Information Structure in Discourse
View/ Open
Date
06/2006Author
Traat, Maarika
Metadata
Abstract
The present dissertation proposes integrating Discourse Representation Theory (DRT),
information structure (IS) and Combinatory Categorial Grammar (CCG) into a single framework.
It achieves this by making two new contributions to computational treatment of information
structure. First, it presents an uncomplicated approach to incorporating information
structure in DRT. Second, it shows how the new DRT representation can be integrated into
a unification-based grammar framework in a straightforward manner. We foresee the main
application of the new formalism to be in spoken language systems: the approach presented
here has the potential to considerably facilitate spoken language systems benefiting from
insights derived from information structure.
The DRT representation with information structure which is proposed in this dissertation
is simpler than the previous attempts to include information structure in DRT. We
believe that the simplicity of the Information-Structure-marked Discourse Representation
Structure (IS-DRS) is precisely what makes it attractive and easy to use for practical tasks
like determining the intonation in spoken language applications. The IS component in ISDRS
covers a range of aspects of information structural semantics. A further advantage of
IS-DRS is that in its case a single semantic representation is suitable for both the generation
of context-appropriate prosody and automatic reasoning.
A semantic representation on its own is useful for describing and analysing a language.
However, it is of even greater utility if it is accompanied by a mechanism that allows one to
directly infer the semantic representation from a natural language expression. We incorporated
the IS-DRS into the Categorial Grammar (CG) framework, developing a unification based
realisation of Combinatory Categorial Grammar, which we call Unification-based
Combinatory Categorial Grammar (UCCG). UCCG inherits elements from Combinatory
Categorial Grammar and Unification Categorial Grammar. The UCCG framework is developed
gradually throughout the dissertation. The information structural component is
included as the final step. The IS-DRSs for linguistic expressions are built up compositionally
from the IS-DRSs of their sub-expressions. Feature unification is the driving force in
this process. The formalism is illustrated by numerous examples which are characterised
by different levels of syntactic complexity and diverse information structure.
We believe that the main assets of both the IS-DRSs as well as the Unification-based
Combinatory Categorial Grammar framework are their simplicity, transparency, and inherent
suitability for computational implementation. This makes them an appealing choice for
use in practical applications like spoken language systems.