Semantic Analyses and Representations - Man-Machine Dialogue: Design and Challenges - page 78

Identifying the implicitness is a challenge for NLP and MMD. Beyond the
cases of ellipsis that can be modeled by exploiting the dialogue history,
implicitness relies on the hypotheses that are hard to formulate. In the case of
a specific task, the inferences are easier to observe than in open domain
because they preferentially orient themselves toward resolving the task. In
open domain, approaches such as relevance theory [SPE 95] provide us with
an interesting framework that is hard to implement.
With or without implicitness, a natural language utterance is often
characterized by the phenomena of ambiguity, that is the possibility of
obtaining various alternative interpretations. This multiplicity can come from
a term or reference that gives the system a choice between several possibilities
even after taking into account the context, such as “I want to leave at six”,
which could be 6 am or 6 pm. It can also come from the sentence structure
which due to the ambiguity of a preposition does not allow the system to link
the right components to a noun or a verb. In “I will take a ticket from Meudon
to Paris”, does the user mean he/she needs a Meudon-Paris single ticket, or
that he/she will take the ticket from Meudon and bring it to Paris? The
challenge for NLP and MMD is to identify the terms and structures which
might generate ambiguities, based, for example, on a grammar-type inventory
[FUC 00], that is a set of rules on specific words and sentence word order.
As we will see in section 5.2, this kind of set of rules can also be used for
sentence syntactic analysis, at the same time or separately from semantic
analysis. The elements mentioned above allow the system to have an idea of
the extent and complexity of processes. One last aspect on which we would
like to emphasize here is the information structure. This notion describes the
fact that some actants are highlighted compared to others according to the
mechanisms that go from syntactic structure, including word order and
specific construction use, to prosody. We can, for example, distinguish
between previously known information and information given by an
utterance, between the topic (that of which the utterance is speaking) and the
commentary (what is said about it), or even between the focalized (the focus,
which is given a prosodic prominence) and that which is not. These
dichotomies lead to ranking the utterance's components, and this ranking
becomes a point of view in relation to the semantic content. Thus we can draw
a line between “I want a first class ticket for Paris”, “I want a first class ticket
for Paris” and “I want a first class ticket for Paris” (the italics indicate the
accent), after, for example, the same utterance without an accent was given to