Can Prosody Aid the Automatic Classification of Dialog Acts in Conversational Speech?

Identifying whether an utterance is a statement, question, greeting, and so forth is integral to understanding and producing natural dialog. Human listeners easily discriminate such dialog acts (DAs) in everyday conversation, responding in systematic ways to achieve the mutual goals of the participants. Little is known, however, about how to build a fully automatic system that can successfully identify DAs occurring in natural conversation. At first blush, such a goal may appear misguided, because most current computer dialog systems are designed for human-computer interactions in specific domains. Studying unconstrained human-human dialogs would seem to make the problem more difficult than necessary, since task-oriented dialog is by definition more constrained and hence easier to process. Nevertheless, for many other applications, as well as for basic research in dialog, developing DA classifiers for conversational speech is clearly an important goal. For example, optimal automatic summarization and segmentation of natural conversations for archival and retrieval purposes requires not only knowing the string of words spoken, but also who asked questions, who answered them, whether answers were agreements or disagreements, and so forth.