REMARKS AND REPLIES 497

The Laryngeal Specifications of

Bert Vaux

Phonologists have traditionallyassumed that the unmarkedlaryngeal state of fricatives is to be unaspirated([ - spread glottis]). However, the data analyzed here, which are drawnfrom Armenian,Greek, Pali, Sanskrit,Spanish, and Thai, suggest thatin theirunmarked state - less fricatives are in fact [ + spreadglottis], whereas voiced fricatives are [-spread glottis].

Keywords:postnasal voicing, laryngealfeatures, fricatives, cluster as- similation, Armenian,Sanskrit

Though laryngeal features have been a central concern of phonologists for more than 25 years (see Kim 1970, Halle and Stevens 1971, Thrainsson1978, Iverson 1983, Keating 1984, Lombardi 1991, Blevins 1993, Kingston and Diehl 1994, Iverson and Salmons 1995), little attentionhas been given to the laryngeal specifications of fricatives. Halle and Stevens (1971) suggest that fricatives are generally [- spread glottis] (roughly equivalent to "unaspirated" in traditional terms), but can exceptionally be [+ spread], as in Burmese. The same view is tacitly accepted by many phonologists and phoneticians(see, e.g., Maddieson 1984, Catford 1988). Recent phonetic work by among others Kingston (1990) and Stevens (1991) has suggested, however, that at least voiceless fricatives are generallyproduced with a spreadglottis. According to Stevens, a voiceless cannot be producedwithout spreadingthe glottis, because this glottal adjustmentis necessary in order to build up sufficient pressure behind the constriction formed by the primaryarticulator for the fricative. On the basis of this observation,Halle (1995) proposesthat all fricativesare [ + spread].Halle's proposalalso encountersa numberof problems, however. It is difficult, for example, to distinguish Burmese plain s from aspiratedSh, both of which must be [ + spread]in his model. Halle's generalizationalso fails to accountfor Catford's observationthat voiced obstruentsare accompaniedby a narrowingof the glottis (1977:112). In this articleI proposea system of laryngealspecifications that accounts both for the phonetic observationsof Kingston, Stevens, and Catfordand for the phonological behavior of fricatives in a number of languages with rich systems of laryngeal contrasts:the New Julfa dialect of Armenian,Pali, Sanskrit,Greek, and Thai. The phonological phenomenato be considered here strongly suggest that the unmarkedspecification for voiceless fricatives is [ + spread], and the specification for voiced fricatives is [- spread].

Many thanks to the following, none of whom necessarily agree with the claims I make here, for helpful comments on earlier versions of this article:Andrea Calabrese, Morris Halle, Kevin Herwig, Javier Martin-Gonzalez,Ken Stevens, PatrickTaylor, and audiencesat the 1996 annualmeeting of the LinguisticSociety of America,the Universityof Pittsburgh, and HarvardUniversity.

Linguistic Inquiry,Volume 29, Number 3, Summer 1998 497-511 ? 1998 by the MassachusettsInstitute of Technology

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438998553833 by guest on 26 September 2021 498 REMARKS AND REPLIES

1 Laryngeal Assimilation in Clusters The first set of data comes from the behavior of consonant clusters in the Armeniandialect of New Julfa (Vaux, forthcoming),which is now spoken only by a few older Armeniannatives of Isfahan.Like Sanskrit,this dialect has a four-waylaryngeal contrast in its stop system, represented in (1).1 (1) New Julfa consonant system b b p p f v m d d t t s z n dz dhz ts ths h h d3 tf t ff 3 g gh k kh X K h fi j r r

1.1. The New Julfa Future Morpheme New Julfa forms the future tense by adding the prefix k- to the present subjunctive.This k- surfacesas k- before vowels andplain voiceless consonants(2a), g- before plain voiced consonants (2b), kh- before voiceless aspiratesand voiceless fricatives (2c), and gh before voiced aspirates (2d). (2) Underlyingform Surfaceform Gloss a. k-erth-a-m kertham I will go k-t-a-m katam I will give k-kien-a-m kakienam I will exist b. k-bzz-a-m gabazzam I will buzz k-l-a-m g3lam I will cry k-zi-a-m gazaram I will bray c. k-thoK-n-ie-m kh3thoiKniem I will allow k_thfaph_ie-m khth faphiem I will measure k-Xnd-a-m kh3X3ndam I will laugh k-savor-ie-m kh3savoriem I will grow accustomedto d. k-bhier-ie-m ghabhieriem I will carry k-g -ho-m g hghom I will come k-dh-n-ie-m ghadhaniem I will put k-dhziev-ie-m ghadhzieviem I will form

r representsa rhotic tap; r is a rhotic trill.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438998553833 by guest on 26 September 2021 REMARKS AND REPLIES 499

What we are dealing with here is clearly a case of assimilation:the future prefix k- assimilates in voicing and aspirationto a following consonant. The fact that the features responsible for voicing and aspirationspread together constitutes strong evidence for the populartheory (e.g., Sagey 1986) that the feature-geometricrepresentation of segments includes a Laryngeal node, which dominates the features [spreadglottis] (responsiblefor aspiration)and [stiff vocal folds] (responsiblefor voicing), as schematizedin (3). (3) Laryngeal

[spreadglottis] [stiff vocal folds] With this structurein hand, we can interpretthe alternationsof the future prefix in (2) as the result of a rule that spreadsthe Laryngealnode to the k- from a consonantto its immediateright, as shown in (4) (intermediatenodes omitted). (4) Laryngealspreading k # [+cons]

Laryngeal This rule precedes a rule of epenthesis that inserts a schwa after unsyllabifiedconsonants. Crucially,the fact that voiceless fricatives cause the prefix to surface as aspiratedkh rather than unaspiratedk (2c) suggests that voiceless fricatives are [ + spread].Voiced fricatives,on the other hand, do not trigger aspiration(2b), suggesting that they are [-spread].

1.2 New Julfa Voiceless Fricatives The same assimilation of laryngeal features in consonant clusters that we observed in the case of the future prefix also occurs in morpheme-internalclusters. New Julfa possesses a rule that deletes unstressedhigh vowels, leading to alternationsof the type in (5). (5) Base form Genitive Gloss ku'ft khaftin side fietku's fietkh sin behind Again, underlying k becomes [+ spread glottis] before voiceless fricatives (cf. Atfaijan 1940: ? 123). Though I have not found any synchronic alternationsof this type with other voiceless stops, the same behavior can be seen in the historical treatmentof original plain voiceless stop + voiceless fricative sequences, as in (6). (6) New Julfa Gloss psak p hasak crown (Turkish)baf0rty phafoithik headscarf gifer khafer evening As in the case of the Laryngealspreading in (4), the processes in (5) and (6) only make sense if

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438998553833 by guest on 26 September 2021 500 REMARKS AND REPLIES

we assume that voiceless fricatives are [ + spread], and this [+ spread] specification spreads to the precedingconsonant. Voiced fricatives again must be [- spread]in orderto avoid triggering the rule. Furtherevidence for the claim that voiceless fricatives are [ + spread]comes from the treat- ment of Greek loanwordsin Classical Armenian.Forms of the type in (7a) show that Greek psi (i.e., the phoneme sequence Ipsl) is renderedin Classical Armenianas a sequence of aspirated ph + s; similarly, (7b) shows that Greek ksi (i.e., the phoneme sequence Iksl) is renderedas aspiratedkh + s. These facts are particularlystriking given that Classical Armenianallows -pS- clusters (e.g., psak 'crown') and generally rendersGreek p and k as Armenianp and k, not ph and kh (7c). (7) Greek Classical Armenian Gloss a. ItXu14EITLXO; (Psammetikhos) Phsametikos Psammetichus 4nxAX,o;(psalmos) phsalmos psalm Ox41vOLv(apsinthion) aphsndin wormwood b. SEptIg (Kserkse:s) Khserkhse:s Xerxes

tEUTTq; (kseste:s) khsest sextary,jar bou, (douks) (

2 Rule (4) must not have been active at the time when psak 'crown' was borrowed into Armenianfrom Middle Iranian. 3 The AtharvavedaPra:tifa:khja 2.6 (Whitney 1862) restrictsthe target of the aspirationrule to word-finalstops. Interestingly,the Taittirija Pra:tifa:khja 14.13 mentions that accordingto the grammarianVa:ctabhi:ka:ra, the aspiration rule does not apply when the fricative is homorganicwith the stop.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438998553833 by guest on 26 September 2021 REMARKS AND REPLIES 501

(8) Underlyingform Surfaceform Gloss bhiFaksi:s-ena bhi akhsi:sena healing lead-instrumental ap-su aphsu water-locative.plural The process of aspirationobserved by the phoneticians can be formalized as a rule that spreadsthe Laryngealnode of fricatives to a preceding stop, as in (9).4 (9) [-son] [-son] [-cont] [+cont]

Laryngeal The theory presentedhere predictsthat voiced fricatives would spreadtheir [-spread] specifica- tion to a preceding stop; in other words, they would not trigger aspiration.Since Sanskritdoes not possess voiced fricatives,however, this predictioncannot be tested. We can conclude, though, thatvoiceless fricativesmust be [ + spread]in orderto triggerthe aspirationrule. The phonological evidence adducedhere accordswith the phonetic observationsof the ancient Indianphoneticians, who state that the Sanskrit fricatives and voiceless aspirates form a class of aspirates, to the exclusion of plain voiceless consonants (e.g., Taittiri.jaPra:tifa:khja 2.11; Whitney 1868:55).

1.3 Fricative Assimilation The final piece of evidence concerning the behavior of laryngeal features in consonant clusters again comes from the Armeniandialect of New Julfa. In this dialect, original plain voiced stops regularlybecome voiced aspirates(lOa), except after nasals (lOb).

(10) Classical Armenian New Julfa Gloss a. berem bhieriem I carry gini ghini wine dew dhev demon dzmern dhzmer winter d3ur dh3ur water b. bambak bhambak cotton gangat g hangat complaint andam andam limb andzrew andzrev rain kamurd karmund bridge

4 The Sanskritfacts do not actually provide any evidence that it is the Laryngealnode ratherthan just the feature [spreadglottis] that is spreading.I assume the strongerhypothesis, though, since there is also no evidence that laryngeal features other than aspirationare not involved in this process.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438998553833 by guest on 26 September 2021 502 REMARKS AND REPLIES

When a voiced fricativeimmediately precedes one of these voiced aspirates,it becomes voiceless, as illustratedin (11).

(11) Classical Armenian New Julfa Gloss anzgam an3sgham wicked zgojf 3sghuf safe zgal 3sghal feel Xeldel XeXdhel strangle albiwr aXbhur fountain zbalum asbhaKum business zldal zaXd23al repent ol3ojn oXd3um greeting Voiced aspiratesare treatedin two differentways in the literature.One camp, composedprimarily of phoneticians,represents voiced aspiratesby means of [murmur],a feature distinct from the conventionalfeature [voice] (Ladefoged 1993:144). In this model, the devoicing of fricatives in (1 1) cannotbe interpretedas a rule spreading[murmur], because the voiceless fricativesproduced by the rule are not murmured.Nor can it be interpretedas a rule of dissimilation,because in this model the feature [voice], which changes in the fricative, bears no phonological relationshipto the feature [murmur],which triggersthe change.5In sum, the murmurmodel fails to accountfor the facts in (11). The othercamp, representedby the work of Halle and Stevens (1971), views voiced aspirates as [- stiff, + spread]segments. In this view, the phoneticmurmur characteristic of voiced aspirates is a by-product of the competition between the phonological specifications [ - stiff] and [ + spread].Halle and Stevens's representationof voiced aspiratesmakes possible a more plausible analysis of the process in (1 1). Let us supposethat there is a rule that spreadsthe featurespecifica- tion [ + spread]from right to left in clusters, as in (12). (12) [+cons] [+cons]

[+spread] When a fricative precedes a voiced aspirate, it should become [+spread] by rule (12). Since voiceless fricatives in the proposedmodel are already [+ spread],rule (12) will have no visible effect in this case. With voiced fricatives, however, rule (12) should produce the configuration [- stiff, + spread],that is, a voiced aspiratedfricative. This type of fricative does not appearto be allowed in naturallanguages, presumablybecause the amountof airflow requiredto cue the presence of the feature [ + spread],and to indicate that the segment is a fricative, is too great for the vocal folds to be able to maintain a [+ stiff] configuration.I propose that the disallowed

5Additional argumentsagainst the feature [murmur]are given in Halle 1973.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438998553833 by guest on 26 September 2021 REMARKS AND REPLIES 503

configuration*[-stiff, + spread]is repairedso as to yield a [+ stiff, + spread]segment, in other words, a voiceless fricative.6This procedureis directly analogous to what we find in Icelandic, where stops spread their specification for the feature [spread glottis] to a preceding sonorant. When the sonorantbecomes [ + spread] as a result of this rule, it devoices (Thrainsson1978). The devoicing of English liquids following voiceless stops in words such as pray [ph rej] can be explained in a similar manner(see Iverson and Salmons 1995). As with the othercases presentedin this section, the behaviorof New Julfafricative devoicing only makes sense if we assume that voiceless fricatives are [ + spread]and voiced fricatives are [- spread].If all fricativeswere [- spread],we would have to say thatvoiceless fricativesdelinked the [ + spread]specification spread by rule (12); voiced fricatives,on the other hand, would have to delink both [ + spread]and [- stiff]. This analysis would miss the generalizationthat fricatives behave as a natural class, since voiced and voiceless fricatives would trigger different repair strategies. If on the other hand all fricatives were [?+spread],we would expect that rule (12) would have no effect on voiced fricatives, which is incorrect.

2 Delinking Furtherevidence for the proposedrepresentation of fricativescomes from the treatmentof fricative + stop sequences in the Middle Indic language Pali. In this language, original Indic fricative + stop sequences(here represented by Sanskritforms) surfaceas voiceless aspiratesin initialposition (13a) and postaspiratedgeminates in medial position (13b).7 (13) Sanskrit Pali Gloss a. Initial skandha khandha- shoulder stana- thana- breast sparfa- phassa- touch b. Medial ha'sta- hattha- hand yaFti- yatthi_ pole The changes representedin (13) result from a general simplification of syllable structurethat occurredbetween Old Indic (Sanskrit)and Middle Indic (Pali, Prakrit,etc.). Whereas Sanskrit allowed complex onsets (14a) and place features in codas (14b), Middle Indic did not (14c,d). Violations of the new syllabic constraintswere repairedby deleting the offending segments. In onset position, this delinking resulted in the total disappearanceof the segment (14c). In coda position, however, the timing unit vacated by the delinked segment associates to the following onset segment, yielding a geminate (14d).

6 The frameworkof constraintsand repairsemployed here is based on Calabrese 1995, to which I returnin section 4. 7 Iverson and Salmons (1995) suggest that there is a mismatchbetween the underlyingand surface representations in postaspiratedclusters of this type: whereas in the acoustic output only the final memberof the cluster is aspirated,in the underlyingphonological structureall members of the cluster are linked to a single [spread glottis] specification. I will not discuss this idea furtherhere, as it is not directly relevant to the theory that I am proposing.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438998553833 by guest on 26 September 2021 504 REMARKS AND REPLIES

(14) Sanskrit Pali Gloss a. kraima- c. kama- step b. uktai- d. utta- spoken What is relevant for our purposes is the fact that the laryngealfeatures of the delinked segments survive. In stana- 'breast'(13a), for example, the initial s delinks (15a), but the floating [ + spread] specificationof the s then attachesto the following segment (1Sb), producinga voiceless aspirate (15c). (15) a. Debuccalization X X X X X

s t a n a

[+spread] b. Laryngealattachment x x x x x

s t a n a

[+spread] c. Final surfaceform xxxx

tha nta

[+spread] A similar process occurs in the synchronicphonology of the Seville dialect of Spanish (Javier Martin-Gonzales,personal communication).As in many dialects of Spanish, s debuccalizes in coda position. Unlike in other dialects, however, the debuccalized s does not simply surface as h when it precedes a stop; rather,it also aspiratesthe following stop, as in (16). (16) Underlyingform Surfaceform Gloss los padres loh pha6reh the parents The Pali and Spanish cases can only be accounted for by assuming that voiceless fricatives are [ + spread]. If the Spanish s were [- spread], for example, we would be unable to account for the [+ spread] specification that appearson the underlyingp in (16).

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438998553833 by guest on 26 September 2021 REMARKS AND REPLIES 505

4 Corroborative Evidence The laryngealspecifications for fricativesproposed in this articlehave the additionaladvantage of allowing us to make sense of importantcomponents of two well-knownbut heretoforeproblematic phonological phenomena,postnasal voicing and tonogenesis. In this section I first consider how to treatpostnasal voicing in languages such as Armenianand Greek, and then turnto tonogenesis in Thai.

4.1 Postnasal Voicing The phenomenonof postnasalvoicing is extremely common crosslinguisticallyand has attracted a great deal of attentionfrom phonologists in recent years (see Pater, forthcoming,for a review of the relevantliterature). However, existing accountshave not provideda satisfactoryexplanation for the failure of fricatives and aspiratesto undergopostnasal voicing. In this section I show how the theory of laryngealspecifications presented in this article can be used to account in a unified way not only for the behavior of fricatives, but also for the behavior of aspirates. A typical case of postnasal voicing occurs in the historicaldevelopment of Modem Greek, where stops and (17a) but not fricatives (17b)8 are voiced after nasal consonants. (17) Classical Greek Modern Greek Gloss a. olumpos olimbos Olympus kentron kendro center ankura angira anchor b. anthro:pos anOro:pos man melankholia melanXolia melancholy Postnasalvoicing is still active in the synchronicphonology of Modem Greek, as shown by the alternationsin (18) (data from Pring 1962:19). (18) Underlyingform Surfaceform Gloss a. Stops ston kipo stongipo in the garden 6en pirazi 6embirazi it doesn't matter en taksi endaksi all right b. Affricates ton psixon tombziXon (all) souls' (day) stin tsanta tis stindzandatis in her bag san ksilo sangzilo like wood Sagey (1986) interpretspostnasal voicing processes of this type as the spreadingof r- stiffl from a nasal consonantto a following consonant (19a). She and other phonologists (cf. Padgett 1991) who have dealt with postnasal voicing assume that it creates a linked structure(in effect, a prenasalizedstop; (19b)). This linked structureis not allowed to contain the featurecombination

8 Note that the outcomes are not *amvoris, *an6ropos, *melanyolia, though Modern Greek has 1ryl,/6/, and /v/ phonemes.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438998553833 by guest on 26 September 2021 506 REMARKS AND REPLIES

*[ + nasal, + continuant],since it is difficult to producecontinuancy when the velum is lowered (l9c). Consequently,postnasal voicing is blocked from applying to [+continuant] consonants. (19) Postnasal voicing in the spirit of Sagey 1986 and Padgett 1991 (relevantfeaturesonly) a. Spreading [+cons] [+cons]

[+nasal] [-stiff] b. Result: linkedstructure [+cons] [+cons]

[+nasal] [-stiffl c. Illicit structure:*[+nasal, +continuant] * [+cons] [+cons]

[+nasa] [- stiff] +ot

Sagey's analysis, as well as the analyses of postnasalvoicing comprehensivelysummarized by Pater (forthcoming),ignore the crucial cases of postnasal voicing, however, which occur in languages with richer systems of laryngeal contrasts. In these languages, such as New Julfa, postnasal voicing is blocked not only with fricatives (20b), but also with aspiratedconsonants (20c). (20) Classical Armenian New Julfa Gloss a. antaj andza gift ankanel anganiel fall gan.g gJan43 fly ajntel andieis there b. insaf justice sunsunakviel starve thanXiel give false hopes semsuri type of melon c. th anthifel thanthfintal mutter famphur famphur spit jawnkh fiunkh eyebrow

The rule of postnasal voicing belongs to the synchronicphonological system of New Julfa, as shown by the fact that it applies in clusters createdby a synchronicrule of vowel syncope (21). (21) Nominative Genitive Gloss kUnik kanga wife ghorth nuk ghorthnga frog

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438998553833 by guest on 26 September 2021 REMARKS AND REPLIES 507

Sagey's analysiscannot account for the failureof voiceless aspiratesto undergopostnasal voicing. Since voiceless aspiratesare [- continuant],we expect them to become voiced, but they do not. The theory developed in this article suggests anothersolution, however. In this model, but not in Sagey's, the fricativesin (17), (18), and (20) sharewith voiceless aspiratesthe featurespecification [ + spread]. Furthermore,the feature combination [ + nasal, + spread] is highly marked,as it is difficult to sustainoral aspirationnoise-the primaryacoustic correlateof [ + spread]-when the velum is lowered. In the terms of Calabrese'stheory of markedness,the articulatorycomplexity and rarityin phonologicalsystems of this particularfeature configuration is capturedby a marking statement of the form in (22), which states that aspiratednasal segments are phonologically marked. (22) *[+ nasal, + spread] Given the assumptionsthat voiceless fricatives are [ + spread]and that Universal Grammarcon- tains the markingstatement in (22), we can now account for both the Greek and the New Julfa facts, with only minimal modification to Sagey's original analysis. In the proposed analysis, postnasal voicing spreads the feature [ - stiff] from a to a following obstruent, creating a linked structure(cf. (19a)). Since linked structuresare treatedby markingstatements as single segments, postnasal voicing makes visible to these marking statements the [spread] specificationof the segment to which the [ - stiff] featureof the nasal has spread.If this segment happens to be [+spread], the linked structurecontains the disallowed configuration*[ +nasal, + spread](cf. (19c)). This disallowed configurationis subsequentlyrepaired via delinking of the [- stiffl feature from the target segment.

4.2 Thai ConsonantClasses The laryngeal specifications proposed in this article also allow us to make sense of one aspect of tonogenesis in the historyof Thai. It is well known thatthe tones thatdeveloped on Thai vowels somewherebetween the fifteenthand seventeenthcenturies (Hudak 1987:763) were determinedby the quality of neighboringconsonants. The Thai consonants fell into three groups with respect to the tones that they engendered.9These groups are schematizedin (23). (23) Thai consonant classes I voiceless aspirates,voiceless fricatives, voiceless sonorants II plain voiceless stops III voiced sounds (vowels, voiced sonorants,voiced aspirates) Note thatvoiceless fricativespattern with voiceless aspirates(group I), ratherthan unaspirated stops; note also that voiced fricativespattern with voiced stops (groupIII), and not aspirates.We can rewrite the groupings in (23) in terms of the naturalclasses in (24).

9 I do not present the tonal outcomes themselves, because they are not directly relevant to the discussion.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438998553833 by guest on 26 September 2021 508 REMARKS AND REPLIES

(24) Thai consonant classes, revised I [ + spread glottis, + stiff vocal folds] II [- spread glottis, + stiff vocal folds] III [- stiff vocal folds] Note that only the theory of laryngeal specifications proposed in this article allows us to group the obstruentscorrectly with respect to their phonological effects in the evolution of Thai tones. Given this theory,we expect the voiceless fricativesto fall into groupI, since they are [ + spread].10 Theories that maintainthat voiceless fricatives are [- spread]cannot accountfor the distribution in (23). By the same token, theories that maintainthat all fricatives are [ + spread]cannot account for the fact that voiced fricatives do not patternwith group I.

5 Conclusions The Armenian, Indic, Spanish, Greek, and Thai facts discussed in this article provide strong phonological evidence that voiceless fricatives are specified as [+spread glottis] and voiced fricatives are specified as [- spreadglottis]. These pieces of phonologicalevidence dovetail well with the phonetic facts mentioned earlier, namely, that voiceless fricatives are producedwith a spreadglottis (Kingston 1990, Stevens 1991) and voiced fricatives are not (Catford 1977). It shouldbe noted thatthese specificationsrepresent the unmarkedstate for fricativesystems, ratherthan an absolute and invariableset of specifications. It is perfectly possible for languages to contrast[spread] values in voiceless fricatives allophonically(as in English) or phonemically (as in Burmese); it also appearsthat in some languages, such as Chinese, both the voiced and voiceless fricative series are [- spread] (Kevin Herwig, personal communication).Systems of the Chinese type are simply more markedthan systems of the Armeniantype, and hence they appearin fewer of the world's languages.In the termsof Calabrese's(1995) theoryof markedness, Universal Grammarprovides a markingstatement of the form in (25), (25) *[ + stiff, - spread]/[-, + cont] which states that [ - spread]voiceless fricativesare phonologicallymore complex than [ + spread] voiceless fricatives(the markingstatement also reflects the fact thatvoiceless vowels aregenerally [ + spread],but I will not consider this fact here). Voiced fricatives that are [+ spread]are ruled out by a markingstatement of the form in (26) (cf. Calabrese 1988:274). (26) *[-stiff, +spread]/[-, +cont] The laryngealspecifications of the fricativesin the languagesconsidered in this articlecannot be derived from the structureof the stop inventory.In a two-series laryngeal system contrasting plain voiced and voiceless aspiratedobstruents, such as we find in StandardWestern Armenian, one could say that the voiceless fricatives are predictably[ + spread] because all voiceless ob-

'0 The fact that the voiceless sonorantsalso belong to group I supportsthe idea that they are [ + spread glottis] as well (cf. Ohala and Ohala 1993, Silverman 1995, Asano 1997).

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438998553833 by guest on 26 September 2021 REMARKS AND REPLIES 509

struents are [+ spread]. This reasoning cannot work for languages like New Julfa or Sanskrit, though, where voiceless obstruents are not predictably [+spread], nor are voiced obstruents predictably[-spread]. It should be noted that the theory presentedhere is primarilya theory of phonological rather than phonetic representations.Though I believe that the phonological representationsproposed here are well groundedphonetically, some phonetic questions remain. For example, both Ken Stevens and Louis Goldstein have independentlypointed out to me that, like voiceless fricatives, voiced fricatives also requiresome spreadingof the glottis in order to produce sufficient airflow to yield frication noise; however, the degree of spreading is not as great as it is for voiceless fricatives. I do not consider this to be a problem,however, given that voiceless fricatives clearly patterndifferently from voiced fricatives with respect to phonological processes. The importantpoint to bear in mind here is that a phonological specification [-X], where X is any feature,does not entail thatthe componentof the vocal tractactivated by [X] is completely inert at the phonetic level. The English phoneme [f], for example, is specified in the phonology as [- round], yet it is often implementedwith some roundingof the lips. Similarly, vowels that are phonologically [- nasal] are often implementedwith some airflow throughthe nasal passage by speakers of American English. The fact that in each of these cases a particulararticulator is phoneticallyactive to a certaindegree does not entail thatthis activity is encoded in the phonologi- cal representation;for example, we do not want to say that English [f] is [ + round]in the phonol- ogy, nor do we want to say that American vowels are phonologically [+ nasal]. Rather, we should acknowledge that phonological features bisect a continuum of phonetic activity. In this view, [ + nasal] for example representslowering of the velum beyond a certain criticalzone; [- nasal] representsany lesser degree of velar lowering. Similarly, [ ? spreadglottis] shouldbe defined in termsof a line drawnsomewhere in the rangeof possible degrees of spreading of the vocal folds (27a), ratherthan in the all-or-nothingterms of spreadingversus no spreading (27b). (27) a. Definition of [?spread glottis] proposed here maximumspreading

[+spreadglottis]

glottis] no spreading [-spread b. All-or-nothingdefinition of [?spread glottis] maximumspreading -

[+spreadglottis]

no spreading - *,-[-spread glottis]

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438998553833 by guest on 26 September 2021 510 REMARKS AND REPLIES

The assumptionthat voiceless fricatives are specified [+ spread glottis] makes it possible to provide a unitary account for a wide range of seemingly disparatephenomena. On the one hand, the phonologicalmodel proposedhere reflects phoneticproperties of fricativesthat hitherto had been observed by phoneticiansbut ignored by phonologists. In addition,it accounts for the phonological behavior of laryngeal spreading, fricative assimilation, and postnasal voicing in Arnenian, aspirationin Sanskrit,debuccalization in Middle Indic and Spanish, and tonogenesis in Thai. Within theories of phonology that do not adopt the representationof fricatives proposed here, the individualphenomena in this set become difficult to explain, andthe commonphonologi- cal threaduniting them is missed.

References

Asano, Makiko. 1997. The problemof NC configurationsand derivationalopacity in Bantu languages.Ms., HarvardUniversity, Cambridge,Mass. Atfarjan,HratJja. 1940. Khnnueiwnnor d3uzaji barbari [Study of the New Julfa Dialect]. Erevan:State University Press. Blevins, Juliette. 1993. Klamath laryngeal phonology. InternationalJournal of American Linguistics 59: 237-279. Calabrese, Andrea. 1988. Towards a theory of phonological alphabets.Doctoral dissertation,MIT, Cam- bridge, Mass. Calabrese,Andrea. 1995. A constraint-basedtheory of phonological markednessand simplificationproce- dures. Linguistic Inquiry 26:373-463. Catford,J. C. 1977. Fundamentalproblems in phonetics. Bloomington:Indiana University Press. Catford,J. C. 1988. A practical introductionto phonetics. Oxford: ClarendonPress. Halle, Morris. 1973. Review of "Preliminariesto linguistic phonetics," by Peter Ladefoged.Language 49: 926-933. Halle, Morris. 1995. Lecture notes, Indo-Europeanphonology. HarvardUniversity, Cambridge,Mass. Halle, Morris, and Ken Stevens. 1971. A note on laryngeal features. In Quarterlyprogress report 101, 198-213. Research Laboratoryof Electronics, MIT, Cambridge,Mass. Hudak,Thomas. 1987. Thai. In The world's majorlanguages, ed. BernardComrie, 757-775. Oxford:Oxford University Press. Iverson, Gregory. 1983. Korean s. Journal of Phonetics 11:191-200. Iverson, Gregory,and Joseph Salmons. 1995. Aspirationand laryngealrepresentation in Germanic.Phonol- ogy 12:369-396. Keating, Patricia. 1984. Phonetic and phonological representationof stop voicing. Language 60:286-319. Kim, Chin-Wu. 1970. A theory of aspiration.Phonetica 21:107-116. Kingston, John. 1990. Articulatorybinding. In Papers in laboratoryphonology I, ed. John Kingston and Mary E. Beckman, 406-434. Cambridge:Cambridge University Press. Kingston, John, and Randy Diehl. 1994. Phonetic knowledge. Language 70:419-454. Ladefoged, Peter. 1993. A course in phonetics. 3rd ed. Fort Worth, Tex.: HarcourtBrace Jovanovich. Lombardi,Linda. 1991. Laryngealfeatures and laryngeal neutralization.Doctoral dissertation,University of Massachusetts,Amherst. Maddieson, Ian. 1984. Patterns of sounds. Cambridge:Cambridge University Press. Ohala, John, and ManjariOhala. 1993. The phonetics of nasal phonology:Theorems and data. In Phonetics and phonology. Vol. 5, Nasals, nasalization, and the velum, ed. Marie K. Huffman and Rena A. Krakow, 225-301. San Diego, Calif.: Academic Press.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438998553833 by guest on 26 September 2021 REMARKS AND REPLIES 511

Padgett,Jaye. 1991. Stricturein featuregeometry. Doctoral dissertation,University of Massachusetts,Am- herst. Pater,Joe. Forthcoming.Austronesian nasal substitutionand other NC effects. In The prosody-morphology interface,ed. Rene Kager,Harry van der Hulst, and Wim Zonneveld.Cambridge: Cambridge Univer- sity Press. Pring, J. 1962. Phonetic introduction.In S. Sofroniou, Teach yourself Modern Greek. Sevenoaks: Hodder and Stoughton. Sagey, Elizabeth. 1986. The representationof featuresand relationsin nonlinearphonology. Doctoraldisser- tation, MIT, Cambridge,Mass. Silverman,Daniel. 1995. Phasingand recoverability.UCLA OccasionalPapers in Linguistics 15. Department of Linguistics, UCLA, Los Angeles, Calif. Stevens, Ken. 1991. Acoustic phonetics. Ms., MIT, Cambridge,Mass. Thriinsson, Hoskuldur. 1978. On the phonology of Icelandic preaspiration.Nordic Journal of Linguistics 1:3-54. Varma, Siddheshwar. 1929. Critical studies in the phonetic observations of Indian grammarians.Delhi: MunshiramManoharlal. Vaux, Bert. Forthcoming.The Armeniandialect of New Julfa. Amsterdam:Rodopi. Whitney, William. 1862. Atharva-VedaPrdtis'akhya. Delhi: Parimal. Whitney, William. 1868. The TcittiriyaPratipikhya and Tribhashyaratna.Delhi: Motilal Banarsidass.

Departmentof Linguistics Harvard University Cambridge,Massachusetts 02138 [email protected]

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438998553833 by guest on 26 September 2021