Page 154 - DCAP506_ARTIFICIAL_INTELLIGENCE
P. 154
Artificial Intelligence
Notes 11.2.4 Discourse and Pragmatic Processing
In the recent decades, the need for an ability to treat discourse dependent factors such as anaphora
resolution has increased in the field of NLP. This calls for a greater understanding of intersentential
dependencies, in terms of how the discourse structure is built both in written narrative discourse
and spoken dialogue. In the Survey of the State of the Art in Human Language Technology
Barbara Grosz (1997) formulates two core questions in research on discourse: (1) What information
is contained in extended sequences of utterances that goes beyond the meaning of the individual
utterances? and (2) How does the context in which an utterance is used affect the meaning of the
individual utterances, or parts of them? The first question thus concerns the issue of inferences,
associations and implicit information, which in different ways can be derived from the utterances,
and from combinations of utterances. Attempts to formalize these relations have been made in
the extensive research on discourse relations (e.g. Hobbs, 1985, Mann & Thompson, 1988, Asher,
1993). Also the issue of knowledge representation is relevant here. The second question concerns
the fuzzier field of domain dependence, i.e. what information is relevant for the specific discourse
setting, what might be presupposed and what kind of conventions are to be applied in the
specific discourse. Also the question of cognitive processing of discourse as well as the situation
setting is relevant here.
When it comes to spoken discourses, there is a need to connect international features with
discourse factors is present in e.g. automatic summarization of spoken language, or in spoken
dialogue systems. The international features are affecting the discourse in dimensions as e.g.
discourse segmenting (prosodic phrasing) and the assignment of discourse relations, such as e.g.
contrast. Also salience, of focality, is affected by the prosody.
This will give an overview of factors relevant in discourse processing. These phenomena include
the concept of salience/focus, anaphora resolution, cohesion, coherence discourse relations and
discourse segments.
What is Discourse?
The term discourse includes both spoken and written forms, as well as both monologue and
dialogue, i.e. “discourse” is taken to be the most super ordinate term. The difference between a
discourse and an arbitrary collection of utterances lies in the phenomenon of coherence.
A discourse is any stretch of coherent utterances.
This means that the utterances together constitute an intentionally meaningful message. The
establishing of coherence can be made in many ways, but the result is always some kind of
relations between items on different levels in the discourse.
Notes However, to assume relations in discourse implies also to assume units between
which these relations holds, and this leads us in to the field of discourse segmenting and
the (hierarchical) discourse structure.
Discourse Specific Issues
Some discourse specific issues are discussed as below:
1. Focus: The first phenomenon to discuss in trying to get hold of coherence in discourse is
focus. Focus is often mentioned feature in research on discourse. The notion of focus is
used for a number of aspects of salience or importance, and the most general description
148 LOVELY PROFESSIONAL UNIVERSITY