Semantic Analyses and Representations - Man-Machine Dialogue: Design and Challenges - page 75

between “par ce chemin” and “part ce chemin”). There are many various
examples, and this is what makes this process difficult. Here, our approach is
not to suggest a meaning of the determination model (there are dozens
already), but to give an idea of the linguistic phenomena that appear in MMD
and of the understanding processes to consider in order to deal with them.
The first section of this chapter thus aims to describe a few phenomena
coming into play in oral and multimodal dialogue (section 5.1), and the
remaining two sections give an idea of the computational processes involved
in determining meaning, with a set of processes that analyze the utterance's
characteristics (section 5.2) and a set of processes that enrich them with
supplementary elements to reach a representation which can be used for
dialogue management (section 5.3).
5.1. Language in dialogue and in man-machine dialogue
5.1.1. The main characteristics of natural language
An utterance is made up of words, each of which has one or more
meanings. In a given situation, some meanings are impossible and we try to
identify the meaning taken by the word in context. Some words come
together to become a constituent characterized by its function in the sentence
(subject and direct object) and by its thematic role (agent and patient). Some
of the constituents refer to specific objects of the context. Even if the meaning
and reference are linked, this specific point is the focus of Chapter 6. Like the
other full words, the verb of the sentence has a meaning, and it is first and
foremost this meaning which will allow us to link all the constituents of the
sentence together and to be able to determine the meaning of the utterance. In
particular, with our example of trains, but in general, the meaning can involve
the notions of time and space, and in that case the understanding of the
utterance involves the knowledge of temporality (notion of date, finding
timetables, length of a trip and placing an event with regard to another) and
the semantics of space (notion of place and notion of journey). The
characteristics of natural language are thus multiple, and we will focus on the
following aspects, each time showing a few examples of how MMD systems
should take them into account: polysemy, metonymy, metaphor, verbal
semantics, implicitness, ambiguity and information structure.