Information Technology Reference
In-Depth Information
Chapter 7
Dialogue Acts Recognition
The utterances “reserve a ticket for Paris”, “how long is this journey?” and
“I am unable to communicate with you” are not only different from a semantic
content point of view, but also by the speech act they carry out: the first is
an order (“telling to”), expressed with the imperative mood and requiring the
system to carry the order out; the second is a query (“asking”) expressed in the
interrogative form and requiring an answer from the system; and the third is an
assertion (“saying that”) expressed as a declaration and requiring the system
to take what is said into account and come to a conclusion, whatever it is.
The nature of these speech acts and their identification mechanisms, classified
here according to the relevance theory point of view [SPE 95], are an aspect
of pragmatics (called third-level), see section 1.2.2, which is essential when
managing dialogue: it is by understanding what speech act the user is carrying
out that the MMD system can determine its own reaction. At least, and we will
see this in this chapter and the following which is linked to it, this is one of
the parameters which allows us the system to decide how to progress with the
dialogue.
Just like for the action model described in the previous chapter, achieving
a speech act involves preconditions and postconditions, and its identification
requires a certain number of parameters. In automatic understanding, the
process responsible for the identification of a speech act requires input
arguments which can be preprocessed, and provides results at the output. At
input, we need here the semantic representation obtained in Chapter 5, with
Search WWH ::
Custom Search