Dev Psych Week 5: Language (Concepts)

Poverty of the input

Definition: speech arrives as a continuous, variable stream that doesn't map neatly onto words or objects, yet infants learn language without being taught

Why language is hard: homophones, idioms, word order, synonyms; no clear gaps between words; the world isn't labeled

How infants manage: prenatal experience, social interaction, infant-directed speech, statistical learning

The textbook child: language milestones

1 month: cooing

6-9 months: babbling (single syllables ~6 months)

6-10 months: first signs of word comprehension; communicative gestures

10 months: word-like babbling with rhythm

11-12 months: first words

16-18 months: vocabulary grows; words combined

24-28 months: grammatical function words

30-36 months: productive language; overgeneralizing ("goed")

Language before birth

Hearing works before birth, so language learning starts in the womb

Newborns prefer a story they heard in the womb and can tell their native language from another by its rhythm

Protoconversation

Definition: pre-verbal back-and-forth between infant and caregiver that mimics a real conversation

Types:

Turn-taking: infants understand back-and-forth interaction early

Pointing: ~6 months; basic = "I want," advanced = "Hey, look" (informing)

Babbling: cooing -> single syllables -> word-like with rhythm; practice + feedback tunes speech

Deaf infants learning sign babble with their hands. Theme: resilience

Infant-directed speech (IDS)

Definition: higher, more variable pitch; slower rate; clearer pronunciation

Why it helps: slower pace gives processing time; pitch and rhythm changes highlight word boundaries

Preference: infants prefer IDS, driven by pitch; large replication found it real but about half as big as thought, stronger in older infants, native language, head-turn method

Theme: active research area

Critical periods for language

Definition: a window when language exposure is necessary for typical learning

First language: deprivation until ~13 means never learning typical grammar (fragility)

Second language: earlier arrival = better grammar; grammar-learning ability stays high until ~17-18, then declines

Resilience: children given flawed input create consistent rules and surpass their input (bias to create rules)

Case studies: one child, no control group, hard to generalize

Three speech perception problems

1. Segmentation problem: where do words begin and end?

2. Invariance problem: how to group variable sounds into one category?

3. Tuning problem: which sound differences matter in my language?

Segmentation problem

Problem: speech is a continuous stream; how do infants find word boundaries?

Solutions:

IDS: prosody (pitch, rhythm), emphasis and spacing mark boundaries

Statistical learning: syllables within words follow each other reliably, across word boundaries they don't; works even when other cues are missing

Invariance problem

Problem: phonemes (smallest units that change meaning: cat/cut/cot) never sound the same twice; they vary by speaker, context, even time of day

Solutions/concepts:

Formants: F1 + F2 = which vowel; F3 + F4 = voice "color" (speaker identity); shifting F1/F2 sounds like a different accent

Vowel clusters overlap, so context is needed

Voice onset time (VOT): gap between lips opening and vocal cords vibrating; short = "ba," long = "pa"

Categorical perception: VOT varies continuously, but we hear "ba" or "pa," never in between; infants do this from very early on (within-category change is the key control)

Tuning problem

Problem: each language uses different sound contrasts; infants must tune to their own

Two theories:

"Start bad, improve": infants start poor at all contrasts and gradually learn native ones

Perceptual narrowing: infants start able to hear all contrasts, lose non-native ones by 10-12 months, and sharpen native ones

Evidence supports narrowing; it also happens for sign language handshapes

Escaping narrowing: only live, interactive exposure keeps non-native contrasts; video/audio doesn't (quality of interaction matters)

Theme: development does not equate to improvement

The mapping problem (word learning)

Problem: learning a word means mapping it onto a concept; a new word could refer to almost anything

Solutions:

Joint attention + ostensive labeling: infants look where others look; parents label what they're looking at (ostensive = naming while pointing/showing)

Short arms: objects held close, so fewer things in view when a word is said

Cross-situational learning: tracking which words and objects keep appearing together across ambiguous situations

Mutual exclusivity: each object has one label, so a new word for a named object = a property; enables fast mapping

Measuring vocabulary

MacArthur-Bates CDI: parent checklist of words the child understands/says -> percentile vs. same-age peers

Versions: Words and Gestures (~8-18 mo); Words and Sentences (~16-30 mo)

+ cheap, fast, many words, whole-day view | - parent memory/bias

Looking-while-listening: two pictures, "Where's the puppy?", track eyes

Measures: accuracy + reaction time (processing speed)

+ no talking needed | - only picturable words

Comprehension comes before production, so counting spoken words underestimates knowledge

First words and variability

First words: frequent, socially important words (people, greetings/routines, animals, foods, toys); things a child can see, point to, and share attention over

Wordbank: open database of CDI data from many languages; first words are similar across languages, but vocabulary growth varies enormously

Processing speed: faster word recognition predicts bigger later vocabulary and IQ

Distinctive contexts: words used in more distinctive places/contexts are learned earlier

Theme: studying variability enhances understanding

Syntax

Definition: how words combine into sentences; each word has a category (noun, verb, determiner) and syntax says how categories combine

Chomsky: syntax and meaning are separate ("Colorless green ideas sleep furiously")

How children learn it: fast, by observing and talking, without being taught

Overgeneralization ("goed," "mouses") shows children extract rules rather than copy

Language and thought

Language is a tool for thinking: lets us be precise, ignore irrelevant details, and use labels as shortcuts that reduce mental load

Categories: words (not tones) are "invitations to form categories," even at 3-4 months

Number: we have exact (up to ~3-4) and approximate number systems; language adds counting for exact large numbers. Number words are a "cognitive technology": without them (no number words, or homesign), people can match exact amounts in view but not hold exact large numbers in memory

Linguistic determinism (Sapir-Whorf hypothesis)

Claim: the language you speak limits the thoughts you can have

Classic example: color words shape which color differences people notice

Current status: early enthusiasm faded; language influences thought but doesn't strictly determine it

Chomsky vs. LLMs

Poverty of the stimulus (Chomsky): the input children hear isn't enough to learn grammar, so grammar must be innate (parameters set by input)

How LLMs learn: next-word prediction on trillions of words, attention, human feedback (alignment), multimodality

Challenge to Chomsky: LLMs learn grammar, even tricky sentences, with no built-in grammar

But LLMs ≠ children:

Data: trillions of words vs. ~10-50 million by age 4

Social cues: children use joint attention

Feedback: children rarely get told what's right or wrong

Theme: nature and nurture