Research Assistantships Available
------------------------------------------------------------------------
(This announcement is available on the web at:
http://tigger.cs.uwm.edu/~mcroy/RA-ad.html)
------------------------------------------------------------------------
Natural Language and Knowledge Representation Research Group
Decision Systems and Artificial Intelligence Laboratory
Department of Electrical Engineering and Computer Science
University of Wisconsin-Milwaukee
------------------------------------------------------------------------
Overview of our Group's Research
A dialog is a two-way interaction between two agents that
communicate. People communicate efficiently and effectively using
dialog. Intelligent dialog systems (IDS) are concerned with the
effective management of an incremental dialog between a computer
system and its user. The general goal of our group's work is to
investigate computational models of dialog that can support
effective interaction between people and computer systems. Our
research aims to provide computational methods for integrating
and using this information to produce relevant utterances and to
identify and resolve communication problems as they arise.
Computers do not typically engage in dialog; they tend to provide
a lot of information all at once, allowing few opportunities for
feedback from the user. Moreover, a computer system will rarely
attempt to understand what feedback it does allow, so that only
feedback anticipated by the software designer affects system
performance. Indirect feedback, such as asking the same question
twice, is not interpreted; thus, the typical computer system will
simply repeat its pre-programmed response, rather than attempting
to address the underlying difficulty. People would rarely make
this mistake (unless they had reason to believe that their
initial utterance was not heard).
An IDS must deal with both input and output and must monitor the
effectiveness of its actions. During an interaction, content to
be presented, as well as the system's model of the user (for
example, the user's apparent level of expertise), changes
dynamically. Intelligent dialog requires the representation of
goals, intentions, and beliefs, including beliefs about the
effectiveness of the interaction. The intelligence of a dialog
system lies in the traditional utility of a dialog model, which
includes the ability to interpret a variety of communicative
forms whose meaning depends on the context, including fragments,
anaphora, and follow-up questions. In our view, dialog systems
must also recognize situations that require dynamic adaptation,
including misunderstanding, nonunderstanding, and argumentation.
Specific Current Research
Our research is funded by the National Science Foundation and
Intel Corporation. Specific research topics we are investigating
include:
o Specifying a uniform, declarative representation of the different
kinds of information that influences people's understanding of an
ongoing dialog and that allow for the detection of potential
problems, such as misunderstanding, misconceptions, and
digressions.
o Reasoning with these representations to achieve the following
tasks.
+ Interpreting utterances and identifying problems as they
arise;
+ Invoking a planner to produce relevant utterances and
repairs;
+ Deciding when and how to take control of the dialog,
allowing that either the system or the user could be in
control.
o Processing multimodal dialog, specifically:
+ Representing and coordinating multimodal information (such
as speech, gesture, pointing) during the processing of
dialog;
+ Evaluating how this information influences people's
understanding of multimodal dialog;
+ Developing computational methods that use these knowledge
sources to identify and resolve multimodal communication
problems as they arise.
o Specifying computational models of argument; users may not agree
with the system, the ability to argue (or recognize the start of
an argument) is necessary for an IDS.
o Specifying a fast template-based natural language generation
system; users will not wait for the system to decide how to say
what it wants to say in English.
We are evaluating this research by building intelligent tutoring
systems in health care domains. We aim to deploy these systems with
end users in the Milwaukee area.
URLs for Additional information
An excellent overview of our group's current research appeared in
ACM SIGART's intelligence magazine 10(1), pg. 14-23, Spring 1999,
and is available online at
http://tigger.cs.uwm.edu/~mcroy/Publications/intelligence.pdf.
The home page for the Natural Language and Knowledge
Representation Research Group is at
http://tigger.cs.uwm.edu/~mcroy/nl-kr.html. Links to all our
projects are available.
This ad (and updates to it) is available at:
http://tigger.cs.uwm.edu/~mcroy/RA-ad.html.
Position Requirements and Application Information
One position is reserved for an assistantship in multimodal
dialog (specifically speech and gesture); remaining positions can
be in any area that contributes to the group's research goals.
These positions require a M.S. or B.S. in Computer Science. These
positions are not budgeted for the postdoctoral level. It is
anticipated that the research associated with these positions
will lead to doctoral dissertations. (The master's program in CS
also includes a thesis.)
These positions are open until filled. Duration is contingent
upon appropriate progress in both research and academic
requirements.
To apply for an RA, you should submit a copy of your academic
transcript, a resume, a 1-2 page statement of interest, the phone
number and email address of 2-3 people willing to serve as
references, and copies of relevant publications. This information
should be directed to the address below.
Offers are contingent upon admission to the EECS graduate program
at UWM. For admission to UWM, a separate application to the
graduate school will be required. UWM has "rolling" admissions.
To obtain an application, phone graduate admissions at (414)
229-4982.
Questions?
Questions about these positions should be directed to the following
address.
Professor Susan McRoy
Computer Science, University of Wisconsin--Milwaukee
3200 North Cramer Street
Milwaukee, 53211
URL: http://tigger.cs.uwm.edu/~mcroy
Email: mcroy@cs.uwm.edu
Phone: (414) 229-6695
------------------------------------------------------------------------