An Assessment of the Evolution of Research and Systems - Man-Machine Dialogue: Design and Challenges - page 10

The computer implementation is being rationalized: the different linguistic
analyzers are implemented as independent modules and a communication
language between the modules is specified. The fact that the modules are
independent allows us to test them, correct them and improve them separately.
Actually, all design follows an exemplary methodology: the authors have
started by collecting and studying the human dialogues focusing on the same
task, that is they carried out a corpus study, the word corpus referring to a
collection of attested linguistic material, and they even implemented a system
simulation experiment (which will later be called a Wizard of Oz), to collect
the data on the user's behavior when faced with the system they imagined.
The fundamental methods of the MMD are set. Obviously, they are applied
with the means at the times, and the computer's sluggishness, for example,
leads to a wait between 10 and 60 s for each utterance, a wait which is taken
into account in the simulation experiment, and is very far from the speed and
naturalness of human dialogue.
1.1.2. First oral and multimodal systems
While it was possible up until now to present the major advances in MMD
through a few emblematic systems, this has no longer been the case after the
1980s. Indeed, this decade saw a multitude of theoretical works that many
researchers discovered, the dialogue and its characteristics, a multitude of
prototypes and MMD systems, and notably the first oral systems and the first
multimodal systems. Moreover, it was also the golden age of video games and
the general public discovered adventure games with textual interaction 3 , that
were the first recreational MMD systems.
3 As an example, the SRAM (Mars backwards) game, published in 1986 by Ere Informatique,
left a strong impression on the 12-year-old player that I was at the time: all the interaction in the
game went through written commands, which led me to type, for example, “I want to go west”
and see (visualization of the analysis steps carried out by the software) the utterance appear on
screen with the words “go” and “west” highlighted in color, and then discover the software's
answer: “you arrive near a waterfall”, with the display of a visual scene in which the player
must look for clues to continue his quest. The techniques implemented here are much simpler
than those in SHRDLU or in GUS, with keyword detection instead of word sequence detection,
and the keywords are almost always found within the verb-complement pattern, but the limits
are not the same: the vocabulary and possibilities are vast, adapted to intensive use, and the
game must also be robust, reliable and interesting.