My research concerns three quite different areas (colour language, syntax, and the cultural evolution of language) but makes connections between all of them. I have used Minimum Description Length (a form of Bayesian inference) to account for the acquisition of syntactic structures in the absence of negative evidence and strong innate biases. I've also modelled the evolution of both colour language and syntax using forms of iterated Bayesian learning. This has led to a new explanation for the typological patterns seen in colour naming, and a better understanding of the phenomenon known as the 'bottleneck effect' in which the transmission of language between speakers via a limited number of example utterances leads to the emergence of regularity.

My colours research is summarised in this poster: A Computational Evolutionary Approach to Colour Term Typology.

Here are notes for a seminar I gave about syntax and minimum description length at the University of Sydney linguistics department in September 1999, and an abstract and some notes which I wrote for the 1999 Australian Machine Learning Workshop.

While I was in Sheffield I worked on the PrestoSpace project, concerning the preservation of audio visual archives. I researched how to automatically divide up news into stories, how to classify it, and how to produce meta-data that could help in searching for specific topics. The same system can also be used to automatically create interactive television news broadcasts by augmenting regular broadcasts with material derived from web pages.

My program for learning syntactic and phonological structure is available here. The software for my models of colour terms can be founded in the appendices of my Ph.D. thesis (see below).