CPSY 4343 Lecture Notes - Lecture 10: Artificial Neural Network, Fault Tolerance, Problem Of Induction
Document Summary
How might words for objects be learned: ostensive definitions & the induction problem, 2 possible answers: Whole object assumption: assume the new word refers to a whole object and not a part or attribute of an object: but if it was innate, it would be universal (chinese children learn words for actions first) The taxonomic (or shape?) constraint: taxonomic groups include objects of the same type (share features that may be perceptual, functional, or more abstract: thematic groups of objects share external features. The mutual exclusivity assumption: if you hear a new word, it refers to the o(cid:271)je(cid:272)t (cid:449)hi(cid:272)h you do(cid:374)"t ha(cid:448)e a (cid:374)a(cid:373)e for. Children pick up the statistical regularities of the words they have learned and use those features to infer the meanings of new words in their language (lecture) Word learning pretty slow initially but then rapidly increases: productive use of morphemes. Acquisition of the past tense u-shaped curve. Review (serial information-processing: knowledge=sequence of states, their appeal.