Semantic Analyses and Representations - Man-Machine Dialogue: Design and Challenges - page 82

and shape recognition algorithms are required. Gestures are also what
transmit a major part of the emotional component of the message: expressive
expressions, such as facial expressions, and paraverbal gestures, such as the
movements that give speech rhythm and emphasize certain words. There are
also the gestures which allow the user to provide information on the referents
in the utterance, whether it is by pointing (pointing or deictic gesture) toward
the referent in question, when it is visible in the shared visual scene, or an
illustrating gesture that indicates a size, shape and action. Finally, there are
also gestures which can carry a dialogue act such as a questioning gesture
(wide eyes and raised eyebrows) or a quotation gesture close to the reported
speech, like pointing to the hearer to mean “as you said earlier” [CLA 96,
p. 258].
In the same way, and even if a gesture is not necessarily present, the
dialogue language can have “deictic” terms, that is the terms or expressions
that refer to the communication situation, with the three categories that are
deictic persons, who refer to the hearers thanks to markers covering personal
pronouns, possessive adjectives and some nouns; spatial deictics such as
“here” or “there” and temporal deictics such as “now”, “tomorrow”, “later” or
“in three days”. For deixis cases, an MMD system has to find a link between
the deictics pronounced and the situational context. This link relies on the
hearers, on the objects in the scene and on the spatial and temporal markers
which will serve as parameters in the semantic and pragmatic representation
of the utterance.
5.2. Computational processes: from the signal to the meaning
The automatic understanding of an utterance starts with the voice
recognition and prosodic analyses, and ends with the enrichment of a
semantic representation through inferences carried out essentially from the
utterance's content. This representation will then be analyzed with a
pragmatic approach that allows us to use it in dialogue management. The
processes involved in building this semantic representation can be deduced
from the content of the previous section: lexical analysis, detection of
polysemy, metonymy and metaphor. While we will not draw that list again,
this section aims to present the general processes by mentioning
methodological and technical solutions and underlining the most crucial
challenges.