LINGUIST List 14.1820

Mon Jun 30 2003

Review: Phonetics/Phonology/Historical Ling: Kavitskaya

Editor for this issue: Madhavi Jammalamadaka <>

What follows is a review or discussion note contributed to our Book Discussion Forum. We expect discussions to be informal and interactive; and the author of the book discussed is cordially invited to join in.

If you are interested in leading a book discussion, look for books announced on LINGUIST as "available for review." Then contact Simin Karimi at


  • Heike Lehnert-LeHouillier Compensatory Lengthening: Phonetics, Phonology, Diachrony

    Message 1: Compensatory Lengthening: Phonetics, Phonology, Diachrony

    Date: Mon, 30 Jun 2003 11:34:49 +0000
    From: Heike Lehnert-LeHouillier <>

    Subject: Compensatory Lengthening: Phonetics, Phonology, Diachrony

    Kavitskaya, Darya (2002) Compensatory Lengthening: Phonetics, Phonology, Diachrony. Routledge, Outstanding Dissertations in Linguistics.

    Announced at

    Heike Lehnert-LeHouillier, University at Buffalo (SUNY)


    The dissertation published in this monograph investigates the two types of compensatory lengthening:

    1) Lengthening of a vowel after loss of a consonant (typically a tautosyllabic postvocalic consonant): CVC > CV: and 2) Lengthening of a vowel after loss of a vowel in a following syllable: CV.CV > CV:C.

    Based on data from 78 languages from 20 different language families, Kavitskaya develops a phonologization model to account for the cross-linguistically attested cases of compensatory lengthening. This phonologization model, which is theoretically based on the idea of listener oriented sound change developed in Ohala (1992) as well as in Blevins & Garrett (1998), intends to provide a better account of diachronic compensatory lengthening than previous accounts, most notably the account by de Chene & Anderson (1979) and the widely accepted account within moraic phonology by Hayes (1989).


    CHAPTER 1 illustrates the two major instances of compensatory lengthening with data from Lithuanian and Serbo-Croatian and lays out the basic tenets of the phonologization model: A. Compensatory lengthening of a vowel induced by the loss of a segment will take place only if the lengthened vowel was previously in an environment that made it phonetically longer. B. The loss of the segment that conditioned the phonetic vowel lengthening leads to a reinterpretation of the phonetically lengthened vowel as phonologically long (i.e. phonologization). Note: that 'loss of a segment' in this model means that the segment was produced by the speaker but misparsed and reinterpreted as missing by the listener.

    In the case of compensatory lengthening of a vowel after loss of a consonant (CVC > CV:), Kavitskaya's model predicts that for example the loss of postvocalic glides will result more easily in compensatory lengthening than the loss of postvocalic obstruents since the relatively long transition from vowel to glide may be re-interpreted as vowel length after the glide is lost. The rather abrupt transition between vowel and obstruent, however, is not a source of phonetic vowel lengthening and the loss of a postvocalic obstruent should, therefore, not result in compensatory lengthening of the vowel.

    The case of compensatory lengthening of a vowel after vowel loss (CV.CV > CV:C) is explained in the phonologization model as follows. Vowels in open syllable are phonetically longer than vowels in closed syllable. The loss of the second vowel in a two-syllable CV.CV sequence results in the resyllabification into a single CV:C syllable of that sequence. The conditioning environment for the phonetic lengthening of the first vowel has now disappeared resulting in the re-interpretation of the phonetic vowel length as phonological.

    CHAPTER 2 reviews the different approaches to compensatory lengthening that have been taken in the literature. The focus of this chapter is on problems that arise for a moraic approach from the analysis of compensatory lengthening.

    The first problem for a moraic account of compensatory lengthening that Kavitskaya identifies is that compensatory lengthening is predicted to occur only in languages that exhibit independently established weight distinction. In other words; there needs to be evidence in a language that coda consonants are moraic, since only the deletion of a weight bearing coda consonant can result in the transfer of the mora from the deleted coda consonant to the previous vowel. However, there exist at least two languages, Ngajan and Piro that don't show any evidence that coda consonants are moraic, but do exhibit compensatory lengthening.

    The second problem for a moraic account is said to be the existence of languages (Samothraki Greek, Romanesco Italian, Onondaga) in which compensatory lengthening is triggered by the deletion of an onset consonant. According to standard moraic theory (e.g. Hayes 1989), the deletion of an onset consonant should never lead to compensatory lengthening, since onsets are never moraic.

    The third problem for a moraic account stated by Kavitskaya is also related to the fact that onsets are weightless. This would predict that in CV.CV > CV:C compensatory lengthening the nature of the intervocalic consonant in the CV.CV sequence cannot have any influence on whether or not compensatory lengthening after loss of the second vowel occurs. The author points out that in Romance and Slavic compensatory lengthening after vowel loss crucially depends on the nature of the intervening consonant; i.e. compensatory lengthening in Polish does occur unless the intervening consonant is a voiceless stop.

    The fourth problem identified is that a moraic account of compensatory lengthening has difficulties to account for cases of compensatory lengthening due to loss of a tautosyllabic consonant that is not adjacent to the lengthening vowel; i.e. CVC1C2 > CV:C1. The author argues that any moraic account that can accommodate these data is not constrained enough otherwise and result in overgeneration.

    The last problem mentioned is that of directionality. The empirical observation that needs to be accounted for is that the element that is lost is always to the right of the vowel that lengthens; i.e. the directionality is right-to-left. Within a moraic approach to compensatory lengthening this generalization is accounted for by two mechanisms: the prohibition of crossing association lines and parasitic delinking (Hayes 1989). The author points out that this account of directionality crucially depends on several theory internal assumptions. First, the principle of no association lines crossing can only account for directionality if vowels and consonants are analyzed as being on the same tier. However, many phonological models analyze consonants and vowels to be on separate tier. Second, the mechanism of parasitic delinking is not independently motivated and hence remains a stipulation. After having identified these five issues, the author goes on to investigate the two types of compensatory lengthening in more detail.

    CHAPTER 3 investigates diachronic CVC > CV: compensatory lengthening by segment type of the lost consonant.

    The loss of glides, which are the most likely consonants to cause compensatory lengthening according to the predictions of the phonologization model, is discussed based on data from Turkish, Kabardian, Ngajan, and Ancient Greek. The claim here is that the transition between vowel and glide is very long and if the glide is lost; i.e. not heard by the listener, the result is a reinterpretation of the preconsonantal vowel as long. It is further illustrated that glides may be more likely not to be heard in certain environments. For example, if they follow a phonetically similar vowel; i.e. if [j] follows a high front vowel or if they precede a sonorant consonant, glides are more likely to be missed since the formant frequencies of the 2. and 3. formant of a high front vowel are very similar to that of a [j] glide.

    The explanation for the loss of liquids resulting in compensatory lengthening is very similar to that of glides; vowel-liquid transitions are very long and after loss of a liquid the vowel can be reinterpreted as phonemically long. However, unlike for glides, the phonetic properties of the liquid itself in addition to its environment determine whether or not the deletion of a liquid will result in compensatory lengthening of the preceding vowel. For example, rhotics that are approximants are more likely to cause compensatory lengthening than rhotics that are trills. Compensatory lengthening after the loss of liquids is discussed on data from Komi, Ngajan, and Turkish.

    Compensatory lengthening due to the loss or reduction of nasals is ascribed to the fact that nasalized vowels tend to be longer than oral vowels. If a post-vocalic nasal is lost and thus the conditioning environment for the phonetic length of the vowel is gone, the length of the vowel can be reinterpreted as phonemic. The loss of nasals seems also phonetically conditioned since in many languages nasals are lost before voiceless fricatives or voiceless obstruents in general. The author makes the interesting observation that compensatory lengthening triggered by nasal loss while occurring in numerous languages, only seems to occur in Indo-European and Bantu languages.

    Fricatives should not result in compensatory lengthening of a preceding vowel since fricatives have usually no lengthening effect on neighboring vowels. The author argues that in all cases in which the apparent loss of a fricative has caused compensatory lengthening, the fricative necessarily undergoes lenition first resulting in an approximant and then is lost causing compensatory lengthening. Since [h] is the most common fricative to cause compensatory lengthening, it is mentioned that [h] may be an approximant in many languages.

    In the section discussing compensatory lengthening after the loss of a stop consonant the author points out that despite the fact that vowels are longer before voiced stops compared to voiceless stops in many language, there don't seem to be cases where the loss of a voiced stop caused compensatory lengthening. For the cases of [g] loss in Turkish and West Saxon discussed here, the author argues that [g] underwent lenition first to a fricative and then to an approximant, so that the compensatory lengthening was the result of approximant loss rather than due to the loss of the stop.

    The last section in this chapter discusses apparent counterexamples, the first one being that glottal stops are among the more common consonants to trigger compensatory lengthening. Kavitskaya argues that only glottal stops that are phonetically realized as having weak constriction of the vocal folds or as laryngealization on the vowel can trigger compensatory lengthening. The second apparent counterexample is morphological compensatory lengthening which is argued to be a ''case of templatic, morphologically conditioned process'' that doesn't need a phonetic explanation. The next apparent counterexample discussed is compensatory lengthening through degemination. Here the author argues that degemination may start out as phonetically motivated process; i.e. degemination of an approximant and then be extended to other geminates via analogical sound change. The last apparent counterexample is compensatory lengthening through onset loss. Since all the cases of onset loss discussed here involve a liquid, it is argued that the same mechanisms that explain compensatory lengthening after the loss of a coda liquid can account for compensatory lengthening after loss of an onset liquid.

    CHAPTER 4 investigates the diachronic development of compensatory lengthening through vowel loss mainly discussing data from Friulian and Slavic. The main argument here is that vowels in open syllable are phonetically longer than vowels in closed syllable. The loss of the second vowel in a two-syllable CV.CV sequence results in the resyllabification into a single CV:C syllable of that sequence and in reinterpretation of the vowel length as phonemic. Especially the discussion of Slavic develops a phonetic account for the fact that some intervening consonants in a CVCV sequence allow compensatory lengthening after loss of the second vowel while others do not. The author argues that the situation can be explained by making reference to the observation that vowels tend to be longer before voiced consonants than before voiceless ones. Kavitskaya claims for Slavic that phonetic vowel duration was longest before glides, less long before other sonorants, short before voiced fricatives, shorter before voiced stops, and shortest before voiceless obstruents. Furthermore, she argues that after the loss of the second vowel in CVCV sequences in Slavic the length of the first vowel was reinterpreted as phonemically long if it preceded some - usually more sonorous consonants - but not others. In order to validate this claim, the author reports an experiment conducted by herself on vowel length in open and closed syllables in Contemporary Standard Russian. In the remainder of the chapter Kavitskaya discusses the prosodic conditioning of compensatory lengthening in Slavic.

    CHAPTER 5 discusses synchronic compensatory lengthening. While it was argued in chapters 3 and 4 that diachronically the two types of compensatory lengthening are similar in that they arise through phonologization of previously phonetic vowel length, this chapter illustrates the asymmetry between CVC and CVCV compensatory lengthening synchronically.

    Since compensatory lengthening through consonant loss (CVC > CV:) is usually phonologically conditioned, the trigger (i.e. the lost consonant) can usually be recovered in all environments that don't cause its deletion. These cases of compensatory lengthening can synchronically be modeled as weight conservation within the syllable. Furthermore, the resulting syllable structure (CV:) is more optimal than the original CVC structure.

    In compensatory lengthening through vowel loss, the lost vowel is usually not recoverable, which often leads to the lexicalization or morphologization of CVCV ~ CV:C alternations and the resulting syllable structure (CV:C) is argued to be cross-linguistically dispreferred. In the very few cases in which compensatory lengthening through vowel loss is phonologically conditioned, the lost vowel can be recovered and these cases are formally similar to compensatory lengthening through consonant loss in that they can be synchronically modeled as a weight conserving process.

    CHAPTER 6 summarizes phonologization model developed in this dissertation and states again that in order to account for all aspects of compensatory lengthening, it is necessary to distinguish between synchronic alternations and diachronic sound changes as well as between compensatory lengthening through consonant loss and compensatory lengthening through vowel loss.


    This work makes a significant contribution to the body of work that tries to further our understanding of how phonetics influences and shapes phonological patterns in the world's languages. It is a great challenge to analyze a huge amount of data - as Kavitskaya did in her dissertation - and to account for cross-linguistic generalizations without neglecting language specific detail. In many instances the author suggests different analyses from those proposed in the sources she used. One such case is Onondaga where the author challenges Woodbury's (1981) diachronic analysis of why Proto-Lake-Iroquoian *CrV sequences developed into CV: sequences in Onondaga, apparently showing compensatory lengthening after deletion of onset-r. According to Woodbury's analysis based on historic documents, Proto-Lake-Iroquoian *CrV sequences were broken up by an epenthetic /e/ (*CerV) with following r- deletion (*CeV) and coalescence of the vowels resulting in a long vowel. Kavitskaya suggests that the epenthetic /e/ was ''actually a phonetic effect of the r itself, perceived by listeners, but not yet phonologized.''

    There are several arguments against such a view. First, the historic sources do not use different symbols to represent epenthetic /e/ and non-epenthetic /e/ which would be expected if they had been phonetically different. In the closely related language Mohawk, the epenthetic /e/ still exists and is not any different in its realization from non-epenthetic /e/. Furthermore, the quality of the lengthened vowel can be easily explained by the coalescence analysis. For example, if the V in the CerV sequence was a back vowel, the outcome is a lengthened front vowel (*Cero > Ce:); if it was a low vowel, the outcome was a lengthened raised vowel (*Cera > Cae:). These changes in vowel quality cannot be easily explained as having been caused by the onset-r, since the 'r' in Iroquoian - where it still exists - has been described by Lounsbury (1978) as a retroflex lateral flap, which we would not expect to cause any fronting. Even though I was not convinced by the author's interpretation of these particular data, overall the phonetic motivation for compensatory lengthening was well illustrated by the author, and I would recommend this book to anybody interested in this are

    REFERENCES Lounsbury, F. 1978. Iroquouian languages. In: Bruce Trigger (ed.) Handbook of North American Indians 15, Smithsonian Institution, Washington, D.C.


    Heike Lehnert-LeHouillier is a PhD student in the linguistics department at the University at Buffalo. She is interested in the phonetic and phonological aspects of word and sentence prosody, laboratory phonology and phonetics as well as in Iroquoian linguistics.