Aspiration and laryngeal representation in Germanic

Aspiration and Laryngeal Representation in GermanicGregory K. Iverson

University of Wisconsin—MilwaukeeJoseph C. Salmons

University of Wisconsin—Madison

1. Introduction.*The phonetic gesture of stop consonant aspiration, which is predictable in a

Germanic language such as English, has been described traditionally as ranging from a“puff of air” upon release of closure (Heffner 1950) to the segmental occurrence of afollowing voiceless glottal approximant /h/ (Trager & Smith 1951). Within thegenerative phonology paradigm, however, aspiration has been construed as a featuralproperty rather than as an independent segment of its own, often casually identifiedsimply as [+aspiration], or, following Chomsky & Halle (1968), as a positivespecification resulting from “heightened subglottal pressure”. We take this kind of viewhere as well, employing a notation with superscripted “h” ([Ch]) to indicaterepresentations in which aspiration is encoded as an integral feature of the segment withwhich it is associated, while we explore the phonological realization of aspiration inGermanic as the reflex manifestation of a spread or open glottis, an idea first advanced inthe seminal work of Kim (1970), and since developed in Anderson & Ewen’s treatmentof ‘|O| languages’ (1987:195-199). Our review of the phonetics and phonology of thislaryngeal gesture brings together a number of superficially unrelated phenomena inGermanic, synchronic (principally, aspiration and sonorant devoicing in English) as wellas diachronic (principally, Grimm’s Law and its exceptions in obstruent clusters). Inconcluding, we outline a phonetic motivation for the shift of voiceless stops to fricativesdiachronically.

Throughout we argue explicitly that the salient property defining the voicingcontrast in Germanic languages like English is the “fortis” feature [spread glottis], not[voice] as in Romance or Slavic. This argument is built directly on and inspired by a longline of work using physiologically-based features for phonation in languages likeGerman and English, especially Kohler’s ‘power’ feature (1984) and earlier work onlenis/fortis, as well as Browman & Goldstein’s gestural approach. We also identify someof the consequences that this distinction in phonetic typology may hold for thephonology, predicting, in particular, that lexical assimilation to voicing occurs only inlanguages for which [voice] is a marked feature. For the English-type of language, then,in which only voicelessness spreads throughout obstruent clusters, laryngeal markednessis the reverse of what has generally been assumed; here the voiced obstruents areunmarked ([ ]), the voiceless marked ([spread glottis]).

This characterization brings laryngeal representation into harmony with theexpected pattern of distribution. That is, the inventory of laryngeal features must include[voice] as well as [spread glottis], given that languages like Hindi and Thai employ bothdistinctively, hence either feature could be expected to be available as the basis formarking in two-way systems. Recent arguments that two-way contrasts as in Germanicare all [voice]-based (as discussed below), however, imply that [spread glottis] iscontrastive only in systems that already exploit [voice]. Such an implicationalrelationship is not supported on phonetic or other grounds: voiceless vs. aspirate systemsare not only well-attested, but represent the second most common laryngeal contrast intwo-way systems, after voiceless vs. voiced (cf. Henton, Ladefoged & Maddieson1992:78-79). In this paper, we maintain simply that either feature may form the basis ofa two-way contrast, just as does the other laryngeal feature, [constricted glottis], inejective vs. voiceless systems.1

Aspiration and Laryngeal Representation in Germanic 2

2. Glottal width and aspiration.Kim’s (1970) pioneering work on laryngeal phonetics was based on Korean, a

language in which there are actually two degrees of aspiration in the three-way distinctivestop stystem. Besides a series of glottally ‘tense’ voiceless unaspirated stops (/p’ t’ cÈ’ k’/),Korean has both heavily aspirated (/ph th cÈh kh/) and lightly aspirated voiceless stops(/p t cÈ k/ = [p‘ t‘ cÈ‘ k‘]) contrasting in syllable-initial position.2 Rather than attributeaspiration just to a delay in the onset of voicing for a following vowel, as suggested bythe cross-linguistic survey of Voice Onset Time (VOT) variations conducted by Lisker &Abramson (1964), Kim proposed that aspiration is the automatic, aerodynamic result of aspread glottis configuration in the larynx. Thus, as abducted vocal folds begin to cometogether after release of oral closure in order to produce voicing in a following vowel, acertain amount of time passes before the vocal folds achieve the adducted state necessaryfor phonation. Up to that point, the vocal folds are not in contact, and air exiting thetrachea during the period of post-release voicelessness is perceived as “aspiration”. InKim’s (1970:77) words, “…it seems safe to assume that aspiration is nothing but afunction of the glottal opening at the time of release. This is to say that if a stop is ndegree aspirated, it must have an n degree glottal opening at the time of release of theglottal closure.”3 In Korean, then, the heavily aspirated series can be characterized ashaving the linguistically maximum degree of glottal width (around 10 mm, based onKim’s cineradiographic tracings, representable as [spread glottis]), whereas the lightlyaspirated or lax series has an only slightly open glottis in which the vocal folds areneither particularly abducted nor adducted (an opening of around 3 mm, glottis neitherspread nor constricted), while the laryngeal configuration of the tense series has the vocalfolds in rather tight approximation (less than 1 mm of opening, [constricted glottis]).

This three-way division of glottal width and the identification of aspiration asensuing from a specification of [spread glottis] have been integrated into the laryngealfeature framework worked out by Halle & Stevens (1971), and continue to form part ofthe distinctive feature stock of current phonological theory (Clements 1985, Goldsmith1990, Kenstowicz 1994). But the “puff of air” of aspiration does not always result from aspecification of [spread glottis]. In particular, as Kim pointed out well prior to thedevelopment of modern autosegmental representation, there is no aspiration producedwithin a syllable-internal cluster of voiceless obstruents in English words like spit, eventhough the glottis is spread. In essence, Kim suggested that the reason English stops arenot aspirated after /s/ is that these clusters contain but a single specification of [spreadglottis], shared between the fricative and the stop:

…the glottal movement for /p/ of /sp/ will start during /s/, i.e., the glottis willbegin to widen. This means that, if the glottis is instructed to open to the samedegree and for the same period for /p/ of /sp/ as it would for initial /p/, the glottiswill begin to close by the time the closure for /p/ is made, and consequently, bythe time /p/ is released, the glottis will have become so narrow that the voicing forthe following vowel will immediately start, and thus we have an unaspirated /p/after /s/. Note that the notion of simultaneous compatibility is crucial here, i.e.,since /s/ is voiceless and does not require the closing of the glottis, the opening ofthe glottis for /p/ does not have to wait for the completion of /s/ but can proceedsimultaneously with the oral articulation of /s/. (1970:80)

Indeed, the peak of glottal opening associated with this gesture in the cluster /sp/ lies in(the latter portion of) the /s/, as will be described below.4 Though the glottisprogressively narrows until achieving voicing in the following vowel, as it does in therelease portion of singleton stops, in /sp/ the period of “aspiration” is consumed in theoral closure phase of the stop member of the cluster. Assuming a largely constant


duration for the specification of [spread glottis] (which equates with the constancy of the“voiceless interval”, or VLI, as laid out for English by Weismer 1980), the absence ofaspiration in /s/-clusters thus reduces to the observation that the narrowing glottis whichcharacterizes the latter portion of the [spread glottis] specification is associated with thestop in the cluster, whereas the presence of aspiration in singleton stops reflectsassociation of a narrowing glottis with the release phase of the stop (equivalently, withthe initial portion of the following vowel):

s p V p V p Vh

[sp gl] [sp gl] [sp gl]

h

=

(1) Cluster: Singleton:

There is considerable phonetic as well as phonological evidence that the [spreadglottis] specification in clusters is shared. Noting that especially sibilant fricatives, like/s/, require a large glottal aperture in order to produce air flow high enough to sustaintheir characteristic turbulence, Kingston (1990) cites experimental work usingphotoelectric glottography on voiceless obstruents, both singly and in clusters, fromseveral languages—Swedish, Japanese, Icelandic, as well as English (Löfqvist &Yoshioka 1981, Yoshioka, Löfqvist, & Hirose 1981). For all of these, in single fricativesthe peak of glottal opening is coordinated with the beginning of oral constriction. Insingle stops the peak occurs later, at the point of oral release, but in clusters there is onlyone such gesture:

Peak glottal opening in clusters of a fricative followed by a stop does not occur atthe same time relative to the oral articulations as it would for either a fricative or astop occurring alone. The most typical point is close to the boundary between thetwo oral articulations, a temporal compromise between the early peak of thefricative and the late peak of the stop. (Kingston 1990:427)

This observation parallels those by Browman & Goldstein (1986a), Anderson & Ewen(1987:195-196), or more recently Goldstein (1990:447) “…that words may begin with atmost one glottal gesture”, and is consistent with a feature representation for obstruentclusters of the sort in (1), in which the same laryngeal gesture has domain over bothmembers of the cluster; however, it runs afoul of traditional phonological accountswherein the [spread glottis] feature (or [aspiration]) is not present in either segment tobegin with, and instead is added by rule to voiceless stops only at the beginning of theword or stressed syllable (see below). The fact is, however, that [spread glottis] cannotbe a property just of word and syllable initial stops, because it also occurs phonetically insingle fricatives and in clusters composed of fricative plus stop, in which case the gesturedoes not occur twice, once for each of the two voiceless segments, but rather is sharedbetween them.

3. Sonorant devoicing.Phonologically, the Obligatory Contour Principle (OCP, McCarthy 1986) would

force tautomorphemic adjacent identical specifications to reduce to one (presumablyshared) specification in any event. That is, a morpheme-internal sequence of /s/ followedby /p/ could contain at most one instance of [spread glottis] per the OCP, hence one of thetwo instances of the feature must be removed, or delinked, in order to avoid independentspecification for each voiceless obstruent in a cluster. Representations like those in (1)would then come about through feature spreading, substantial support for which lies in


the well-known English evidence for “sonorant devoicing”. Under our assumptions,[spread glottis] comes to be shared even with adjacent sonorant consonants, which are notspecified one way or the other for this feature to begin with:

(2) plan [plfiæõn] crow [k®fio] slip [slfiíp]shrimp [ß®fiíõmp] sneeze [sn‹iz] fleet [flfiit]

On the interpretation then that aspiration equates with voicelessness, i.e., that bothphenomona are realizations of an open glottis, a unified explanation becomes availablefor the compound fact that, in initial clusters, (a) stops are not aspirated after /s/, and (b)sonorants are voiceless following voiceless obstruents, viz.:

(3) In obstruent-initial consonant clusters, the feature [spread glottis] is shared.

In spit the single [spread glottis] gesture in the syllable onset is shared between the /s/ andthe /p/, as it is between the /s/ and the /l/ in slip; in the former case, “aspiration” isabsorbed by the voiceless stop’s oral closure, in the latter case it is manifested asvoicelessness in an otherwise spontaneously voiced sonorant.5 The basic constancy ofthe English VLI affects longer clusters, too, as in street or split; but here the duration ofthe single [spread glottis] gesture is generally consumed by the two obstruent members ofthe cluster—as verified in the experimental work of Repp (1984), the following sonorantremains voiced ([st®it], [splít], not *[st®fiit], *[splfiít]).6 The constraint expressed in (3)accounts automatically for this distribution, in contrast to proposals which posit forEnglish a progressive assimilatory rule of sonorant consonant devoicing (Hoard 1971,Kiparsky 1979, Nespor & Vogel 1986, Jensen 1993), because such a rule does notdistinguish voicelessness in initial stops (which otherwise results in aspiration) from itsoccurrence in clusters (which does not). This empirical difficulty does not arise if thevoicelessness of the liquids and nasals in the words of (2) is taken to be due to the sharingof a [spread glottis] gesture which inheres in voiceless obstruents.7

In languages whose voiceless stops are uniformly unaspirated and in which[spread glottis] accordingly plays neither a phonemic nor a phonetic role, e.g., Spanish(or Polish, French, Czech, Hungarian), sonorant consonants remain predictably voicedafter initial voiceless obstruents, as in [tres] ‘three’, [flori∂a] ‘Florida’, etc. Thisdifference now derives not from the presence of a sonorant devoicing rule in Englishversus its absence in Spanish, but rather from the general dynamics of [spread glottis]realization, a feature which simply is not represented in Spanish. On this account, thereis no rule of aspiration per se in English (or, of course, Spanish), either—instead, theaspirated quality of initial voiceless stops in English ensues from their basic [spreadglottis] property, which, by universal phonetic implementation, trails off into a narrowedglottis configuration toward the end of its occurrence, after the release of oral closure insingleton stops but essentially simultaneous with it in obstruent-initial clusters. Thus,assumption of a fundamentally constant duration for the specification of [spread glottis]results in the variable VOT distributions commonly observed in English and otherlanguages, i.e., no delay upon release of closure in obstruent clusters (unaspirated stops),but substantial lag upon release of singletons (aspirated stops). In languages such asSpanish, on the other hand, for which [spread glottis] is not distinctive, there is esentiallyno lag in VOT upon closure release, and hence no aspiration.

4. English aspiration distributions.The conventional understanding of aspiration in English, however, is that the

feature which characterizes it (say, [+aspiration] or [+spread glottis]) is added by rule tovoiceless stops in certain contexts, while elsewhere they remain unaspirated. And since


the influential metrical work of Kiparsky (1979), the two main contexts in whichaspirated stops are found in English—at the beginning of a word or at the beginning of astressed syllable—have been unified into a single environment, viz., the initial position ofa stress foot. Nespor & Vogel (1986:90-91) cite the examples given in (4), in which thestop /t/ is aspirated in American English, as well as those in (5), in which it is not. Thedifference directly correlates with independently determined foot structure, which Nespor& Vogel label with ‘Σ’:

(4) a. time → [th]ime [time]Σb. tuna → [th]una [tuna]Σc. toucan → [th]oucan [tou]Σ [can]Σd. typhoon → [th]yphoon [ty]Σ [phoon]Σe. terrain → [th]errain [te]Σ [rrain]Σf. detain → de[th]ain [de]Σ [tain]Σg. detention → de[th]ention [de]Σ [tention]Σh. entire → en[th]ire [en]Σ [tire]Σi. curtail → cur[th]ail [cur]Σ [tail]Σj. satire → sa[th]ire [sa]Σ [tire]Σk. reptile → rep[th]ile [rep]Σ [tile]Σl. infantile → infan[th]ile [infan]Σ [tile]Σm. longitude → longi[th]ude [longi]Σ [tude]Σn. tree toad → [th]ree [th]oad [tree]Σ [toad]Σo. sweet tooth → sweet [th]ooth [sweet]Σ [tooth]Σ

(5) a. sting → *s[th]ing [sting]Σb. abstain → *abs[th]ain [ab]Σ [stain]Σc. austere → *aus[th]ere [au]Σ [stere]Σd. after → *af[th]er [after]Σe. alter → *al[th]er [alter]Σf. satyr → *sa[th]yr [satyr]Σg. shatter → *sha[th]er [shatter]Σh. hospital → *hospi[th]al [hospital]Σi. night owl → *night[th] owl [night]Σ [owl]Σj. flat iron → *flat[th] iron [flat]Σ [iron]Σ

The metrical foot in English, following now familiar principles (Hayes 1981,Halle & Vergnaud 1987), is figured from the right edge of the word, and (usually)consists of a stressed syllable along with any following unstressed syllables; hospitalconstitutes but a single foot while infantile, with two stressed syllables, contains two feet.The general theory of prosodic phonology requires that all syllables be parsed intometrical constituents, though, even unstressed syllables left stranded by the basic footingprocedures; hence a single word-initial unstressed syllable will also be incorporated into ametrical foot, albeit a “degenerate” one, as in the first syllable of bipedal forms liketerrain and abstain. The unified generalization which characterizes the distribution ofEnglish aspiration then emerges as follows:


(6) Voiceless stops are aspirated in foot-initial position, elsewhere they are not.

Accordingly, /t/ is aspirated in the foot-initial positions of the words in (4), but not in thefoot-medial or foot-final positions of the words in (5). Kiparsky (1979) considers that the‘devoicing’ of sonorant consonants exemplified in (2) is constrained by the metrical foot,too, for it appears that assimilatory devoicing takes place following voiceless obstruentswithin feet (Islip [ayslfiíp]Σ, eye slip [ay]Σ [slfiíp]Σ ) but not between feet (ice lip[ays]Σ *[lfiíp]Σ ). Still, sonorant devoicing does not occur following voiceless obstruentsin the foot-internal environments of words like Atlas (Jensen 1983:130), apricot,acclimate, acrimony, etc. Words like these involve sequences of stop plus sonorantconsonant that do not appear at the beginnng of the stress foot, the only position wherevoiceless stops are aspirated by the terms of (6). The characterization of “sonorantdevoicing” in terms of the spreading of aspiration from a preceding stop accounts for theabsence of devoicing in these forms without stipulation, however: Sonorant devoicingoccurs just after those stops which would be aspirated if they were followed by a vowelrather than a (now voiceless) sonorant consonant. That is, sonorant devoicing and stopaspiration are instantiations of the same phenomenon, the implementation of [spreadglottis] in—as will be argued presently—the onset of a metrically prominent syllable.8

By the proposals advanced here, then, no rule to the specific effect of (6) need beposited since aspiration is merely the phonetic implementation of a [spread glottis]phonological specification. Aspiration is inherent in the voiceless obstruents of Englishbecause the [spread glottis] feature forms part of their basic representation. This gestureis present even in foot-initial /s/-clusters, as in (5a-c) [sting, abstain, austere], but thereits effects do not include aspiration because, following Kim’s (1970) originalobservation, the open glottis has already begun to close during the articulation of the stopin the cluster, achieving an adducted state of the vocal folds by the time of closurerelease. In (5d) [after], similarly, there is at most one [spread glottis] specification to beshared between the two segments /f/ and /t/, as per the observation in (3) (as well as theOCP), with the result that aspiration is again masked by the closure phase of the cluster.9In the stops of the words in (5e-j) [alter, satyr, etc.], on the other hand, the glottis willhave narrowed considerably since no aspiration occurs, and there is no other obstruentsegment with which a [spread glottis] specification could be shared. Here, amongvoiceless stops medial in the foot, the glottis appears no longer to be sufficiently spreadto induce an aspiration effect, or a devoicing effect in a following sonorant.

This reduction in glottal width is reminiscent of the glottal narrowing which takesplace among the lax obstruents in intervoiced contexts in Korean. In this language, as inEnglish, the fricative /s/ is articulated with an open glottis throughout in word-initialposition, but the phonemically lax stops /p t cÈ k/ are produced word-initially with onlyslightly abducted vocal folds, i.e., they are moderately aspirated. In word-medialposition, the space between the vocal folds is reduced in all the lax obstruents, to aboutthe same degree for both the fricative and the stop (cf. Iverson 1983b). With its glottismore widely open to begin with, this narrowing action results in a less “forceful”, orsomewhat breathy, fricative since glottal airflow has been reduced appreciably; yet thefricative remains voiceless because the vocal folds still do not come into contact. The laxstops, on the other hand, become voiced since the same degree of narrowing for themdoes result in approximation of the vocal folds, i.e., in closing of the glottis. In English,it would appear a similar “laxing” of the voiceless stops characterizes phonation in foot-medial position, both because aspiration is absent there and, in American dialects, /t/ issubject to voicing (flapping).10 In elaborating their general paradigm for “phoneticknowledge”, in fact, Kingston & Diehl (1994:431) summarize the recent research onglottal abduction relative to word position in English phonetics in exactly these terms:


Glottal opening is simply smaller intervocalically than initially and beforeunstressed than before stressed vowels, and this smaller opening leads to shortervoicing lags (VOTs) and thus less aspiration.

They go on to underscore the foot-based character of English aspiration, pointingout its sensititvity to both word position and stress:

…intervocalic stops have much longer VOTs before a stressed vowel than beforean unstressed vowel, while [word-]initial stops have relatively long VOTsregardless of the stress of the following vowel. This result is surprising given thatinitial position as well as the following stress increases the glottal opening’s sizeand duration.

In place of the conventional provision of the feature [spread glottis] just to foot-initial stops, however, we suppose that the feature is present to begin with in English andmost other Germanic languages. But this feature is implemented in its fullest realizationonly in foot-initial position—elsewhere the vocal fold abduction associated with a[spread glottis] specification is of such lesser degree that aspiration either does not occurat all or is sharply reduced:

(7) [spread glottis] is implemented with fully abducted vocal folds only in foot-initial position.

Accordingly, /t/ in the words of (4) exhibits its [spread glottis] property maximallybecause in those forms it is in the onset of a foot-initial syllable. In the words of (5a-c)[sting, abstain, austere], the feature [spread glottis] is also realized to full extent in a foot-initial onset, but there it is part of a cluster of obstruents and so is interpreted as per (3),i.e., the feature is shared, hence aspiration is absent on release of the stop, as describedabove. Elsewhere, as in the remaining words of (5), or in words like Atlas, the /t/ is notpart of the onset of a foot-initial syllable, and [spread glottis], while still present in thesegmental representation, consequently is not implemented in its strongest, aspiration-inducing form.11

The manifestation of phonological structure in fullest form precisely inprosodically prominent positions (word-initial, beginning of a stressed syllable) accordsentirely with the general principles and preference laws for syllable structure developedby Vennemann (1988), specifically, the Head Law, which Vennemann has more recentlyraised to the level of the phonological word:

The optimal phonological word form…is one in which the initial syllable, and inaddition any stressed syllable, has a very strong consonant, while all othersyllables have weak consonants… (1993:323-324)

We suggest, however, that the realization of [spread glottis] in English is even morestraightforward—yet also more variable—than is represented in (7). In particular, thoughaspiration in English is not contrastive, there are (as in Korean) multiple degrees ofaspiration, or varying extents of glottal width, among the voiceless stop articulations inthis language. By the terms of (7), aspiration occurs in the prosodically doublyprominent position of the onset of a syllable at the beginning of a foot, where a stop isboth syllable-initial and foot-initial; but as has often been observed (cf. especiallyKingston & Diehl 1994:431), aspiration is strongest (longest, noisiest) in this position ifthat is also the syllable on which primary stress falls. That is, there is greater word-initialaspiration in the primary-stress forms of (4a-c) [time, tuna, toucan] than in the unstressed


syllables of (4d-e) [typhoon, terrain], while the primary-stress word-medial aspiration in(4f-i) [detain, detention, entire, curtail] is about as strong as the primary-stress word-initial aspiration in (4a-c); similarly, the weaker word-initial aspiration in the unstressedsyllables of (4d-e) is like that in the secondarily stressed word-medial forms of (4j-m)[satire, reptile, infantile, longitude]. If we assume three general degrees ofaspiration—strong, weak, and none, as will be developed below—and thus three degreesof glottal width, as per Kim (1970), then these distinctions can be read directly off ametrical grid of the sort utilized by Halle & Vergnaud (1987). Here the metricalprominence of a syllable is ranked (0, 1, 2) according to how many asterisks it isaccorded in the grid (1, 2, 3):

(8) 2 (Word) * * * *1 (Foot) * * * * * * *0 (Syllable) * * * * * * * *

[tu na]Σ [te]Σ [rrain]Σ [en]Σ [tire]Σ [sa]Σ [tire]Σ

The metrical prominence of the syllable containing /t/ in its onset is greater in tuna thanin terrain, and so is the strength of the /t/’s aspiration; similarly, the metrical prominenceof the word-medial syllable containing onset /t/ is greater in entire than in satire, as is thedegree of aspiration of /t/. Outside of the syllable onset (night owl), or in the onset of asyllable with the lowest degree of metrical prominence (hospital), aspiration is at itslowest level, zero, though phonologically the stop itself remains characterized by thefeature [spread glottis].

The phonetic realization of [spread glottis] in English is thus more accuratelydescribed by the statement in (9), where a syllable’s “metrical prominence” correlateswith the number of asterisks in metrical grid representation. The degree of glottalopening upon closure release, and so the extent of aspiration, then varies directly with thelevel of metrical prominence (1 = moderately open glottis, 2 = fully open glottis):

(9) Vocal fold abduction in syllable onsets is enhanced in relation to metricalprominence.

In the onset of a primary-stressed syllable, glottal width will be greater than in theonset of a secondary-stressed one, and greater still than in the onset of an unstressedsyllable. In the syllable coda, irrespective of metrical prominence, [spread glottis] alsowill be implemented weakly since it is not in the onset. But on the assumption that thisfeature is a basic rather than derived property of voiceless obstruents in English,particularly of stops, there is no requirement (or justification) in the language for a rule ofaspiration per se. Rather, the extent of the aspiration which inheres in voiceless stopsvaries according to the metrical prominence of the syllables they initiate, as per theprinciple in (9), and according to whether the [spread glottis] feature is independent orshared, as per the principle in (3).12 In short, aspiration is a matter of degree, correlatingdirectly with degree of stress.

5. Germanic and other parallels.Other Germanic languages exhibit similar variation in the degree of aspiration. In

instrumental studies of Danish, Hutters (1985:16-18) has confirmed that, as expected, theduration of aspiration decreases along with the duration of oral closure in faster speech;13

however, the extent of aspiration also correlates positively with degree ofstress—syllables with main, reduced, or weak stress are initiated by voiceless stops with,respectively, heavy, light, or (nearly) zero aspiration. In an extensive impressionistic


analysis of German, Lotzmann (1975) found essentially the same distribution, with heavy(“positive”), light (“indifferent”), and zero (“negative”) degrees of aspiration. Similarly,Keating (1984:306-308) showed graphically that aspiration in German as well as Englishis stronger for syllables with main stress than with secondary, and that voiceless stopsbeginning word-initial stressed syllables are more heavily aspirated than those beginningword-medial syllables. She also recorded a three-way, stress-sensitive split in degree ofaspiration in English. The most recent experimental work on English also verifies themetrical variation of aspiration as exemplified in (8). In experiments with nonsensedisyllabic words, Cooper (1994) shows statistically significant positive correlationsbetween extent of aspiration and degree of stress in word-medial environments, and alsofinds word-initial voiceless stops to be substantially aspirated even in the absence ofstress. Cooper (1994:21) concludes that “it appears that these I[ndices] O[f] A[spiration]are distributed differently in different syllable positions and that stress, or perhaps someother prosodic variables, also contribute to the contextual variation of the IOAs.”14

The representation of English voiceless stops in terms of [spread glottis] also tiesin closely with the ‘lenis-fortis’ analysis of obstruent manner distinctions promulgatedespecially by European phoneticians working on the Germanic languages (cf. Kohler1984, Hutters 1985, others). Thus, Kohler (1984:153) takes aspiration and voicingmerely to be “…glottal reinforcements of the fortis and lenis actions of the oral valve…”,concluding as well that there is a close connection between ‘glottal opening’ and theprocesses of both sonorant devoicing and stop aspiration (1984:158). A fairly complexsystem of gestures and correlates is employed in defining the composite characteristics offortis vs. lenis articulation, but we consider the feature [spread glottis] to be primary, atleast phonologically, for the Germanic voiceless obstruents phonemes under discussionhere. In fact, Kohler (1984:160-161) expresses a similar view in explaining part of theanatomical basis for fortis, or “power”, articulation:

An abduction of the vocal folds can also lead to a higher power because, withvolume × pressure = power, a rise in airflow at a constant subglottal pressureincreases the air-stream power. Thus the dynamics of the supraglottalarticulators…and/or the air-stream, regulated by the glottal valve, imply greatereffort in the fortis obstruent through higher power.

In our terms, broadly abducted vocal folds equate with a specification of [spread glottis],the primary gesture with which fortis articulation is then associated.

We follow Kohler as well in ascribing the cluster of fortis properties to basicarticulation mode in the Germanic languages. This means, specifically, that the two-waylaryngeal contrast among the obstruents of English (German, Danish, etc.) is encoded as afortis versus lenis distinction, or spread versus nonspread glottis, not as merely voicelessversus voiced. On the other hand, we represent the two-way laryngeal contrast inlanguages like Spanish or Polish (cf. Keating, Mikos‰, & Ganong 1981:1268) by means ofthe feature [voice] rather than [spread glottis], the phonetic result of which is that thephonologically voiced stops are voiced even during the closure phase, whilephonologically voiceless ones are simply voiceless, not also aspirated. This familiartypological difference between the majority of Germanic (weakly or passively voiced“voiced” stops, aspirated voiceless stops) and the Romance and Slavic languages(thoroughly voiced voiced stops, unaspirated voiceless stops) is thus made fundamental, apart of the phonological representation itself. The proposal has merit beyond the level ofcomparative phonetics, however.

If we pursue this idea in the context of suggestions put forth in underspecificationtheory (Kiparsky 1982), English voiceless or fortis obstruents would be represented with


the feature [spread glottis], but the language’s voiced or lenis obstruents would beunderspecified for laryngeal features, i.e., not specified at all, which is tantamount toclaiming that it is the voiceless obstruents of English that are “marked”, the voiced ones“unmarked”. This break with tradition yields fruit on a number of counts, phonologicalas well as morphological. For example, irregular monosegmental inflectional suffixes, asin the preterite form meant (cf. mean with long vowel), are voiceless in English, butregular suffixes are voiced (weaned, creamed, etc.). Under the fortis articulationproposal, morphological markedness matches up directly with markedness in thephonology, because the irregular or marked inflections are also the phonologicallyspecified ones ([spread glottis]), while the regular, unmarked inflections arephonologically underspecified for laryngeal features ([ ]).

In the phonology per se, it is also the case that voicelessness is shared in obstruentclusters in English (per the principle in (3)), hence the plural form cats terminates in an[s] rather than in a [z] (cf. boys, girls, etc.). Interpreted in conventional assimilatoryterms as feature spread, this means that the voiceless quality of the stem-final /t/ in cathas to extend into the suffix /z/ in order to make it into [s], which is possible only ifvoicelessness is specified in some way to begin with. As underspecification theoryregularly posits [+voice] rather than [-voice] to be the specified value, however (orsimply [voice] if the feature is construed as “privative”, following Mester & Itô 1989),only voiced values can be represented phonologically; accordingly, the feature whichspreads cannot be [-voice] because this value is not available to the phonology. Thoughthis confounds the description of assimilatory devoicing in English, it does appear thatjust [+voice] (or privative [voice]) is the required representation in other systems with atwo-way laryngeal contrast, as argued in the widely cited demonstration Itô & Mester(1986) give of the Lyman’s Law phenomenon affecting “rendaku” compounding inJapanese. There the otherwise general attachment of the feature [voice] to the initialobstruent in the second element of compounds is interrupted by the presence of othervoiced obstruents (kita+kaze ‘(freezing) north wind’, not *kita+gaze), but [voice]attachment freely passes over both sonorant consonants (origami < ori+kami‘fold+paper’) and voiceless obstruents (nekojËita ‘aversion to hot food’ < neko+sËita‘cat+tongue’). Taking [voice] to be specified in voiced obstruents but not marked at allin sonorants (where it is predictable) or in voiceless obstruents (where it is absent) willaccomodate this pattern of rule-blocking in Japanese within a standard autosegmentalframework. But if [voice] is privative in Japanese, Itô & Mester reason, so must it be inall languages, because the phonological features are assumed to be universal. Obviously,this conclusion invalidates any statement crucially dependent on the absence of [voice],as progressive assimilatory devoicing has generally been described for English.

Approaches to assimilation in which devoicing does take place must thereforeaccount for the phenomenon in different terms, generally by appealing to phonologicalneutralization in syllable codas (German fragt [fRakt] < /fRag+t/ ‘asks’) and to “universaltautosyllabic devoicing” via the principle that voiced obstruents may not lie outside ofvoiceless ones in the same syllable (English cats [khæts] < /kæt+z/; cf. especially Cho1990, but also Lombardi 1991, 1995). The latter assumption does not in itself determinea unique resolution of the voicing heterogeneity, however, because tautosyllabicsequences like /tz/ could be ‘repaired’ in any number of ways, including vowel epenthesisto interrupt and resyllabify the cluster, or deletion of the offending segment, or evenvocalization of one or the other of them. On the other hand, if Germanic (but notJapanese) voiceless obstruents are represented with [spread glottis], then an appropriatelaryngeal feature is available for spreading, and assimilatory devoicing (as in English,German, Swedish, Danish, etc.) can be characterized directly as the assimilation that itappears to be.15


Languages like Spanish, Polish, or Japanese, then, in which voiced stops arevoiced throughout while voiceless stops are generally unaspirated, employ the privativefeature [voice] to distinguish voiced from voiceless obstruents; hence in these it is thefeature [voice] which is available for assimilatory spread.16 But in (most) Germaniclanguages, voicing in obstruents is passive, or redundant, and the laryngeal mannerdistinction in obstruents is encoded instead via the feature [spread glottis], contrastingunmarked voiced with marked voiceless obstruents. In these languages, “voicelessness”,not voicing, is available for assimilatory spread, hence German fra[kt], but unden[kb]ar‘unthinkable’, not *unden[gb]ar.17 With bidirectional effect, the same spreading ofvoicelessness rather than voicing appears in Swedish kö[pt] < köp+d (participle)‘bought’, not *kö[bd], or lå[kt] < låg+t ‘low’ (neuter), not *lå[gd]. There is one majorGermanic language in which the voicing property does spread, however, namely, Dutch(Booij & Rubach 1987), e.g., za[k]en ‘pockets’, za[gd]oek ‘handkerchief’ etc.18 We takeit as significant confirmation of the approach to obstruent representation outlined herethat Dutch differs from the other Germanic languages also in the phonetic nature of itsobstruents, having Romance-like voiced in contrast with unaspirated voiceless stops(Lisker & Abramson 1964). Thus Dutch is phonetically in the camp of languages inwhich [voice] rather than [spread glottis] would be the feature appropriate tophonological representation—and in Dutch (along with its close relative Afrikaans andYiddish), unlike the other Germanic languages, voicing spreads from one obstruent toanother.19

The scheme of laryngeal representation we suggest for two-way contrasts alongthe voice-onset continuum therefore differs from language to language, depending on thephonology and the phonetics. In the simplest systems, of course, like Hawaiian, there areno VOT contrasts at all, and hence no marked laryngeal features, either; but in two-wayVOT contrastive systems, either the feature [voice] is contrasted with an absence oflaryngeal specification (Romance, Slavic, Japanese, Dutch), or the feature [spread glottis]is so contrasted (English, German, Swedish, Icelandic). Both features are employed inthree-way VOT-contrastive systems, (e.g., Thai), as well as in more complex four-waysystems such as in Hindi. In other three-way systems, however, [constricted glottis] isinvoked rather than [voice] (Korean, Quechua), or these two features together define asystem without [spread glottis], as in Hausa, whose glottally constricted articulations areactually either laryngealized (labials, coronals) or ejective (velars) (Ladefoged 1973).There are numerous two-way systems as well in which [constricted glottis], rather than[voice] or [spread glottis], is the marked laryngeal feature (e.g., K’ekchi [Mayan]), andthere are also four-way systems which exploit [constricted glottis] in various ways (Yuchi[Macro-Siouan], Sedang [Austro-Asiatic], Zulu [Bantu], Kullo [Afro-Asiatic]).20

Moreover, as Ladefoged’s (1973) pioneering survey of laryngeal typology first laid out(cf. also Iverson 1983a, Maddieson 1984), there are a number of five-way laryngeallycontrastive systems (e.g., Sindhi [Indic], Siswati [Bantu]), and even six-way systemshave been reported (Beja [Cushitic], Igbo [Kwa]). In our terms, these fill out theremaining combinatorial possibilities of the simplex laryngeal features, taking intoaccount that [spread glottis] may not cooccur with [constricted glottis] for obviousanatomical reasons and that all complex systems (three or more contrasts) seem toemploy at least one VOT distinction. The table in (10) thus exemplifies knowncombinations, arranged in increasing order of presumed complexity as reflected in thenumber and kind of contrasts.


(10)/p/ /b/ /ph/ /bh/ /p’/ /bÔ ~ £/

Hawaiian [ ]English [ ] [spr gl]Spanish [ ] [voice]K’ekchi [ ] [cnstr gl]Thai [ ] [voice] [spr gl]Korean [ ] [spr gl] [cnstr gl]Hausa [ ] [voice] [cnstr gl]Hindi [ ] [voice] [spr gl] [spr gl]

[voice]Yuchi [ ] [voice] [spr gl] [cnstr gl]Sedang [ ] [voice] [spr gl] [cnstr gl]

[voice]Zulu [ ] [spr gl] [cnstr gl] [cnstr gl]

[voice]Kullo [ ] [voice] [cnstr gl] [cnstr gl]

[voice]Sindhi [ ] [voice] [spr gl] [spr gl] [cnstr gl]

[voice] [voice]Siswati [ ] [voice] [spr gl] [spr gl] [cnstr gl]

[voice]Beja [ ] [voice] [spr gl] [spr gl] [cnstr gl] [cnstr gl]

[voice] [voice]

In all cases, save ordinary Germanic as represented by English, the unmarked typeof stop falls in the voiceless (unaspirated) series. These segments can be expected then toexhibit the familiar phonological qualities of the unmarked, serving as the site to whichrather than from which other features spread, and standing in as the product that resultsfrom nonassimilatory neutralizations. Such properties are in fact characteristic of the so-called voiced stop series in English and most other Germanic languages, where in word-initial position the phonemically voiced stops are phonetically only moderately voiced, ifat all, while initial voiceless stops are aspirated according to the pattern laid out in theprevious section. Support for this representation of laryngeal structure in most of theGermanic languages is thus substantial, but it has recently nonetheless been specificallyargued against on the basis of one further phonetic effect, the raising and lowering offundamental frequency (Fo) in neighboring vowels.

Kingston & Diehl (1994) observe that voicing in obstruents generally correlateswith a lowering of pitch in neighboring vowels. This effect was noted in, among otherlanguages, Spanish, French, Italian, Portuguese, and Japanese, all of which would berepresented with phonological [voice] under the scheme developed here. As voicephonation generally entails a slackening of vocal fold tension (Halle & Stevens 1971), aprimary means by which lower pitch is achieved, the correlation between phonological[voice] and Fo values falls into place. But a similar lowering of adjacent Fo values wasalso recorded in English, German, Swedish, and Danish, languages which, as representedin (10), have laryngeally underspecified voiced obstruent phonemes, and so offer nofeature ([voice] or any other) on which the pitch lowering effect could be defined.Kingston & Diehl conclude, therefore, that even these Germanic languages withphonetically often unvoiced lax obstruents should oppose them to the phonemicallyvoiceless, phonetically aspirated class via the feature [voice]. The lowering en passant ofFo in a neighboring vowel would then be due not to any strictly phonetic property of


voicing itself, but rather to the phonological presence of the feature [voice]: Only if[voice] is available in the phonological representation, obviously, is it possible to definevocalic pitch lowering on the feature [voice].21

However, Kingston & Diehl themselves also point toward work, specifically onaspirated stops in German (pp. 439-440), which shows that spread glottis phonation has acomplementary, raising effect on Fo in neighboring vowels. Since the relation betweenpitch values in vowels next to (partially) voiced stops versus voiceless aspirated stops isrelative rather than absolute, the comparatively lower Fo values associated with voicedphonemes in the Germanic cases could be attributed instead to the higher Fo values thatare found in vowels next to voiceless aspirated stops. That is, Fo correlates positivelywith glottal aperture in two-way contrastive systems of the Spanish and English type:narrow glottal aperture (voicing) has a lowering effect, wide glottal aperture (aspiration)has a raising effect. Accordingly, Fo is lower in vowels next to voiced versus voicelessphonemes in either type of representation, irrespective of whether the system marks[voice] but underspecifies voicelessness or marks [spread glottis] but underspecifiesvoicing.22

This is true as well of languages in which the phonemic status of the [spreadglottis] feature is not in dispute given the contrasts that exist between aspirated andunaspirated stops, such as in Korean, Thai, or Hindi. In the three-way system of Korean,moreover, the feature [voice] is everywhere redundant—not just in sonorants, but also inthe obstruents (cf. Iverson 1983b, 1989), because though the phonemically lax series isvoiceless (and even slightly aspirated) in word-initial environments, its allophones arevoiced in context between sonorants, which are themselves redundantly voiced.Following general principles laid out in the theory of Lexical Phonology (Kiparsky1982), Structure Preservation in particular (which prohibits phonological reference tophonetically redundant features), the appropriate laryngeal representation for the laxseries in Korean is total underspecification, as in (10). Nevertheless, as Kingston &Diehl (1994:434-435) further observe, Fo values in vowels next to the phoneticallyvoiceless initial lax stops are low relative to the glottally constricted and the aspiratedseries. Tensing of the vocal folds in the glottally constricted series will account for theraising of vowel pitch relative to the unmarked lax series, and the open glottal aperture ofthe aspirated stops has an effect parallel to that noted for English or German. This isreminiscent of the general Germanic situation, then, in which Fo values are lowest next tothe class of laryngeally underspecified obstruents by virtue of the raising effect that otherphonation types have on fundamental frequency. The view that only a phonological[voice] specification can result in relative lowering of Fo, conversely, would force therepresentation of [voice] onto the phonology of Korean, too, where it clearly isrepresentationally inappropriate (see, most recently, Jun 1994).23

In Thai and Hindi, where [voice] actually is phonological, Kingston & Diehlreport that Fo values are lower too in vowels next to voiced stops than next to voicelessunaspirated (like in Spanish) or voiceless aspirated ones (like in Korean or German). Butvoiceless unaspirated stops in these languages also have an Fo raising effect, one whichexceeds that even of the aspirated stops. This leads us to infer that voiceless unaspiratedstops here are not “lax” as in Korean and Germanic, but rather are redundantly “tense”,i.e., produced with an increased stiffness of the vocal folds relative to their counterpartsin Korean and Germanic. Absence of such stiffness induces the passive voicing whichthe unmarked stops undergo in Korean and Germanic, but its presence in Thai and Hindi,and perhaps other languages, inhibits the redundant voicing of laryngeally unmarkedstops.24


That voiceless unaspirated stops in such languages can be considered in somesense tense is supported by phonetic evidence from both Thai and Hindi. For Hindi,Davis (1984:190-191) defines the unaspirated stops as having the goal of a closed glottisat time of release; more specifically, Hindi [k] may have (limited) glottal opening duringoral closure, but the glottis returns to closed position before oral release. She furtherfollows M. Ohala’s (1983) description of aspirates and breathy voiced (murmured) stopsbeing grouped together under a feature [+reduced glottal resistance after release], thusimplying greater glottal tension or resistance in the “plain” voiceless stops. In Thai, thesituation is more complex. The voiceless unaspirated stops are usually described ashaving “tense phonation”, traditionally and impressionistically attributed to simultaneousoral and glottal closure and release. Gandour & Maddieson (1976:263), however, showthat the glottis is not closed at release in this stop series and ascribe the clearly tensequality of the voiceless unaspirated stops to pharyngeal rather than glottal constriction.Rischel (1985:52-53) indicates that this is specifically “a narrowing in the low pharynx,appearing as a retraction of the epiglottis, i.e. a (low) pharyngealization”. Tongue rootretraction is documented in Thai, especially in the anterior stops /p t/, but Rischelcautions that constrictions in the back oral cavity and pharynx are too acousticallydeceptive to allow conclusions without physiological data. Rischel notes further(1985:73 and elsewhere) that even within dialects of a single language, conflictingevidence exists about the pitch perturbing effects of consonant articulation, which hespeculatively attributes to the fact that voiceless unaspirated stops may be produced in avariety of different ways, including with accompaniment of the pharyngeal tensionnoted.25 Most recently, Abramson & Erickson (1992) bring forth evidence indicatingthat, while /p/ and /p·/ pattern together on some measures, changes in fundamentalfrequency play no major role in the perception of the aspirate. Nonetheless, the evidencedoes indicate that voicless unaspirated stops in Thai and Hindi are not of the same laxquality as in languages (Korean, Germanic) in which these segments are susceptible topassive voicing, or have an Fo lowering rather than raising effect relative to other stops.

6. Historical implications: the exceptions to Grimm’s Law.The assumption of [spread glottis] as a contrastive feature in most of Germanic

has important diachronic consequences as well. In particular, it is well-known that theshift of Indo-European voiceless stops to Germanic voiceless fricatives was blocked byan immediately preceding fricative, without regard to whether that fricative was inheritedor itself came about as a product of the application of Grimm’s Law, as illustrated in (11-13) (following Lehmann 1986).

(11) Shifted: p, t, k → f, †, x

IE *pëteÌÂr → Old Icelandic fa∂ar ‘father’IE *pelu → Gothic filu ‘very, much’IE *tak- → Gothic óahan ‘to be silent’IE *tewtaÌ → Old Icelandic ójoÌ∂ ‘people’IE *kap- → Gothic hafjan ‘to lift’IE *kai-lo → Gothic hails ‘hale, healthy’IE *deks → Gothic taihswa ‘right’


(12) Unshifted: p, t, k in /s/-clusters

IE *(s)pyaw- → Gothic speiwan ‘(to) spit’IE *(s)ter- → Gothic stairno ‘star’IE *(s)kel- → Gothic skulan ‘to owe’

nwIE *peyskus → Gothic fisks ‘fish’IE *esti → Gothic ist ‘is’ (3rd sg.)

(13) Unshifted: t in double-stop clusters (-pt- → -ft-, -kt- → -xt-)26

nwIE *kap-to- → Gothic hafts ‘captured, prisoner’IE *skap-t- → Old English sceaft ‘shaft, pole’IE *nok‚t- → Gothic nahts ‘night’IE *oktoÌw- → Gothic ahtau ‘eight’

The “exceptions” in (12) and (13) have been the subject of considerablespeculation. Almost all traditional accounts tie the fricativization element of the shiftexemplified in (11) to aspiration: “Again and again, a series of aspirated voiceless stopshas been assumed as an intermediate stage in the shift of the voiceless stops” (Schrodt1976:218).27 Unshifted forms then reflect phonetically unaspirated stops in the proto-language, parallel to the lack of aspiration among stops in s-clusters in the variousdaughters (cf. e.g., Grammont 1933). However, this approach implies that the partiallyshifted cluster in Germanic forms like Gothic ahtau ‘eight’ arose from IE *okto via [kht].With aspiration in the first segment rather than in the second, intervocalic clusters likethese seem highly implausible phonetically, and conflict in any case with the generalassimilatory and neutralization processes operating in Indo-European, includingBartholomae’s Law, which shifts “aspiration” to the end of the cluster (see below).Variants of the aspiration theory include the Brugmann & Delbrück (1930:699) proposalthat the fricativization of the first stops in stop+stop clusters may have been a very earlyphenomenon, well prior to the actual sound shift, a view taken up by Streitberg(1963:115) but which has not found general favor since.28 Others, such as Keller(1978:86), suggest vaguely that these arose “presumably because there was neutralizationand subsequent reassignment”. Even some attempts at basic observation proveinadequate: Meyer (1894) proposed a rule prohibiting sequences of two fricatives inProto-Germanic, which naturally fails on examples such as Gothic taihswa, from Proto-Indo-European *deks-.

The most recent account to rely on aspiration as a trigger for Grimm’s Lawspirantization is the proposal by Garrett & Hale (1993) that *okto- would have syllabifiedas o.kto- because kt- was a legal word-initial cluster in Proto-Indo-European. Ifaspiration accrued only to syllable-initial stops, the fricativization of just the /k/ in /kt/would then be explained on the assumption that all and only aspirated stops underwentthe shift. This kind of syllabification is very unlikely, however, particularly for Indo-European and Germanic in view of Vennemann’s (1988:43-44) demonstration that, inthese families, “Internuclear clusters of two are heterosyllabic after short vowels.” Andthe idea that the existence of word-initial clusters should imply their same syllabificationword-medially is greatly weakened by the work of Treiman, Gross, & Cwikiel Glavin(1992), who show that allowable English word-initial clusters like sp- are consistently notinterpreted as tautosyllabic medially, but rather split between coda and onset. That is, thelegal medial syllable onsets in English constitute a proper subset of the word-initialpossibilities, just as the [-ks†s] in ‘sixths’ is possible word-finally but not in word-medialcodas.


We too postulate that fricativization under Grimm’s Law was related toaspiration, but only in the sense that the shift took place whenever the old stop wasarticulated with a spread glottis, as voiceless stops generally still are in the modernGermanic languages, aside from Dutch. Again in parallel to the modern languages, wefurther suppose that obstruent clusters shared a single laryngeal gesture (as describedabove in section 2), with peak glottal width occuring toward the end of the first obstruent,narrowing to the point of closure during the course of the second obstruent. A phoneticprerequisite for the shift of voiceless stops to fricatives then was a substantially openglottis, present in the articulation of singleton voiceless stops, or in stops which formedthe first (but not the second) half of a stop cluster. Thus the [spread glottis] specificationdid not necessarily result in aspiration; in particular, the [k] in a [kt] cluster need not havebeen aspirated at all (indeed, could not have been if, as in English today, it typically wasnot released in that position; cf. active [æk\tìv], etc.).29 The [k] in *okto (unremarkablysyllabified as *ok.to) shifted, in other words, because it was produced with an openglottis, but the [t] did not, because it was produced with a narrow(ing) glottis.30

The present analysis also brings the fricativization element of Grimm’s Law intoconformity with other aspects of Indo-European phonology. The regressive voicingassimilation exemplified in (14) and (16) indicates that intermorphemic clusters in theproto-language also had shared laryngeal properties, presumably involving only a singleglottal gesture even heterosyllabically, and (15) illustrates that the process remainedmorphophonemically active into some of the attested languages (Szemerényi 1989:108).

(14) IE *yeug+to Sanskrit yuk-tá ‘joined’ < ‘to tie, bind’Greek zeuk-tós ‘joined’

*weid+to Sanskrit vitta- ‘seen, known’, via *wid-to ‘to see, know’

(15) Latin rego ‘lead, direct’ 1.sg.pres. vs. rectus perf.pass.part.lego ‘gather, read’ 1.sg.pres. vs. lectus perf.pass.part.

(cf. “Lachmann’s Law”)

(16) IE *sed- ‘sit’: *ni+sd+os > *nizdos ‘nest’*pad- ‘foot’: Avestan fra-bda- ‘fore part of the foot’

In most Indo-European daughters, the same sharing of laryngeal structure alsoapplies to contact between reflexes of a traditional “voiced aspirate” and a voiceless stop(e.g., *legh- ‘to lie’ appears without aspiration in Greek léktron ‘bed’ < *lékh-tron, etc.).But in Indic, the only family in which the voiced aspirate articulation type was retained,an apparently progressive assimilation known as Bartholomae’s Law took place, wherebysequences of voiced aspirate plus voiceless stop became simple voiced stop plus voicedaspirate. This too reflects the sharing of larygneal features within a cluster, of thegestures for both voicing and aspiration (data from Szemerényi 1989:106-107):31

(17) *bheudh + ta → Sanskrit buddha ‘the awakened one’*labh + ta → Sanskrit labdha ‘grasped’ (adj.)*augh + ta → Avestan augzÈa ‘he said’

Voiced aspirates are actually murmured or breathy stops, characterized in theHalle & Stevens (1971) feature system by the property of [slack vocal folds] (or simply[voice], as we shall refer to it) coordinated simultaneously with a specification for [spreadglottis]. This results in a glottal configuration quite different from the [spread glottis]gesture in voiceless sounds, because in murmur the vocal folds are held in approximationthroughout the anterior extent of the glottis, vibrating in order to produce voicing, but


they are moderately spread at the posterior portion, held apart by the arytenoid cartilagesin order to maintain the relatively turbulent airflow which is characteristic of aspiration(Ladefoged 1993). In voiced aspirates, in other words, the feature of [voice] combineswith [spread glottis] to form a laryngeal state distinct from either of these twospecifications alone, resulting in the breathy kind of voicing known as murmur. Thesharing in clusters of this special voicing (in the Indo-European daughters which retainedit) gives rise to Bartholomae’s Law, which we take to be the phonetic implementation ofmurmur as extended over two obstruents in succession, shared the same way as are otherlaryngeal gestures in clusters:

(18) “Bartholomae’s Law”

b h

sp glvoice

t=

sp glvoice

b dh

Extending throughout the cluster, this combination of features results in a distinctiveburst of “voiced aspiration” upon release of the second closure. The implementation of[spread glottis] in combination with [voice] thus differs from the essentially aerodynamicmanifestation of [spread glottis] in the absence of any vocal fold vibration: the former,murmur, is a largely steady-state configuration in which the major portion of the glottis isalready closed throughout the articulation; the latter, voicelessness or aspiration,progresses over time from a wide open to a closed state of the glottis.32 But both gesturetypes were shared between the members of an obstruent cluster, and both the progressiveand regressive laryngeal assimilations of Indo-European are instantiations of the samefeature-sharing phenomenon as explains the apparent exceptions to fricativization underGrimm’s Law.33

The present account of aspiration in modern Germanic languages and of thefricativization exceptions to Grimm’s Law also provides a motivation for the process bywhich especially aspirated voiceless stops tend to shift to fricatives. Kohler (1984:164)observes that voiceless aspirated stops generally have a “weaker and shorter” closure thanunaspirated stops (a substantial portion of an aspirated stop’s duration is taken up in therelease phase, whereas the entirety of an unaspirated stop is contained in the closure).And as Ohala (1990:438) has recently observed, “…the shorter the closure, the morelikely it is to be incomplete, thus a spirant…” (cf. also Grammont 1933:167-168). Inaddition to this articulatory “weakness” of oral closure as a motivation for their shift tofricatives, it can be observed that the wide glottis of aspirated stops parallels the laryngealconfiguration required for strident voiceless fricatives like /s/ (Kingston 1990:427). InIndo-European, where /s/ was the only fricative obstruent, the [spread glottis] gesturepresumably would have been particularly salient, and rather easily identified as theprimary manner property of the weakly occluded, aspirated stops, thus providingperceptual support as well for their shift to fricatives.

7. Conclusion.This paper has argued for the phonological representation of voiceless stops in

most Germanic languages via the feature [spread glottis]. Together with the observationthat that gesture is shared in obstruent clusters, a simpler understanding emerges ofaspiration and sonorant devoicing in English, accounting for a broader range of facts andinvoking less machinery in the process. The analysis also relates the varying degrees ofaspiration to the independently required weightings of metrical structure, and achieves arapprochement with the “fortis/lenis” distinction which has recently been advocated for


Germanic by several phoneticians. Thus a typology emerges between languages whichemploy [voice] (Romance, Slavic) and those which employ [spread glottis] (Germanic) intwo-way obstruent systems, though in more complex systems, as in Hindi, both featuresare used contrastively. The diachronic implications of this view are also significant.Based on the reasonable assumption that Proto-Germanic had essentially the samephonetics as most of its daughters, the long-unexplained set of exceptions to Grimm’sLaw falls out naturally as a consequence of the [spread glottis] feature having beenshared in obstruent clusters, a pattern which also fits in with a range of rules or lawspointing toward the sharing of laryngeal features within obstruent clusters in otherbranches of Indo-European.


NOTES

* We thank Robert Channon, Phillip Connell, Stuart Davis, Randy Diehl, Daniel Dinnsen,Fred Eckman, Colin Ewen, Jack Gandour, Kee-Ho Kim, John Kingston, EdithMoravcsik, Mary Niepokuj, Monica Macaulay, Richard Page, Robert Port, Hyang-SookSohn, and Ronnie Wilbur for various comments on and discussions of this project,though of course none of them is responsible for any errors or infelicities of expressionthat may follow. We are also grateful to three anonymous reviewers for this journal whocontributed significantly to our views on the problems discussed below.1 Recent work in phonetics makes clear that multiple phonetic cues are employed tosupport phonological distinctions: ‘…some contrasts, most notably those for phonation,use a large number of cues for each distinctive feature’ (Henton, Ladefoged & Maddieson1992:96; see also Kingston & Diehl 1994 and Kohler 1984 for similar views). In seekingthe phonological basis of (especially two-way) laryngeal contrasts, we start from the roleof glottal width and glottal tension but will deal also with other key cues throughout, suchas VOT and the fundamental frequency of vowels following different stop types.2 In syllable-final position, laryngeal manner contrasts are neutralized to the unreleasedstops [p\ t\ k\]; additionally, the lightly aspirated series is voiced to [b d jÈ g] inintervoiced contexts—cf. Iverson (1983a, 1989).3 Similarly, Ladefoged (1993:142) holds that “In general, the degree of aspiration (theamount of lag in the voice onset time) will depend on the degree of glottal apertureduring the closure. The greater the opening of the vocal cords during a stop, the longerthe amount of the following aspiration.”4 A similar wide glottis configuration is known to attend the fricative /s/ in Korean evenas a singleton (following fiberscopic data reported by Kagaya 1974; cf. also Iverson1983b).5 Robert Port has pointed out to us that the extent of sonorant devoicing followingfricatives is likely less than that following stops, a circumstance which, in the presentframework attributing devoicing to a fixed duration of the [spread glottis] feature, is anautomatic consequence of the longer closure period in fricatives relative to stops.6 In formal elicitations conducted under laboratory conditions, Repp (1984:118-119)found that voicelessness does not extend into the liquid in three-member /s/-clusters (e.g.,the /p/ in splash exhibited “…an initial 10-ms release burst, which preceded the firstglottal pulse of the [l] segment”). Perhaps in faster speech, with shorter segmentdurations, the [spread glottis] gesture may have domain over the initial portion of thesonorant member in a triconsonantal cluster, but Repp did not find this experimentally.The extent of the spread of this feature appears to be just a function of the compositeduration of the segments, whatever their number, that comprise the cluster with which thefeature is associated.7 Cho (1991) presents an overview of languages in which (unlike English) voicing insonorants is contrastive, arguing as here that the appropriate representation of voicelesssonorants is with [spread glottis], not the absence of [voice].8 The (partial) sonorant devoicing observed in Islip [ayslfiíp]Σ is then due to some metricalprominence in the second syllable, which retains the full vowel [I]; but sonorantsfollowing foot-medial voiceless fricatives do not devoice if the syllable containing the


sonorant has no prosodic prominence (and the unstressed vowel is schwa), e.g., Kessler[khèslë®].9 Such sharing of a single laryngeal gesture even in medial clusters of voicelessobstruents has been reported by Munhall & Löfqvist (1992), and is also the basisproposed by Cooper (1994) for lack of aspiration in words like after.10 Similarly, in Kiparsky’s (1979) account of English aspiration foot-internal voicelessstops become [+lax] after nonconsonantal segments; [-lax] stops then acquire the feature[+aspiration], though this would need to be restricted to foot-initial position since the /t/in sting, for example, is neither lax nor aspirated.11 In (5d) and (5h), the stop forms a heterosyllabic cluster with a preceding stridentfricative, a segment type which generally retains a more substantially open glottisthroughout its articulation in order to maintain turbulent airflow (Kingston 1990); in thesecases, absence of aspiration in the stop member follows the familiar pattern of otherfricative plus stop clusters.12 Heterosyllablically, the two principles may intersect, so that [spread glottis] is sharedbetween a coda stop and a metrically prominent onset, as in (a[kth]ivity ); aspiration stillaccrues to the onset stop, presumably because the short duration of the coda stopconsumes only a small portion of the laryngeal gesture (cf. intervocalic clusters ofphonetically longer fricative followed by unaspriated stop, e.g., de[sp]icable).

13 Length of closure in aspirated stops in a constant VLI language like Danish is inverselyproportional to the duration of aspiration (Weismer 1980, Fourakis & Iverson 1985), e.g.,[p] has the longest closure and shortest aspiration while [t] has the shortest closure andlongest aspiration.14 VOT is another IOA. Variation in VOT is generally a function of glottal width andother physiological/dynamic factors, but in Icelandic (among other Scandinavianvarieties, including Faroese), it appears to be controlled independently. Thus a single[spread glottis] gesture results either in preaspiration (from underlying geminate stops) orpostaspiration (underlying singletons); in sonorant consonant plus voiceless stop clusters,however, if the sonorant devoices (as is usual in casual speech) then the stop is notaspirated, e.g. vanta ‘lack’ [van›ta], not *[van›tha], and if the sonorant is voiced then thestop is aspirated: [vantha], not *[vanta] (cf. Thráinsson 1978, Kingston 1990). Thesefacts are easily accomodated if just a single instance of [spread glottis] is assigned to thecluster, varying in rate or style of speech as to the time at which it is initiated.15 Icelandic follows this pattern as well, with prototypical representation of [spreadglottis] throughout the obstruent system: the stop series written as voiced is in factvoiceless unaspirated (sometimes called ‘lenis’) [p, t, k], and the traditional voicelessstops are aspirated even in codas and in unstressed syllables in some dialects ([ph, th, kh])(Thráinsson 1994).16 For languages that employ [voice] rather than [spread glottis], we follow Cho (1990) incharacterizing regressive assimilation to voicelessness in obstruent clusters in terms ofthe delinking (or underparsing) of [voice] in position before another obstruent; thus /g+t/→ [kt] via delinking, /k+d/ → [gd] via spread (and, with vacuous effect, /g+d/ → [gd] viaboth delinking and spread).


17 In normal spoken German (Umgangssprache), words like undenkbar are typicallypronounced with voicelessness permeating the cluster, although the morpheme-initial /b/retains lenis characteristics: unden[kb›]ar (cf. Ramers & Vater 1991:92, Kloeke 1982).Such “devoicing” falls out automatically in virtue of the fact that the [spread glottis]specification in /k/ inhibits passive voicing of the following /b/, much as word-initialposition also inhibits passive voicing.18 Dutch also appears to assimilate voiced fricatives to voiceless regressively, as intre[k]+vogel > tre[kf]ogel ‘migrant’ (Booij & Rubach 1987). Following Cho (1990), thiswould be the result of a neutralization that delinks [voice] in fricatives following asyllable-final obstruent. See also Lombardi (1995). De Schutter (1994:448, 451),however, reports that many Dutch speakers devoice fricatives in word-initial positiongenerally, irrespective of assimilatory considerations (and some varieties—selectednorthern dialects and Surinam Dutch—devoice fricatives in all environments). On theapparent mismatch between the presence of phonetic voicing and judgments of voicingby native speakers, see also Slis (1986) and van den Berg (1988).19 Both Dutch and German ‘devoice’ syllable-final obstruents. We take this to be aprosodically determined phenomenon to the effect that laryngeal features are not licensedto occur in the syllable coda, whether [voice] (Dutch) or [spread glottis] (German).Obstruents without phonological specification for laryngeal features are then interpretedphonetically as voiceless, generally without aspiration.20 This does not necessarily preclude there being languages which utilize more than onefeature in a two-way system. If we interpret prenasalized stops as ‘hypervoiced’(Henton, Ladefoged & Maddieson 1992), then the Bantu language Chagga offers a likelyexample, since there prenasalized stops contrast with (aspirated) voiceless stops.21 Browman & Goldstein (1986b:341), however, point to evidence that Danish Fo valuesactually do not conform to the predictions of Kingston & Diehl’s voicing analysis.Namely, both aspirated and unaspirated initial stops in Danish appear to show highfalling Fo patterns.

22 Ohde (1984), in fact, presents data showing that unaspirated voiceless stops in Englishs-clusters show Fo effects like those of voiceless aspirated stops, not like the ‘voiced’stops. He concludes that these results are “generally consistent with a vocal cord tensionhypothesis” of Fo effects, tension which is further consistent with forward positioning ofthe hyoid bone and raising of the larynx. His hypothesis that the vocal cords becometense during /s/ and “remain in this physiological state through the stop and vowelarticulations” (1984:228) parallels our characterization of laryngeal feature sharing insuch clusters.23 This in fact is the characterization of Korean that Kingston & Diehl (1994:435) areforced to accept in order to account for the relatively lower Fo values associated with laxstops under the assumption that a phonological specification for [voice] is the propertyresponsible for these pitch effects.24 Kingston & Diehl (1994:434-435) point out that several forms of Chinese have just atwo-way laryngeal contrast between voiceless stops (unaspirated versus aspirated), andhere too the unaspirated series has an Fo elevating effect relative to the aspirated series.We attribute this raising as well to vocal fold stiffness in phonation of the voicelessunaspirated stops.


25 Rischel’s skepticism in this passage about drawing simple correlations between statesof the larynx and pitch strikes us as well-founded.26 We leave aside here the broader Northwest Indo-European development of *-tt- to -ss-throughout Celtic, Germanic and Italic, as well as to -st- in a number of other daughters.See Streitberg (1963:115-116), Prokosch (1938:85) and Szemerényi (1989:108-109) forgeneral discussions of this development in an areal and general Indo-European context.27 “…immer wieder wird als Zwischenstadium der T[enues]-Verschiebung eine Reihevon T[enues] A[spiratae] angenommen.”28 The suggestion is based on certain parallels in other branches of Indo-European,including developments like Old Irish recht ‘law’ < *-kt.29 We surmise that Indo-European did not evince clusters of the form [kth] as found inactivity, where metrical prominence evokes aspiration in the second element. This viewconforms to the accentual typology developed by van Coetsem, Hendricks & McCormick(1981), who distinguish between “non-dominating” and “dominating” systems, theformer showing no accent-related reductions, the latter reducing metrically lessprominent syllables. On the common assumption then that Pre-Germanic maintained theIndo-European “pitch” or “non-dominating” accent, implementation of [spread glottis]would have been essentially constant from syllable to syllable, so that (9) presumablyplayed no role in the grammar of Indo-European and Pre-Germanic.30 The only previous observation close to this is the intriguing remark by Prokosch(1938:60), who suggests that shift in double stop clusters did not occur “apparentlybecause the surplus expiration was absorbed by” the preceding fricative.31 The feature sharing analysis also conforms to the view that Bartholomae’s Law was atwo-stage process of complete laryngeal assimilation which later comes to have murmurrealized only on the second stop, viz., /bh+t/ → /bhdh/ → /bdh/. See Collinge (1985:8-10) for a review of such proposals, including Schindler’s (1976) rule for Proto-Indo-Iranian, which has /dht, tdh, ddh/ all becoming /dhdh/.32 Synchronically, as Davis (1994) has persuasively argued for Hindi, voiced aspiratesmay not manifest breathy voice per se at all, but the four-way laryngeal contrast isnonetheless phonetically accurately captured with the features [voice] and [spreadglottis]. Both aspirate series in Hindi are produced with a relatively wide glottis, thoughvoiceless aspirates show a longer lag before voicing due to greater glottal width, voicedaspirates a shorter lag due to a less spread glottis. Phonetic use of breathy voice thenserves to enhance the perceptual saliency of the voiced aspirate series, which will berepresentationally more marked ([spread glottis, voice]) than either the voiceless aspirates([spread glottis]), the simple voiced ([voice]), or the voiceless unaspirated series ([ ]).33 Additional evidence for the sharing of laryngeal features in Indo-European lies in theprocess known as “Siebs’ Law” (Collinge 1985:155-157), which neutralized laryngealdistinctions in initial position following the sole fricative in the system, /s/. Thus, *s+k-and *s+g- yield *sk-, *s+gh yields *sk: Dutch stoom ‘steam’ vs. doom ‘steam, fog’, etc.,Middle High German schellen vs. Old High German gellan ‘to make noise, a racket’.(We leave aside the far less certain non-Germanic instantiations of this rule.)

23

REFERENCES

Abramson, Arthur S. & Donna M. Erickson. 1992. “Tone Splits and Voicing Shifts inThai: Phonetic Plausibility.” Haskins Laboratories Status Report on SpeechResearch SR-109/110.255-262.

Anderson, John M. & Colin J. Ewen. 1987. Principles of Dependency Phonology.Cambridge: Cambridge University Press.

Berg, R. J. H. van den. 1988. “The Perception of Voicing in Dutch Two-ObstruentSequences: A Comparison of Synthetic and Natural Speech,” Journal ofPhonetics 16.171-180.

Booij, Geert, & Jerzy Rubach. 1987. “Postcyclic versus Postlexical Rules in LexicalPhonology,” Linguistic Inquiry 18.1-44.

Browman, Catherine P., & Louis Goldstein. 1986a. “Toward an ArticulatoryPhonology,” Phonology Yearbook 3.219-252.

Browman, Catherine P., & Louis Goldstein. 1986b. “Representation of voicing contrastsusing articulatory gestures,” Journal of Phonetics 14.339-342.

Brugmann, Karl & Berthold Delbrück. 1930. Grundriß der vergleichenden Grammatikder indogermanischen Sprachen. 1. Band: Einleitung und Lautlehre. Berlin:Walter de Gruyter. [Originally published 1886.]

Cho, Young-mee Yu. 1990. Parameters of Consonantal Assimilation. Ph.D.dissertation, Stanford University.

Cho, Young-mee Yu. 1991. “‘Voiceless’ Sonorants are Aspirates,” Paper presented atthe 65th Annual Meeting of the Linguistic Society of America, Chicago.

Chomsky, Noam, & Morris Halle. 1968. The Sound Pattern of English. New York:Harper & Row.

Clements, George N. 1985. “The Geometry of Phonological Features,” PhonologyYearbook 2.225-254.

Collinge, N.E. 1985. The Laws of Indo-European. (Current Issues in Linguistic Theory,No. 35.) Amsterdam and Philadelphia: John Benjamins Publishing Co.

Cooper, André M. 1994. “Aspiration in English: How Should it be Defined and WhereShould it be Described?,” Manuscript, University of Michigan, Ann Arbor.

Davis, Katharine. 1994. “Stop Voicing in Hindi,” Journal of Phonetics 22.177-193.

De Schutter, Georges. 1994. “Dutch,” in Ekkehard König & Johan van der Auwera,eds., The Germanic Languages. New York: Routledge, pp. 439-477.

Fourakis, Marios, & Gregory K. Iverson. 1985. “On the Acquisition of SecondLanguage Timing Patterns,” Language Learning 35.431-442.

24

Gandour, Jack & Ian Maddieson. 1976. “Measuring Larynx Movement in Standard ThaiUsing the Cricothyrometer,” Phonetica 33.241-267.

Garrett, Andrew, & Mark Hale. 1993. “The Phonetics and Phonology of Grimm’s andVerner’s Laws,” Paper presented at the 67th Annual Meeting of the LinguisticSociety of America, Los Angeles.

Goldsmith, John. 1990. Autosegmental and Metrical Phonology. Cambridge, MA:Blackwell Publishers.

Goldstein, Louis. 1990. “On Articulatory Binding: Comments on Kingston’s Paper,” inJohn Kingston & Mary Beckman, eds., Papers in Laboratory Phonology I:Between the Grammar and Physics of Speech. Cambridge: CambridgeUniversity Press, pp. 445-450.

Grammont, Maurice. 1933. Traité de phonétique. Paris: Delagrave.

Halle, Morris, & Kenneth Stevens. 1971. “A Note on Laryngeal Features,” QuarterlyProgress Report 101.198-192 (Research Laboratory of Electronics, MIT).

Halle, Morris, & Jean-Roger Vergnaud. 1987. An Essay on Stress. Cambridge, MA:MIT Press.

Hayes, Bruce. 1981. A Metrical Theory of Stress Rules. Ph.D. dissertation, MIT.Published by Garland Press, New York, 1985.

Heffner, R.-M. S. 1950. General Phonetics. Madison: University of Wisconsin Press.

Henton, Caroline, Peter Ladefoged & Ian Maddieson. 1992. “Stops in the World’sLanguages,” Phonetica 49.65-101.

Hoard, James E. “Aspiration, Tenseness, and Syllabication in English,” Language47.133-140.

Hutters, Birgit. 1985. “Vocal Fold Adjustments in Aspirated and Unaspirated Stops inDanish,” Phonetica 42.1-24.

Itô, Junko, and R. Armin Mester. 1986. “The Phonology of Voicing in Japanese:Theoretical Consequences of Morphological Accessibility. Linguistic Inquiry17.49-73.

Iverson, Gregory K. 1983a. “On Glottal Width Features,” Lingua 60.331-339.

Iverson, Gregory K. 1983b. “Korean s,” Journal of Phonetics 11.191-200.

Iverson, Gregory K. 1989. “On the Category Supralaryngeal,” Phonology 6.285-304.

Iverson, Gregory K. & Joseph C. Salmons. 1992. “The Phonology of the Proto-Indo-European Root Structure Constraints,” Lingua 87.293-320.

Jensen, John T. 1993. English Phonology. (Current Issues in Linguistic Theory, No.99.) Amsterdam and Philadelphia: John Benjamins Publishing Co.

25

Jun, Sun-Ah. 1994. “The Status of the Lenis Stop Voicing Rule in Korean,” in Young-Key Kim-Renaud, ed., Theoretical Issues in Korean Linguistics. Stanford: CSLI,pp. 101-114.

Kagaya, R. 1974. “A Fiberscopic and Acoustic Study of the Korean Stops, Affricatesand Fricatives,” Journal of Phonetics 2.161-180.

Keating, Patricia A. 1984. “Phonetic and Phonological Representation of Stop Voicing,”Language 60.286-319.

Keating, Patricia A., Michael J. Mikos‰, & William F. Ganong, III. 1981. “A Cross-Language Study of Range of Voice Onset Time in the Perception of Initial StopVoicing,” Journal of the Acoustical Society of America 70.1261-1271.

Keller, R.E. 1978. The German Language. New Jersey: The Humanities Press.

Kenstowicz, Michael. 1994. Phonology in Generative Grammar. Cambridge, MA:Blackwell Publishers.

Kim, Chin-Wu. 1970. “A Theory of Aspiration,” Phonetica 21.107-116. Reprinted inSojourns in Language I (Collected Papers by Chin-W. Kim). Seoul: Tower Press,1988, pp. 73-82.

Kingston, John. 1990. “Articulatory Binding,” in J. Kingston & M. Beckman, eds.,Papers in Laboratory Phonology I: Between the Grammar and Physics ofSpeech. Cambridge: Cambridge University Press, pp. 406-434.

Kingston, John & Randy L. Diehl. 1994. “Phonetic Knowledge,” Language 70.419-454.

Kiparsky, Paul. 1979. “Metrical Structure Assignment is Cyclic,” Linguistic Inquiry10.421-441.

Kiparsky, Paul. 1982. “Lexical Morphology and Phonology,” in I.-S. Yang, ed.,Linguistics in the Morning Calm. Seoul: Hanshin Publishing Co., pp. 3-91.

Kloeke, W. U. S. van Lessen. 1982. Deutsche Phonologie und Morphologie.(Linguistische Arbeiten, No. 117.) Tübingen: Niemeyer.

Kohler, Klaus. 1984. “Phonetic Explanation in Phonology: The Feature Fortis/Lenis,”Phonetica 41.150-174.

Ladefoged, Peter. 1973. “The Features of the Larynx,” Journal of Phonetics 1.73-83.

Ladefoged, Peter. 1993. A Course in Phonetics. 3rd ed. Fort Worth: Harcourt BraceJovanovich College Publishers.

Lehmann, Winfred P. 1986. A Gothic Etymological Dictionary. Leiden: Brill.

Lisker, Leigh, & Arthur Abramson. 1964. “A Cross-Language Study of Voicing inInitial Stops: Acoustical Measurements,” Word 20.384-422.

26

Lombardi, Linda. 1991. Laryngeal Features and Laryngeal Neutralization. Ph.D.dissertation, University of Massachusetts at Amherst. Published by GarlandPress, New York, 1994.

Lombardi, Linda. 1995. “Laryngeal Neturalization and Syllable Wellformedness,”Natural Language & Linguistics Theory 13.39-74.

Löfqvist, Anders, & H. Yoshioka. 1980. “Laryngeal Activity in Swedish ObstruentClusters,” Journal of the Acoustical Society of America 68.792-801.

Löfqvist, Anders, & H. Yoshioka. 1981. “Interarticulator Programming in ObstruentProduction,” Phonetica 38.21-34.

Löfqvist, Anders. 1980. “Interarticulator Programming in Stop Production,” Journal ofPhonetics 8.475-490.

Lotzmann, Geert. 1975. Zur Aspiration der Explosivae im Deutschen: Einsprechwissenschaftlich-phonetischer Beitrag zur Deutschen Hochlautung.(Göppinger Arbeiten zur Germanistik, No. 156.) Göppingen: Verlag AlfredKümmerle.

Maddieson, Ian. 1984. Patterns of Sounds. Cambridge: Cambridge University Press.

McCarthy, John. 1986. “OCP Effects: Gemination and Antigemination,” LinguisticInquiry 17.207-263.

Mester, R. Armin, & Junko Itô. 1989. “Feature Predictability and Underspecification:Palatal Prosody in Japanese Mimetics,” Language 65.258-293.

Meyer, Richard M. 1894. “Eine urgermanische Inlautregel,” Anzeiger für deutschesAltertum und Literatur 38.53-54.

Munhall, K., & A. Löfqvist. 1992. “Gestural Aggregation in Speech: Laryngealgestures,” Journal of Phonetics 20.111-126.

Nespor, Marina, & Irene Vogel. 1986. Prosodic Phonology. Dordrecht: Foris.

Ohala, John J. 1990. “The Generality of Articulatory Binding: Comments onKingston’s Paper,” in John Kingston & Mary Beckman, eds., Papers inLaboratory Phonology I: Between the Grammar and Physics of Speech.Cambridge: Cambridge University Press, pp. 435-444.

Ohala, Manjari. 1983. Aspects of Hindi Phonology. Delhi: Motilal Banarsidass.

Ohde, Ralph N. 1984. “Fundamental Frequency as an Acoustic Correlate of StopConsonant Voicing.” Journal of the Acoustic Society of America 75.224-230.

Prokosch, Eduard. 1938. A Comparative Germanic Grammar. (William DwightWhitney Linguistics Series.) Baltimore: Linguistic Society of America.

Ramers, Karl-Heinz & Heinz Vater. 1991. Einführung in die Phonologie. (KölnerLinguistische Arbeiten–Germanistik, No. 16.) Köln: Gabel Verlag.

27

Repp, Bruno H. 1984. “Effects of Temporal Stimulus Properties on Perception of the[sl]-[spl] Distinction,” Phonetica 41.117-124.

Rischel, Jørgen. 1985. “An Appraisal of Research in the Phonetics and Phonology ofThai,” Annual Report of the Institute of Phonetics, University of Copenhagen19.43-93.

Schrodt, Richard. 1976. Die germanische Lautverschiebung und ihre Stellung im Kreiseder indogermanischen Sprachen. 2nd ed. Vienna: Karl M. Halosar.

Schindler, Jochem. 1976. “Diachronic and Synchronic Remarks on Bartholomae’s andGrassmann’s Laws. Linguistic Inquiry 7.622-637.

Slis, I. H. 1986. “Assimilation of Voice in Dutch as a Function of Stress, WordBoundaries and Sex of Speaker and Listener,” Journal of Phonetics 14.311-326.

Streitberg, Wilhelm. 1963. Urgermanische Grammatik: Einführung in dasvergleichende Studium der altgermanischen Dialekte. 3rd ed. Heidelberg: CarlWinter. [Originally published 1896.]

Szemerényi, Oswalt. 1989. Einführung in die vergleichende Sprachwissenschaft. 3rded. Darmstadt: Wissenschaftliche Buchgesellschaft.

Thráinsson, Höskuldur. 1978. “On the Phonology of Icelandic Aspiration,” NordicJournal of Linguistics 1.3-54.

Thráinsson, Höskuldur. 1994. “Icelandic,” in Ekkehard König & Johan van der Auwera,eds., The Germanic Languages. New York: Routledge, pp. 142-189.

Trager, George & Henry L. Smith. 1951. An Outline of English Structure. (Studies inLinguistics, Occasional Papers No. 3). Norman, OK: Battenburg Press.

Treiman, Rebecca, Jennifer Gross, & Annemarie Cwikiel Glavin. 1992. “TheSyllabification of /s/ Clusters in English,” Journal of Phonetics 20.383-402.

van Coetsem, Frans, Ronald Hendricks, & Susan McCormick. 1981. “Accent Typologyand Sound Change,” Lingua 53.295-315.

Vennemann, Theo. 1988. Preference Laws for Syllable Sructure and the Explanation ofSound Change—With Special Reference to German, Germanic, Italian, and Latin.Berlin: Mouton de Gruyter.

Vennemann, Theo. 1993. “Language Change as Language Improvement,” in CharlesJones, ed., Historical Linguistics: Problems & Perspectives. London: Longman,319-344.

Weismer, Gary. 1980. “Control of the Voicing Distinction for Intervocalic Stops andFricatives: Some Data and Theoretical Considerations,” Journal of Phonetics8.427-438.

Yoshioka, H., A. Löfqvist, & H. Hirose. 1981. “Laryngeal Adjustments in theProduction of Consonant Clusters and Geminates in American English,” Journalof the Acoustical Society of America 70.1615-1623.

Aspiration and laryngeal representation in Germanic

Documents

Transcript of Aspiration and laryngeal representation in Germanic