Dialogue Understanding based on Situation Semantics and Pragmatics

Hidetosi SIRAI, Ken-ichiro SHIRAI, Imani IKUMI and Toshiyuki TONOIKE

School of Computer and Cognitive Sciences, Chukyo University
Tokodate 101, Kaizu-cho, Toyota, 470-03, JAPAN
e-mail: sirai@sccs.chukyo-u.ac.jp

We have been studying a dialogue understanding model based on Situation Semantics. The object is to investigate how the context including the beliefs of conversational participants are developed under the grammatical and pragmatic constraints.
Last year, we studied the constraints of so-called presupposition. We analyzed the problems around presupposition, especially on the projection problem which is the relation between the presupposition of the whole sentence and those of its parts. Furthermore, we formalized the presupposition from the point of inverse information, and described some related phenomena.
This year, we studied `definiteness' and `negation' in Japanese, because our team's common theme is `Representation and Understanding of Concepts', and definiteness and negation are the concepts which are hard to understand and should be investigated in more detail. Furthermore, we studied on the descriptions of conceptual structures in lexicon in order to make a dialogue system.
Our contributions are summarized as follows:

{1. Research of Definiteness and Presupposition}
In this research, the notion of definiteness in Japanese is investigated employing recent outcomes of dynamic semantics (Dynamic Predicate Logic and Update Semantics). Based on the dynamic treatment of definiteness from the information-theoretic perspective of meaning, existential presupposition of Japanese noun-phrases will be given a more discourse-oriented account.

{2. Research of Negation in Japanese}
The purpose of this research is to clarify how meanings of negative expressions are constrained by situations in which they are used, and to discuss what constraints there are. Negation has traditionally been analyzed by syntactic or semantic approaches. However, it seems that only with these approaches we cannot explain why (1) sounds strange.

John-wa rappa-wo 2-kai narasa-nak-atta. (1)
John-TOP trumpet-ACC 2-times blow-NEG-PAST
`John did not blow the trumpet twice'

The strangeness of (1) is not from syntactic or semantic reasons but from some pragmatic reason, because (1) sounds natural if it is used in appropriate situations such as the one in which although John had been ordered to blow the trumpet 12 times, he did it only 10 times. This indicates that we have to investigate what constrains induce or block the reading of propositional or predicative negation. The following results have been obtained so far.
Let S be a sentence with negation and quantifier(s), and S' a sentence which is the same as S, except that it is lack of negation.

  • It is easy to construe S as having the reading of propositional negation if there is a situation in which S' is not supported by it.
  • It is easy to construe S as having the reading of predicative negation if there is a situation in which helps us calculate a logical relation between negation and quantifier(s). We classified situations of this sort into three types.


    In each case, we can see situations as helping calculate the content of S.

    {3. Research on Lexical Conceptual Structure}
    In this study, we have considered how to compile a lexicon for a natural language in general, and lexical extensions in particular. Most lexical items can be polysemous; some of their meanings are lexicalized, i.e., registered in the lexicon, some lexical items can be used in an extensive way, often in a heavily context-dependent way. we have tried to capture such extension possibilities and proposed to store such knowledge as heuristics, as extension-patterns of semantic features in the lexicon. With the aid of such heuristics, we can obtain a lexicon which is generative in the sense that when we cannot interpret utterances by static lexical descriptions, we can generate tentatively extended meanings. For example, a group of nouns which are place names (LOC-ation) can refer to people live there (ORG-ination). Also we assume lexical conceptual structures which consists of a verbal element, nouns/noun phrases and possibly adverbial elements.
    In the next fiscal year, we will consider the relationships between lexical descriptions and lexical conceptual structures and examine the structured lexicon and the mechanism of forming and processing concepts.

    Keywords: definiteness, presupposition, dynamic semantics, negation, quantifier and situation, generative lexicon