Vocabulary acquisition via cross-situational learning

A child learning language must determine the correct mappings between spoken words and their referents. It has been proposed that representations of word meanings can be built up over time if learners can track co-occurrences of unknown words and perceptual information surrounding use of these words, and can spontaneously retrieve this information and use it to infer word meanings. Indeed, a recently developed research paradigm has yielded data from adult and infant learners lending support to this hypothesis. The cross-situational word-learning paradigm tests the ability of learners to acquire correct word-to-world mapping for small novel vocabularies within the course of short training sessions. A training session is divided into many separate learning trials. During each trial, several objects are depicted, while simultaneously, the words that correspond to these objects are spoken in an arbitrary order. Although it is impossible to deduce the correct word-to-referent mappings within the course of a single trial, adult learners perform significantly better than chance in a post-training vocabulary test. I will report the results of three series of original experiments based on the cross-situational word-learning paradigm; these aim at characterizing the roles of known vocabulary in acquiring new words and of perceptual exposure to word or referent stimuli prior to attempting to learn mappings, and the relative influences of joint versus conditional probability structure in environmental structure. Through analysis of these data, along with computational model simulations, I provide a detailed description of how each of several variables influence inferences made by subjects about word meanings in the cross-situational learning paradigm, and discuss extensions to word learning in childhood