Semantic Analyses and Representations - Man-Machine Dialogue: Design and Challenges - page 81

which would not be the case in a polylogue. We have to just plan a process
so that either they are ignored in the syntactic structure of the utterance or a
positive or negative qualifier is taken from them and used to react accordingly.
In general, the way in which the user addresses the machine can take on
various forms, due to the lack of social status that the machine has: “give me
a train ticket for Paris” (imperative, easier to use with a machine than a human
hearer), “reserving a single ticket for Paris” (gerundive form which is specific
to MMD, in French it would be the infinitive form, “réserver un aller pour
Paris”), “ticket for Paris please” (elliptic form), “we are going to reserve a
single ticket for Paris”, “and then I get a ticket for Paris”, etc. Moreover, the
utterances are not always complete sentences (see Chapter 10 of [COH 04])
which cause various issues in MMD. On the one hand, there are issues for the
detection of the end of the utterance, since only the prosodic criterion may
help the system decide. On the other hand, there are issues for the syntactic
analysis: if it produces anything, the result is an incomplete structure, which
might be false when the analyzer is not adapted. Adaptation is also necessary,
as [VIL 05] shows us, for example. Finally, an utterance can be so incomplete
thatitcanonlybeinterpretedinrelationtothepreviousutterance.Thesearethe
answers to questions, these are also the non-sentencial (NSU), such as “Ok”,
“thank you” and “sorry”, which are notably studied by [GIN 12].
5.1.4. Language and conversational gestures
In a human dialogue carried out face to face, gestures play an important
role both in the organization of the interaction, in the transmission of emotions
or modalities and in the designation of various elements of the shared visual
scene [KEN 04].
Language and gesture supplement each other in spontaneous
communication [LAN 06], which leads us to favor, when technical means
allow us to, the multimodal MMD over the oral MMD. The gestures thus
allow the user to speak, keep his/her speech turn and let the machine know
that he/she understands what the system is telling him/her. These
synchronizing gestures can be interpreted without the need for an utterance to
be pronounced at the same time. In MMD, a recording device like a camera