marc brunelle university of ottawa [email protected] institute of phonetics, cologne, june 7...

39
Tonal coarticulation in Northern and Southern Vietnamese Marc Brunelle University of Ottawa [email protected] Institute of Phonetics, Cologne, June 7 2010

Upload: alvin-rufus

Post on 15-Dec-2015

216 views

Category:

Documents


2 download

TRANSCRIPT

  • Slide 1

Slide 2 Marc Brunelle University of Ottawa [email protected] Institute of Phonetics, Cologne, June 7 2010 Slide 3 The Northern Vietnamese (NVN) tone system Slide 4 The Southern Vietnamese (SVN) tone system Back to discussion Slide 5 Tonal coarticulation The physical realization of a tone varies depending on its environment Neighboring tones Intonation Vowels and consonants Example: In Vietnamese, a mid-level tone starts higher after a rising tone than after a falling tone Slide 6 Progressive coarticulation in tone huy n Female SVN Subject Slide 7 Language-specific hypotheses Coarticulation should be bidirectional as in other tone languages (Han and Kim 1974, Shen 1990, Gandour et al. 1994, Brunelle 2003) There should be more progressive than regressive coarticulation in NVN (Han and Kim 1974, Brunelle 2003) Similar results in Thai (Gandour et al. 1994) Contours should be relatively stable (Han and Kim 1974, Brunelle 2003) Slide 8 If a phonetic dimension is crowded, it should vary less (evidence from V-V and nasal coarticulation) Supporting evidence (Manuel and Krakow 1984, Magen 1984, Cohn 1990, Choi 1995, Manuel 1999) Conflicting evidence (Clumeck 1976, Han 2007) If this is true of tonal coarticulation as well? Pitch targets are less important in NVN than in SVN, because voice quality plays a perceptual role role in NVN tones (V 1981, 1982, Brunelle 2006) Because of the lesser role of voice quality in SVN, contours should be less variable Hypotheses based on models of coarticulation Tone 1 Tone 2 Tone 1 (modal voice) Tone 2 (creaky voice) F0 Dialect ADialect B Slide 9 The recordings 5 NVN speakers (3 women, 2 men) 6 SVN speakers (3 women, 3 men) Read the syllable /ma/ with all tones after the vowel / / bearing all tones. Frame sentences are half-realistic as strange first names were coined for the experiment. Meaningful: ti ni ch m xem ng c hiu khng. Let me say the word rice seedling to see if that man understands. Borderline: ti cho s Ma xem ng y c nh ti khng. Let me greet monk Ma to see if he remembers me. 36 (NVN) or 25 (SVN) frame sentences read 10 times each. Slide 10 Results: direction of coarticulation Impressionistically, there is more progressive than anticipatory coarticulation in all speakers Anticipatory, in ngangProgressive, in ngang Female SVN speaker Slide 11 Results: A tone in which voice quality is crucial When voice quality is a central phonetic cue, pitch varies more (extreme example) Anticipatory, in nngProgressive, in nng Male NVN speaker Slide 12 Quantification of coarticulation The vowels / / and /a/ and the intervening /m/ are measured at 5 equidistant points General linear model analysis for each dialect (modified from Gandour et al. 1994) Dependant variable F0 of 5 measurement points of each tone before all tones (anticipatory coarticulation) F0 of 5 measurement points of each tone after all tones (progressive coarticulation) Factors Speaker F0 at edge of adjacent vowel onset of V2 for anticipatory offset of V1 for progressive Slide 13 Strength of coarticulation (in F values) NVN SVN AnticipatoryProgressive Slide 14 Summary of results (language-specific hypotheses) Stronger coarticulation in NVN than in SVN Bi-directional coarticulation, assimilatory in both directions Different from Thai (Gandour et al. 1994) Stronger progressive than anticipatory coarticulation in both dialects, but: Much more short-distance progressive than anticipatory coarticulation Slightly more long-distance anticipatory than progressive coarticulation Would the effect be stronger in real speech? (wordlist effect) Slide 15 Discussion: the role of voice quality Voice quality in NVN tones allows more variation in pitch without risk of confusion Hence more coarticulation The strength of coarticulation seems predictable from patterns of contrast (Manuel and Krakow 1984) Slide 16 Discussion: Anticipatory vs. Progressive Two types of coarticulation (Perkell and Chiang 1996) Long-distance anticipatory coarticulation is due to planning on the part of the speaker Start early but dont blur tonal contrasts! Short-distance progressive coarticulation is due to immediate physical constraints You cant jump from 100 to 200 Hz in 5 milliseconds! Other types of phonetic dimensions and other languages favor anticipatory coarticulation. X ? Slide 17 Discussion: Why is there more progressive coarticulation? Rises and drops in pitch are often delayed (Ohala 1978) Universal constraints A foresighted speaker could plan ahead and anticipate Tone onsets are less distinct than tone offsets, so blurring is less costly from a communicative point of view A language with more variation at tone onsets than tone offsets should exhibit stronger anticipatory coarticulation Tone charts Slide 18 Conclusion Two types of assimilatory co-articulation in NVN and SVN Long-distance anticipatory co-articulation The speaker is anticipating the following tone Short-distance progressive co-articulation Physical constraints on pitch production; transitions cannot be too abrupt The functional load of pitch determines the extent of coarticulation NVN has pitch and voice quality: more co-articulation SVN has pitch only: less co-articulation The direction of coarticulation is determined by the tonal targets Slide 19 Marc Brunelle [email protected] H Kiu Ph ng [email protected] Martine Grice [email protected] Institute of Phonetics, Cologne June 7, 2010 Slide 20 An old question... Communicative functions Markers (morphemes, particles, ) Sentence restructuring Intonation How do you realize intonation when lexical tone is already making use of f0? A bit Eurocentric, yet very relevant typologically 19 Slide 21 Intonation in East Asian tone languages H LL HH L HL X XX X X X X X H%H XX Strategy 1: Boundary tones alternate with lexical tones All tones in the same tier 20 Slide 22 Intonation in East Asian tone languages H LL HH L HL %q-raise X X XXX XXX Strategy 2: Superposition of intonation over lexical tone Upward/Downward shift Expansion/Compression These effects could be either global or local 21 Slide 23 Mandarin Both boundary tones and superposition (Peng et al. 2005, Shih 1988) Superposition only (Xu 1999, Yuan et al. 2002, Yuan 2004, 2006) Cantonese Boundary tones only (Wong et al. 2005) Superposition only (Fox et al. 2008) Thai Evidence for boundary tones, but overriden by lexical tones (Pittayawat 2007) 22 We need data on more languages if we want to do serious typology Slide 24 Final particles (common in East Asia) Grammatical functions Yes-no questionTrang i lm khng? ImperativeTrang i lm i! Pragmatic functions ConfirmationTrang i lm y. SuggestionTrang i lm nh! Paraphrasing and context As in any other language 23 Slide 25 Global f0 variation Globally lower/higher f0 ( et al. 1998; Nguyn and Boulakia 1999) Local f0 variation Pitch range expansion in stressed syllables ( et al. 1998) Higher pitch on focussed constituents (Jannedy 2007, 2008) Higher pitch on sentence-final question markers (Nguyn and Boulakia 1999; V et al. 2006) Analysed as boundary tones (H and Grice 2010) Intensity (Nguyn and Boulakia 1999) Imperatives louder than declaratives Duration (Nguyn and Boulakia 1999) Questions shorter than declaratives Non-instrumental observations on Northern (Thompson 1965) and Southern Vietnamese (Trn 1967) 24 Slide 26 Why one more study? Previous studies looked at frame sentences not controlled for segments, tones or syntactic structures ( et al. 1998) controlled for tone and segments, but only partly for syntactic structure (Nguyn and Boulakia 1999) controlled for tones, but not for segments and syntactic structure (V et al. 2006) Our study uses simpler sentences, but controls for tones, segments and syntactic structures. 25 Slide 27 Which intonational cues are predominant in Northern Vietnamese? Pitch Global or local? Intensity Duration Is intonation realized through the addition of tones or through superposition (or both)? How much inter-speaker variation do we find? 26 Slide 28 6 sentences, all 4-words long First 3 words: Always SVO with constant tone Khng [xo m ] is always the last word ( only or yes- no question marker) The sentences can have 4 different meanings, depending on the intonation Ex: Ty n c m khng Declarative : Ty only eats rice. Annoyed declarative: Ty only eats rice. (I just told you!) Yes-no question: Does Ty eat rice? Command question: Ty, will you eat your rice?! Declar. Questions Unmarked Emphasis 27 Slide 29 8 native speakers recorded in Hanoi 3 men, 5 women Speakers were requested to produce the sentences appropriately according to given contexts Not a trivial task! Each sentence was recorded three times 6 tones X 4 communicative functions X 3 repetitions: 72 sentences 28 Slide 30 Sentences labelled and measured in Praat Each sentence divided into 4 syllables Each syllable measured at 5 equidistant points Data inspected for doubling, halving and irregular vocal fold vibrations 29 Slide 31 Statistical analysis: GLMs in PASW (SPSS 18) Conducted independently for each subject Dependant variables at each sampling point Duration f0 Intensity Independent variables Tone Communicative function Interaction 30 Slide 32 No clear difference between comm. functions f0 range expands from 1 st to 3 rd syll. Likely due to coarticulation rather than intonation 31 Slide 33 No special emphasis Emphasis No clear contrast between questions and declaratives Questions are high Declaratives are low Questions have a final rise Declaratives have a final fall Emphatic functions higher than their non- emphatic counterparts 32 Slide 34 Global effect: Higher overall f0 for questions than for declaratives: 5/8 speakers Frequency code (Ohala 1983) for emphatic than for unmarked: 6/8 speakers Effort code (Gussenhoven 2004) Local effect on last syllable Clear rise at the end of the question marker khng : 3/8 speakers Higher pitch at the end of khng in questions: 1 more speaker (Total 4/8) Clear fall at the end of the declarative: 4/8 speakers 33 Slide 35 Higher intensity in emphatic functions Effort code again Consistent for 4/8 speakers Incr. contrast towards end of sentence No systematic divide between questions and declaratives 34 Annoyed declarative higher than normal declarative Command question higher than normal question Slide 36 There are stat. sign. durational differences in all speakers but they are not consistent across speakers Significant differences are not always on the same syllables The differences do not always go in the same direction 35 Slide 37 Strategies for marking communicative functions are variable across Hanoi speakers Robust Higher overall pitch marks either questions or emphasis Common Final rise for questions/Final fall for declaratives Higher intensity (esp. sentence final) marks emphasis Inconsistent Duration seems speaker-specific 36 Slide 38 Which intonational cues are predominant in VN? Pitch Optional intensity contrasts Is intonation realized through the addition of tones or through superposition (or both)? Superposition is systematic Focal tones are common, but optional How much inter-speaker variation do we find? A lot. Why? 37 Slide 39 Various intonational strategies available for communicative functions Nonetheless, intonation plays a more subtle role than in non-tonal languages Almost null for 2/8 speakers despite the exaggerated nature of the experimental task Intonation is not grammaticalized. Rather, universal tendencies emerge Frequency and effort codes 38 Slide 40 Where it all links up Intonation in Southern Vietnamese f0 seems more important in tonal contrasts: could that affect intonation? Less leeway for f0 variation??? Fewer and less frequent final particles Intonation could be expected to be more important for marking communicative functions 39