Extended Deadline: May, 28, 1999
Conference: August 23 - 25, 1999
Workshop: August 26, 1999
University College, Chester, England
The translation of spoken utterances (MI, machine Interpreting) is one
of the most challenging tasks within the field of MT. There are still
several phenomena which are unknown or insufficiently described and
are not accounted for in the systems currently available. These
phenomena affect all stages of processing within MT systems. To name
just a few, the translation of spoken language has to take into
account:
- Prosody. While prosodic features have been integrated into MI
systems over the last few years, their application has largely been
restricted to disambiguation tasks (to detect boundaries of various
kinds and for classification). The influence of specific conditions,
e.g. stress, on the prosodic realization of texts is more or less
unknown.
- Basic characteristics of speech as opposed to written language.
Hesitations, repairs, breaking offs and others fall into this
class. Their occurrence if often unexpected, the recognition and
utilization of these seems underdeveloped
- Characteristics of lexica and grammars. Clearly, the knowledge
sources for systems dealing with spoken input has to be different from
those analyzing written language. Especially spontaneous speech can
usually not be covered using a standard grammar constructed with
written texts in mind.
- User-Machine interfaces. Using speech as input and output of machine
translation systems poses new and sometimes unexpected problems for
the design of the interface through which an user interacts with a
system. These range from designing a high quality of synthesis to the
assessment of the reactions to sometimes long waiting periods during a
translation.
As a consequence, a MI system does not only contain the additional
components needed to analyze and synthesize speech, but the core
modules have to be adapted as well to the demands of speech-to-speech
communication. The collaboration between researchers working in the
traditional core field of MT and in the more speech-oriented fields
is highly advantageous for a successful design and implementation of a
speech-to-speech translation application.
The workshop ``Constraints on Machine Interpretation'' calls for
contributions relevant for the translation of spoken language,
including, but not limited to:
- Speech recognition, language recognition, speaker identification
- Prosody and its use in MI systems
- Analysis, transfer and generation of spoken input utterances
- Architecture and design of MI systems and their user interfaces
- Dialog behavior, design and analysis
- Evaluation of MI systems
The workshop complements the main theme of TMI'99 (Modalities and MT:
Where can MT be used?) by concentrating on one important modality,
namely speech. It will be held as a one day workshop on Aug, 26th
1999, right after the main conference.
IMPORTANT DATES:
Paper submissions due May, 28, 1999
Acceptance notification June, 15, 1999
Final copies due July 2, 1999
Conference dates Aug 23-25, 1999
Workshop Aug 26, 1999
SUBMISSION GUIDELINES
The format of submissions for the workshop should follow the
general guidelines of the main conference.
The submissions should be sent to one of the contact addresses
below. Also note that the schedule for submissions, notification
and final versions is different from the schedule of the main conference.
CONTACT
Susanne J. Jekat
University of Hamburg
Computer Science Department
Vogt-Koelln-Str. 30
225 27 Hamburg
Germany
Phon: + 49 40 5494 - 2520
Fax: + 49 40 5494 - 2515
e-mail: jekat@informatik.uni-hamburg.de
web: http://nats-www.informatik.uni-hamburg.de/~jekat/
Jan W. Amtrup
Computing Research Laboratory
New Mexico State University
Las Cruces, NM 88003
USA
Phone: (+1 505) 646.2965
Fax: (+1 505) 646.6218
e-mail: jamtrup@crl.nmsu.edu
web: http://crl.nmsu.edu/~jamtrup/