Re: [Communicator-user] Communicator and VoiceXML?

[Dave S. <sta...@bb...>]

Bob Carpenter wrote:
> 
> Two responses in one:
> 
> > from: Dave Stallard
> > Exactly.  In *theory*, you could develop a Communicator system using a
> > state machine formalism, but in practice the number of states and links
> > becomes prohibitive.  MIT had a paper at an ACL Conversational Systems
> > workshop
> 
> Does anyone have an exact reference to this?

Seneff S., and Polifroni, J. (2000) Dialogue Management in the Mercury
Flight Reservation System. ANLP Conversational Systems Workshop.

> If by a "Communicator system", you mean a system that takes
> customer calls to reserve airline flights, then you can
> also use a state machine in practice.  In fact, we have.
> With a very high task completion rate, I might add.
> But our systems probably wouldn't qualify as "communicator systems".

As I recall, at one time you guys were supposed to serve as a "pacing
horse" for the program, but that fell through somehow.

> You could argue that our reservations systems weren't "interesting"
> from a research point of view, and I'd agree with one caveat.
> They demonstrate what you can do robustly with today's technology
> with a high task completion rate.  From an engineering standpoint,
> no other technology has been demonstrated to work with even
> close to the same accuracy and task completion rates as our
> system-initiative, heavily hand-tailored, human voice talent
> prompted, state-based systems.  

I agree totally.  I'm not sure when, if ever, mixed-initiative systems
will pan out in terms of better task completion and user satisfaction
than system-initiative ones.  The motivation for many people working on
the problem is the pure existential goal of "talking to a machine".  
Who knows if that ultimately can be achieved, and if people will
actually want to do it if it is. 

BTW, what are your task-completion rates?  Communicator was topping out
at 80-90%.  

    Dave