Electrophysiological evidence for prelinguistic infants

1
Electrophysiological evidence for prelinguistic
infants word recognition in continuous speech

• Valesca Kooijman, Peter Hagoort, Anne Cutler
• Cognitive Brain Research 24 (2005) 109 116
• Jill Warker
• Psych 593 9/15/2005

2
Recognizing Words in Speech
3
The Trials and Tribulations of Word Segmentation

• Why is it so difficult?
• Speech is extremely fast (20 phonemes/sec or
  faster)
• Phonemes dont always sound the same

4
(No Transcript)
5
The Trials and Tribulations of Word Segmentation

• Why is it so difficult?
• Speech is extremely fast (20 phonemes/sec or
  faster)
• Phonemes dont always sound the same
• in isolation
• in context
• No pauses between words
• Sounds and words run together
• I scream or ice cream

6
Sohow do we do it?

• Adults can make use of meaning and vocabulary
• But how do infants who dont yet have a lexicon
  do it?
• At what age do we begin to segment words from
  continuous speech streams?

7
When do we begin to segment?

• Infants begin to show segmentation abilities
  around 7.5 months
• Become less sensitive to non-native language
  sounds
• Start to have the beginnings of phonetic
  categories
• 
  
• 
  (Jusczyk, 1999)

8
Headturn Preference Procedure
9
Infant Behavioral Studies

• Prefer continuous stream passages containing
  target words (familiarized with targets in
  isolation)
• Prefer target words in isolation (familiarized
  with targets embedded in passages)
• Respond to more than acoustic qualities
  (familiarized with tup, heard cup in passage
  but did not prefer cup passage)
• Possible beginnings of a lexicon (familiarized on
  targets but tested 24 hours later)

10
Sources of Information

• Stress patterns, rhythmic cues
• prosodic cues like pitch
• Distributional properties and statistical
  regularities (Saffran et al., 1996)
• Phonotactics
• fat cat not fa tcat /tk/ cant begin a
  syllable
• Allophonic cues
• night rates vs nitrates
• 
  (Jusczyk, 1999)

11
Going beyond HPP

• How quickly do infants segment?
• the proposed answer ERPs
• can look more closely at the timecourse
• look at time needed to segment and recognize a
  word after hearing it in the stream
• look at whether word recognition is done in whole
  words or if recognition begins before the word is
  over

12
Procedure

• 28 10-month olds
• 20 experiment blocks
• No less than 9 blocks
• Familiarized with 2-syllable word in isolation
• eg. python
• Test phase four sentences with familiar word and
  four without

13
ResultsERPs collected during familiarization
14
Positivity diminishes with familiarization
significant effects over left right frontal
quadrants
familiarity effect starts early on in the word
(around 160 ms)
15
Results ERPs collected during testover left
hemisphere, fam. words gt negative deflection from
350-500ms
response to familiarity begins around 340-370 ms
16
Results Recap

• Infants showed a recognition effect (reduced
  positivity) to a word within half a second during
  the continuous stream
• About 180 ms longer to recognize word in stream
  than in isolation
• harder to determine onset
• Difference in distribution effects
• suggests that different processes may be at work
  during recognition and during segmentation

17
Their Conclusions

• Infants start the recognition response before a
  word is over
• Begin segmentation and recognition processes by
  the end of the first syllable
• NOT using whole word templates
• May be accessing memory representations from
  familiarization and matching them to initial
  portion of word in stream
• Able to generalize across sound tokens
• Further evidence that they are forming phonetic
  categories

18
Discussion

• Can the hemispheric differences in the results
  really be attributed to different processes?
• How much do you think segmentation contributes to
  the formation of the lexicon?
• Given the right hemispheres role in activation
  of broad semantic meaning, do adults show more
  activity in the right hemisphere during word
  recognition?
• Are ERPs sensitive enough to provide a way to
  investigate at what point in time the different
  segmentation cues come into play as well as how
  influential they are?

19
Mismatch Negativity (MMN) paradigm

• Passive oddball paradigm unexpected stimuli
  change results in negative-going increase in
  amplitude
• Typically used to study auditory tones, phonemes,
  and syllables in isolation