A Few Dialogue Strategies - Man-Machine Dialogue: Design and Challenges - page 128

prevent the speaker from going on with his/her intervention. This can thus
lead to voices being superimposed. In MMD, some studies have integrated the
generation of such utterances for the system's behavior [WAR 03, EDL 05],
but there are many technical issues: speech recognition works at the same
time as the system is making noise, which can lead to lower performances; a
control utterance generation can happen at the moment when the user was
about to stop talking, which can create a small moment of uncertainty and, in
general, generating a control utterance is not always done at the most relevant
moment, due, for example, to the slight temporal delay with regard to the best
places, the TRP (see section 2.2.1). However, it is true that giving the system
the ability to use the backchannel increases the realistic aspect of the
dialogue, providing it with the ability to occupy the field when the user is not
saying anything, that is to generate a reviving message when the user is not
answering, to generate a dialogue maintenance message when the user does
not know what to say and even to generate a waiting message when a process
may take more than a few seconds.
8.1.3. Interpretation and inferences
To react in a completely relevant manner to an utterance, understanding is
obviously necessary as well as understanding its implicitations and
explicitations (see section 5.3). Determining the semantic content and speech
acts is the first step, but it is also necessary to make the right inferences,
which are based on the utterance and its context and allow the human hearer
to understand the allusions and other implicit contents immediately. In this
case, Grice [GRI 75] also suggested a terminology and general principles
which allow us to better understand the dialogueic phenomena. Using the
term of implicature for pragmatic inferences, he draws a distinction between
conventional implicatures which are triggered by a conventional use of
language, conversational implicatures, which are triggered by the utterance's
link in the context of its enunciation, and generalized conversational
implicatures, which are triggered in a contextual manner without any help
from the utterance's linguistic elements. These types of implicature allow us
to explain understanding and cooperation in a dialogue. It is when the speaker
seems to violate one of the maxims that we have to look to implicatures.
Calculating an implicature is to determine what the speaker implicitly
supposed, to keep the cooperation principle intact [DEN 08, p. 18].