Causality and Point of View in Japanese Complex Sentences
<P>
Hiroshi NAKAGAWA 
<P>
Dept. of Electronics and Computer Engineering,
Yokohama National University<BR>
156 Tokiwadai, Hodogaya, Tokohama, 240, JAPAN<BR>
e-mail: nakagawa@naklab.dnj.ynu.ac.jp
<P>
We focus here on the semantics of complex sentence that expresses a
causality relation. Fortunately in complex sentence we can usually
identify the discourse relation between subordinate and main clause in
virtue of conjunctive particles. So, we pursue the coreference problem
by fully utilizing the discourse relation known from these conjunctive
particles.
<BR>
The main target of this research is a complex sentence that describes
a situation in which a human feels something emotionally or physically
and then responds to it by doing some action or having a new emotion. 
Namely what we deal with is a complex sentence which describes a
causality relation and uses a conjunct like {node} (`because' or
`since' in English) in Japanese. To capture this kind of causality
relation, the following notion is needed.
<BR>
Definition:
{Motivated} is a person who is so strongly effected by the situation
described by the subordinate clause that s/he acts as described by the
main clause.
<BR>
Usually a {motivated} which is one of the roles appearing in the
subordinate clause corresponds to the matrix subject.  The problem we
pursue in the rest is who is preferred as a {motivated} from among
roles appearing in the subordinate clause. For this, we will examine a
number of examples hereafter.
<BR>
We start with an example which does not have a modality part in
subordinate clause.  Note that intuitive interpretations follow just
after a sentence, henceforth.
<BR>
<P>
$[\phi_{sub-s}$ & kurusii &  node], <BR>
               & feel bad & because<BR>
$\phi_{sub-m}$ & kusuri-o & non-da.<BR>
              & medicine & took<BR>
<P>
`Since $\phi_{sub-s}$ felt bad, $\phi_{sub-m}$ took a medicine.'<BR>
$\phi_{sub-s} = \phi_{sub-m}$}
<BR>
This example does not have a modality part. That means that no other
role are introduced in the subordinate clause besides the semantic role
{experiencer} of feeling pain or bad. And in intuitive
interpretations, these semantic roles corefer with the matrix subjects
respectively. In general, in this type of subordinate clause, a {
motivated} is the semantic role {experiencer} of proposition part
that often corresponds to the grammatical subject of subordinate clause.
<BR>
<P>
$[\phi_{sub-s}$ & kurusi & -gat-ta & node],<BR>
               & feel bad & behaved like & because <BR>
 $\phi_{sub-m}$ & kusuri-o & atae-ta.<BR>
               & medicine-ACC & gave <BR>
<P><BR>
`Since $\phi_{sub-s}$ behave like feeling bad, $\phi_{sub-m}$ gave a medicine.'<BR>
$\phi_{sub-s} \neq \phi_{sub-m}$}
<BR>
In (\ref{40}), modality expression
{-gat} (`behave like --ing') are used. Then these expressions
introduce a pragmatic role {observer} who observes the situation
described by the subordinate clause. {Observers} are not explicitly
expressed in the subordinate clauses. Nevertheless in intuitive
interpretations of (\ref{40}), it appears as the matrix
subjects. In our term, as the referent of {motivated} a semantic
role of experiencer of subordinate clause is overridden by a pragmatic
role {observer} introduced by the modality part {gat}.
Let's see the cases in which the point of view expression
is used in a subordinate clause.
<BR>

<P>
$[\phi_{sub-s}$ & tasuke & -tekure-ta & node],<BR>
               & help   & -GIVEN-PAST & because<BR>
 $\phi_{sub-m}$ & totemo & kansha-si & -teiru.<BR>
               & very   & thank     & -PERFECT<BR>
<P><BR>
`Since $\phi_{sub-s}$ helped, $\phi_{sub-m}$ felt the great thanks.'<BR>
$\phi_{sub-s} \neq \phi_{sub-m}$}
<BR>
Japanese point of view expression {-tekurer} introduces a {
beneficiary} who gets a certain benefit from the action described by the
verb phrase to which {-tekurer} is attached. And in intuitive
interpretation, the introduced {beneficiary} comes to be a {
motivated} which actually is the matrix subject. In fact, an {
affectee} of passive coincides with {beneficiary} of {-tekurer}
in our framework. 
In (\ref{50}), the {beneficiary} coincides with the point of view. 
However in {-teyar} (`GIVE') case, the point of view is a subject,
and the {beneficiary} is an object. In this kind of cases, of
course, a {motivated} is not a point of view but a {beneficiary},
for instance shown in the intuitive interpretations of the following
examples.
<BR>
<P>
$[ \phi_{sub-s}$ & ronbun -o & otosi & -teyat-ta <BR>
                & paper-ACC & reject & -GIVE-PAST<BR>
node], & $\phi_{sub-m}$ & okot-ta.<BR>
because &              & got angry<BR>
<P><BR>
`Since $\phi_{sub-s}$ intentionally rejected $\phi_{ben}$'s paper, $\phi_{sub-m}$ got angry.'<BR>
$\phi_{sub-s} \neq \phi_{sub-m} = \phi_{ben}$}
<BR>
In Japanese, the combination of point of view expressions and passive
form, for instance, verb{-are-teyaru} (passive morphoneme {-are}
plus -teyar) are often used. In this kind of case, a pragmatic role
introduced by the last expression is taken as a {motivated}. Namely
in {-are-tryar} case, it is a beneficiary of {-teyar}.
<BR>
So far, we have dealt with interpretations based only on voice, modality
and point of view. But we have to think of the priority among
interpretations based on these linguistic factors and those based on
commonsense knowledge.
<BR>
<P>
$[ \phi_{sub-s}$ & takusan & sigoto-o & sita <BR>
                & a lot of & work-ACC & did<BR>
node $]$, &  $\phi_{sub-m}$ & tukare-ta.<BR>
because   &                & got tired<BR>
<P><BR>
$\phi_{sub-s} = \phi_{sub-m}$}
<BR>
Here the intuitive interpretation comes from the commonsense knowledge.
However in the following example,
<BR>
<P>
$[ \phi_{sub-s}$ & takusan & sigoto-o & si <BR>
                & a lot of & work-ACC & did<BR>
-tekure-ta & node $]$, &  $\phi_{sub-m}$ & tukare-ta.<BR>
-BE GIVEN & because    &                & got tired<BR>
<P><BR>
$\phi_{sub-m}$ = Beneficiary of the subordinate clause.}
<BR>
we try to seek a background in which the matrix subject, namely {
motivated}, can be interpreted as the beneficiary of {-tekure}, because
for Japanese natives the interpretation of $\phi_{sub-s} = \phi_{sub-m}$ is
pragmatically unacceptable. As such a background, for instance, we think
of the situation like the following. $\phi_{sub-s}$ is a boss of
$\phi_{sub-m}$. Therefore the boss's working for $\phi_{sub-m}$ makes
$\phi_{sub-m}$ feel nervous, and $\phi_{sub-m}$ mentally gets tired. From
these observations, we know that Japanese point of view expression like
{-tekurer}, {-teyar}, etc. have so strong linguistic power that
an interpretation based on commonsense knowledge can be overridden.
<BR>
Finally we show a statistical data for Japanese. We gathered 205
Japanese complex sentences conjuncted by {node} from Japanese novels
written by Souseki Natsume, Yokio Mishima, Sin-ich Hoshi, Banana
Yoshimoto, Osamu Souda and Teru Miyamoto etc, and from the articles of
Japanese journal ``Asahi Journal''. Within them, 56 sentences are of
type that a subordinate clause describes a human's emotional experience
and the main clause describes the action which is done by a certain
human. In all the sentences of this type, the referents of motivated
predicted by our theory coincides with those we get intuitively. At the
judgement for interpretations, if a zero pronoun is used, we fill it by
asking Japanese natives. If proper name is used, we just compared it
with the prediction of our theory. Since in these examples, we find no
exceptional cases, we confirmed the validity of our theory. 
<P>
Keywords: Japanese, complex sentences, point of view, causality