Abduction, Belief and Context in Dialogue: Studies in Computational Pragmatics Harry Bunt and William Black (editors) (Tilburg University and UMIST) Amsterdam: John Benjamins (Natural language processing series, edited by Ruslan Mitkov, volume 1), 2000, vi+471 pp; hardbound, ISBN 1-55619-794-2 and 90-272-4983-0, $90.00
Matthew D. Stone · Computational Linguistics · 2002
The problem of pragmatics arises as soon as we move beyond the linguistic analysis of an utterance and ask what the speaker meant by it.Now, speaker's meaning is a particular case of intention, and all intentions are complex mental attitudes that tie an agent's actions to its goals, its background beliefs, and its appraisal of the context in which it acts.Inferring the speaker's meaning, then, is a matter of recognizing how the speaker might have represented the utterance as linking up with the current discourse context and thereby furthering the goals of the conversation.This recognition of intention is reasoning to the best explanation, or abductive reasoning, as popularized in computational linguistics by Hobbs and colleagues (1993).So there they are: abduction, belief, and context-Harry Bunt and William Black's ABC of computational pragmatics-three concepts that rightly frame the diversity of current research in problems of pragmatic interpretation.In this volume (ABC for short), Bunt and Black collect 15 chapters that grow, by and large, out of the ESPRIT project PLUS (Pragmatics-Based Language Understanding System).This coherence lends the volume strengths not often found in collections of research papers; ABC fits together to give a broad picture of computational pragmatics as an interdisciplinary enterprise in which a multitude of different investigations can be brought to bear constructively on a common project.To start, ABC offers generally consistent terminology and perspective, laid out in a 150-page three-chapter overview of computational pragmatics in PLUS.As outlined by Bunt and Black in Chapter 1, dialogue system design in PLUS centered on the representations and inference required for recognizing the communicative intentions behind users' utterances.Jens Allwood's Communicative Activity Analysis (Chapter 2) provided the theoretical framework for this design, while Bunt's Dynamic Interpretation Theory (Chapter 3) bridged this dialogue theory and specialized models of meaning and context from computational linguistics and from computer science more generally.Concretely, this approach is distinguished by a wide view of context, including social and physical dimensions as well as linguistic ones, and a wide view of agency, mandating considerations of ethics and trust in cooperation from the start.It is also guided by some more practical working assumptions:• a focus on dialogues of information exchange in which domain reasoning (particularly reasoning about users' domain plans and domain-specific communication strategies) can be sharply circumscribed;