Smith, Simon Graham Jeremy (2003)
Ph.D. thesis, University of Birmingham.
A body of work exists on the classification, by prosodic analysis and other means, of utterance types and dialogue moves in spoken corpora. Much of this output, while often linguistically well motivated, tends to rely on hand-crafted rules. This thesis presents a data-driven approach to the classification of utterances, using a novel combination of existing algorithmic approaches. Previous work has generally classified utterances according to such categories as wh- question, yes/no question, acknowledgement, response and the like; in general, the audio data used has been specially commissioned and recorded for research purposes. The work presented here departs from this tradition, in that the recorded data consists of genuine interaction between the telephone operator and members of the public. Moreover, most of the calls recorded can be characterized as queries. The techniques presented in this thesis attempt to determine, automatically, the class of query, from a set of six possibilities including "statement of a problem" and "request for action". To achieve this, a scheme for automatically labelling utterance segments according to their prosodic features was devised, and this is presented. It is then shown how labelling patterns encountered in training data can be exploited to classify unseen utterances.
This unpublished thesis/dissertation is copyright of the author and/or third parties. The intellectual property rights of the author or third parties in respect of this work are as defined by The Copyright Designs and Patents Act 1988 or as modified by any successor legislation. Any use made of information contained in this thesis/dissertation must be in accordance with that legislation and must be properly acknowledged. Further distribution or reproduction in any format is prohibited without the permission of the copyright holder.
Repository Staff Only: item control page