Language, Mind, and Brain: Experience Alters perception Chapter 8 The New Cognitive Neurosciences M. Gazzaniga (ed.) Sep 7, 2001
Relevant points from Stein et al. (Chap. 5) AES functions as an association area, and feeds back to the superior colliculus (SC) Multisensory neurons prefer spatially and temporally coincident events Neonatal SC neurons do not receive cortical input. By around 20 weeks postnatal: Percentage of multisensory neurons increase from ~ 0% to ~60% RFs for the SC neurons shrink Mutisensory neurons become spatially aligned and integrated
Relevant points (contd.) One dominant modality (e.g. vision) may first develop and influence the development of other modalities (e.g. audition) Sensory integration seems to occur after development of individual modalities => Initially, the modalities are developmentally related but not integrated End of critical development for cats: ~ 135 days ~ 20 weeks 20 weeks x 7 human days/cat day = 140 days = ~ 5-6 months (about 0.6% - 1% of lifespan) Speculation: Basic perceptual abilities of infants may be developed by 5 months postnatal, and integration starts occurring.
Overview of the paper Development versus learning Language as a case study Developmental changes in speech Crossmodal influences The perceptual magnet effect Acoustic structure of vowels Experience alters perception A theory of speech development (Native Language Magnet) Argument for NLM Nature of language input Vocal learning Brain correlates Conclusion
Development vs. Learning An organism s behavior changes over time. Are the changes caused by: Unfolding of the genetic program (e.g. growing a leg)? A general learning mechanism that detects and adapts to patterns in the world (e.g. a baby learns to use a walker) A combination of both (how are they related?) Kuhl sfour types of models: A) Development and Learning are independent B) Learning <=> Development C) Development enables learning D) Development enables learning, and learning triggers development
Issue: What is development? All learning involves physical changes to the brain Which physical changes are due to development, as opposed to learning? Possible distinction: developmental changes are internally triggered Problem: Kuhl s eventual position is that learning sets up developmental changes => externally triggered Possible distinction 2: Aspects of behavior that were not environmentally specified are developmental. Problem: We are defining development wrt behavior, not physical changes in the brain.
Language as a case study Chomsky (Development + Selection) Skinner (Learning +Instruction) A child grows language, just as he grows arms and legs. Argument Grammatical structure is too complex to be learnt Don t need a lot of input, just some of the right kind of input to set parameters for the developing language A child learns language by associating sounds with objects and events. Argument Language is learnt by the same mechanisms that are used to learn other patterns in the world (learn associations with feedback from a teacher ) Debate applies at three levels: phonology (sounds of speech) word structure (ordering of speech sounds) grammar (ordering of words)
Language as a case study: Developmental changes Cooing 1-4 months Vowel-like sounds Able to discriminate between most speech contrasts Canonical babbling 5-10 months Consonant-vowel strings ( bababa ) First words Two-word utterances 10-15 months 18-24 months Consistent use of a word to refer to objects Usually verb + noun pairs Able to discriminate between only native language speech contrasts Meaningful speech 15 months onwards Long intonated utterances Deaf children follow a similar developmental pattern
Language as a case study: Crossmodal influences The McGurk Effect Auditory information for [b] (closure at front of the mouth) + visual information for [g] (closure at back of the mouth) perception of [d] (closure in the middle of the mouth) 18-20 week old infants prefer correlated visual+speech signals Robust influence of vision on speech perception Polymodal representation of speech
The perceptual magnet effect: Acoustic structure of vowels Each vowel has characteristic resonance frequencies (spectral peaks) Vocal tract Spectrum Formants = spectral peak Quality of a vowel approximately given by locations of the first two Formants (F1 and F2). F1 F2 F3 from P. Ladefoged, Elements of Acoustic Phonetics, 2 nd ed.
The perceptual magnet effect: Language experience alters perception Discrimination is worse near the prototypes of phonetic categories The better the prototype, the worse the discrimination near it (as if the prototype attracts the percepts towards itself) Effect seen in humans but not in monkeys 1-dim MDS solutions (vertical distance arbitrarily chosen to prevent overlap) from Iverson & Kuhl, JASA 97(1)
The perceptual magnet effect: Effect of native language Adults exhibit the PME only for native language speech contrasts 6-month old infants exhibit a similar native language effect
The perceptual magnet effect: Theoretical implications Language input scults the brain [to highlight] contrasts used in language, while de-emphasizing those that do not, and this happens prior to word learning The PME may assist infants in chunking the sound stream into words => Learning (PME) promotes development (word acquisition)
Issue: Paradoxical nature of learning Speech sounds: discriminating among frequently-encountered stimuli is more difficult Face recognition: discrimination among frequently-encountered stimuli is much easier When does learning involve loss of ability, and when does learning sharpen ability? Are both loss and sharpening caused by the same learning/ developmental process? What kind of neural mechanisms could provide this kind of learning? (competitive-learning neural nets?)
The Native Language Magnet model of speech development Phase 1 (ability at birth): Infants can hear almost all speech contrasts Predisposition may be due to auditory abilities Phase 2 (ability at 6 months): Infants start representing native language information Language-specific perceptual maps start to form Phase 3 (stabilization of perceptual maps): Native speech distinctions are maximized Non-native speech distinctions are minimized
NLM
Argument for NLM: Critical Periods Critical periods are not be strictly timed. During sensitive periods : Exposure to some kinds of information may be more effective, and thus alter the duration of the sensitive period Why are 2 nd languages difficult to learn? Maturationally defined temporal window of learning Neural commitment resulting from early learning Reason for neural commitment: Interference Brain plasticity is governed by statistical sufficiency Once the speech prototypes stabilize, further data are ignored for processing efficiency Thus old patterns interfere with learning new patterns. Speculation
Argument for NLM: Nature of language input to the child Motherese is hyperarticulated Possibly to give clearer prototypes of speech sounds to infants and thus promote PME
Argument for NLM: Vocal learning Production becomes discrete at around the same time PME is first seen By ~20 weeks, infants also more likely to imitate correctly
Argument for NLM: Vocal learning Possible explanation for discrete productions: Due to perceptual magnet effects, infants speech representations become more discrete: => they hear more discretely and so produce more discretely => Perceptual representations serve as guides for production
Argument for NLM: Vocal learning
Argument for NLM: Brain correlates Disassociation for speech and nonspeech signals Phonetic processing engaged the left hemisphere Pitch processing of same stimuli engaged the right hemisphere Speech becomes lateralized by ~ 4 months
Conclusion: Learning and development are interrelated Development (genetic program) sets up the constraints for effective learning: Auditory sensitivity and natural discontinuities the neural mechanisms for pattern detection Possible explanation: Experience with linguistic input causes the perceptual magnet effect, which causes the formation of discrete perceptual representations, which allows for word acquisition, which triggers the left-hemisphere dominance, which allows better representation and processing?