Ioana ChitoranIoana ChitoranDartmouth College, USA

L i i th t l i l Lezgi in the typological context of vowel devoicingcontext of vowel devoicing

Conference on Caucasian Languages MPI, Leipzig, 13-15 May 2011MPI, Leipzig, 13 15 May 2011

Work in collaboration with:Work in collaboration with:

• Khalil Iskarous, Haskins Laboratories & University of Southern California, USAy f f ,

• Egidio Marsico, Laboratoire “Dynamique du L ” CNRS L FLangage”, CNRS, Lyon, France

2

Goal of today’s talkGoal of today s talk

• To examine a well-documented process of vowel loss in Lezgivowel loss in Lezgi

• To propose vowel devoicing (VDev), b d i d l idbased on acoustic and perceptual evidence

3

OutlineOutline

1. A typology of Vdev phenomena2 Th l d i L i2. The relevant data in Lezgi3. The interpretation of acoustic evidence4. The interpretation of perceptual evidence5 Implications for sound change5. Implications for sound change

4

OutlineOutline

1. A typology of Vdev phenomena2 Th l d i L i2. The relevant data in Lezgi3. The interpretation of acoustic evidence4. The interpretation of perceptual evidence5 Implications for sound change5. Implications for sound change

5

Most common conditions for VDevMost common conditions for VDev

In a C1VC2 sequenceIn a C1VC2 sequence

a. Predominantly high vowela. Predominantly high vowelb. At least one voiceless consonant (especially

voiceless fricatives and aspirated stops)voiceless fricatives and aspirated stops)c. Unstressed/unaccented voweld U d d ld. Unrounded vowel

Cho 1993, Gordon 1998, Chitoran & Marsico 2010

6

Some examples:Some examples:

Quebec FrenchQuebec Frenchtisse ['tis] ‘s/he weaves’

ti [ i' ]tissu [ti'sy] ‘fabric’

Turkishtüfek [ty'fek] ‘rifle’

JapaneseJapanese[ɕikíso] ‘pigment’

[ té ́][sɯtérɯ́] ‘to throw’7

PropertiesProperties

• VDev is phonetically common• Highly variable: within and across • Highly variable: within and across

languages and speakers• Distinguish positional (word phrase utterance final) • Distinguish positional (word-, phrase-, utterance-final)

vs. non-positional devoicing (Chitoran & Marsico2010)

• Often reported as a particular manifestation of vowel reduction or deletion.

8

Vowel devoicing database (Chitoran & Marsico 2010)

Th iThree main sources:1. Gordon 1998 (55 languages)

2. UPSID 451 (Maddieson 1984; Maddieson & Precoda 1990)

3 Addi i l d f i l 100 l3. Additional data from approximately 100 languages (grammars and articles)

We retained 39 languages with devoicing.

9

Genetic and geographic distribution of h lthe sample

10

Non-positional devoicing is more common

Of the 39 languages:

• Positional devoicing only 12• Non-positional devoicing only 22• Both types of devoicing 5Both types of devoicing 5

11

Phonetic accountsPhonetic accountsNon-positional VDev is understood as an passimilatory process

• Aerodynamic voicing constraint (Ohala 1983)• Aerodynamic voicing constraint (Ohala 1983)– Insufficient transglottal pressure differential– Narrow constriction of high V impedes air flowg

• Glottal gesture overlap (e.g., Jun & Beckman 1993)Absence of stress shortens V increasing overlap – Absence of stress shortens V, increasing overlap between C1 and C2

– Glottal opening gesture of C may extend over the V gesturegesture

12

OutlineOutline

1. A typology of VDev2 Th l d i L i2. The relevant data in Lezgi3. The interpretation of acoustic evidence4. The interpretation of perceptual evidence5 Implications for sound change5. Implications for sound change

13

The Lezgi factsg

• High vowels [i, y, u] disappear in pre-stress position, after a voiceless obstruent – “syncope /

d i / d l i ”reduction / deletion”• May be perceived as secondary articulations on

C1C1

Uslar 1896, Talibov 1980, Kodzasov 1990, Haspelmath 1993- Daghestan dialect

Babaliyeva 2007 - Azerbaijan dialect

14

Morphological alternationsMorphological alternations

Monosyllabic rootsMonosyllabic roots

absolutive singular absolutive pluralabsolutive singular absolutive plural(root stress) (no root stress)

sík’ sik’ ár ‘f ’sík sik - ár ‘fox’tʃhúf tʃhuf - ár ‘cloud’thúph thup ár ‘ ’thúph thup - ár cannontʃhýkh tʃhykw - ér ‘flower’

All data from Azerbaijan dialect15

Disyllabic rootsDisyllabic roots

No stress alternation

khitáb khitáb - ar ‘book’k táb k táb ar book

thykwén thykwén - ar ‘shop’h ’ál h ’ álthup’ál thup’ ál - ar ‘ring’

tʃhuk’úl tʃhuk’úl - ar ‘knife’Can be reflected in orthography: ktab, ktabar, Qsar (Qusar)(for more examples see Haspelmath 1993: 36)

16

OutlineOutline

1. A typology of VDev2 Th l d i L i2. The relevant data in Lezgi3. The interpretation of acoustic evidence4. The interpretation of perceptual evidence5 Implications for sound change5. Implications for sound change

17

Qualitative acoustic descriptionQualitative acoustic description

Data from 7 speakers recorded in Azerbaijanf p jDuring the vowel portion:– No periodic voicingNo periodic voicing– Unclear formant structure

Strong frication noise– Strong frication noise

E l tʃhuk’úl ‘k if ’ Examples: tʃhuk’úl ‘knife’ sik’-ar ‘fox’ pl.

18

[tʃhuk’úl] ‘knife’[ ʃ ] f

19

[ … a # tʃh u k’ u l # s…]

[sik’-ar] ‘fox’ pl. [ ] f p

20

[ s i k’ a r ]

Presence of a vocalic gesturePresence of a vocalic gestureEvidence from secondary labialization of (non-labial) C2

• Voiced C1 – variable labializationlytkhe ~ lytkhwe ‘boat’lytk e lytk e boat

ʁud ʁut – ar ~ ʁutw – ar ‘fist’

• V i l C1 r t ti l bi li ti n• Voiceless C1 – more systematic labializationsingular plural(more regular)k’ k’ khuk’w ‘ k’k’uk’ khuk’w – ar ‘peak’tyd thytw – er ‘throat’tʃhykh tʃhykw – er ‘flower’

21

ʃ y ʃ f

Acoustic evidence (Chitoran & Iskarous 2008)Acoustic evidence (Chitoran & Iskarous 2008)

• Hypothesis:If V gesture is still present similar fricative-V coarticulation patterns will be found in both stress coarticulation patterns will be found in both stress contexts.

C i f DFT t f [ ] diComparison of DFT spectra of [s] precedingstressed and unstressed V

sík’ – sik’ár vs. súth – sutár vs. sáf – safár‘fox’ ‘measure of land’ ‘sieve’

22

f f

Local spectral propertiesLocal spectral properties

• Data from 7 speakersData from 7 speakers• Two windows extracted from each fricative:

- 2/3 into [s] (40 ms)- last 1/3 of [s] (40 ms)

Differences among [i y u] are visible in the Differences among [i,y,u] are visible in the energy between 4 and 9 kHz

23

Energy between 4 - 9 kHz, averaged across f frequency

[sík] [sikár]

[sáf] [safár]

[sík] [sikár]

[sút] [sutár]

24

ResultsResults

• Coarticulation patterns in [sik’, suth] (full V) are similar to those in [sik’ar sutar] V) are similar to those in [sik ar, sutar] (non-full V)S f V i b h • Suggests presence of V gesture in both stressed and unstressed contexts

25

Acoustic duration of [s]Acoustic duration of [s]

• Hypothesis: Hypothesis: If V is present but devoiced, [s] will be longer before non-full Vs (sup-ar, sut-ar)longer before non full Vs (s p ar, s t ar)than before full Vs (saf-ar, sal-ar, sam-ar)

• Interpretation:Interpretation:The longer [s] duration corresponds to the devoiced vocalic portion, visible as devoiced vocalic portion, visible as increased frication due to a highly constricted V gestureg

26

[s] duration (ms) is longer before [u] h b f [ ] (non-full V) than before [a] (full V)

250

200

100

150[su]

50

100 [sa]

0sp1 sp2 sp3 sp4 sp5 sp6

27

sp1 sp2 sp3 sp4 sp5 sp6

OutlineOutline

1. A typology of VDev2 Th l d i L i2. The relevant data in Lezgi3. The interpretation of acoustic evidence4. The interpretation of perceptual evidence5 Implications for sound change5. Implications for sound change

28

PerceptionPerception

Hypothesis:Hypothesis:• If the V gesture is present in the stressed

environment only: environment only: – coarticulation effects on the fricative should

be present only in that environment be present only in that environment – the identification rate for the V should be

higher in the stressed environment higher in the stressed environment (cf. Beckman & Shoji 1984 for Japanese)

29

Experiment F d h i id ifi iForced choice identification

Sti li [ ] i d f l L i d • Stimuli – [s] excised from real Lezgi words, in stressed and unstressed context

h is(ík’), s(úth), s(áf) s(ik’ár), s(utár), s(afár) • Three response choices:p

“si”, “su”, “sa”• Participants: Native speakers of French (11) • Participants: Native speakers of French (11),

Japanese (9), Lezgi (2)

30

StimuliStimuli

For each fricative, 6 portions were presented, For each fricative, 6 portions were presented, randomized:1 - first third of [s]1 first third of [s]2 - second (middle)3 - third (end)3 - third (end)4 - first + second third5 d + thi d5 - second + third6 - full fricative

f h ll d d5 repetitions of each, all randomized31

Identification rate of Lezgi Vs as a function of tstress

1

0.7

0.8

0.9

1

0.4

0.5

0.6

nostress

0.1

0.2

0.3stress

0sa si su sa si su sa si su

J speakers F speakers LZ speakers

329 Japanese listeners 11 French listeners 2 Lezgi listeners

ResultsResults

• Different identification patterns for [sa] vs Different identification patterns for [sa] vs. [si, su]

• [si su] have comparable (relatively high) • [si, su] have comparable (relatively high) identification rates in both stressed and unstressed environmentsunstressed environments

• Identification of [u] is the least affected by t b t ti ff t f l bi litstress robust acoustic effects of labiality

Support for the presence of a vowel gesture

33

OutlineOutline

1. A typology of VDev2 Th l d i L i2. The relevant data in Lezgi3. The interpretation of acoustic evidence4. The interpretation of perceptual evidence5 Implications for sound change5. Implications for sound change

34

• The speech production system seen as a p p ydynamical system

• A change in progress is a transition state g p g• From instability to stability:

– of categories, allophonic variationg , p– gradient, phonetic variation is not necessarily

eliminated (“stable variation”)

• VDev is phonetically common but p yphonologically rare (allophonic variation in Japanese, Korean)

35

Relevant propertiesRelevant properties

• The variation in Lezgi is gradient The variation in Lezgi is gradient, quantitative variation resulting from variable overlapvariable overlap.

• The role of morphologyE id f h h• Evidence from orthography

• Initial C clusters – very few (ʧka ‘place’ < ʧi’ka)• Very slow progress (by comparing Uslar 1896 and

Haspelmath 1993)

36

PredictionsPredictions

• V loss for polysyllabic roots only except V loss for polysyllabic roots only, except for:– Secondary labialization in the case of [u]– Secondary labialization in the case of [u]

• No loss for monosyllabic roots, protected b h l i l lt tiby morphological alternation– Possibly maintain VDev or phonemic

secondary articulations (especially labialization)

37

ConclusionsConclusions

• Some evidence from production and Some evidence from production and perception: vowel devoicing may be present in Lezgipresent in Lezgi.

• The interaction of phonetic and morphological factors predicts that the morphological factors predicts that the language may reach a stage of allophonic variation rather than complete vowel lossvariation, rather than complete vowel loss.

38

Thank youThank you

Acknowledgments:

- All native speaker participants- Richard Barton (Dartmouth), Emmanuel Ferragne (DDL,

L )Lyon)- Kim Betts, research assistant- The Dickey Center for International Understanding The Dickey Center for International Understanding,

Dartmouth College

39

References• Babaliyeva, A. 2007. Présentation du dialecte Lezgi de Yargun (Azerbaïdjan):

grammaire, textes glosés et traduits. Mémoire principal de Master II, Ecole Pratique d H E d P ides Hautes Etudes, Paris.

• Beckman & Shoji. 1984. Spectral and perceptual evidence for CV coarticulation in devoiced /si/ and /syu/ in Japanese. Phonetica 41. 61-71

• Chitoran & Iskarous. 2008. Acoustic evidence for high vowel devoicing in Lezgi. In g g gSocks et al. (eds.) Proceedings of ISSP 8, Strasbourg, France, December 2008. 93-96

• Chitoran & Marsico. 2010a. Vowel devoicing – an updated phonetic typology. Paper presented at BLS 36. February 2010

• Chitoran & Marsico 2010b A perceptual test of language-specific temporal Chitoran & Marsico. 2010b. A perceptual test of language specific temporal organization. Poster presented at LabPhon 12, U of New Mexico, July 2010.

• Cho. 1993. The phonology and phonetics of voiceless vowels. BLS 19. 64-75• Gordon. 1998. The phonetics and phonology of non-modal vowels: a cross-

linguistic perspective BLS 24 93 105linguistic perspective. BLS 24. 93-105• Haspelmath, M. 1993. A Grammar of Lezgian. Mouton de Gruyter• Jun & Beckman. 1993. A gestural-overlap analysis of vowel devoicing in Japanese

and Korean. Paper presented at January 1993 LSA Meeting, Los Angeles, CA• Kodzasov, S. 1990. Fonetika. In A.E. Kibrik & S.V. Kodzasov (eds.) Sopostovitel’noe

izuchenie dagestanskix jazykov Moscow. 311-347• Ohala. 1983. The origin of sound patterns in vocal tract constraints. In MacNeilage

(ed.) The Production of Speech. 189-216( ) p• Talibov, B.B. 1980. Sravnitel’naja fonetika lezginskix jazykov Nauka: Moscow.• Uslar, P.K. 1896. Etnografija Kavkaza. Jazykoznanie. VI. Kjurinskij jazyk. Tbilisi.

40