Semantic Analyses and Representations - Man-Machine Dialogue: Design and Challenges - page 85

for each word, the set of semantic features which describe the meaning of the
word and thus allow the creation of a lexicon, which is relatively close to the
way a dictionary works. In MMD, this is a completely realistic approach from
the moment when we place ourselves in a closed domain, that is with a
limited list of words. The lexical database approach like in WordNet allows us
to obtain an organized list of words, without necessarily equivalent definitions
(a thesaurus is drawn up rather than a dictionary), but with many oriented
links between the words: hyponomy (“is a kind of”, e,g, from “TGV” to
“train”), metonymy (“is a part of”, e.g. from “coach” to “train”), troponymy
(“X is Y in a way”, one of the kinds of implications between two verbs),
antonymy (opposite), synonymy, etc. These relationships, especially since
they have a direction, allow us a greater analysis possibility than a purely
component approach would: we can thus model the fact that a “long” journey
is “not short”, but that a “not long” journey is not necessarily “short”. Finally,
the approach of derived lexicon, built from corpus or resources such as those
obtained through the previous approach by following the principles of a
specific lexical theory, allows us to obtain richer data structures. A famous
example that has often been used in MMD, within the framework of a limited
task, is that of conceptual graphs [SOW 84]. This is a kind of formalism that
allows us to go further in the representation of knowledge, with multiple
relations between concepts, and which leads to the notion of ontology as a
representative data model of a set of concepts in a domain, allowing us to
reason on the objects falling within the scope of this domain. An MMD
system focusing on the train ticket reservation needs not only language
resources, for example a semantic lexicon, but also resources on the world of
trains, transportation and reservation, with the matching ontology, whether it
is represented by a conceptual graph or a set of feature structures.
Moreover, some syntactic formalisms use lexical data such as that
constituting a semantic lexicon, which leads to implementing lexicalized
grammars [ABE 07], and thus mix lexicon, syntax and lexical semantics. One
of the MMD challenges is to create and use these semantic syntactic lexicons,
especially in an open domain MMD. We have mentioned earlier the WordNet
initiative, and can mention here, as one of the paths that is currently being
investigated by linguistics and NLP in general, the FrameNet initiative that
consists of cataloging sentence constructions, by linking them to each other as
soon as they describe a similar meaning. It is a kind of semantic syntactic
lexicon leading toward the automatic identification of thematic roles, but not
exclusively, and this is the type of initiative that MMD can use.