Basic Research for Modeling of Spoken Dialogue Understanding Using Prosody<P>
--On the Supra-segmental Features of Spoken Language and Sign Language--
<P>
ICHIKAWA Akira and YAGI Kenji
<P>
Department of Information and Computer Sciences,  Chiba University<P>
1-33 Yayoi-cho, Inage-ku, Chiba-shi, Chiba 263, Japan<P>
e-mail: ichikawa@ics.tj.chiba-u.ac.jp
<P>
To clear the properties of natural dialogue languages and to find
clues of actualization of a spontaneous spoken dialogue interface
system, we made a plan to compare the spoken language with the sign
language as the natural dialogue languages.

In this first report, it is claimed that natural dialogue language has
important characteristics which differ from written languages to
comprehend its utterances easily under the real time communication
conditions. The characteristics should have a function that indicates
the semantic structure of the utterance. The function of spoken
language exists in the supra-segmental features, ex. prosody, but that
of sign language has not yet investigated. So we assume that there are
some supra-segmental features in the sign language, and confirm these
hypotheses.

Before analyzing the supra-segmental information of the real spoken
dialogue language, we survey two types of experiment for those of the
written style utterances. First one is the comparison between the
results of the comprehension test of reading and those of hearing for
Japanese garden-pass sentences.  Second is the prosodical analysis of
the announced discourses of the radio news.

As for the sign language, we try to classify the sign language word
from the viewpoint of the supra-segmental structures.  We got six
types of the structures.  Those suggest that the structure of the sign
language word is similar to those of the KANJI characters due to
direct visual understandability.  Next, time structures of the sign
language sentences were analyzed. It becomes clear that the
distribution of the length of the pauses between words in a sentence
is highly correlative to the structure of the sentence.

These results suggest that using the supra-segmental information and
parallel processing are essential to comprehend dialogue utterances,
and the process extremely differs from that for the written sentences.
<P>
Keywords: spontaneous spoken dialogue interface system, prosody,
supra-segmental features, sign language, comprehension test, garden-
pass sentences<P>