ERCIM News No.26 - July 1996 - INRIA
Man-Machine Dialogues and Language
by Jean-Marie Pierrel and Laurent Romary
The Dialogue project at CRIN-CNRS and INRIA Lorraine aims at designing
man-machine dialogue systems based on a real linguistic communication. To
do so, we are conducting studies on the kind of relations which may link
language with the overall environment in which the dialogue actually occurs.
The approach adopted by the Dialogue project is based upon pluridisciplinary
research in the domain of pragmatics, which attempts to combine linguistic
descriptions with the constraints of a real implementation. As we are still
far from the realisation of systems which would cover the whole scope of
natural language, one of the difficulties that has to be faced is to evaluate
how much a user will accept, within a given context, to interact with a
system which can only understand a subset of a given language.
We are thus addressing two complemen-tary aspects:
- reference interpretation, which is of the utmost importance for task-oriented
dialogues
- the definition of software architectures to implement such dialogues.
Reference and Dialogue
Our aim is to provide a user with the linguistic means needed to communicate
instructions to a system. In this context, two essential problems have to
be solved:
- reference to actions in order to know what the user intends to do
- reference to objects to anchor the operation to be performed to the
task at hand.
These two questions have to be combined with the very nature of the kind
of applications we consider. In general, the users have to deal with information
which is presented to them through a graphical interface; this has encouraged
us to carry out specific research on:
- the role of language to refer to space either from the point of view
of objects (eg the leftmost window ) or that of actions (eg move the green
window to the right)
- the consequences of allowing multimodal utterances combining speech
and gesture, since this might facilitate some referring acts, especially
when the graphical space is crowded with objects of the same kind. Here
again, such references can be used to point to objects (eg this window )
or to define some actions (move the window there).
In this framework, we have proposed an original model for reference interpretation
based on a contrastive perspective which can provide some information concerning
the underlying intention of the user behind different referring acts. In
addition, work has been conducted on the analysis and interpretation of
gestural trajectories associated with linguistic expressions. We are also
carrying out some specific studies in lexical semantics to investigate whether
it is possible to design a generic framework for the description of the
underlying lexicon, despite the ever changing nature of the tasks we are
working with.
Software architecture for dialogue management
The results obtained from the basic studies described above have been used
to design new software architectures in order to integrate contextual constraints
that guide the recognition phase of a given utterances. These architectures
have been tested within several national (contracts with the French company
Thomson-DASM and the DRET) and international (European Esprit Projects Multiworks
and ROARS) projects.
Further information on the Dialogue project can be found at our web site:
http://www.inria.fr/Equipes/DIALOGUE-eng.html
Please contact:
Jean-Marie Pierrel - INRIA-Lorraine
Tel: +33 8359 2001
E-mail: jmp@loria.fr
or Laurent Romary - INRIA-Lorraine
Tel: +33 8359 2037
E-mail: romary@loria.fr
return to the contents page