My current work is focussed on the structure of natural spoken dialogue and in the area of managing, annotating and exploiting large corpora of real language. In the past I have also worked extensively in a number of other areas of natural language processing including parsing, metatheory, evaluation, statistics, speech recognition and lexical access. Work on producing natural spoken dialogue corpora has led to an interest in replacing the existing top-down theory of dialogue micro-structure (who talks when) with a theory grounded in the detailed timing information only now becoming available. Issues in annotation of such dialogue corpora, along with collection and production of a number of large and in some cases multilingual text corpora has led to my interest in architectures and tools for supporting the creation, management and exploitation of large SGML-encoded document collections.