Post-vocalic spirantization: Typology and phonetic motivations Alan C-L Yu University of California, Berkeley 0. Introduction Spirantization involves a stop consonant becoming a weak fricative (e.g., B, D, Ä) or an approximant (e.g., w, l). As frequently noted by phonologists (e.g. Bauer 1988, Foley 1977, Hyman 1975), non-vowel height conditioned spirantization exhibits an asymmetry in that it is most commonly found in post-vocalic contexts but rarely in syllable-initial position. It should be noted that the term postvocalic is intended as a cover term for environments such as preconsonantal and intervocalic positions. Vowel height conditioned spirantization (e.g., ti > tsi > si), specifically, will not be discussed here. Historical (1) and synchronic (2) examples can be found in languages as diverse as Italian, an Indo-European language, and Salar, a Turkic language. (1) Sanskrit mata > Mid. Indo-Aryan [maja] pacified Latin faba > Italian [fava] bean Old Turkic adaq > Salar [ajax] foot (2) Tiberian Hebrew /gaùdlu/ [gaùdlu ] be great 3 perf. /yigdaùluù/ [yiäda ùluù] be great 3mp impf. Despite the naturalness of this phonological process, as suggested by the relative frequency of post-vocalic spirantization found in the world s languages, the actual physical mechanism of such a change, that is, the change in the manner of articulation of a segment, is by no means clear (cf. Bauer 1988). The purpose of this paper is to provide a preliminary examination of this post-vocalic spirantization process. The result of a typological study of more than 200 languages is presented in section 1, showing the various realizations of postvocalic spirantization in the world s languages. By so doing, the generalizations of post-vocalic spirantization are clearly laid out. Following that, I will present a theory on the phonetic origin of post-vocalic spirantization in section 2. To support the theory, some phonetic evidence drawn from research in both the motor and perceptual domains is given. 1. Typology 1.1. Sources This typological study draws from two main sources: The Stanford Handbook of Phonological Data (Crothers et al. 1979) which is a survey of 197 grammars of languages representative of the major language families of the world, and a list of 272 lenition patterns found in Robert Kirchner s 1998 UCLA dissertation. These lenition patterns, in turn, come from Lavoie s 1996 survey of 165 languages and Kirchner s own survey of grammars. Kirchner s list is a collection of general
lenition patterns such as intervocalic voicing and pre-vocalic spirantization, in addition to a collection of post-vocalic spirantization patterns. Only the postvocalic spirantization patterns are included in the final list of patterns for the present study. 1.2. Methodology The collection of post-vocalic spirantization is first arranged into three different categories according to the environment(s) in which the spirantization applies. These categories are coda position, intervocalic position, or both simultaneously. Coda position includes patterns that only apply to word final position. Morphological boundaries are not taken into account for the purposes of this study. That is, the category for intervocalic spirantization includes contexts where the stop in question is in word final position but is preceded by a vowel and followed by a vowel initial word (e.g. V # V). The list is further arranged by the place of articulation of the stops that undergo spirantization. Three regions of place of articulation are posited: alveolar, labial and velar. The alveolar category includes the spirantization of retroflex and dental stops, in addition to the general alveopalatal ones. The category velar also includes uvular stops. In certain languages, more than one set of stops undergo spirantization. For example, both /b, g/ spirantize intervocalically to [v, Ä] in Kanuri, a Nilo-Saharan language spoken in Nigeria (Lukas 1967), while the alveolar /d/ remains a stop. In order to capture this, four additional categories have been added according to the logical combinatory possibilities among the three series of different places of articulation. These categories are collections of rules that spirantize only velar & labial, velar & alveolar, alveolar & labial, and obstruents of all three regions of place of articulation. 1.3. Results and generalizations The results of this survey are summarized in (3a). To illustrate more clearly, the graphic representation of (3a) is given in (3b). The statistics in (3a) are arranged by the place of articulation of the segments involved and the prosodic positions where the spirantization applies. (3a) Place Coda Intervocalic Both Total Alveolar/Retroflex 2 6 3 11 Labial 2 7 1 10 Velar/Uvular 5 13 5 23 Velar & Alveolar 1 2 0 3 Velar & Labial 0 5 2 7 Labial & Alveolar 0 1 0 1 All Places 3 24 7 34 Total: 13 58 18 89
(3b) 25 20 15 Coda Both Intervocalic 10 5 0 Intervocalic Alveolar/Retroflex Labial Velar/Uvular Velar+Alveolar Velar+Labial Place of Articulation Labial+Alveolar All Places Coda Both Number of Patterns Found (3a, b) show that there is a significantly high occurrence of spirantization intervocalically. More than 85% of the spirantization patterns found only occur intervocalically or in both intervocalic and coda position. About 35% of spirantizations occur in coda position or in both intervocalic and coda positions, while only 20% appear in both coda and intervocalic positions. (3a, b) also show that velars and uvulars are the most common segments to undergo spirantization. In fact, about 75% of the patterns involve velar and/or uvular segments. Labials and alveolars, on the other hand, have a relatively equal amount of occurrences, 57% and 55% respectively. It is noteworthy that alveolar stops, unlike the bilabials and velars, have strong strident counterparts (e.g., s, z) but always become weak fricatives (e.g., D, T) or approximants (R, l), never stridents. This reinforces the assertion that vowel height conditioned spirantization, which typically turns aspirated stops into strident fricatives, is motivated by a different phonetic mechanism. It is important to point out that only one pattern, from the language Dahala, a Cushitic language spoken in Kenya (Tosco 1991), is found where labials and alveolars undergo spirantization but not velars. This is a significant contrast with the other two combined categories, velar/labial and velar/alveolar, which have seven and three attested patterns respectively. Facts on the behavior of the manner of articulation are also explored in this study. Nine out of seventy six intervocalic spirantization patterns have voiceless stops in the input, yet six of them spirantize and become voiced fricatives rather than voiceless fricatives. For example, in Auyana, a Trans-New Guinean language of Papua New Guinea, /p, k/ becomes [B, Ä] intervocalically, not [, X] (Crothers et al. 1979). This suggests that intervocalic voicing might also be involved.
With respect to the coda position, six out of thirteen languages turn voiced stops into voiceless fricatives. For example, in Uyghur, a Turkic language spoken mainly in China, /G/ becomes [X] instead of [Ò] syllable finally. As for the patterns applying in coda and intervocalic environments, eight out of thirty three patterns turn voiced stops into voiceless fricatives. This suggests that, regardless of how spirantization behaves when it applies in both intervocalic and coda position, coda devoicing might be relevant. These facts revealed by the typological study present some interesting problems to the validity of strict implicational universals and the notion of lenition in general. Foley (1977) proposes that the natural progression of spirantization begins with g, continues to d, and finally reaches b. Yet this study reveals that although velars tend to spirantize much more commonly than other stops, the generalizations about alveolar and bilabial stops do not favor Foley's implicational universals. That is, the alveolars do not spirantize any more often than the bilabials. In fact, the direction seems to be the reverse. Labial spirantization, despite the small difference, is found more often than alveolar spirantization. It is even more revealing when one looks at the facts from the combined categories according to place of articulation. More spirantization solely affecting velars and labials is found than affecting only velars and alveolars. In light of these facts, I propose avoiding strict implicational universals and instead refer only to statistically significant tendencies. The notion of lenition seems unclear in light of the facts presented above. The scenario that is most worthy of discussion is the tendency for voiced stops to spirantize and devoice in coda position (a similar discussion is also mentioned in Lavoie 1996). When voiced stops spirantize, it is generally assumed that this is a lenition process. However, when voiced fricatives devoice, by most definitions of lenition, it would be a fortition process. It is also worth pointing out that since the devoicing occurs in the exact same environment as spirantization, the process of devoicing should be classified the same way as spirantization. However, if one adhered to a strict notion such as lenition, then one would fail to capture an important and curious generalization. To this end, I suggest the adoption of Steriade's (1994) position-based approach, in which phonological processes are explained in terms of their respective phonetic environments. For a more detailed exposition of this approach, I refer the reader to Steriade 1994 & 1997. 2. Phonetic motivations With the results of the typological study laid out above, the job of the experimental phonologist is to find out what motivates these tendencies of sound change and alternations (cf. Ohala 1982, 1993). To this end, a phonetic theory of the origin of the tendencies of post-vocalic spirantization is presented below. Supporting phonetic evidence is also presented in order to motivate the theory concretely.
2.1. Previous account: Beckman, De Jong, Jun, and Lee 1992 Beckman et al. (1992) present the following theory of spirantization: If the undershoot [of the articulatory gesture] results in a less perfect seal, a seal through which air can leak enough air pressure may be vented noiselessly through the leak to prevent a strong stop burst on release...[t]he listener might misinterpret the lack of a strong burst as an intentional feature of a weak non-sibilant fricative or even of an approximant. (Beckman et al. 1992: 49) Beckman et al. only present some anecdotal evidence in support of their hypothesis. For example, they cite Fant & Kruckenbery 1989 as they note that, in a database of short read texts in Swedish, there is frequent incomplete oral closure of voiced stops & nasals which become reduced to voiced continuant and nasalized vowels. Beckman et al. also find noticeable frication in productions of intervocalic /b/ and /d/ in their own data from Tokyo Japanese, particularly data of Shitamachi speakers. Unfortunately, despite the fact that this evidence suggests the articulatory origins of the spirantization, that is, the incomplete closure of the voiced stops that might give rise to an approximant percept, they nonetheless remain anecdotal. To illustrate and support this hypothesis, more phonetic evidence supporting the theory is presented below. 2.1.1. Byrd 1996 The best articulatory evidence for coda reduction is given in Byrd 1996. Byrd conducts a series of experiments using electropalatography (EPG) to study the articulatory time in English consonant sequences. She shows that stop articulation has less lingua-palatal contact in coda position than in onset position in the heterosyllabic stop-stop sequences, such as d#g & g#d. This result suggests that coda reduction is a general phenomenon, as English is not known to have any coda alternation at the phonemic level. Byrd also demonstrates that lingua-palatal contact for the velar stop is smaller and tends to be shorter in coda position than in onset position. This again suggests that the seed for the predominant tendency of velars to undergo spirantization in coda position is a result of the propensity for velars to reduce their articulatory gesture more syllable finally than initially. Byrd cites other experiments that report articulatory data, showing reduction of both labial and coronal stop reduction in coda position (e.g., Browman & Goldstein 1995, Fromkin 1965). 2.1.2. Redford, MacNeilage, and Davis 1997 More evidence for the general articulatory asymmetry related to the prosodic location of a given segment is reported in Redford et al. 1997. They provide statistical evidence from their corpus of baby babbling data, demonstrating that
the apparent asymmetry found in post-vocalic spirantization might have a genuine motor-muscular origin. Redford et al. find that fricatives are produced much less frequently in initial position (7%) of a CVC syllable than in final position (24%). They assert that this peculiar disproportion of fricatives in final position is due to the reduction in the level of energy provided to the speech production system at the termination of an utterance. This hypothesis, they claim, has received some evidence from other studies, which show that there is a terminal energy decrease in the respiratory and/or phonatory components of early infant vocalizations. However, it should be emphasized these characteristics of infant speech do not necessarily parallel those of the adults (however, see Davis & MacNeilage 1995 for arguments for drawing this parallel). My motivation in presenting this here is to provide some empirical studies that are in line with the hypothesis put forth by Beckman et al. 1992. 2.1.3. Segmental duration In her survey of earlier studies of syllable juncture, Lehiste (1960) measures the duration in spectrograms of intervocalic consonants that serve contrastively as either syllable-final or syllable-initial. Initial consonants are usually found to be significantly longer than final consonants. Lehiste (1970) has also shown that voiced segments are generally shorter than voiceless ones. Ohala (p.c.) speculates that this difference might also have contributed to the spawning of spirantization. That is, the shorter the closure duration, the more likely a listener is to miss the closure duration all together and misinterpret it as some sort of approximant. Ending on this note, the next section is a discussion of the acoustic perceptual evidence of the development of postvocalic spirantization. 2.2. An extended theory: Speech-rate conditioned sound change Despite the fact that Beckman et al. s hypothesis takes both the listener s and the speaker s perspectives into account, suggesting that the phonetic seed of spirantization is the interface between the articulatory undershoot of the speaker and the mis-parsing of the listener, it remains doubtful that the absence of the stop burst alone is enough to trigger the mis-perception of stops as approximants. For example, the most common type of segment to participate in spirantization, voiced unaspirated stops, do not have a very strong burst to start with; some other cues must be involved if a uniform analysis is to be achieved. I propose that the ultimate phonetic cause for the mis-perception arises from the variation caused by the differential speech rate effects. This means that if the listener fails to normalize and assumes that the stop-approximant boundary remains at the point of normal speaking rate, then s/he might mis-interpret the transition as a cue for approximancy, rather than stops (the relation of speech rate variation and other sound changes is also argued in SoleÛ 1995 and SoleÛ & Ohala 1991). This hypothesis is strongly supported by the experiment presented by Miller and Liberman (1979). However, in order to understand this study and
appreciate the argumentation presented below, some background information regarding the general cues for the stop-approximant contrast must first be explicated. 2.2.1. Liberman, Delattre, Gerstman, and Cooper 1956 The general cues for the stop-glide contrast are the duration, rate, and extent of formant transition (Liberman et al. 1956). More specifically, rapid formant transitions (about 40 msec.) are perceived as stops, and those with slow formant transition rates (about 80 msec.) are perceived as glides. The rate of formant transition correlates with the rate of the articulatory movement. 2.2.2. Shinn and Blumstein 1984 Shinn and Blumstein found that a rapid increase in amplitude in the vicinity of the release of consonants will be perceived as the class of noncontinuants, and those speech sounds with a gradual increase in amplitude in the vicinity of the release will be perceived as a class of continuants. Shinn and Blumstein eloquently explain the reason for this pattern: For stops, there is a complete closure of the vocal tract with a resultant increase in air pressure behind the constriction. With the release of the closure, there is an abrupt and transient change in pressure. The consequence is a rapid rise of acoustic energy at the release of a stop consonant. In contrast, glides and weak fricatives are produced with only a partial constriction in the vocal tract and a gradual constriction out of the configuration of the preceding vowel and a gradual release into the configuration of the following vowel. As a result, for the glides, the relative change in amplitude at the release is considerably less and more gradual than that of stops. (Shinn and Blumstein 1984: 1243) Most peculiarly, they show that subjects in a free-identification task classify all glide tokens as stops when the glide s amplitude envelope is replaced by a stop amplitude envelope. However, they identify stops with the glide s amplitude envelope as fricatives rather than as glides. This asymmetric response of the subjects provides a plausible explanation as to why spirantization turns voiced stops into weak fricatives more often than into glides. Now that the relevant facts regarding the perception of the stopapproximant contrast have been introduced, we shall return to the discussion regarding the effect speech rate has on spirantization. 2.2.3. Miller and Liberman 1979 Miller and Liberman (1979) found that the longer the syllable duration, the longer the transition required to perceive [w]. In the experiment, they first showed that, by lengthening or shortening the steady-state portion of the vowel in synthetic [ba, wa] syllables, a rate normalization of the formant rate boundary is observed. They then synthesized [bad, wad] syllables of equal duration as the lengthened
[ba, wa] syllables mentioned earlier by adding the VC transitions for [d]. Thus, they were able to demonstrate that the [b, w] transition speed boundary shifts to a faster speed. That is, the added VC transitions for [d] produce the effect of a seemingly shorter CV formant transition duration for the previous segments. With this in mind, it is argued here that the shorter the syllable, the shorter the transition that is required to perceive [w]. The reason for this is the transition cue for the [b, w] distinction is essentially temporal, which means that it should be perceived in direct relation to the speech rate. This would suggest that in fast speech, the boundary between a stop and a glide is relatively narrow, providing more opportunities for a stop to be confused as a glide or approximant. 3. Conclusion As noted early on, this paper is meant to be taken as a preliminary look at the generalizations regarding post-vocalic spirantization. It undoubtedly does not do justice to the complex and interesting issues regarding some of the details of the different specific phonetic instantiations of the post-vocalic spirantization processes found in the world s languages. Nevertheless, this study has shown that many previous assumptions about post-vocalic spirantization are unwarranted. The typological study reveals that there is no strict implicational universal; instead, there are only statistically derived tendencies. By synthesizing research from both the articulatory and perceptual domains, it is shown that a sound phonetic theory of post-vocalic spirantization is possible. As a side development of this paper, it is shown that lenition as a cover term for a group of rather heterogeneous processes does not capture any significant generalization. In some cases, it might even obscure other important and interesting ones. References Bauer, L. 1988. What is lenition? Linguistics 24.381-392. Beckman, M., K. De Jong, S-A Jun & S-H Lee. 1992. The interaction of coarticulation and prosody in sound change. Language and Speech 35.45-58. Browman, C. P. & L. Goldstein. 1995. Gestural syllable position effects in American English. Producing speech: contemporary issues for Katherine Safford Harris, ed. by F. Bell-Berti and L. Raphael. Woodbury. NY: AIP Press. Byrd, D. 1996. Influences on articulatory timing in consonant sequences. Journal of Phonetics 24.209-244. Crothers, J., J. Lorentz, D. Sherman, & M. Vihman (eds.). 1979. Handbook f phonological data. Vol. 1, 2. Stanford University: Department of Linguistics. Davis, B. L. & P. F. MacNeilage. 1995. The articulatory basis of babbling. Journal of Speech and Hearing Research 38.1199-1211. Fant, G., and A. Kruckenberg. 1989. Preliminaries to the study of Swedish prose reading and reading style. Speech Transmission Laboratory Quarterly Progress Report 2.1-83. Foley, J. 1977. Foundations of theoretical phonology. Cambridge: Cambridge University Press. Fromkin, V. 1965. Some phonetic specifications of linguistic units: an electromyographic investigation. UCLA Dissertation. Hyman, L. 1975. Phonology: theory and analysis. New York: Holt, Rinehart & Winston. Kirchner, R. 1998. An effort-base approach to consonant lenition. UCLA Dissertation. Lavoie, L. 1996. Consonant strength: Results of a data base development project. Working Papers of the Cornell Phonetics Laboratory 11.269-316. Lehiste, I. 1960. An acoustic-phonetic study of internal open juncture. Basel: Karger.
Lehiste, I. 1970. Suprasegmentals. Cambridge, MA: MIT Press. Liberman, A. M., P. C. Delattre, L, J. Gerstman, & F. S. Cooper. 1956. Tempo of frequency change as a cue for distinguishing classes of speech sounds. Journal of Experimental Psychology 52.127-137. Lukas, J. 1967. A study of the Kanuri language: Grammar and vocabulary. London: Dawsons of Pall Mall, reprint of 1937 original. Miller, J. L. & A. M. Liberman. 1979. Some effects of later-occurring information on the perception of stop consonant and semivowel. Perception and Psychophysics 25.457-465. Ohala, J. 1983. The origin of sound patterns in vocal tract constraint. The production of speech, ed. by P. F. MacNeilage, 189-216. New York: Springer-Verlag. Ohala, J. 1993. The phonetics of sound change. Historical linguistics: problems and perspectives, ed. by Charles Jones. London: Longman. Redford, M. A., P. F. MacNeilage, & B. L. Davis. 1997. Production constraints on utterance-final consonant characteristics in babbling. Phonetica 54.172-186. Shinn, P. & S. E. Blumstein. 1984. On the role of the amplitude envelope for the perception of [b] and [w]. Journal of Acoustic Society of America 75.1243-1252. Sole, M. J. 1995. New ways of analyzing sound change: speech rate effects. Belgian Journal of Linguistics 9.21-44. Sole, M. J. & J. J. Ohala. 1991. The phonological representation of reduced forms. Proceedings of the ETRW Phonetics and phonology of speaking styles. 49.1-49.5. Barcelona, Spain. Steriade, D. 1994. Positional neutralization and the expression of contrast. UCLA ms. Steriade, D. 1997. Phonetics in phonology: The case of laryngeal neutralization. UCLA ms. Tosco, M. 1991. A grammatical sketch of Dahala. Hamburg: Helmut Buske Verlag.