Discussion
While there are still serious technical issues remaining to
be overcome, dialogue-based user interfaces are showing
promise. Once they reach a certain level of basic competence,
they will rapidly start to revolutionize the way the
people interact with computers, much like the directmanipulation
interfaces (using menus and icons) revolutionized
computer use in the last decade.
We are close to attaining a level of robust performance
that supports empirical evaluation of such systems. For
instance, we performed an experiment with an earlier dialogue
system that interacted with the user to define train
routes (Allen et al., 1996). For subjects, we used undergraduates
who had never seen the system before. They
were given a short videotape about the routing task, taught
the mechanics of using the system, and then left to solve
routing problems with no further advice or teaching, except
that they should interact with the system as though it were
another person. Over 90% of the sessions resulted in successful
completion of the task. While the task was quite
simple, and some of the dialogues fairly lengthy given the
task solved, this experiment does support the viability of dialogue-based interfaces, and validated the claim that such
systems would be usable without any user training.
In the next year, we plan a similar evaluation of the
TRIPS 911 system, in which untrained users will be given
the task of handling emergencies in a simulated world. This
experiment will provide a much more significant assessment
of the approach using a task that is near to the level of
complexity found in a wide range of useful applications