Problems in Creating Tagged Orthographical Transcriptions of
Spoken Dialogs in Japanese
<P>
Syun Tutiya
<P>
Department of Philosophy, Chiba University<P>
1-33 Yayoi-cho, Inage-ku, Chiba, 263 Japan<P>
tutiya@cogsci.l.chiba-u.ac.jp
<P>
Problems in creating orthographical transcriptions of spoken
dialogs in Japanese are discussed.  Sociolinguists' and
conversation/discourse analysts' methods are well motivated
and well meant but lack digitization and portability. The TEI
encoding scheme of spoken language is introduced and
discussed, along with the annotative tags for the
morphological categories.  TEI encourages the use of header
information as an inline means of cataloging. In the case of
speech corpus, the information internal to it, like the
speakers and situations, are easily and consistently
expressible the header part of the document.

Dialogs are full of interruptions.  Turn takings do not take
place as described in drama scripts.  Sociolinguistic
transcriptions of the phenomena are readable by humans but
it can not be encoded in any straightforward manner in to
electronic form.  We propose to accept the TEI style
encoding which, by utilizing empty tags with pointers,
enables us to encode non-linear phenomena like overlapping,
interruption and interleaving.  The mechanism requires
further elaboration in developing and implementing handling
tools but that part of elaboration will be made use of in
linking locations to digital sound frames.  "Pretty
Printing" of the fully tagged  texts will be available.

Japanese orthographical transcription poses an interesting
challenge to the corpus encoders.  Japanese orthography does
not have a builtin mechanism with which to make word
boundaries visible. I.e., words, or whatever corresponds to
the Western ``words,'', are not separated by spaces.  This
fact requires the encoder to mark some morphological
segmentation in an elegant and manageable way so that search
for a "word" may not pick up all the possible strings which
contains the "word."  Segmentized units are marked with
category indicators in a way which is conformat to SGML/TEI.
This will facilitate the use of spoken dialog corpora for
various different purposes.  A detailed account of
morphological units and their categories are given in terms
of the written corpus being develop by the author.
<P>