The Comprehensive Study of Japanese Response Markers on the basis of 
a Knowledge Model
<P>
Satoshi KINSUI and Toshiyuki SADANOBU
<P>
Faculty of Letters, Kobe University<BR>
Nada-ku, Hyogo, 657, JAPAN<BR>
e-mail: kinsui@icluna.kobe-u.ac.jp
<P>
  We examine the discourse management function of Japanese Response
Markers -- interjections, sentence-final particles, yes-no words etc. 
-- which have been so far regarded as noises for the spoken dialogue
processing.  For this purpose, we posit a cognitive interface between
language and knowledge-base. This interface contains pointers or
indices which control the access paths to the knowledge-base and the
temporary memory-base set up for each dialogue session. Utterances in
a dialogue can be seen as instructions for operations on this
interface: registering, searching, copying, and inferring. We examine
the nature of these operations by analyzing various discourse
particles and interjections, which, we claim, work as mental
monitoring devices for these operations.  Furthermore we hypothesize
that in the memory-base there are two domains, direct experience
domain (D-domain), which is linked to the long term memory and
indirect experience domain (I-domain), which is linked to the working
area set up for the purpose of each discourse. Information presented
in a discourse is distributed in the appropriate domains. The
properties of the two domains are summarized below:
<BR>
<P>
<LI>    D-domain (linked to long term memory)
<BR>
     :direct (assimilated, digested) information obtained by direct experience
      and past experience are stored
<BR>
     :directly accessible
<BR>
<LI>    I-domain (linked to the working area)
<BR>
     :indirect (not yet assimilated, digested) information obtained by hearsay, 
      inference, hypothetical information are stored 
<BR>
     :only indirectly accessible
<BR>
<P>
<BR>
  We will call linguistic forms which set elements into the D-domain the
direct forms, and those which set them into the I-domain the indirect form.
<BR>
   The distinction between directly acquired information and 
indirectly acquired 
information is part of a more general distinction in Japanese, namely, the 
information already processed and incorporated in the memory or belief system 
and the information still being processed. Thus the information in the sentence
(ia) with the modal showing how the information is acquired can be 
assimilated in the memory base and can some time later be reported by 
the direct form (ib).
<BR>
<BR>
 (ia). kare wa kekkon-siteiru sooda.
       (I hear he is married.)   <BR>
 (ib). kare wa kekkon-siteiru.
       (He is married.)
<BR>
<BR>
  While, Takubo(1992) said that sentence-final particle YO can be seen
as the instructions to register the information into I-domain.
<BR>
    In Takubo(1984,1989a,1990), it is argued that the same distinction is at
work in the behavior of proper nouns and the third person pronouns in 
Japanese. Proper nouns, the third person pronouns (KARE,KANOJO,KARERA) and
KO/A-series of demonstratives are considered direct forms, while proper nouns
and SO-series of demonstratives are considered indirect forms generally.
<BR>
   Then, the following problems arise.
<BR>
<P>
<LI> Is there a parallelism between indirect forms of NPs and those of 
    sentences?
<BR>
<LI> How do modals differ from YO as the indirect forms of sentences ?
<P>
<BR>
  To solve these problems we distinguish two types of indirect forms
functionally. The one is used to show the hearer that the path through
which the speaker has accessed the date is indirect (type I), and the
other is used to show the date is a new information for the hearer
(type II). Though we cannot find morphological difference between type
I and type II in NPs, we must use different sentence-final forms when
the sentential informations are indirect. Namely, modals correspond
type I and sentence-final particle YO corresponds type II.
<P>
Keywords: discourse management, sentence-final particles, D-domain, I-domain
modals