ECAI-96 Workshop on
DIALOGUE PROCESSING IN SPOKEN LANGUAGE
SYSTEMS
Budapest, Hungary, August 13, 1996
INTRODUCTION
The development of dialogue components for interactive systems that
employ speech as input and/or output modality has to take into account
problems that are specific for the treatment of spoken language. Among
these problems are the following:
Segmentation of dialogue contributions into basic units
While for dialogue systems that cope with written / typed language a
sentence can serve as basic unit for dialogue modeling, this approach
cannot be applied for the treatment of spontaneous speech or even read
speech, where phrasing of sentences may differ from speaker to
speaker. Spoken input is often incomplete, incorrect and contains
interruptions and repairs; full sentences occur only very
occassionaly. Therefore, new basic units for the development of
dialogue models have to be proposed in order to also capture
fragmentary input.
Related to this problem is the determination of the boundaries that
exist between the various dialogue units in longer single-speaker
dialogue turns. While for written language punctuation and
paragraphing serve as indicator for segmentation, reliable cues for
the segmentation of spoken language still have to be determined.
It can be expected that dialogue models that build on such a new
notion of basic dialogue units differ significantly from dialogue
models that treat only written language. A contrastive examination of
the differences between dialogue models that treat spoken and written
dialogue contributions is a point of future research.
Interaction of prosody and dialogue processing
For some of the above-mentioned issues the consideration of prosody
can contribute to a solution of the problems. Prosody can perform many
functions such as chunking turns into smaller units, emphasizing
important information, indicating discontinuities (e.g. interruptions,
corrections), expressing intention and emotion. Therefore components
which make prosodic information accessible to dialogue processing
become more and more important.
Robustness
Robustness of all components is an important issue in the design and
the development of spoken language systems. With respect to dialogue
components robustness is related to the following topics:
- recognition errors and missing information
- unexpected input
- clarification
- disfluencies
Evaluation
So far the evaluation of spoken language systems has been focusing on
the quality of the speech components. Evaluation criteria for the
dialogue components of such systems are still to be developed.
A careful evaluation can contribute to the improvement of the system
with respect to what a user expects from the machine and how she
adjusts to its abilities. Therefore criteria like user acceptance and
user satisfaction have to be taken into account.
These topics are addressed by papers presented in the course of this
workshop.
ORGANIZING COMMITTEE
Elisabeth Maier
DFKI GmbH
maier@dfki.uni-sb.de
Marion Mast
IBM Scientific Center
mast@heidelbg.ibm.com
Susann LuperFoy
MITRE
susann@azrael.mitre.org
PROGRAM COMMITTEE
Nick Campbell, ATR, Japan
Morena Danieli, CSELT, Italy
Norman Fraser, VOCALIS, UK
Julia Hirschberg, AT&T, USA
Susann LuperFoy, MITRE, USA
Elisabeth Maier, DFKI GmbH, Germany
Marion Mast, IBM Scientific Center, Germany
PRELIMINARY PROGRAM
09:00 - 09:15 Welcome
Session 1: Theoretical and Sociological Aspects of Spoken Dialogue
System Design
09:15 - 09:45 Detlev Krause (University of Hamburg, Germany)
Social Research in Context of Speech Systems.
The Case of VERBMOBIL
09:45 - 10:15 David Traum (TECFA, Universite de Geneve, Switzerland)
and Peter Heeman (University of Rochester, USA)
Utterance Units in Spoken Dialogue
10:15 - 10:30 Summary and discussion
(Rapporteur: Nils Dahlbaeck, Linkoeping University,
Sweden)
10:30 - 11:00 Break
Session 2: Prosodic Aspects of Spoken Dialogue Processing
11:00 - 11:30 Bernd Tischer, (Ludwig-Maximilians-Universitaet
Muenchen, Germany)
The syntax of Self-Corrections in Dialogues
11:30 - 12:00 Mark Seligman (Berkeley, USA), Junko Hosaka (University
of Tuebingen, Germany), and Harald Singer (ATR, Japan)
``Pause Units'' and Analysis of Spontaneous Japanese
Dialogues: Preliminary Studies
12:00 - 12:30 Brigitte Grote (FAW, Germany), Eli Hagen (Technical
University of Darmstadt, Germany), Adelheit Stein
(GMD-IPSI, Germany), and Elke Teich (University of the
Saarland, Germany)
A Generation Perspective on Speech Production in
Dialogue
12:30 - 12:45 Summary and discussion
(Rapporteur: N.N.)
12:45 - 14:00 Break
Session 3: Spoken Dialogue Systems - Design and Implementation
14:00 - 14:30 Alicia Abella, Michael K. Brown and Bruce Buntschuh
(AT&T Research, USA)
Development Principles for Dialog-Based Interfaces
14:30 - 15:00 Niels Ole Bernsen, Laila Dybkjaer and Hans Dybkjaer
(Roskilde University, Denmark)
User Errors in Spoken Human-Machine Dialogue
15:00 - 15:30 R.J. vanVark, J.P.M. de Vreught, L.J.M. Rothkranz
(Delft University of Technology, Netherlands)
Classification of Public Transport Information
Dialogues Using an Information-Based Coding Scheme
15:30 - 15:45 Summary and discussion
(Rapporteur: Jacques Siroux, ENSSAT and IUT,
Lannion, France)
15:45 - 16:15 Break
Session 4: Evaluation of Systems/Algorithms
16:15 - 16:45 Donna Gates, Alon Lavie, Lori Levin, Alex Waibel,
Marsal Gavalda, Laura Mayfield, Monika Woszczyna
and Puming Zhan (CMU, USA and University of Karlsruhe,
Germany)
End-to-end Evaluation in JANUS: a Speech-to-speech
Translation System
16:45 - 17:15 Masahiro Araki and Shuji Doshita (Kyoto University,
Japan)
Automatic Evaluation Environment for Spoken Dialogue
Systems
17:15 - 17:45 Teresa Sikorski und James Allen (University of
Rochester, USA)
A Task-Based Evaluation of the TRAINS-95 Dialogue
System
17:45 - 18:00 Summary and discussion
(Rapporteur: Massimo Poesio, Edinburgh University, UK)
18:00 - 18:30 Final discussion and closing statements
PARTICIPATION
Since workshop attendance will be limited to maximally 40 people,
persons without a paper should contact the organizers as soon as
possible. Preference will be given to people who present a paper, to
their co-authors and to persons who submitted a paper. Other persons
interested in attending are asked to provide a one-page description of
their background and of their interest in the workshop.
REGISTRATION
This workshop takes place directly before the general
ECAI-conference. It is an ECAI policy that workshop participation is
not possible without registration for the general
conference. Registration information can be obtained from the
following address:
http://www.dfki.uni-sb.de/ecai96/registration/Registration.html
Workshop registration includes admission to the workshop, the workshop
notes and refreshments. Note: Individuals must pay the ECAI-96
technical programme registration fee in addition to the workshop fee
of 50 ECU.
NOTE: Workshop information can also be obtained from
http://www.dfki.uni-sb.de/~maier/ecai.html