Integrative Techniques in ASR & Spoken Dialogue Systems

Dr. Eric Fosler-Lussier
OSU AI Group

Friday, Nov. 7th
3:30pm; 263 Dreese Labs
All interested parties are invited.
Refreshments will be served.

Abstract
The traditional approach to both automatic speech recognition and
spoken dialogue system design has been to consider the process as a
pipeline, with one module handing off information to the next. This
modularity has allowed researchers to focus on one particular aspect
of the language processing problem. However, as spoken language
technology advances, it is becoming clear that we can no longer ignore
the interconnection between different parts of the system (or,
analogously, different representational levels of the dialogue
process). Furthermore, as a field we have seen that multiple
representations of the same linguistic process can help improve
modeling; thus there are connections within the same representational
level that have to be taken into account.

All of this, of course, is just a lot of mumbo-jumbo saying that
there's a lot of information out there in spoken dialogue processes,
certainly more than we're currently looking at. On Friday, I'll give
a short overview of the canonical recognition/dialogue process, and
outline how people are starting to make some new connections between
previously unrelated modules that seem really interesting, with some
illustrations from my own work (which might not be interesting ;) ).
I'll also talk about some proposed research which will push the
technology in various ways.

The talk is meant to be informal, and I invite all of you to derail it
as much as possible.