Semantic Analyses and Representations - Man-Machine Dialogue: Design and Challenges - page 89

“to want” or “to be able to” as well as the presence of adverbs such as
“maybe”, “possibly” or “apparently” are triggers to automatically identify a
modality and use it to enrich the semantic representation.
Ironyis the typical example for which the utterance's semantic content does
notmatchthecontentexpressed.Itisaphenomenaclosetotheunderstatement,
and can be hard to detect, even for a human hearer. As always in MMD, it is
desirable to be able to detect when the user expresses this type of behavior, so
as to bring the dialogue back toward a more neutral path, but we can consider
that it is not necessarily a priority, especially in closed domain.
Taking salience into account, profiling according to the cognitive linguistic
approach [LAN 87, p. 39], allows us to rank the different elements not only in
the utterance, for example the number of tickets, the destination and the type of
seat in “I want a first-class ticket for Paris” depending on the accents, but also
between sentences presented slightly differently, which all lead to the same
semantic content: “it is a ticket that I want, a first-class one for Paris”, “it is for
Paris that I want a first-class ticket”, etc. In these utterances, the cleft pronoun
“it is” has the consequence of putting some words in salience, and this salience
has to be a part of the description of the utterance's meaning. The challenge
for the MMD is to implement a model describing all the salience factors, and
for each utterance to calculate the saliences linked to each element, so as to
rank them and take this ranking into account when determining the system's
reaction.
In the same kind of idea, the approaches based on focus, i.e. on salience
with a point of view that is mainly prosodic, a point of view that translates the
importance of this factor in the English language, exploring the factors which
will help identify the focus in an utterance as well as the mechanisms through
which the focus effect spreads from a word to a wider linguistic segment
(issue of association with the focus). Some approaches, such as the approach
of Beaver and Clark [BEA 08], consider that a focalized element highlights a
set of alternatives which are calculated in a compositional manner as
focalized or alternative meaning, i.e. as a set of propositional contents. The
challenge for the MMD is to automatically identify this alternative content
which adds itself to the semantic content coming from more classical
analyses.
To continue on this path consisting of determining additional propositional
content, an essential aspect of the interface between semantics and pragmatics