Infant Speech Perception. LSCP Infant Lab

Transcription

1 Infant Speech Perception LSCP Infant Lab

2 Outline Introduction to Phonology Problem of Speech Perception Testing two theories of speech perception Infant Categorical Perception Cross-species categorical perception The use of visual cues

3 Universal Structural Design

4 English Consonants

5 15 1. Introduction to Phonology

6 Phonemes Vowels: unimpeded sound through vibrating vocal cords Vary by: Placement of tongue ee is high front ah is low back Rounding of lips

7 Phonemes Consonants: Sound is impeded in some way Vary by: 1. Place of articulation m, p, b n, t, d ng, k, g

8 Consonants vary by Phonemes 2. Kind of articulation: stop: p, b, d nasal: n,m fricative: ch, th, f 3. Voicing, Nasality, Creaky Voice, Clicks voiced: b, d, z... voiceless: p, t, s...

9 Phonemes are bundles of features P: bilabial, voiceless, stop B: bilabial, voiced, stop S: dental, voiceless, fricative

10 Phonemes are bundles of features p: bilabial, voiceless, stop s: dental, voiceless, fricative

11 Evidence Phonological features participate in speech errors Phonological rules typically refer to features: plural marker assimilates the voicing of the consonant before it cup/s/, beet/s/, back/s/ cub/z/, bead/z/, bag/z/

12 Phonemes? Phonemes

13 2. A Brief Tour of Speech Perception 15 10:25

14 The String Fallacy Speech sounds seem separable and sequential: like beads on a string Reality: Speech sounds overlap (co-articulation) Each speech sound is affected by the elements around it

15 Two problems of speech Segmentation perception how do we discover where one phoneme ends and the next begins Invariance: How can we identify a phoneme when it s sound changes depending on context? We won t be answering these questions today

16 One Hint: some complex features of a speech sound may stay constant

17 Speech sounds vary along a continuum

18 Yet we perceive them as discrete categories B D G

19 Categorical Perception: Identification % identification

20 Categorical Perception: Discrimination

21 3. Where does this ability come from? 12 10:40

22 Two theories The Motor Theory: perception is informed by our innate knowledge of articulation The Auditory Theory: speech perception is based solely on auditory properties of speech

23 Motor Theory We have innate knowledge of how articulation changes depending on context

24 Motor Theory Innate knowledge of articulation Used to predict changes in sound wave and recover phonemes

25 Auditory Theory Auditory system breaks speech into phonemes innate property of sensory system or perceptual learning

26 Contrasting the two theories The Motor Theory 1. Perception is based on production Process: Determine what articulatory gestures a speaker made Then identify phoneme The Auditory Theory 1. Perception by generic auditory mechanisms Process Auditory system transforms sound wave so phonemes available

27 Contrasting the two theories The Motor Theory 2. Perception is species specific Speech production and speech perception evolved together Only humans speak, so only humans have SP system The Auditory Theory 2. Not species specific Production system evolved to make use of existing auditory capacities No special adaptations to speech

28 Contrasting the two theories The Motor Theory 3. SP is innate Tacit knowledge of articulation given by evolution The Auditory Theory 3. SP may be innate Innate properties of the auditory system may be adequate to isolate phonemes Or Perceptual Learning may be required

29 4. Is Speech Perception Innate? 18 10:52

30 Do newborns perceive phonemes categorically? Predictions: If categorical perception requires knowledge of contrastive phonemes: NO If it requires exposure to language: NO If it is an innate ability: YES But how can you test immobile infants?

31 High Amplitude Sucking Procedure Infant given a pacifier Contains sensor to monitor sucking rates Each time the infant sucks the paci, a stimulus is played LSCP Infant Lab

32 High Amplitude Sucking Procedure Initially sucking rate increases (novelty) Then it decreases This decline in response is habituation LSCP Infant Lab

33 High Amplitude Sucking Procedure When sucking rate declines to set point (habituation criterion) The computer switches the auditory stimulus LSCP Infant Lab

34 High Amplitude Sucking Procedure If sucking rate increases Then we know the infant has detected the change The renewed response is dishabituation LSCP Infant Lab

35 Stimuli for the Eimas Study Voice Onset Time: time btw consonant release and vocal cord vibration S1 S2 S VOT in milliseconds

36 Stimuli for the Eimas Study Phoneme Category Boundary at 25ms BA PA 1 PA VOT in milliseconds

37 Predictions Innate Categorical Perception Untuned Sensitivity Insensitive Within Category remain habituated dishabituate remain habituated Between Category dishabituate dishabituate remain habituated

38 Stimuli from different phonological categories are distinguished SPM sucks per minute PA 1 BA Eimas, et al., 1971 Time

39 But stimuli from the same phonological category are not SPM sucks per minute PA 1 PA 2 Eimas, et al., 1971 Time

40 Speech Perception is Innate Predicted by Motor Theory Speech Perception driven by innate knowledge of articulation Consistent with Auditory Theory Speech perception due to innate structure of auditory system

41 5. Is Speech Perception Species Specific? 10 11:10

42 Do other animals perceive phonemes categorically? Kuhl & Miller, 1978: test chinchillas and humans with identical stimuli Human Task: identification (b or p) Chinchillas: avoidance conditioning

43 Avoidance Conditioning Procedure Shock paired with speech sound at one end of continuum (A) 0 PA80 VOT in milliseconds

44 Avoidance Conditioning Procedure Sound at other end (B) paired with safety BA0 80 VOT in milliseconds

45 Avoidance Conditioning Procedure Animal learns to run to other side of cage when it hears sound A but stay after B What will they do for sounds in between? stay??? run

46 Predictions Percent labelled [b] Voice Onset Time (ms) Categorical Perception Graded Perception

47 Kuhl & Miller, 1978 Percent labelled [b] Voice Onset Time (ms) English Speakers

48 Kuhl & Miller, 1978 Percent labelled [b] Voice Onset Time (ms) English Speakers Chinchillas

49 Not all aspects of speech perception are species specific Contrary to Motor Theory Claim: only humans have knowledge (innate or learned) of articulation Consistent with Auditory Theory General auditory abilities adequate for (some aspects of) speech perception

50 6. Is Speech Perception Affected by Knowledge of Articulation? 10 11:20

51 McGurk Effect Vary visual information about place of articulation palatal: GA dental: DA bilabial: BA Vary auditory stimulus (BA DA GA) Task: phoneme identification

52 McGurk Effect Finding: phoneme category boundary affected by visual information sound BA + visual GA = percept DA Adult speech perception affected by visual cues Did they learn it?

53 Do infants know visual cues to articulation? No evidence of visual cues influencing categorization But can infants match sound with the correct mouth shape?

54 Preferential Looking (Kuhl & Meltzoff, 1982 [a]-face [i]-face

55 a...a...a...a..a [a]-face [i]-face

56 Speech Perception is influenced by information about articulation Predicted by Motor Theory Innate connection btw production and perception Inconsistent with Auditory Theory Though a visualauditory theory might explain this Would require perceptual learning

57 Provisional Conclusions Speech Perception makes use of auditory mechanisms which evolved prior to language These abilities are innate Speech Perception also makes use of our knowledge of articulation This ability MAY be innate

58 McGurk_large.mov McGurk Effect Revisited

59 Development of Phonological Representations

60 Working toward a model. Syntactic Lexical Constructed Phonological Auditory Innate? Phonetic Articulatory

61 What s innate? Auditory abilities Articulatory rudiments Not developed Desire to coo and babble even in deaf infants Auditory Articulatory Innate

62 Newborns are universal listeners Infants perceive speech categorically Newborns are sensitive to every phonological distinction yet tested What happens to the categories that are not used in the language that they learn?

63 English VOT perception

64 English VOT production Not uniform 2 categories

65 English s - sh 2 categories Notice also a shift in curve with a or u Cue trading

66 English versus Japanese r - l Developmental changes

67 English versus Spanish VOT Developmental changes

68 Testing Across the Lifespan Habituation paradigms work only with young infants Adults & children can be asked to detect a change The Conditioned Head Turn Paradigm parallel task for older infants

69 Conditioned Head Turn Paradigm Kuhl Lab, U Washington, 1992

70 Conditioned Head Turn Paradigm Kuhl Lab, U Washington, 1992

71 Conditioned Head Turn Infant Trained with 2 clearly different auditory stimuli (bell and whistle) each time the sound changes the toy is activated next toy is activated only if the sound changes AND the baby turn to look at it

72 Conditioned Head Turn Child hears Stimulus 1 (/ba/) repeatedly Then Stimulus 2 is presented (/da/) If child detects difference, they should turn to look at the toy when the stimulus changes If they do not detect it, they shouldn t turn until after the toy is activated

73 Contrasting Views

74 Maintenance or Loss Model If you don t use a perceptual ability, you lose it Parallel to aspects of early visual development

75 Maintenance or Loss Model Problems: Children older than 1yr can acquire a new language with no accent Many of the relevant sounds appear in child s input but not meaningful Adults can be trained to make distinctions Perceptual distinction is readily available for non-linguistic tasks

76 Functional Reorganization The newborn has perceptual categories Those which are meaningful in the native language become speech categories The remainder are perceived but not recruited in speech perception

77 Speech Perception Before Reorganization Behavior Auditory Phonetic Articulatory Innate & Universal

78 Creation of Phonological Representations Behavior Constructed & Language Specific Phonology Auditory Innate Phonetic Articulatory

79 Functional Reorganization Syntactic Appears around 14m Lexical Constructed Appears around 10m Phonological Auditory Innate Phonetic Articulatory

80 Critical period Foreign accent syndrome Why are foreign languages hard to perceive? Differences in phonological categories and phonological rules.

81 Epenthetic Vowels French allows consonant clusters Japanese does not Japanese loan words add vowels to break up clusters

82 Epenthetic Vowels Japanese speakers tend to produce and perceive additional vowels in L2 Add epenthetic vowel /u/ to break up clusters For French: ebuzo & ebzo distinct Not for Japanese learner of French

83 Behavioral results Japanese can t distinguish presence or absence of these vowels ebzo ebuzo Dupoux, Fushimi, Kakehi & Mehler, 1999

84 ERP experiment habituation, switch paradigm ebuzo ebuzo ebzo ebzo ebzo.ebuzo record EEG & average French have early, middle, late mismatch responses Japanese only have late response Jacquemot, Pallier, LeBihan, Dehaene & Dupoux, 2002

85 Early Response

86 Middle Response

87 Late response

88 Problem Japanese should detect the difference at phonetic level But supress at the lexical level and in decision process Why do they only have the late response? Could the phonetic response be the late response?

89 When are epenthetic vowels added? But gone by here Syntactic Constructed Distinction should be available here. Lexical Phonological Auditory Innate Phonetic Articulatory