2. 1 Cross-language evidence for three factors in speech perception Sandra Anacleto uOttawa

3. 2 Article Werker, J. & Logan, J. (1985). Cross- language evidence for three factors in speech perception. Perception & Psychophysics 37 (1), 35-44. Janet Werker Ph.D. (Indiana) UBC Department of Psychology http://infantstudies.psych.ubc.ca/ John Logan Ph.D. (Indiana) Carleton University Cognitive Science and Psychology

4. 3 Why do this study? Controversy of whether speech perception could be best explained by single factor psychoacoustic models, single factor specialized linguistic processor models, or dual factor models (phonetic & psychoacoustics)

5. 4 Background Werker & Tees (1984b) Found the single-factor and dual- factor models could not account for their findings Results showed that the English children 6-8 months can discriminate the non-English distinctions as well as native Hindi and Thompson speakers. But English speaking adults and infants 10-12 months cannot.

6. 5 Background Cont. The length of the ISI was shown to distinguish phonemic from phonetic processing.

7. 6 Why consider a three factors in speech perception? 1.When subjects perceive stimuli according to native language phonological categories they are demonstrating “phonemic perception”. 2. When subjects show a sensitivity to phonetic distinctions that are used in some other, (not their native) languages they are using phonetic perception.

8. 7 Why consider a three factors in speech perception? 3. P sychoacoustic or auditory level processing is demonstrated only when subjects show a sensitivity to acoustic differences that do not correspond to phonetic boundaries that function phonologically (to contrast meaning) in any of the world’s languages.

9. 8 Goal of the experiment This experiment was designed to test the proposed three-factor hypothesis against a modified dual- factor model by attempting to determine whether phonemic, phonetic, and auditory processing could be differentiated as independent processing factors under varying experimental conditions.

10. 9 Cross-Language Differences English vs. Hindi alveolar [d] retroflex [D] ?

11. 10 What About Non-Humans? Chinchillas show categorical perception of voicing contrasts!

12. 11 Experiment 1 AX task: two-choice button press response 1500msec, 500msec, and 250msec ISI condition Uses a within-, rather than a between-, subjects design Subjects were tested only on the Hindi retroflex/dental contrast

13. 12 Subjects 30 adults (15 males and 15 females) w/ normal hearing (how do they know?) Psychology students Unilingual English speakers

14. 13 Stimuli Three types: (1) physically identical (PI) pairings (2) name-identical (NI) pairings (3) different (DIFF) pairings Hindi place of articulation contrast (contrast the voiceless, unaspirated retroflex, and dental consonants /  / vs. /t/

15. 14 Three-factor model: Predictions

16. 15 Stimuli

17. 16 Results: Exp. 1

18. 17 Results: Exp 1 Not expect! Subjects used a phonetic processing strategy in the shortest ISI and possibly using a phonetic and auditory strategy in the longest ISI What Now?

19. 18 Results: Exp. 1 ANOVA assess possibility that in a within groups design the potential effects of ISI were lessened by the context in which the ISI conditions were presented. 1. Looked at the effect of the position 2. Effect of the order

20. 19 Conclusion This interaction suggests that order of presentation interacted with both ISI and pairing type in affecting subjects’ scores. Within-subjects design was not appropriate for assessing the proposed three factors.

21. 20 Experiment 2 The purpose: eliminate the problem of interaction of Exp 1. Tested subjects on their ability to discriminate the several Hindi retroflex and dental exemplars in a between-subjects design. Three groups of subjects: one in 250, 500, and 1500msec in ISI

22. 21 What’s different in Exp. 2? More testing trials [480 vs. 192] The proportion of PI pairings was increased to be equal to that of NI pairings Measure of reaction time (RT) was included in addition to type of response

23. 22 Method Subjects: 30 adults (15 female and 15 male) –10 subjects were tested in all three ISI –all unilingual Stimuli: –same 4 retroflex and 4 dental from Exp 1 –for each of the 3 ISI conditions, the pairings were randomized into five blocks of 96 trials. –the within-category trials contained: 24 PI stimulus pairs and 24 NI stimulus pairs

24. 23 Results: Exp. 2

25. 24 Results: Exp. 2

26. 25 Results: Exp 2 This suggests that the three ISI conditions affect performance differentially and is consistent with previous work showing that ISI affects the use of phonetic vs. auditory levels of processing.

27. 26 Results: Exp 2 Comparing the average reaction time (RT) for each type of pairing in each ISI condition: two significant interactions: one between type of response and type of pairing [p <.001] and the other among ISI, type of response, and type of pairing [p <.001].

28. 27 Results: Exp 2

29. 28 Conclusion: The strong prediction of “phonetic” processing was not supported in the data collected from English listeners in Experiment 2, but their data were in the predicted direction in the last three trial blocks in the 250-msec ISI condition, in which the difference between proportion “same” responses to NI and P1 pairings was less than that between NI and DIF pairings.

30. 29 Experiment 3: Purpose? Determine how native Hindi speakers would respond to Hindi syllables when tested in an AX task at the longest ISI. This would provide data on the natural categories used by native speakers

31. 30 Method: Exp. 3 AX procedure with the same stimuli A single, long ISI was chosen for use in this experiment Subjects: –4 native Hindi speakers (3 males and 1 female) –All 4 spoke English

32. 31 Results: Exp. 3 ANOVA: a significant effect for type of pairing [p <.001]. The classification of P1 and NI pairings as “same” an equally high proportion of the time; The DlF pairings were consistently perceived as “different”. This clearly indicates that Hindi subjects were using a single processing strategy in perceiving these syllables: “phonemic processing”.

33. 32 Final Conclusions The data from all three experiments together, indicating the presence of a phonetic level as well a phonemic and auditory levels, suggest that dual-factor models may also be inadequate to explain the data pattern.

34. 33 Final Conclusions The subjects tested in the 1500msec condition in Exp. 2 showed a data pattern consistent with the prediction for “phonemic” perception in the first two trial blocks. Thus, it appears that, without practice, subjects rely on phonemic categories when responding to speech syllables in paradigms that have high memory requirements.

35. 34 Final Conclusions The strong prediction of “phonetic” processing was not supported in the data collected from English listeners in Experiment 2, but their data were in the predicted direction in the last three trial blocks in the 250-msec ISI condition Support for the phonetic level was provided in Exp. 1 in both the 250 and 500msec ISI conditions. Clear support for this universal “phonetic’ level was supplied by the Hindi subjects in Exp. 3.

36. 35 Final Conclusions In recap, subjects: (1) classify the syllables according to familiar phonemic categories, (2) show a perceptual sensitivity to nonnative, phonetically relevant category boundaries, and (3) discriminate syllables on the basis of any acoustic variability between individual exemplars.

37. 36 THE END Thank You!