Semantic Analyses and Representations - Man-Machine Dialogue: Design and Challenges - page 87

DRT, which describes how discourse representation structures are built
[KAM 93], is especially the focus of many extensions or adaptations for
computational implementations, for example Discourse Representation
Language (DRL), see [KAD 01]. For MMD, an important extension is
Segmented DRT (SDRT) [ASH 03], which reviews all the aspects of the
dialogue, and the implementations that have already been carried out,
especially in some QAS, or in the Verbmobil project, see [COL 98]. Other
extensions take into account the underspecification phenomena found in oral
language, such as Compositional DRT (CDRT) [MUS 96] or Underspecified
DRT (UDRT). Moreover, DRT even warrants an extension for multimodal
dialogue, with the integration of linguistic aspects and visual and gesture
aspects in the same formal framework. This is, Multimodal DRT (MDRT)
[PIN 00].
A completely different path is the use of probabilities so as to obtain a
probabilistic semantic grammar. This is what has been done in the Tina
Project [SEN 95]. Other approaches use Hidden Markov Models by adding a
ranking structure to combine the advantages of a semantic grammar with
those of statistics. Jurafsky and Martin [JUR 09, p. 859] thus present Hidden
Understanding Models (HUM).
The semantic analyses can thus happen in various ways and various
approaches can be combined. Two remarks to close this section: first, as we
underlined with the importance of local or partial syntactic analysis compared
to a global and comprehensive syntactic analysis, the semantic analysis can
also remain incomplete. As Enjalbert [ENJ 05, p. 303] writes, “we absolutely
need to let go of the idea of 'complete' understanding, which is incompatible
with the extreme variety of issues to be dealt with. The understanding goals
must be brought back to a specific task, which will direct, limit the analysis
and provide additional information. Moreover, is it not thus that the human
reader works, recording some information in a text, depending on his
interests, on the goals for his reading - even if he has to come back later for a
more in-depth reading?” Second, determining the meaning of an utterance is
not only a semantic issue: it is also a pragmatic issue, obviously, with the
determination of referents (the variables in the logical forms) and the
identification of the speech acts. The latter allows us to understand elliptic
utterances and non-sentence utterances. Pragmatic analysis is thus useful for
semantic analysis, and semantic analysis is one of the settings to identify the
pragmatic characteristics of an utterance. Semantic analysis and pragmatic
analysis thus go hand in hand [COL 98, p. 189].