Mike Dowman

School of Information Technologies,

University of Sydney,

NSW2006,

Australia

This research has investigated whether Bayesian inference can provide a plausible explanation of how people learn the meanings of words. Bayesian inference is a statistical procedure which allows inferences to be made on the basis of observations. Tenenbaum and Xu (2000) showed how a Bayesian model was able to account for the acquisition of the meanings of concrete nouns. They hypothesised that people learning these words would observe which objects they were used to refer to, and generalise from those examples to determine the nouns' correct denotations. Psycholinguistic experiments showed that the model learned similar denotations to those learned by people when presented with the same examples, and so were very supportive of the Bayesian approach.

This talk concerns a Bayesian model which learns the meanings of basic colour terms by generalising from examples in a similar way. Colour terms are prototype categories (Taylor, 1995), as each term has a single best example near to the centre of its denotation, with colours becoming members of the category to a lesser extent towards the category's fuzzy boundaries, where exactly which colours are members of the category becomes unclear. The model learns a word's meaning from examples of the specific colours which it was seen being used to identify. It then calculates how likely each part of the colour space is to come within the extension of the colour term. Prototype phenomena are emergent properties of the categories learned in this way, as it is possible to be most certain that colours near to a category's centre are within the term's extension, but membership is less sure for colours towards the category boundaries. This suggests that linguistic categories may have prototype properties simply because they are learned empirically from examples, and that prototypes may play no role in defining their extent.

Key words: Bayesian inference, semantic acquisition, prototype categories, basic colour terms, fuzzy sets.

Taylor, J. R. (1995). Linguistic Categorization: Prototypes in Linguistic Theory, Second Edition. Oxford: Oxford University Press.

Tenenbaum, J. B. & Xu, F. (2000). Word Learning as Bayesian Inference. In L. R. Gleitman & A. K. Joshi (Eds.) Proceedings of the Twenty-Second Annual Conference of the Cognitive Science Society. Mahwah, NJ: Lawrence Erlbaum Associates.