Module 4 - L8-9 Statistical Learning

At birth, infants….

• prefer forward-going speech to backward-going speech

• prefer their own language to another language

• can discriminate function words (it, this, in, some) from content words (baby, table, eat) - diff acoustic properties

• prefer mother’s voice to other women’s

• discriminate most sound contrasts used in today’s languages

• learn language IN THE WOMB

At 1 year old, infants…

Learning has slowed (they are picking up on patterns in behaviours)

• only make contrasts present in own language

  • e.g. Hindi has 2 ‘d’ sounds, 8-month old Hindi infants can discriminate this but not English-speaking 8-month olds

• Are learning social knowledge through patterns

  • e.g. they look to the mouth when a cup is raised (expect the movement to finish)

Nativism (Noam Chomsky)

• innate grammar due to POVERTY OF THE STIMULUS

  • input of language available to the child was insufficient to be learnable)

• Universal grammar

  • all languages had shared principles of grammar that are innate → children master grammar by 3 years

• Learning theory (Behaviourism)

  • parents reward children for correct grammar and punish for incorrect (THIS WAS DISPROVED → led to universal grammar beliefs as alternative)

DISPROVED: cannot identify grammatical structures common between different languages

RENEE Baillargeon

• social knowledge (Theory of Mind) is INNATE

  • motivations behind things - why does he pick up the pen? because he wants it and knew that’s where it would be

Statistical Learning & Grammar (frequent frames)

Nouns and Verbs occur at predictable/the same places in sentences → teaches babies about the grammar

Verb

“You ______ it”

• put, see, did, want, have, get

Noun

“The _______ and”

• tractor, shark, roof, house, chair, shirt

Environment is providing stimuli for children to learn

Statistic learning (Language) - Word Segmentation

Identification of a ‘word’ come from the TRANSITIONAL PROBABILITIES between syllable pairs (how often are syllables paired together)

They expect syllables that co-occur with greater frequency

Statistical Learning in Sleeping Newborns (Teinonen et al. 2009)

Even when asleep, they are using familiarity to group sounds into words.

Experiment:

sleeping infants listened to nonsense syllables for 15 minutes

awake → listened for 45 minutes

old and new works

• brain activity was different for old and new words

Evans et al (2009) - Statistical vs Specific Language Impairments

If statistical learning is important for language development, those with SLIs should have worse statistical learning

Experiment:

• ages 5-14, (typical group, SLI group)

• children listened to

  • nonsense syllables 21 min

  • nonsense syllables 42 min

  • tones 42 min

• children asked which of two combos had hear before

Is difficulty SPECIFIC to language? Or just

RESULTS

• SLI kids struggled with BOTH words and tones

• indicates issues with statistical learning

• BETTER STATISTICAL LEARNING → BETTER LANGUAGE ABILITY

Specific Language Impairment (SLI)

• use short sentences, small vocab, word finding problems, difficulty learning new words

• 50% go on to have reading difficulties, dyslexia

Statistical Learning and Language: Summary

• statistical learning assists

  • Grammar acquisition

  • Word segmentation

    • both key components of language acquisition

Social Knowledge (Theory of Mind) & Statistical Learning

Understanding of goals, desires, and beliefs (internal)

Aided by statistical learning of PATTERNS IN BEHAVIOUR

Baldwin et al (2001) - Rug mid-pull

Two videos:

• when rug is IN grasp

• Mid-bend to pick up rug

OR

• Reaching for basket

• with laundry basked IN hand

RESULTS

• infants looked longer at MID-ACTION sequence

  • ONLY when preceded by full video

• They are expecting action sequence to be completed

Inference:

• Babies learn to predict behaviour patterns

• → workout mental states behind behaviour (desires, beliefs, etc)

Ruffman et al. (2023) - Babies & Repeated Behaviour

• head camera recorded everyday activities at home

• Repetition of behaviour

RESULTS:

• Observing MANY repeated behaviours per hour

  • folding washing

  • washing dishes

  • drinking from a cup

  • pushing a car back and forth

Seeing behaviour repeated over 500,000 times in the 1st year of life

Correlation: Repeated Behaviours & Mental State Vocabulary

Repeated behaviours convey mental states such as desires → correlate with mental state vocabulary

• child’s survival depends on learning about events in the world so they can predict outcomes and understand motivations