An Assessment of the Evolution of Research and Systems - Man-Machine Dialogue: Design and Challenges - page 15

MMD systems that started to emerge in the 2000s. As an example, to compare
with the figures mentioned previously in this chapter, Ritel's vocabulary has
65,000 words, which approximately matches the number of entries in a
language dictionary. Another example from the 2000s, the Amitiés system
[HAR 06], a closed-domain MMD system provides us with a way to compare
previous systems of the same type as well as an open-domain system such as
Ritel. Amitiés was designed from an in-depth corpus study of about 1,000
dialogues all belonging to the financial domain on which one of the tasks is
focused. The figures corresponding to this material are as follows: 30,000
sentences for approximately 8,000 words of vocabulary. This is much more
than GUS could do, but is still very far from the 65,000 words of a language.
Finally, in the 2000s (and in the following decade), as we saw at the very
beginningofthischapter, thefirstgeneralpublicMMDsystemshaveappeared,
incorporated to various Websites, electronic diaries, geolocation systems and
other personal digital assistants. Even if the quality is not there yet, we can
imagine that it will help encourage the scientific community's efforts.
1.2. A list of possible abilities for a current system
At the level of general public systems, as we mentioned, we are still far
from a natural dialogue in natural language. A few tests of systems, called
voice-controlled or voice-recognition systems, allow us to quickly verify this.
For example, the geolocation systems and cell phones are still at a keyword
detection level: city names for the first and recipient names for the second. We
arestillveryfarfromtheautomaticunderstandingofutterancessuchas“Iwant
to go to Grenoble by bypassing Lyon and avoiding the highway between Saint-
Etienne and Lyon”, in which the user mentions a point of passage and different
preferences for two parts of the journey all at once (a much quicker request to
say than to program directly into the system, if at all possible). We must still
admit that from examples such as these, voice control is not often adapted
to the computer system user: it is often noisy, we are never sure of being
understood properly, and we are always convinced of being more efficient by
directly manipulating the system with a classic MMI. Contrary to what various
researchers claimed in the 1980s, one cannot say that because there are more
and more computers and more and more data accessible that the MMD will
impose itself as a new communication mode. As Vilnat [VIL 05, p. 5] states,