Modelling incremental self-repair processing in dialogue

Self-repairs, where speakers repeat themselves, reformul ate or restart what they are saying, are pervasive in human dialogue. These phenomena provide a wind o into real-time human language processing. For explanatory adequacy, a model of dial gue must include mechanisms that account for them. Artificial dialogue agents also need this c apability for more natural interaction with human users. This thesis investigates the structu e of self-repair and its function in the incremental construction of meaning in interaction. A corpus study shows how the range of self-repairs seen in dia logue cannot be accounted for by looking at surface form alone. More particularly it analy ses a string-alignment approach and shows how it is insufficient, provides requirements for a sui table model of incremental context and an ontology of self-repair function. An information-theoretic model is developed which address es these issues along with a system that automatically detects self-repairs and edit terms on transcripts incrementally with minimal latency, achieving state-of-the-art results. Additi onally it is shown to have practical use in the psychiatric domain. The thesis goes on to present a dialogue model to interpret an d ge erate repaired utterances incrementally. When processing repaired rather than fluent tterances, it achieves the same degree of incremental interpretation and incremental repr esentation. Practical implementation methods are presented for an existing dialogue system. Finally, a more pragmatically oriented approach is present ed to model self-repairs in a psycholinguistically plausible way. This is achieved through extending the dialogue model to include a probabilistic semantic framework to perform incremental inference in a reference resolution domain. The thesis concludes that at least as fine-grained a model of c ontext as word-by-word is required for realistic models of self-repair, and context mus t include linguistic action sequences and information update effects. The way dialogue participa nts process self-repairs to make inferences in real time, rather than filter out their disfluency eff ects, has been modelled formally and in practical systems. Submitted for the degree of Doctor of Philosophy Queen Mary University of London