Modelling the Acquisition of Colour Words: A Bayesian Approach to Fuzzy Sets

How Bayesian inference might be used as the basis of a system for learning and representing the meanings of colour words in natural languages was investigated. The paper is primarily concerned with cognitive modelling, but has potential applications in natural language processing. A Bayesian cognitive model was constructed to test the hypothesis that people learn language, and in particular the meanings of colour words, using Bayesian inference. The model learned the range of colours which could be named by a particular colour word from examples of colours which could be denoted by that word, and was able to do so accurately even in the presence of large quantities of random noise in the input data. The resulting meaning representations display many of the properties of colour words in natural languages, in particular prototype properties. This suggests that the Bayesian approach closely models the mechanism which people use to learn colour words, and that this approach may be useful in constructing language technology systems which require accurate representations of the words of natural languages. 1