Information Technology Reference
In-Depth Information
allow for both types of interaction, for example some companion robots can
both give their users information and carry out simple tasks on demand, such
as walking or dancing. Other systems do not aim to provide information or
carry out specific tasks. These are, for example, purely recreational systems,
with the example of conversation robots on the Internet. The examples given
here are taken from general public MMD systems, or at least systems that are
destined to be such. But do they really work? In fact, what we can note when
using this type of system is that it is quite difficult to establish a proper
dialogue. When a word is recognized and understood, which is not
systematic, the machine tries to give an answer based on this word or attempts
to restart the dialogue in its own way, which is rarely a relevant one. As Vilnat
[VIL 05, p. 5] states, the MMD systems only work in a very imperfect
manner and thus are greatly criticized, up to the point where “it will never
work” is often heard. The criticism comes, first and foremost, from the users
who notice that there is a wide gap between what they test and what they hope
for, and they often believe that a classic MMI is quicker, more efficient and
even easier, or less confusing, to use. The criticism also comes from
researchers and developers in the MMD field. Indeed, the amount of work
required to achieve a system is such that there is a lot of discouragement. The
amount of work corresponding to a doctoral dissertation is not sufficient, at
least when trying to achieve an innovative system. As an example, Guibert's
[GUI 10, p. 60] discouragement when designing a system called A is striking:
“following the termination of the development of this system A, taken as an
example among others, this body of work is actually the chronicles of the
foretold failure of current dialogue systems”.
We will see that when the dialogue is directed by a clearly defined task, it
is possible to design a performing MMD and this design has actually greatly
progressed in the past few years. After discussing a few historical landmarks
(section 1.1), we will quickly cover the functionalities that are more and more
present on current systems (section 1.2), and from this we will deduce a
primary list of potential challenges for the years to come (section 1.3).
1.1. A few essential historical landmarks
The dialogue between human being and machine is a key field in computer
science: a kind of quest for the Holy Grail, which was the source of computer
science developments and researcher vocations. As it so happens, the first
system to become a landmark, ELIZA [WEI 66], is also a huge subterfuge
Search WWH ::
Custom Search