Evidence for the centrality of experience Evidence for

Evidence for the centrality of experience

Evidence for the centrality of experience Psycholinguistic Studies of Entrenchment Catherine L. Harris (1997) p Constrains on Statistical learning Jenny R. Saffran (2002) p

Outline of Lecture The big picture p Idiom-level representation p Collocation level representation p Statistical learning p Conclusions p

The big picture • Psycholinguistics • Computation

Evidence for the Centrality of Experience Entrenchment p Statistics p Performance p

Entrenched expressions Words (Carr 1986) p Common word combinations (collocations) p Multi word idioms p

Does “Idiom Level” of Representation exist ? (Harris 1997) Priming p Semantic priming p n n Spreading activation Semantic integration Great minds think alike Great minds alike p Minds think alike p

Idiom level p Great minds think alike 42 - 4 words idiom n n Contiguous with target Noncontiguous with target p p p 30 – first two words best eliciting 12 – middle two words best eliciting. Associative and Semantic trios n n n baby, cradle -> bottle ear, foot -> mouth doctor, nurse -> surgeon -semantic -associative -both

Results p Degree of priming n n p Semantic – 57 Associated – 42 Both - 65 Idioms – 44 No difference between the groups

Results Prime was: Best elicited idioms

Conclusions so Far… Idiom Level of Representation exist p Syntactic schema p

Does Collocation Level of Representation exist ? (Harris 1997) p Is priming good enough? n p Processing letters n p Cradle -> baby Word Superior Effect -Carr(1986) Random word pairs vs. collocation

Collocation X G TAX BILL

Collocation Single Collocation Nonword neighbor collocation letters Tax bill Tax bell Tax deep XG Night club Night clue Night wall NE world Free world Tree world Open world R U

Collocation p p p Letter detection was better in the collocation than in other stimuli. “sophisticated guessing” Collocation “friends” n p Bog down/bow down Superiority of detection of words in collocation remains

Collocation p p Are collocations activated in response to partial input? Trick items n n p Tag bill Eight club Accuracy impaired n n 65% trick condition 90% collocation condition

Conclusions so Far… Idiom Level of Representation exist p Syntactic schema p Collocation level representation exist p n n n Polysemous words interpretation Child language acquisition Computation

Constraints on Statistical Learning (Safran 2002) Predictive dependencies affect learnability of sequential structures p Domains p

Constraints on Statistical Learning (Safran 2002) Combinatorial explosion p Source of the constraints p Linear input to Nonlinear structure p n n (the professor (graded (the exam)) Innate knowledge Dependency relations between categories

Word categories

Language P vs. Language N

Rules for both languages

Results Sequential/ Stimuli simultaneously Linguistic? result Sequential Auditory Yes/no P>n>½ Sequential Visually Yes/no P=n>½ simultaneously Visually Yes/no P>n>½

Results Adults p > n p Children p>n p Other domains p

Conclusions Language evolved to fit the human learner p Similarities among human languages may reflect constrains to fit human learner p Bridge between nature and nurture p

Summary p Entrenchment n n n p Statistics n n p Representation of idioms Representation of collocations Representation of words Between word categories Phonemes Letters Performance n n Acquiring predictable language using idioms or collocations