An Assessment of the Evolution of Research and Systems - Man-Machine Dialogue: Design and Challenges - page 14

- the increase in system assessment methodology (see Chapter 10);
- the multiplication of communication modalities with the machine, and
thus of the models and techniques of multimodal dialogue: force-feedback
gesture or haptic gesture, gestures and postures caught on camera, taking into
account the eye direction, lip reading, etc. [LÓP 05];
- the implementation of links with other scientific domains, such as
robotics (see [GAR 11, Chapter 10]), and other fields of NLP, for example
machine translation within MMD systems, being able to go from one language
to another [LÓP 05];
- the increase in toolkits for a quick prototyping of MMD systems, for
example the well-known VoiceXML, a standardized language for relatively
simple, from a linguistic point of view, voice applications;
- the integration of MMD in wide intercommunication platforms, whether
we are referring to ambient intelligence or other aspects, for example linked to
software architectures [ISS 05];
- the rise of the embodied conversational agents (ECA) field that takes into
account the emotions;
- the rise in the QAS field.
About this last item, for example, the point of integrating dialogue abilities
to a QAS is to allow it to carry out exchanges to specify bit by bit the query
[VIL 05, p. 48]. From a (single) QAS that is content with finding the result to
a query, like the database managers do or like IBM Watson that is still limited
by the rules of a game show, we move on to a system of questions and
answers (plural), in which the dialogue allows for clarifications, precision,
and especially follow-up questions on the same subject: “does this journey go
through Meudon?”, “is this the shortest journey to get to Paris?”, “when does
it leave?”.
An example for such a system using MMD and QAS is the Ritel project
[VAN 07, ROS 08]. The system's architecture highlights question
management, with modules devoted to topic detection, user return
management, dialogue history management, question routing, implicit
confirmation management and additional query management. The project's
goals clearly highlight the QAS performances as much as the MMD
performances, and the project is therefore a significant step for open-domain