Predictive Modelling For Topic Handling Of Natural Language Dialogue With Virtual Agents
Abstract
In this thesis, we aim to contribute to ongoing research in the field of human- computer dialogue and help move closer to the goal of having more realistic human-computer dialogue. We address the current challenge of topic handling in human-computer conversation by proposing a Topic Handler model that is designed in such a way that is flexible and compatible with third party dialogue systems. This model builds off of previously proposed dialogue grammars and systems and is based on speech act theory and conversation analysis. By employing feature engineering of existing dialogue act corpora and using this data in machine learning experimentation, we successfully demonstrate that not only does speech act and semantic annotation data improve the performance of classifiers for the task of identifying appropriate points of topic change, we also demonstrate how the proposed Topic Handler model can provide needed inputs to assist in topic handling when used in parallel with dialogue systems.