabic (No classical words are considered) Syllable weight: Long vowels and closed syllables are heavy Extrametricality: Final consonant of a word (indicated by *+) Direction of Parsing: Left to Right Main stress: End Rule Right (X ) ( X) ( X ) ( X ) ( X ) ( X ) (x) (x) (x) (x) (x) (x.) (x .)(x .) mmm mmm mmm mmmm m mmmm mmmm be:ta´*k+ gato´:kata´b*t+ muda´rri*s+ mudarrı´si*t+ katabı´tu ‘your (m.sg.) house cake’ ‘I wrote’ ‘teacher’ ‘teacher (f. construct)’ ‘she wrote it (m.)’ Moraic trochee (Wargamay: Pama-Nyungan language of Australia) Syllable weight: Long vowels are heavy Direction of parsing: Right to Left Main stress: End Rule Left (X) (X ) (X) ( X ) (x) (x .)(x .) (x .) (x .) (x .) mmm mmmm m mm mmm mm mu´:bag´ıJ-awu`lu gaga´ra J-ura´gaymı`ri ‘stone fish’ ‘freshwater jewfish’ ‘dilly bag’ ‘Niagara-Vale-from’

(29) Latin stress revisited Syllable weight: Long vowels and closed syllables are heavy Extrametricality: Final syllable Direction of Parsing: Right to Left Main stress: End Rule Right (X ) ( X ) (X ) (X ) (x) (x) (x) (x .) (x) mmmmm mmmmmm mmmm mmmmm re fe´c *tus+ de¯ le´c *tat+ vo´ lu *cres+ mu´rmu*ris+

(30) Constraining gemination in OE Gemination blocked: (x .) (x .) ([mm] m) [mmm] m ([m m] m) [m mm] m H L LLL wı¯ tje Ͼ *wı¯tte æ pe lje Ͼ *æ pel le ‘punishment dat. sg.’ ‘noble dat. sg.’ Gemination permitted (x .) (x) (x .) (x ) (x .) ([mm] m) m ([mm]) ([mm] m) ([m m]) ([mm] m) HLL H H L LL H L we¯ ste nje Ͼ we¯ sten ne cy nje Ͼ cyn ne ‘desert dat. sg.’ ‘race dat. sg.’ processes, we will draw on evidence from seg- branch. An example of a foot based process mental rules and stress in Germanic to pro- is West Germanic gemination, which is sim- vide support for asymmetric trochees. ply a process by which all consonants are doubled when followed by a front glide /j/. It 3.5. Foot-based phonological processes is constrained only when the head becomes Phonological processes can be sensitive to the trimoraic. The following Old English nouns foot. The foot relevant for Germanic was a illustrate gemination, where the head of the resolved moraic trochee, which is essentially foot is circumscribed by square brackets (cf. an asymmetric trochee, where the head must (30)). 96. Metrical patterns 1357

The form *wı¯tte is impossible because the Vowels are often lengthened to meet a head cannot be trimoraic. Similarly, *ATelle minimal word requirement. In Bengali, for is disallowed because the weak branch of the instance, a vowel in a monosyllabic word is head is strengthened and again the head be- always lengthened unless the vowel nucleus comes trimoraic. Not just strengthening pro- has a diphthong (Fitzpatrick-Cole 1994). cesses, but deletions can also be sensitive to (32) Vowel lengthening in Bengali foot structure. For instance, high vowels in a. tsa: ‘want, ask for 2p.familiar Old English were deleted in the weak branch imperative / tea’ of a foot (Dresher & Lahiri 1991). In the b. tsaϩi Ͼ tsa«i ‘want, ask for ϩ1p. present’ following examples the underlined vowels are c. tsa:ϭi ‘teaϭonly’ deleted (cf. (31)). d. na:k ‘nose’ The [u] in lofu is not deleted because it is e. nakϩi ‘noseϩadjectival suffix / within the head. In contrast, in wordu and nasal’ fAreldu, the [u] is in the weak branch of the f. na:kϭi ‘noseϭonly’ foot, and hence is deleted. Note that a tri- moraic head in fAreld is permitted because The morpheme /tsa/ can be both a verb root there was no choice to begin with. However, ‘to want’ or the noun ‘tea’. The ‘ϩ’ boundary a process like gemination is prohibited from indicates a suffix while the ‘ϭ’ sign marks a creating one as we saw in the case of *ATelle. clitic. The different suffixation and cliticised The foot also accounts for stress: the head of forms show the vowel length alternation. It the foot in each word bears main stress. should be noted that Bengali does not have contrastive vowel length. At first glance the 3.6. Minimal word and the foot lengthening of the vowel in (32 c, f) seems The phonological word is the next constitu- to be a counterexample to the minimal word ent above the foot. Just like the syllable and requirement. In fact, (32e) shows that a de- the foot, most languages adhere to con- rived word which is disyllabic does not straints which try to maintain a ‘minimal lengthen a vowel similar to the monosyllabic word’. Most languages have a minimal word word in (32b). However, the final vowel in requirement which is closely related to a foot. (32f) is not a suffix but a clitic, just as in The minimal word must be at least a foot, (32c). Clitics are added to a word and hence, or two syllables, or bimoraic, or some other the minimal word requirement must be met prosodic constraint. Our interest here is pri- before the clitic is added. We see a difference marily on the correlation between minimal in (32b) and (32c) where the former is a suf- word requirements and metrical coherence. fixed word and the resulting diphthong satis- We will, therefore, briefly illustrate the role fies the minimal word requirement. In (32c), of the minimal word in prosodic phonology however, the final vowel is a clitic and again, and morphology. the initial vowel is lengthened.

(31) Foot based syncope in OE (X ) (X ) X ) (X ) (x .) (x .) (x .) (x .) ([mm] m) m ([mm] m) ([m mm] m) ([mm] m) m H LL H L LHL HLL he¯afuß de wor duß fæ rel duß clı¯ we nu he¯afde word færeld clı¯wenu ‘head’ ‘word’ ‘journey’ ‘ball of thread, clew’ dat. sg. nom.pl. nom.pl. nom. pl. (X ) (X ) (X ) (X ) (x .) (x) (x .) (x ) (x ) ([mm] m) [mm] ([m m] m) ([m m]) ([m mm]) H LH LLL LL LH he¯afuß des we ru duß lo fu su num he¯afdes werud lofu sunum ‘head’ ‘troop’ ‘praise’ ‘son’ gen.sg. nom.pl. nom.pl. dat.pl. 1358 XII. Phonology-based typology

These facts are not unusual. Many lan- In the last two examples, the base forms are guages, including Germanic languages like not disyllabic since the suffixation shows that English, Dutch and German, also have min- they are consonant final stems (cf. the exam- imal word requirements. No content word ples above). However, to meet the minimal of these languages can end with a single lax word requirement, a final vowel is added to vowel: *[si] or *[bi] would be completely the base stem or the uninflected form to en- impossible words in these languages. They sure that it surfaces as disyllabic. should either have a long vowel as in sea/see [si:], or should be closed as in sit [sıt]. The 3.7. The foot and typological premises research of McCarthy & Prince (1990) de- How do the above analyses fit into the usual monstrates that the minimal word plays an typological premises made when referring to important role in the prosodic morphology stress? We have argued that stress is not a of languages. In their discussion of this phe- feature on a vowel, but rather is the linguistic nomenon in Arabic, they give examples of a manifestation of rhythmic structure. As such, small number of nouns (usually related to although one could state that a given syllable body parts and kinship terms) which disobey in a word bears the main stress, this is not the bimoraic, minimal word requirement (fi- the best way to account for stress rules. An nal consonants are extrametrical and hence alternative and better way is to construe do not count for weight): [?ab] ‘father’, (?ax] stress placement as the parsing of a word into ‘brother’ etc. However, when these nouns metrical feet. This does not preclude the pos- serve as the basis of regular word formation sibility that there are languages with fixed processes, they acquire an extra consonant, main stress either on the initial syllable or on thus fulfilling the minimality requirement: the final syllable. Out of 300 languages, Hy- [?ab] ‘father’, but [?abaw-iy] ‘paternal’; cf. man (1977) noted that 114 languages have [masør] ‘Egypt’, [masør-iy] ‘Egyptian’. initial stress, 97 final stress, 77 penultimate Minimal word requirements are also often stress and only 12 have stress on the second reflected in blocking the application of rules syllable. Such a statement, however, says that may shorten a word beyond the mini- nothing about how secondary stress could mum. For instance, Lardil has a disyllabic work. Under a metrical foot analysis, a lan- word minimum. Apocope applies freely to guage with final syllable stress could easily trisyllabic or longer stems, but it is blocked have either a moraic trochee or an iamb, if in disyllables since it would shorten a word all final syllables happened to be heavy and beyond the acceptable minimal word require- parsing was from right to left. Predictions for ment (Kenstowicz 1994). In the following ex- secondary stress, however, would be dif- amples we see that the final stem vowel is al- ferent. Consider the following hypothetical ways deleted in the uninflected form, except example. in the last two words which are disyllabic. (35) Final syllable stress: iamb or trochee? (33) Lardil apocope Parsing Right to Left, Moraic tro- uninflected inflected gloss chee, End Rule Right yalul yalulu-n ‘flame’ (X) mayar mayara-n ‘rainbow’ (x) (x .) (x) (x) karikar karikari-n ‘butterfish’ s˘ s¯ s˘ s˘ s¯ s˘ s¯ mela mela-n ‘sea’ Parsing Right to Left, Iamb, End witøe witøe-n ‘inferior’ Rule Right (X) Minimality constraints can also add a mora (. x) (. x)(. x) or a syllable when the base has less than the s˘ s¯ s˘ s˘ s¯ s˘ s¯ weight required to satisfy the minimum word requirements. Such a process is also evident The final syllable obtains stress in both in- in Lardil (Kenstowicz 1994). stances. But parsing into metrical feet pre- dicts that the third syllable from the begin- (34) Addition of a mora in Lardil ning could bear secondary stress only if the uninflected inflected gloss foot is a moraic trochee, and not if it is an kentapal kentapal-in ‘dugong’ iamb. Thus, broad typological statements yaraman yaraman-in ‘horse’ such as main stress is final, can be mislead- yaka yak-in ‘fish’ ing. This does not mean that stress cannot be øteraø ter-in ‘thigh’ fixed regardless of the type of syllable. Usu- 96. Metrical patterns 1359 ally this happens if stress, or rather the metri- the relevance of metrical constituents outside cal parameters, are in some way morpholog- phonology, (b) the coherence of metrical units ically governed. For instance, with English within a given language, and (c) correlations productive affixation like be´autiful, stress is drawn between metrical constituents and other insensitive to rhythmic patterns. Stress falls phonological and morphological patterns. on the antepenultimate syllable because it happens to be the initial syllable of the stem. 4.1. Metrical constituents outside Another way of looking at it is that in de- phonology rived adjectives the final syllable is extra- Research in prosodic morphology (cf. Mc- metrical. Morphological effects of stress can Carthy & Prince 1986, 1993) indicates that be also found when certain suffixes always metrical categories required in phonology are bear stress as in German -ier, (e. g. Juwelı´er), the same that are necessary for morphol- or English -ee, (e. g. devotee). However, ogical processes like reduplication. Typologi- even with morphologically stress, the window cally, therefore, the implication is that if a within which stress falls is usually constrained. new category is found to be necessary to de- Hence, when typological correlations are scribe either stress or any other phonological drawn with respect to stress, it is worthwhile process, it ought to be found relevant for a to be more precise about the rhythmic organ- morphological process as well. The following isation and the type of foot. We will discuss two examples support the view that both syl- this in more detail in § 4. lables and feet are relevant for morphology. In this section, we have covered a wide In Mokilese, the progressive is expressed range of facts involving metrical structures. by a form of reduplication (Harrison & Al- We have briefly discussed various phonologi- bert 1976; McCarthy & Prince 1986). Sam- cal processes sensitive to metrical structure, ples of the data are given below. including shortening and lengthening pro- (36) Mokilese progressive cesses, segmental alternations, repair strate- a. pc.dok pcd-pc.dok ‘plant’ gies for preferred syllables, etc. The central b. pa paa-pa ‘weave’ goal was to show that along with stress as- di.ar dii-di.ar ‘find’ signment, phonological rules do not only c. scc.rck scc-scc.rck ‘tear’ operate in local segmental contexts, but that caak caa-caak ‘bend’ hierarchical structures like syllables and feet d. an.dip an.d-an.dip ‘spit’ also constrain representations and processes. o.nop on.n-o.nop ‘prepare’ At first glance, it seems as if the progressive 4. Typology and Metrical Structures is formed by some sort of a prefix which is equivalent to a syllable. However, the nature Having established the necessity for metrical of the syllable differs for each word type. The constituents like syllables and feet, we are reduplication is essentially the prefixation of now in a position to address issues of typo- a bimoraic syllable [smm]. In (36a) the initial logical implications. In the preceding sec- syllable of the stem is monomoraic and hence tions, we outlined different types of syllable the following onset is included to form the structures and feet which languages appear prefix. In (36b), where the stem is monosyl- to have. We have not discussed, however, any labic or where the second syllable has no particular correlations between such struc- consonantal onset, the vowel is lengthened to tures nor any possible relationship between satisfy the bimoraic requirement. In (36c) the the existence of different types of metrical bimoraic requirement is met by simply taking structures within a given language. For in- the initial long vowel. The most interesting stance, is it possible that all types of feet co- case is (36d) where [an] or [on] would surely exist in a single language? Would it be pos- have met the bimoraic requirement. How- sible to infer the preferred syllable structure ever, the output form would then have been of a language if its type of foot is known? *[on-onop], which would then be syllabified Are there any implications to be drawn from as *[o.no.nop] and the initial prefix would be preferred metrical structures of language and monomoraic. Thus, the constraint is that the the types of rules that are permitted? Such morphological prefix must be an entire bi- questions are rarely if at all addressed within moraic syllable, and simultaneously the onset phonology. We will draw attention to three of the next syllable must be maximised. For possible typologically interesting issues: (a) the latter, if the stem begins with a vowel, 1360 XII. Phonology-based typology either the onset of the second syllable is used metrical coherence to morphology as well. As even if it has to be doubled. Hayes (1995) points out, usually in any given In the next example we see that a morpho- language, the kind of foot used for stress is logical process can be sensitive to a foot. The the same as that used in morphology. We will data is from Ulwa (McCarthy & Prince address this issue first from a purely phono- 1990). logical perspective and then briefly address the notion of coherence within prosodic pho- (37) Ulwa construct state (3 sg. possessed) nology and morphology. a. kii (kii)-ka ‘stone’ bas (bas)-ka ‘hair’ 4.2.1. Coherence in stress and sana (sana)-ka ‘deer’ phonological processes sapaa (sapaa)-ka ‘forehead’ amak (amak)-ka ‘bee’ There is nothing discussed in the earlier sec- b. suulu (suu)-ka-lu ‘dog’ tions that suggests that metrical coherence is baskarna (bas)-ka-karna ‘comb’ a must or even preferred. One could imagine siwanak (siwa)-ka-nak ’root’ that metrical structures are counting devices anaalaaka (anaa)-ka-laaka ‘chin’ such that one type of foot is used for stress karasmak (karas)-ka-mak ‘knee’ and another could be used as the context for other phonological rules. For instance, Key- The data in (a) is straightforward Ϫ a suffix ser and O’Neill (1976) suggested that in Old [ka] is added to the stem. However, the data English, stress required an initial quantity in (b) shows that the [ka] behaves like an in- insensitive left-headed foot while a quantity fix and the way in which the stem is divided sensitive right-headed iamb was necessary to up appears to be different in each case. In the account for the rule of high vowel deletion, first two examples, the [ka] is added to the where R ϭ rhyme first syllable, while in the others it is added after two syllables. If we compare the two (38) Feet in Old English following Keyser sets, we can see that [ka] is added after two and O’Neill syllables only when the first syllable is light a. Word stress: left-headed unbounded and the second is heavy or light Ϫ [a.naa] Ϫ foot but not, when the first syllable is heavy and the second is light as in [suu.lu]. Thus, it is not a syllable which is the relevant constitu- ent but a foot and a typical iamb [smsmm]. The morphological process is thus the following: add the suffix [ka] to the leftmost iambic b. High vowel deletion: following a foot. right-headed quantity sensitive foot Thus, the metrical constituents including in an open syllable syllables, feet and minimal word that are used for the description of stress and are sensitive to other phonological processes are argued to be the same for morphological processes. The obvious question that now comes to mind is whether within a given language, the same metrical constituent is used both for stress and other phonological processes, and in ad- Under such an analysis, headedness is com- dition whether morphology and phonology pletely arbitrary. For the purposes of stress, share precisely the same type of constituent the left head of a foot is the strongest, while as well. We now turn to these issues. for syncope, the right head of the foot is strong. However, such a system would be 4.2. Metrical coherence incredibly difficult for the language learner to Typically languages adhere to the same foot acquire. As shown in § 3.5., this analysis is for stress as well as for other phonological not the only one that could account for the processes. A language following such a prin- data. Germanic is essentially metrically co- ciple would be judged as being metrically herent, and both stress and deletion of high coherent and arguably such a language would vowels can be accounted for by a single foot also be easier to learn (cf. Dresher & Lahiri type. Other languages have similar proper- 1991). One could also extend the principle of ties. One example is Unami, an Eastern Al- 96. Metrical patterns 1361 gonquian language described in Hayes (1995: cess of reduplication is described in McCar- 211Ϫ213) based on Goddard (1979). Hayes thy & Prince (1986: 39). argues that stress in Unami is accounted for (40) Manam Stress and Reduplication by forming iambs left to right with foot ex- a. mo´.tu ‘island’ trametricality and End Rule Right. More- ma.na´m ‘Manam island’ over, voiceless consonants other than [h] are wa.rı´.ge ‘rope’ geminated after a strong vowel which would ma.la.bo´n ‘flying fox’ be the head of a foot. ?i-po.a.sa.ge´.na ‘we are tired’ (39) Unami stress and gemination b. la´?ola?o- la´?o ‘go’ (X) moı´ta mo-ita-ı´ta ‘knife’ (. x) *(. x)+ malabo´n mala-bom-bo´n ‘flying fox’ ne me t eme6 ‘I follow a trail’ J Stress is assigned by a moraic trochee parsed from right to left. The reduplication also (X) clearly refers to the same foot. However, there (. x) *(. x)+ is nothing inherent about metrical structures ne me tt eme6 that requires the same foot to be used for Compare/meteme6w/ J [mete´me6(w)] phonology and morphology. One glaring ‘he follows a trail’ counterexample in the literature is Axininca In general, phonological rules sensitive to Campa. While stress is iambic (Payne 1981; foot structure are usually deletion, weaken- Spring 1990a; McCarthy & Prince 1993), ing or strengthening processes as we have there is disagreement in the literature about seen in earlier sections. In most instances, the foot type(s) needed for morphology they apply to either repair metrical structures (Spring 1990a, 1990b; McCarthy & Prince or are invoked to build preferred structures. 1993). For instance, Spring (1990b) claims Therefore it is not surprising that the foot re- that while the foot necessary for genitive allo- quired for stress and other phonological pro- morphy is the moraic trochee, the foot for cesses would be the same. What is more inter- verb reduplication is the iamb. esting is the notion of metrical coherence out- However, other analyses of Axininca side phonology to which we turn next. Campa suggest that the situation is not as complicated as suggested by Spring. While 4.2.2. Metrical coherence and morphology the distinctive base of both genitive allomor- As we mentioned earlier, the set of metrical phy and reduplication is characterised by a constituents that are relevant for phonology bimoraic foot, the reduplicant is a different are the same that are used in prosodic mor- constituent with a strong tendency towards phology. For instance, often, suffix allomor- disyllabicity. Metrical coherence may be main- phy is governed by a metrical unit. In Dutch, tained for Axininca Campa if the bimoraic the nominalising suffix {-aar} adds only to bi- foot is regarded as ‘minimal iamb’, while the syllabic verbs, while {-er} is added to mono- disyllabic foot is considered to be a ‘maximal syllabic verbs: luister ‘to listen’ Ͼ luister-aar iamb’. Only a small sample of the crucial ‘listener’; bel ‘to ring’ Ͼ bell-er ‘ringer’. Booij data is presented to show the analysis (cf. (1997) argues that the underlying reason for (41)). this allomorphy is to maintain a proper The descriptive generalisation for the geni- trochee on the surface. The form *luisterer tive allomorphy is that if the stem contains would not form a proper trochee while only two moras, the suffix /ni/ is taken, other- luisteraar ends up as two acceptable feet. wise the suffix is /ti/. This led Spring (1990b) Our interest here is whether the type of to suggest that the genitive allomorphy was foot necessary for a language’s morphological sensitive to a trochee. However, a bimoraic system is the same as that required for stress. foot could be both a trochee as well as an That is, can we extend the notion of metrical iamb, the difference being the headedness. coherence to morphology as well. It appears Since there is no evidence of headedness in that almost always the foot used for stress is this context, the analysis could just as well the same as that used for prosodic morphol- be that if the foot is a minimal iamb, the ogy. One example is Manam (Austronesian suffix is /ni/. The verb reduplication is more language, spoken in New Guinea). The de- straightforwardly an iamb. If the non-pre- scription of stress is given in Kenstowicz fixed stem is a single syllable, the reduplica- (1994: 614, 659) and the morphological pro- tion includes the prefix. That the reduplicant 1362 XII. Phonology-based typology

(41) Axininca Campa a. Stress (following McCarthy & Prince 1993, Spring 1990a, 1990b): Iamb, Left to Right, Final light syllable extrametrical. (ki.mı´).ta.*ka+ ‘maybe’ (o.cı´).(to.mo´).*ko+ ‘monkey’ (i.ra´a).(wa.na´).*ti+ ‘su caoba’ b. Genitive allomorphy (data Spring 1990a) /sima/ no-sima-ni ‘my fish’ /mii/ no-mii-ni ‘my otter’ /sawoo/ no-sawoo-ti ‘my cane’ /maini/ no-maini-ti ‘my bear’ /cokori/ no-cokori-ti ‘my armadillo’ c. Verb reduplication-prefixed pattern (Payne 1981) /kiNtha/ non - kintha - kintha ‘tell’ /kawosi/ non - kawosi - kawosi ‘bathe’ /naa/ no - naa - nonaa ‘chew’ /na/ no - na - nona ‘carry‘ /osaNpi/ n - osampi - sampi ‘ask’ /apii/ n - apii - napii ‘repeat’ is a canonical iamb is seen in the last exam- ture, pitch accent and mora-counting, while ple. Normally, the stem initial vowel is not flective languages have complex syllable struc- reduplicated as in [n-osampi-sampi]. How- tures, stress accent with reduction of un- ever, when the stem itself is disyllabic as in stressed syllables, and syllable-counting. Plank /apii/, the reduplication does not ignore the (1998) provides a detailed account of the var- initial vowel. The reduplicated form is [n- ious attempts in the literature to draw corre- apii-napii] and not *[n-apii-pii] showing that lations within phonological constituents as there is strong preference to have a canonical well as between phonological and morpho- iamb if possible (Black 1991; Loewe 1996). logical categories. The results are highly un- Thus, even in a complex set of interactions satisfactory, and as Plank points out, often as in Axininca Campa, there is no clear evi- contradictory. Our interest here is primarily dence that entirely different feet are required on metrical constituents and we will discuss for morphology and phonology. Since in most briefly some of the proposals put forward in instances, morphology requires either a disyl- the literature. labic or bimoraic foot with no clear headed- Donegan & Stampe (1983) argue that ness preferences, or just a minimal word (cf. rhythmic properties determine different mor- § 3.6.), the stress facts do not clash with mor- phological types and syntactic word order. phological processes (cf. McCarthy & Prince The central notion is that since accent is the 1990 for various examples). It seems, there- only factor which is pervasive through all fore, that metrical coherence can be extended levels of language, it is the only meaningful to morphology as well. determinant in connecting the different levels of language. For instance, there is a direct re- 5. Universals, implications and lationship between word order and phrasal correlations accent. In a sentence, the operand or head is given information, while the operator or In the typological literature, little has been modifier is asserted and hence bears the main said concerning the details of metrical con- accent, regardless of the relative order of the stituents and implications thereof. The focus two parts. They make a very strong claim of attention has been either to associate that rising (final) vs. falling (initial) phrase morphological types like agglutination vs. accent is the primary variable and that oper- flection with stress-timing or syllable-timing, and/operator order follows from it. In fact, or with general rhythmic patterns like iam- primary phrase accent determines more than bic and trochaic with possible syllable types just word order: syllable and word canons, and types of clusters. For example, Lehmann phonological segments, as well as timing are (1973, 1978) claims that agglutinating lan- closely related. For instance, initial phrase ac- guages correlate with simple syllable struc- cent correlates with trochaic word accent, 96. Metrical patterns 1363 syllable-timing or mora-timing, a preference phrasal rhythm as not being a dependable for (C)V(C) syllables and geminate clusters. variable to support typology. The difference Final phrase accent and iambic word accent between Donegan & Stampe and Nespor et goes hand in hand with stress-timed lan- al. is that the former connect phrasal rhythm guages, with (C)V or (C)(C)V(Glide)(C) syl- with word rhythm, while for the latter word lables and non-geminate clusters. The stress is an independent variable. trochaic pattern goes with agglutinative mor- Second, a correlation is made between word phology while the iambic pattern goes with rhythm and other phonological constituents more flective morphology. like syllable structure, stress/syllable timing, Gil’s approach to prosodic typology is and morphological structure. Here, as we also based on rhythm (Gil 1986), but it seems mentioned earlier, Donegan & Stampe, and to make somewhat opposite predictions. Al- Gil sometimes make opposite predictions. though Gil agrees with Donegan & Stampe For instance, Donegan & Stampe associate that trochaic rhythm patterns with syllable iambic rhythm with complex syllable struc- timed languages and iambic rhythm with ture while Gil has it the other way around. stress timed languages, he supports the view Other authors who have attempted these that agglutinative languages prefer iambic correlations do not always agree either (see rhythm, are stress timed, have a high conso- Plank 1998: 216 for details). What then do nant-vowel ratio, and have a simple syllable these typological correlations mean? We will structure. In contrast, flective languages pre- concentrate on the correlations drawn on the fer trochaic rhythm, are syllable-timed, have basis of word rhythm since this has been the complex syllable structure, and have a low primary focus in this paper. consonant-vowel ratio. Thus, the correlation The authors (Donegan & Stampe, and Gil) between agglutinative/flective morphology are basing their hypotheses and analyses on with trochaic/iambic rhythm is the opposite. samples of data from which they have ob- Gil (1987) however, lays less emphasis on the served certain patterns. But the conclusions typological prominence of phrasal rhythm, do not appear to be based on detailed prop- since, he states, most languages tend to have erties of metrical constituents as we have dis- iambic patterns on the phrase and clause cussed so far. It is true that if a language pre- level. What he had in mind is perhaps that fers an iambic foot, it is almost certain that an intonation phrase is invariably divided up it would have long vowels since an iamb is into head+nucleus where the nucleus con- by nature asymmetric in quantity. However, tains the most important information (cf. there is nothing to prevent a language from Hayes & Lahiri 1991). However, it is not the preferring an iambic foot where the quantity case that within the nucleus, languages always is determined solely on the basis of closed vs. prefer iambic patterns. Banking on the differ- open syllables. But more important, prefer- ence between iambic and trochaic rhythm ring an iambic word rhythm, does not neces- within phonological phrases, Nespor, Guasti sarily mean that the language prefers stress & Christophe (1996) argue that language to be at the right edge of a word, which is learners correlate this difference with the what the correlations seem to imply. Whether branching nature of syntax. Thus, a trochaic the right or the left edge of a word bears rhythm correlates with left-branching struc- main stress depends on the End Rule and not tures and an iambic rhythm with right- only on the foot type. Consider once again a branching structures. The syntactic branch- hypothetical example. ingness correlates with heaviness. Crucially, (42) Iambs: End Rule Right/Left the authors make no attempt to link word Language A stress with phrasal stress, the former being an Parsing Right to Left, Iamb, End independent variable. Rule Right In sum, in the attempts to draw corre- (X) lations between rhythm (which is of interest (. x) (. x) (. x) to us since it is related to metrical structures) s˘ s¯ s˘ s˘ s¯ s˘ s¯ and other phonological, morphological, and Language B syntactic structures, there are two major Parsing Right to Left, Iamb, End approaches. First, a correlation is established Rule Left between phrasal rhythm and syntactic struc- (X ) ture (Donegan & Stampe 1983, Nespor, Gu- (. x) (. x) (. x) asti & Christophe 1996), but Gil (1987) views s˘ s¯ s˘ s˘ s¯ s˘ s¯ 1364 XII. Phonology-based typology

If each parsing represented a language, then guage requires iambs for stress, then vowel for the given sequences of light and heavy lengthening or gemination rules would apply syllables, in Language A the main stress to convert a minimal iamb [smsm] to a canon- would fall on the word final syllable, while in ical one [smsmm]. Such iambic lengthening Language B, main stress would fall on the examples are frequently mentioned in Hayes second syllable of the word. Although these (1995). Similarly, as we saw in § 2., epenthesis are hypothetical examples, they are by no or deletion processes are often invoked to ob- means exceptional. Consider the examples tain preferred syllable structures. One could given in (28). Both Cairene Arabic and War- construct a set of strategies that correlate gamay have a moraic trochee, but the direc- with preferred metrical structures as in (44). tion of parsing as well as the End Rule are (44) Correlations of preferred metrical different. As a result, although two and three structures and repair strategies syllable words look very similar for stress as- Preferred Strategies signment, four syllable words are different. structures We repeat the crucial examples here. (a) Languages prefer coda deletion, (43) Moraic trochees in Cairene and War- open syllables vowel epenthesis gamay (b) Languages prefer apocope or syn- Cairene Arabic: Parsing Left-to- closed syllables cope; no coda Right, End Rule Right deletion, vowel (X ) ( X ) epenthesis (x) (x .)(x .) (c) Iamb Iambic length- mmm mmmm ening be:ta´*k+ katabı´tu (d) Moraic trochee Trochaic short- your (m.sg.) house she wrote it (m.) ening Wargamay: Parsing Right to Left, (e) Minimal Word Vowel lengthen- End Rule Left ing, epenthesis (X ) (X ) ( X ) Unfortunately, a constant problem in pho- (x) (x .)(x .) (x .) nology is that surface output forms can easily mmm mmmm mmm be opaque with respect to the metrical struc- mu´:bag´ı-awu`lu gaga´ra J tures of the language. Thus, just as iambic stone fish freshwater dilly bag lengthening is frequent, deletion of un- jewfish stressed syllables are equally frequent, reduc- Cairene tends to have main stress towards ing a disyllabic iamb [smsmm] to a monosyl- the end of a word (katabı´tu), while Wargamay labic one [smm]. This type of opacity has been has stress at the beginning of the word (gı´Ja- addressed repeatedly throughout the history wulu). of generative phonology. Thus, before we attempt to establish cor- In recent research, opacity is elegantly relations between word rhythm and other captured in terms of constraint interaction phonological constituents, we have to first within the framework of Optimality Theory establish whether we are dealing with iam- (OT, cf. Prince & Smolensky 1993; McCar- bic/trochaic feet or whether we are referring thy & Prince 1995; for a recent survey, see to merely word edges. Since we are dealing Kager 1999). In OT, the explanatory burden with foot type, direction of parsing, End Rule is shifted from processes to output candi- (plus extrametricality) as parameters for as- dates. The central claim is that universal signing stress, any correlation concerning grammar is made up of a set of constraints, word rhythm could refer to all of them collec- all of which are available to a given language. tively or any one of them individually. The grammar generates a potentially infinite What types of correlations and implica- set of output candidates for each input, tions can we then draw given what we know which are then evaluated based on the con- about metrical patterns? One sort of correla- straint system of the language. The candidate tion could be connected with metrical coher- which best fits the constraint system is the ence. If languages prefer to stick to a given victor. Languages differ in the way the con- type of foot for stress, phonological processes straints are ranked. For our purposes, since as well as morphology, one might think that constraints are violable and can easily conflict, various types of rules would conspire to one of the most interesting consequences is achieve a preferred foot. Thus, if the lan- that an output form need not conform to all 96. Metrical patterns 1365 the constraints of the language. Let us con- Metrical constituents are however, not sider some of the preferences stated in (44) only relevant for stress but for other phono- and examine how they can conflict. A lan- logical and morphological phenomena. We guage could have the following preferences have argued that typologically, languages translated into constraints: adhere to metrical coherence within both the phonological and morphological systems of (45) Possible ranked constraints a given language. However, when it comes a. Last syllable cannot be stressed. to drawing correlations between metrical b. Foot is not right-headed; i. e. it is structure and other aspects of phonology and iambic. morphology, the typological literature is c. Main stressed foot aligns with the somewhat uncertain in drawing any conclu- right edge of a word. sions. Part of the reason is that only fixed The constraints (a) and (c) are in direct con- templatic structures have been taken into ac- flict. Constraint (c) says that the head of the count and metrical patterns are not viewed as foot should coincide with the right edge of organising principles of grammatical systems. the word. Since the foot is an iamb, the last However, since we do have a well understood syllable stress seems to be preferred. How- set of metrical patterns and are aware of ever, there is a direct conflict with (a) which possible processes and constraints, future re- says right syllables must not be stressed. In search will undoubtedly try to lay out mean- (46) we provide two possible candidates given ingful typological implications in this area. the above ranking. (46) Possible candidates 7. References i. (. x) (. x) (. x) s˘ s¯ s˘ s¯ s˘ s¯ Black, Andrew. 1991. “The optimal iambic foot ii. (. x) (. x) and reduplication in Axininca Campa”. In: Phonol- ભ s˘ s¯ s˘ s¯ s˘ s¯ ogy at Santa Cruz 2: 1Ϫ18. Blevins, Juliette. 1995. “The Syllable in Phonologi- Since (43a) is ranked above (43c), the last cal Theory”. In: Goldsmith, John (ed.), 206Ϫ244. constraint will lose out and (46 ii) would be Booij, Geert. 1979. “Allomorphy and autonomy of the preferred candidate. The output is then Ϫ opaque to the fact that the head of the foot morphology”. Folia Linguistica 31: 25 56. should preferably line up with the right edge Borowsky, Toni & Itoˆ, Junko & Mester, Armin. of the word. 1984. “The formal representation of ambisyllabi- What would this mean for the possible city: evidence from Danish”. North Eastern Lin- guistic Society Ϫ typological correlations suggested in (44)? 14: 34 48. Given that preferences and constraints can Chomsky, Noam & Halle, Morris. 1968. The sound be conflicting which predicts that the output pattern of English. New York: Harper and Row. may be opaque, we can only state the corre- Clements, G. Nick & Hume, Elizabeth V. 1995. lations we have as preferences and not abso- “The internal organisation of speech sounds”. In: lute. Obviously, the preferences remain as Goldsmith, John (ed.), 245Ϫ306. open questions. Donegan, Patricia J. & Stampe, David. 1983. “Rhythm and the holistic organisation of language structure”. In: Richardson, John F. & Marks, 6. Conclusion Mitchell & Chuckerman, Amy (eds.), Papers from the Parasession on the Interplay of Phonology, Mor- In this article, we have established that there phology and Syntax, 337Ϫ353. Chicago: Chicago are systematic organising principles which Linguistic Society. construct metrical patterns in natural lan- Dresher, B. Elan & Lahiri, Aditi. 1991. “The Ger- guages. These organising principles combine manic Foot: Metrical coherence in Old English”. various units of prosodic structure like syl- Linguistic Inquiry 22: 251Ϫ286. lables and feet. There is a limited inventory Eisenberg, Peter & Ramers, Karl Heinz & Vater, of syllable and foot types for all languages. Heinz (eds.). 1992. “Silbenphonologie des Deut- Stress, which marks relative prominence schen”. Studien zur deutschen Grammatik 42. Tü- within metrical constituents, is a result of bingen: Gunter Narr Verlag. various independent parameters including Fitzpatrick-Cole, Jennifer. 1994. The prosodic do- the type of foot and the edge of a word and main hierarchy in reduplication. PhD dissertation, direction of parsing. Stanford University. 1366 XII. Phonology-based typology

Fitzpatrick-Cole, Jennifer. 1996. “Reduplication Hyman, Larry M. 1997. “On the nature of linguis- meets the phonological phrase in Bengali”. The tic stress”. In: Hyman, Larry M. (ed.) Studies in Linguistic Review 13: 305Ϫ356. Stress and Accent. Southern California Occasional Ϫ Ghini, Mirco. 2001. Asymmetries in the phonology Papers in Linguistics.4:37 82. of Miogliola. (Studies in Generative Grammar.) Jacobs, Haike. 1989. Historical studies in the non- Berlin: Mouton de Gruyter. linear phonology of French. PhD dissertation, Uni- versity of Nijmegen. Giegerich, Hans. 1992. “Onset maximisation in German: the case against resyllabification rules”. Jacobs, Haike. 2000. “The revenge of the uneven In: Eisenberg, Peter & Ramers, Karl Heinz & trochee: Latin main stress, metrical constituency, Vater, Heinz (eds.), Silbenphonologie des Deut- stress-related phenomena and OT”. In Lahiri, schen, 134Ϫ171. Aditi (ed.). 333Ϫ352. Gil, David. 1986. “A prosodic typology of lan- Kager, Rene´. 1989. A metrical theory of stress and guage”. Folia Linguistica 20: 165Ϫ231. destressing in English and Dutch. Linguistic models 14, Dordrecht: Foris. Gil, David. 1987. “On the scope of grammatic the- ory”. In: Modgil, Sohan & Modgil, Cecilia (eds.), Kager, Rene´. 1995. “The metrical theory of word Ϫ Noam Chomsky: Consensus and Controversy,119Ϫ stress”. In: Goldsmith, John (ed.), p. 367 402. 141. Barcombe: Falmer Press. Kager, Rene´. 1999. Optimality Theory. Cambridge: Cambridge University Press. Goddard, Ives. 1979. Delaware Verbal Morphology. New York: Garland publishing. Kahn, David. 1976. Syllable-based generalizations in English phonology. Cambridge, Mass.: MIT PhD Goldsmith, John (ed.). 1995. The Handbook of dissertation. Distributed by Indiana University Phonological Theory. Oxford: Blackwell. Linguistics Club. Greenberg, Joseph H. & Kashube, Dorothea. 1976. Kenstowicz, Michael. 1994. Phonology and genera- “Word prosodic systems: A preliminary report”. tive grammar. Oxford: Blackwell. Working Papers on Language Universals 20: 1Ϫ18. Keyser, Samual J. & O’Neill, Wayne. 1976. Rule Gussenhoven, Carlos. 1986. “English plosive allo- generalization and optionality in language change. phones and ambisyllabicity”. Gramma. 10.2: 119Ϫ Dordrecht: Foris. 141. Kiparsky, Paul. 1979. “Metrical structure is cyclic”. Haas, Mary. 1977. “Tonal accent in Creek”. In: Linguistic Inquiry 10: 421Ϫ441. Hyman, Larry (ed.) Studies in stress and accent. Lahiri, Aditi (ed.). 2000. Analogy, Levelling, Mark- Southern California Occasional papers in Linguis- Ϫ edness: Principles of change in phonology and mor- tics 4, 195 208. Los Angeles: University of South- phology. Berlin: Mouton de Gruyter. ern California, Department of Linguistics. Lahiri, Aditi. 2000. “Hierarchical restructuring in Halle, Morris & Idsardi, William. 1995. “General the creation of verbal morphology in Bengali and properties of stress and metrical structure”. In Germanic: evidence from phonology”. In: Lahiri, Goldsmith, John (ed.), 403Ϫ443. Aditi (ed.), 71Ϫ123. Harrison, Sheldon P. & Albert, Salich Y. 1976. Lahiri, Aditi & Dresher, B. Elan. 1999. “Open syl- Mokilese Reference Grammar. Honolulu: Univer- lable lengthening in West Germanic”. Language sity of Hawaii Press. 75: 678Ϫ719. Hayes, Bruce. 1981. A metrical theory of stress Lahiri, Aditi & Koreman, Jacques. 1988. “Syllable rules. Cambridge, Mass.: MIT PhD dissertation. weight and quantity in Dutch”. West Coast Confer- Distributed by Indiana University Linguistics Club ence on Formal Linguistics 7: 217Ϫ228. and published by Garland Press in 1985. Lahiri, Aditi & Riad, Tomas & Jacobs, Haike. Hayes, Bruce. 1989. “Compensatory lengthening in 1999. “Diachrony”. In: van der Hulst, Harry (ed.) moraic phonology”. Linguistic Inquiry, 20: 253Ϫ Word Prosodic Systems in the Languages of Europe, 306. 335Ϫ422. Berlin: Mouton de Gruyter. Hayes, Bruce. 1995. Metrical Stress Theory: Prin- Lehmann, Winfred. 1973. “A structural principle ciples and Case Studies. Chicago: University of of language and its implications”. Language 49: Ϫ Chicago Press. 47 66. Hayes, Bruce & Lahiri, Aditi. 1991. “Bengali into- Lehmann, Winfred P. 1978. “Conclusion: Towards national phonology”. Natural Language and Lin- an understanding of the profound unity underlying Syntac- guistic Theory 9: 47Ϫ96. languages”. In: Lehmann, Winfred P. (ed.) tic Typology: Studies in the Phenomenology of Lan- Hulst, van der, Harry. 1984. Syllable structure and guage, 395Ϫ432. Austin: University of Texas Press. stress in Dutch. Dordrecht: Foris. Liberman, Mark. 1975. The intonational system of Hulst, van der, Harry. 1985. “Ambisyllabicity in English. Cambridge, Mass.: MIT PhD dissertation. Dutch”. In: Bennis, Hans & Beukma, Frits (eds.) Distributed by Indiana University Linguistics Linguistics in the Netherlands. Dordrecht: Foris. Club. 97. Tone systems 1367

Loewe, Martha. 1996. Prosodic structures. MA the- Plank, Frans. 1998. “The covariation of phonology sis, University of Konstanz. with morphology and syntax”. Linguistic Typology, McCarthy, John & Prince, Alan. 1986. “Prosodic 2, 195Ϫ230. Morphology”. Manuscript, University of Mas- Prince, Alan & Smolensky, Paul. 1993. “Optimality sachusetts, Amherst and Brandeis University. theory: Constraint interaction in generative gram- McCarthy, John & Prince, Alan. 1990. “Foot and mar”. Manuscript, Rutgers University and Univer- Word in Prosodic Morphology: the Arabic broken sity of Colorado at Boulder. plurals”. Natural Language and Linguistic Theory Rubach, Jerzy. 1990. “Final devoicing and cyclic 8: 209Ϫ282. syllabification in German”. Linguistic Inquiry 21: McCarthy, John J. & Prince, Alan. 1993. “General- 79Ϫ94. ized alignment”. Yearbook of Morphology,79Ϫ153. Spencer, Andrew. 1996. Phonology. Oxford: Black- McCarthy, John J. & Prince, Alan. 1995. “Prosodic well. Morphology”. In: Goldsmith, John (ed.), 318Ϫ 366. Spring, Cari. 1990a. Implications of Axininca Nespor, Marina & Guasti, Therese & Christophe, Campa for Prosodic Morphology and Reduplication. Anne. 1996. “Selecting word order: the rhythmic PhD dissertation, University of Arizona, Tucson. activation principle”. In: Kleinhenz, Ursula (ed.) Spring, Cari. 1990b. “How many Feet per Lan- Interfaces in Phonology. Studia Grammatica 41, guage?” In: The Proceedings of the Ninth West 1Ϫ26. Berlin: Akademie-Verlag. Coast Conference on Formal Linguistics, 493Ϫ508. The phonology and morphol- Payne, David. 1981. Vennemann, Theo. 1972. “On the theory of syllabic ogy of Axininca Campa. Publication of the Summer phonology”. Linguistische Berichte 18: 1Ϫ19. Institute of Linguistics and the University of Texas at Arlington. Plank, Frans. 1997. “Linguistic Typology: intro- Aditi Lahiri, University of Konstanz ducing a new journal”. Linguistic Typology,1,1Ϫ3. (Germany)

97. Tone systems

1. Defining “tone” 1973) definition is as good as most: “A tone 2. Typologizing tone systems by phonological language is a language in which both pitch contrasts phonemes and segmental phonemes enter 3. Typologizing tone systems by function into the composition of at least some mor- 4. Typologizing tone systems by tone rules 5. Tone and/or accent phemes.” Thus, tone is clearly indicated in 6. Tone-marking conventions the case of such pairs as Pawaian [Oceanic] 7. References su´ ‘tooth’ and su` ‘road’ and Mende [Sierra Leone] pi´li´ ‘house’ and bi`li` ‘trousers’ (see § 6. for tone-marking conventions). While Pike 1. Defining “tone” (1948) had suggested that a tone language has “contrastive, but relative pitch on each Within the phonological realm, few typologi- syllable”, Welmers improves on this definition cal issues have generated as much discussion by recognizing the existence of toneless mor- (and controversy) as the issue of tone. It is phemes, especially grammatical morphemes generally assumed that as many as half of which take their tone from the surrounding the world’s languages are “tonal”. While context. If we reinterpret Welmers in modern most language families in the world have one terms to mean that the pitch phonemes must or more tonal offsprings, including those in be presented in “the underlying representa- North and South America, Europe, and Oce- tions of at least some morphemes”, this will ania, languages with fully developed tone sys- allow for the Mende toneless postpositions tems are highly concentrated in Subsaharan -hu ‘in’ and -ma ‘on’, which copy their tone Africa, Southeast Asia, and Mexico. Beyond from the preceding nominal, e. g. pi´li´-hu´, these generalities, the typological study of pi´li´-ma´;bi`li`-hu`,bi`li`-ma`. It is also possible tone systems has at times faltered on the very for lexical morphemes to be underlyingly basic question of what constitutes a “tone” toneless. Hyman (1981) has analyzed tone in and hence a “tone system”. Welmers’ (1959, Somali as being largely predictable on the