Consonantal Effects on Pitch in Tonal Languages

Transcription

1 Consonantal Effects on Pitch in Tonal Languages Qian Luo Department of Linguistics and Languages Michigan State University

2 INTRODUCTION

3 Consistent Consonantal Effects on F0 Consistent Voicing Effect: F0 after voiceless obstruents is usually higher than F0 after voiced ones F0 (Hz) p b ms Time (ms) 3

4 Consistent Consonantal Effects on F0 Consistent Voicing Effect: F0 after voiceless obstruents is usually higher than F0 after voiced ones. Effect Low F0 High F0 Neutral Consonant type voiced Spanish, French, Italian, Portuguese, Hindi, German, Swedish, English, etc. Spanish, French, voiceless Italian, Portuguese, Hindi, German, Swedish, English, etc. 4

5 Inconsistent Consonantal Effects on Pitch Inconsistent Aspiration / Sonorancy Effects: Inconsistent patterns: either raise or lower F0; Inconsistent reports from the same language. Effect Low F0 High F0 Neutral Consonant type aspiration Mandarin (Xu & Xu 2003), Cantonese (Francis 2006), Danish (Jeel 1975) Mandarin (Chen 2011), Cantonese (Zee 1980), Taiwanese (Lai et al. 2009) Danish (Fischer- Jørgensen 1968) Burmese (Maddieson Thai (Gandour 1974), sonorant 1984); Danish (Jeel Tibetan (Kjellin 1977), 1975), Gã and Yoruba Danish (Thorsen (Painter 1978), Hindi 1974), Bade (Tang (Ohala 1980) 2008) 5

6 Functional vs. Physical Accounts Enhancing contrasts of laryngeal features a hybrid account (Kingston 2007; Chen 2011) Physical properties of laryngeal articulation Functional Account The enhancement of contrastive features causes the F0 differences by controlled articulations (Kingston and Diehl 1994). Difference in onset F0 Physical Account F0 difference is an unintended side-effect of differences in physiological and aerodynamic properties (Whalen and Levitt 1995; Connell 2002) 6

7 When the prosodic context plays a role Previous studies show perturbation effects can be conditioned by: Cantonese Participants! Pitch-accent (Jun 1996; Kingston 2007); Focus (Chen 2011); Intonation (Kirby and Ladd 2016)! Lexical tones (Xu and Xu 2003) 7

8 Adding Tone into the Picture Enhancing contrasts of laryngeal features Physical properties of laryngeal articulation Functional Account Difference in onset F0 Physical Account Enhancing contrasts of lexical tones shared physical properties Physical properties of tone articulation 8

9 Tone Dispersion across Languages Theory of Adaptive Dispersion (Liljencrants and Lindblom, 1972): the size of a tonal inventory affects acoustic tone-space size Alexander (2010) found that tone-space size was fixed across level-tone and contour-tone at midpoint and offglide However, languages with smaller tone inventories had larger tone spaces at onset. 9

10 Tone Dispersion across Languages Alexander (2010): Languages with smaller tone inventories had larger tone spaces at onset. 10

11 Functional Account: Enhance the Tone Contrast If languages with bigger tone inventories have smaller tone spaces at onset, it is more risky to stretch the pitch with perturbation in a smaller space, the stretching can cause trespassing into other tonal categories, and thus threatens tone contrasts. This predicts that perturbation effects would be weaker in languages with bigger tone inventories than those with smaller ones. 11

12 Research Questions 1 Are consonantal effects on F0 conditioned by tones in tonal languages? 2 If yes, is it due to enhancement of contrastive tones (functional) or Intrinsic properties of tone articulations (physical )? 12

13 Predictions by Functional vs. Physical Accounts Prediction I: when tone inventories differ Prediction II: same inputs but different surface Prediction III: different inputs but neutralized surface Functional Physical Weaker effects in languages with bigger inventory. The inventory size does not matter. Same underlying inputs lead to similar effects, despite difference in phonetic realizations. 13

14 When tones crash into the party the Chaozhou experiment to test the inputoutput predictions (II, III) Outline 1 INTRODUCTION 2 Experiment I 3 Experiment II 4 Wrap-up the Mandarin and Cantonese experiment to test the dispersion prediction (I) Conclusion & Discussion 14

15 EXPERIMENT I Mandarin and Cantonese 15

16 Target languages: Cantonese and Mandarin Mandarin mandatory in schools, government and official media Cantonese spoken in Guangdong, Guangxi, Hong Kong and Macau 16

17 Tone Inventory of Cantonese Six contrastive long tones; The on-going merger between T2(35) and T5(23) Two-way laryngeal contrasts: aspirated vs. unaspirated obstruents Figure from Fung and Wong (2010) T1(55) T2(35) T3(33) T4(21) T5(23) T6(22) Onset F0 high mid mid low low low Shape level rising level falling rising level 17

18 Tone Inventory of Mandarin! Four contrastive long tones! Two-way laryngeal contrasts: aspirated vs. unaspirated obstruents T1(55) T2(35) T3(214) T4(51) Onset F0 high mid low high Shape level rising fall-rise falling 18

19 L2:Mandarin native T1 (55) 普 Bilingual Patterns Speech Learning Model (Fledge 1987, 1995, 1997, 1999, 2002 ) : Difference between L1 and L2 may create new non-native sound categories, instead of assimilating to native phonological categories. 粤 L1:Cantonese T1 (55) T2 (35) T2 (35) T3 (213) T4 (51) T3 (33) T4 (21) T5 (23) T6 (22) No need to maintain contrast across L1 & L2? The need to maintain contrast within L1 or L2. 19

20 Repeating Prediction I related to dispersion Functional Physical Prediction I: when tone inventories differ The perturbation effect in Cantonese T2(35) is weaker than that in T2(35) in native and non-native Mandarin. It s unlikely for tones with similar onset pitch and tone shapes to have significant different perturbation effects. 20

21 Exp. 1: participants Mandarin Cantonese Cantonese Participants 2F 4M 4F 2M All speak standard Mandarin and no other Chinese dialects; aged All speak standard Cantonese natively, and Mandarin nearnatively. aged All took the Mandarin experiment, for the interest of bilingual patterns. 21

22 Exp. 1: stimuli Onsets: Aspirated (p h, t h, k h ), Unaspirated (p, t, k), Sonorant (m, n, l) Vowels: /a, e, o/ Monosyllabic words in CV form presented Chinese characters Mandarin (n=86) Covered all four Mandarin tones Carrier phrase: [wɔ213 ʂwɔ55 tsɯ51 san55 ts h i51] I say for three times. Cantonese Participants Cantonese (n=87) Covered all six Cantonese tones Carrier phrase: [ŋɔ23 kɔŋ35 tsi22 sam55 ts h i33] I say for three times. 22

23 Exp. 1: procedure Stimuli are randomly presented through PsychoPy (Pierce 2007). Cantonese Participants 3 repetitions of each sentence at a normal pace Cantonese speakers participated in the Cantonese experiment first, and then the Mandarin experiment after a break. Native Mandarin speakers participated in the Mandarin experiment 23

24 Exp. 1: measurement and analysis Measurement Cantonese Participants F0 extracted at every 5ms within the first 50ms of the vowel The average values of the first 20ms were analyzed. Analysis Normalization: Hz were converted to cents ANOVA and paired t-tests were performed for data analysis. Sonorants are used as baseline to interpret the results (Hanson 2009; Kirby and Ladd 2016). 24

25 Segmentation example: p h a1(55) Cantonese Participants p h a1(55) in Mandarin. c = closure; r = release; a =the vowel /a/ 25

26 Segmentation example: pa1(55) Cantonese Participants pa1(55) in Mandarin. c = closure; r = release; a =the vowel /a/ 26

27 Exp. 1 results: Cantonese 27

28 T1(55): Aspirated > Sonorant Unaspirated * * Exp. 1 results: Cantonese 28

29 Exp. 1 results: native Mandarin 29

30 Exp. 1 results: native Mandarin o o o T1(55): Aspirated Unaspirated > Sonorant T2(35): Unaspirated Sonorant > Aspirated T4(51): Aspirated Unaspirated > Sonorant * * * * * * 30

31 Exp. 1 results: Cantonese Mandarin 31

32 Exp. 1 results: Cantonese Mandarin o No significant difference in any tonal context 32

33 Exp. 1: summary of the results Lexical tones Onset F0 values (20ms) Cantonese T1(55) Aspirated > Sonorant Unaspirated Mandarin T1(55) Aspirated Unaspirated > Sonorant Mandarin T4(51) Mandarin T2(35) Unaspirated Sonorant > Aspirated 33

34 Exp. 1 results: T1(55) in different languages o o Cantonese: Aspirated > Sonorant Unaspirated Both Mandarin: Aspirated Unaspirated > Sonorant 34

35 Exp. 1 results: T2(35) in different languages o Weaker effects in Cantonese than in Mandarin -- predicted by the functional account 35

36 Intra-speaker Inconsistency o Participant 2, Female, age: 26 36

37 Exp. 1: Discussion Further analysis is needed to check whether physical properties of T1(55) / T2(35) are similar between Cantonese and Mandarin. Cantonese Participants The physical account can also explain the results: Manuel (1999): parallel to vowel coarticulation, which shows that low vowels allow the greatest variability due to wider range of variances of F1 and F2. If Mandarin has larger tonal space, does it indicate larger variances and thus greater variability for the perturbation to occur? The inconsistency issues still remain: Why F0 after aspirated obstruents is lower than F0 after unaspirated only in Mandarin T2(35)? 37

38 EXPERIMENT II In Chaozhou Tone Sandhi 38

39 Arguments for Enhancing the Laryngeal Contrast Kingston & Diehl 1994 F0 does not depend on phonetic differences in voicing in Tamil stops. Variable effects on F0 following English [s] + stop clusters, for [voice] is neutralized Input 1 Neutralized allophones may induce variable effects. Input 2 Output 39

40 If it were underlying tones that matter for the effects Input tone 1 Input tone 2 Input tone 1 Output tone Neutralized tones may induce variable effects. Output tone1 Output tone 2 Different surface tones may have similar effects. neutralized context allotonic context 40

41 Chaozhou Tone Sandhi Relevant tones: T1(33), T2(53), T5(55), T7(11) Except T1(33) and T7(11), all tones undergo sandhi in non-phrase final position Neutralized context Input form T5(55)+T1(33)/T5(55)! T7(11)+T1(33)/T5(55)! Sandhi form 11 derived1 +T1(33)/T5(55) 11 derived2 +T1(33)/T5(55) Neutralized tones may have variable effects. Input form T2(53)+T1(33)" T2(53)+T5(55)" Allotonic context Sandhi form 24 derived +T1(33) 35 derived +T5(55) Different surface allotones may have similar effects. 41

42 Exp. 2 result: the neutralized context Citation form T5(55)+T1(33)/T5(55)! T7(11)+T1(33)/T5(55)! Sandhi form 11 derived1 +T1(33)/T5(55) 11 derived2 +T1(33)/T5(55) 42

43 Exp. 2 result: the allotonic context Citation form T2(53)+T1(33)" T2(53)+T5(55)" Sandhi form 24 derived +T1(33) 35 derived +T5(55) 43

44 WRAP-UP 44

45 Consonantal Effects on Pitch in Tonal Languages Enhancing contrasts of laryngeal features Physical properties of laryngeal articulation Functional Account Difference in onset F0 Physical Account Enhancing contrasts of lexical tones shared physical properties Physical properties of tone articulation 45

46 ACKNOWLEDGEMENTS Michigan Karthik Durvasula, Yen-Hwei Lin Audience at Phono Group Edinburgh James Kirby, Bert Remijsen, Misnadin, Patrick Honeybone, Bob Ladd, Audience at P-Workshop Tonal Aspects of Languages Carlos Gussenhoven, Yiya Chen, Christian DiCarnio 46