An Assessment of the Evolution of Research and Systems - Man-Machine Dialogue: Design and Challenges - page 26

we are starting to see systems integration abilities belonging to robotics and
MMD abilities, preferentially in a multimodal manner [GOR 11]. A second
example is that of MMI when we try to give them speech, while keeping, on
the one hand, the possibilities of directly manipulating the HM and, on the
other hand, the MMI advantages in terms of ergonomics, plasticity: adapting
to the user, the terminal, the environment.
1.3.4. Facilitating the implementation
At the level of system development, the technical challenges are found in
the facilitation of development processes. A first step on this path is the
multiplication of toolkits devoted to MMD. VoiceXML is a basic example,
but there are many other platforms devoted, for example, to helping design
multimodal dialogue systems [LÓP 05]. A second step would be the
implementation of a library offering a rich and performing panel of NLP tools
and dialogue managers. This is an important challenge and was tentatively
introduced by products such as Apache's OpenNLP for some aspect of
written NLP. An OpenDial library would probably be useful and would help
focus efforts elsewhere rather than on the components that all systems have in
common. Finally, a third step in the same direction would be the
materialization of a whole set of services linked to vocal recognition, text to
speech, prosodic, syntactic and semantic analyses in a software layer such as
middleware, or better yet, in a computer extension card, such as graphics
cards for 3D visualization. This challenge, if it happens someday, would
allow it to be exceptionally easy to develop a system: all processes would be
carried out in hardware rather than software, which increases the speed, and it
would really open the door to systems usable in real time. Obviously, this is
not a simple challenge, and if we compare it to 3D, for which the graphics
card works much more during the design than the final product, which needs
overspecific and overdelicate processes, we could imagine that a dialogue
card, at first, would accelerate and simplify the design of systems without
carrying out the full development.
1.4. Conclusion
The quest for a machine able to understand human language and answer
its user as well as a human interlocutor has gone on for more than 50 years.
The issues arising from natural language processing have not allowed us to