Semantic Analyses and Representations - Man-Machine Dialogue: Design and Challenges - page 86

5.2.3. Semantic analyses
Semantically analyzing an utterance first means determining the meaning
of each full word used, which a semantic lexicon allows us to do, and also
means building the sentence's meaning. This can be done from the structure
identified by the syntactic analysis: following the relationships between
components allows us to build semantic relationships. The analysis thus uses
a semantic syntactic lexicon and is based, for example, on the semantic
features and grammatical functions to determine the thematic roles of the
different sentence components. This is one of the key roles of a semantic
analysis, with notably the works of C. Fillmore highlighting this aspect in
case grammar, following the observation that the syntactic structure is not
enough to explain the links between a verb and its actants (in general, see
[ENJ 05]). Another role is to choose among the possible meanings of a word
by confronting the semantic features of the elements present, i.e. solving the
polysemy cases and other aspects of language mentioned in section 5.1.1.
For a closed domain MMD the semantic analysis can almost stop there,
because the determination of verbal semantics and the identification of
the actants allow us to guess the semantic content of the utterance, at least in
the simpler cases such as “I would like to go to Paris” or “I would like to
reserve a single ticket”. For an open domain MMD or for systems which try
to achieve a fine understanding of linguistic phenomena, another role of the
semantic analysis is to express the meaning in a logical manner so that
inference calculations can be run on this logical form and on those that have
already been recorded as the dialogue progressed, and thus model a part of
the implicitness. One method consists of following the mathematical logic
principles and to carry out a calculation of predicates, i.e. the formalization of
the utterance's content with variables, relations, predicates, logical connectors
(conjunctions, disjunctions and implications) and quantifiers (universal, such
as “all the trains have first-class coaches”, or existential, such as “a train has
broken down near the Palaiseau station”). The challenge is then to formalize
the natural language with the constraints of logics, which creates an
incredible amount of questions and issues. This is how modal, temporal and
hybrid logics can be applied to MMD.
Beyond the boundaries of the utterance, it is on this path that theories such
as the File Change Semantics (FCS) and the Discourse Representation Theory
(DRT) will suggest formal frameworks for utterance interpretation [KAD 01].