Speech development Language & Psycholinguistics BCS 152 / LIN 217 / ASL 260 / PSY 152
Development of Speech Perception and early Word Recognition Categorical perception of phonemes Cross-language effects: the role of experience Segmenting words from fluent speech Reference: which sounds go with which objects?
Categorical Perception Generate a synthetic continuum of speech sounds: from /ba/ to /pa/ Voice onset time (VOT) Equal physical steps Collect labeling (identification) data Collect discrimination data
Eimas et al. (1971)
Remember from Lecture on speech perception Synthetic continuum: /ba/ 0 msec - /pa/ 80 msec
Labeling (identification) 100 50 /ba/ /pa/ 0 0 10 20 30 40 50 60 70 Voice onset time (msec)
Discrimination 100 50 /ba/ /pa/ 0 0 10 20 30 40 50 60 70 Voice onset time (msec)
Labeling & Discrimination 100 50 /ba/ /pa/ 0 0 10 20 30 40 50 60 70 Voice onset time (msec)
Design of Eimas HAS study 100 50 /ba/ /pa/ 0 No Yes No 0 10 20 30 40 50 60 70 Voice onset time (msec)
High Amplitude Sucking (HAS)
Cross-Language Categories in Adults
Discrimination data from infants +25 msec
Discrimination data from infants +4 msec
Discrimination data from infants -25 msec
Consensus view about infant speech perception Early discrimination (including categorical-like effects) are based on acoustic differences, not phonemes Categorical-like effects are a property of the mammalian auditory system, not a special speech mechanism (cf. Chinchilla study in lecture on speech perception) English (and other languages) typically use acoustic categories to form phonemes
(cont d) By 6 months, infants are better at speech discrimination than adults; they can discriminate any speech contrast that may be relevant in a natural language When do infants tune in to their native language?
Werker & Tees (1984) 100 80 Hindi /t/-/t/ Salish /k/-/q/ Percent meeting criterion 60 40 20 0 English infants English infants 7 9 11 12 Age (months) Hindi infants Salish infants
Role of experience Species-specific face discrimination Ö6-month-olds Ö9-month-olds Ö6-month-olds 9-month-olds Pascalis et al. (2002)
What if label monkey faces? Only if mothers label at INDIVIDUAL level do infants maintain their ability to discriminate monkey faces between 6- and 9-months of age. Scott (2009)
Is acoustic input sufficient? Kuhl et al. (2004)
Speech discrimination predicts word learning (2004)
Can infants recognize words in fluent speech?
Mandel & Jusczyk (1997) Preference for first words emerges at 4½ months
Familiarization - Test paradigm (Jusczyk & Aslin, 1995) His bike had big black wheels. The girl rode her big bike. Her bike could go very fast. The bell on the bike was really loud. The boy had a new red bike. Your bike always stays in the garage. The feet were all different sizes. This girl has very big feet. Even the toes on her feet are large. The shoes give the man red feet. His feet get sore from standing all day. The doctor wants your feet to be clean. TEST ITEMS: bike, feet (familiar); cup, dog (novel)
Post-familiarization Test Jusczyk & Aslin (1995) 12 11 p <.005 Familiar Novel Looking Time (sec) 10 9 8 7 6 5 4 1 (18/24 familiar) Did not confuse with: gike (bike) zeet (feet) tup (cup) bawg (dog)
Word segmentation: How do infants know where one word ends and the next word begins? Where are the silences marking word boundaries?
Cues to word-segmentation isolated word presentation prosodic (pitch, rhythm) information statistical information
One type of statistic: Transitional Probabilities Syllable-to-syllable TPs within a word > between words Look at the happy baby Look at the happy kitty hap-py > py-bay py-kit < kit-ty
Artificial language consists of concatenated CV syllables pabikutibudogolatudaropi pabiku tibudo golatu daropi
Experiment 2: Transitional Probabilities Design: Saffran, Aslin & Newport (1996) TPs = 1.0 1.0.33 1.0 1.0 GO LA TU DA RO PI word part-word Test Items
HeadturnPreference Procedure
Sequence of syllables: A-B-C-D-E-F-G-H-I-J-K-L... Test triplets: D-E-F vs. I-J-K 9 8 Words Part-words Listening Time (sec) 7 6 5 4 1
Learning words: Development of the Lexicon How are phonetic skills, once words are segmented from fluent speech, used to store auditory word-forms in the mental lexicon?
Can infants learn minimal pairs? Teach: gak dak Test: Where s the gak?
Result (10-18 months): dak gak Shvachkin, 1948
Werker and Cohen Switch procedure Familiarization bih bih bih bih bih bih bih bih bih bih bih bih bih bih dih dih dih dih dih dih dih dih dih dih dih dih dih dih
Werker and Cohen Switch procedure Test: switch dih dih dih dih dih dih dih dih dih dih dih dih dih dih Test: control dih dih dih dih dih dih dih dih dih dih dih dih dih dih Result: most 14-month-olds fail to detect switch, but some 17-month-olds do
Why so difficult? Objects are novel Words are novel combinations of sounds Yet, sounds themselves are familiar And sounds are easily discriminable Perhaps method is insensitive? Eye-tracking as new measure
Spoken word recognition in infants: Known words Where s the apple? Where s the ball?
IR camera Loudspeaker
Proportion of trials on which infant moves eyes Eye-tracking data: 14-month-olds 1 0.75 Move to target 0.5 Move to distractor 0.25 0 0 500 1000 1500 2000 msec from target onset start on distractor start on target target offset Swingley & Aslin (2002)
Mispronunciation study: items Where s the... dog tog ball gall kitty pity baby vaby apple opple car cur
Proportion of trials on which infant moves eyes 1 0.75 0.5 0.25 Time course of fixation All four conditions different from chance 0 0 500 1000 1500 2000 msec from target onset Correct pronunciation start on distractor start on target Mispronunciation start on distractor start on target Swingley & Aslin (2002)
Acoustic similarity with known word interferes with learning new words meb Ø shang Ø tog dog gall ball Swingley & Aslin (2007)
Are mispronounciations OK? bettle?? 18-month-olds Look longer at match Even if match was just mispronounced block bleck Same for Unlabeled White & Aslin (2011)
Are statistical words meaningful? Does word segmentation precede sound-referent mapping? Graf-Estes et al. (2007): 17-month-olds go la bu - da ro pi golabu more easily attached to object than budaro
Simultaneous word-segmentation and sound-meaning mapping? 6-month-olds 5-syllable sentences Statistical word: AB Two prosodic groupings: [xa] - [Byz] [xab] - [yz] Shukla, White & Aslin (2011)
A B C D E F A B C G H I 1. + stats, + prosody 2. + stats, prosody 3. stats, + prosody 4. stats, prosody The baby ate his food. The baby drank her milk. Test: baby vs. by-drank or baby vs. by-ate
Tobii Eye-tracker IR lights to create corneal reflections LCD display Hi-res video camera
Contingent eye-tracking paradigm Three objects present during learning phase Sentences + one target in motion
Testing phase All three objects in motion Stat word Stat part-word
Results Prosodic word Straddling word Proportion looking over baseline p <.01 for both comparisons and for ANOVA interaction + stats stats
Conclusion Infants both segment and map word candidates to referents 6 months of age is very early Context that more closely matches real world language learning, even though more complex, is successful