Studies of Cognitive Semantic Model and Analytic Methods for Discourse Understanding

Shun ISHIZAKI, Shigenori TANAKA and Mutumi IMAI

Faculty of Environmental Information, Keio University
5322 Endoh, Fujisawa 252, Japan

In this study, we adopted a process view of conversational process, according to which conversation is a two-way collaborative process, and the linguistic unit of conversation is not a sentence but a fragmentary and meaningful chunk. This study attempts to construct a "grammar" of ordinary language. We believe that the grammar should first and foremost satisfy the condition that we must be able to express freely without being bothered by the constraint of grammatical well-formedness. With the framework as briefed as above, we attempted to identify the unit of chunk through analyzing conversational data (approximately 30,000 running-word data). Last year, we suggested that the unit of chunk be operationally defined on the basis of the three criteria: structural, idiomatical and physical criteria.
One of the major aims of this study is to build a computational model with a flexible mechanism for discourse understanding. Last year, we introduced a metric space among concepts (thing concepts) in a concept dictionary by using MDS method. A dynamic feature of such concepts' semantics is examined by using SD method to apply the features to metaphor analysis. This year, the following two topics are studied for a flexible discourse understanding.

{(1) A meaning-making process in daily conversation: the role of because}
This study emphasizes the notion of meaning-making in the analysis of discourse in daily conversation. Tanaka and Ishizaki(1994a) suggested that meaning is interactively produced through the mechanism of chunking, where a chunk or an utterance unit successively triggers another chunk, developing a stream of meaning or what is called 'discourse'. Within the framework of meaning-making. Tanaka and Ishizaki(1994b) did a conversational analysis of the functional role of two discourse markers, 'you know' and 'I mean'. In this study, we focussed on because, another discourse marker. In the standard grammar of English, because, a conjunction of subordination, introduces a causal subordinate clause, which is governed by the main clause. In the meaning-making process of daily conversation, however, the main-subordinate relationship is not always clear-cut in terms of their clause boundaries. We take it that the preceding chunk triggers 'because', with a perceived need to justify what is said. In other words, the triggered
'because' play a functional role to provide a discourse space for the speaker to continue. Our conversational data showed two things: (1) the scope of 'because' is not always easy to delimit, and (2) a 'because' clause is not always well-formed with 'structural disturbances'. We interpreted these findings as supporting a chunking model of daily conversation.

{(2) Dynamics of word meaning caused by context: Interaction between verbs and objects (nouns)}
In this project, the structure of verb's polysemy is studied to obtain the mechanism of dynamics of verb's semantics and to construct its cognitive semantic model. The objective of this study is to use such a mechanism and a model for constructing a concept dictionary and also for programming algorithms of semantic analysis and conceptual analysis of discourse. Many semantic features are listed this year to represent dynamics of verb's semantics which interacts with that of verb's object (noun) in a sentence. Increasing the number of verbs and nouns (objects), the characteristics and describability of their semantic features are studied by using cluster analysis.

Keywords: cognitive semantics semantic space, dynamics of semantics, concept dictionary, cluster analysis, SD method, chunk, adverbial clause, discourse marker, because, trigger