Man–Machine Dialogue Fields - Man-Machine Dialogue: Design and Challenges - page 41

French) in the oral language. For an MMD, we can thus draw a predefined list
of the uses of “of which” (“dont”), which will simplify the preparation of
linguistic resources, at least on that point. In a general manner,
[BLA 10, p. 11] shows that if the frequencies (and uses) differ, it is indeed a
single syntactic framework, which allows us to account for the oral or written
language. The main difference is in fact in morphology: written language
involves spelling corrections, which are rich in well-known rules and
exceptions, whereas oral language involves a specific morphology, focuses,
for example, on the pronunciation of connections between words. As for
syntax, this observation has important consequences on the design of MMD
systems. In certain oral systems, the speech recognition module produces a
written transcription of the user's utterances, and the transcription is then
processed by a morphosyntactic analyzer. The latter relies on the morphology
of the written language, and that of the oral language is simply ignored. In
most cases, it is taken into account in the transcription itself, since
pronunciation helped determine this transcription. But in other cases, the
written transcription introduces ambiguity, such as with the French word
“plus” which can mean “more” or “no more” but which is never ambiguous
when spoken, since the meanings have different pronunciations. Ideally, either
the morphosyntactic analysis is carried out directly on the oral language, or
annotations are added, describing the pronunciation to the transcription, or
back-and-forth motions between the modules are implemented, with, in this
case, an analysis module detecting ambiguity and requesting the recognition
module to solve it without starting on a fresh transcription and analysis.
In a dialogue, the language is dependent on speech turns and interactive
aspects. A written sentence is not interrupted, or it is deliberately so, whereas
interrupting an interlocutor happens during a dialogue, whether it is oral or
written. Moreover, the dialogue is characterized by opening and closing
utterances, that is key-moments in which the messages follow predefined
codes. These are greetings, farewells, polite forms of address, or excuses.
Finally, a dialogue is studied as a relevant succession of utterances, which
means there are valid reasons for an utterance, “here are the possible
journeys” to follow the previous utterance, “I would like to go to Paris”. We
can refer to these reasons as coherence, thematic continuity, or reactive action
following an initiating action. In any case, a dialogue is analyzed as a set of
utterances, which some refer to as a discourse. Cole [COL 98, p. 198]
reminds us that research on MMD has historically followed two paths drawn
by research on human dialogue: first, the discourse analysis path, from speech
acts theory [SEA 69], which sees dialogue as a rational cooperation and thus