Analysis of Paraphrases and Cohesion in Dialogues at Seminars -- Notional Relationships --

Kikuko NISHINA, Yooko SASAGAWA, and Mitsuru DOI

International Students Center, Tokyo Institute of Technology
2-12-1 Ookayama Meguro-ku Tokyo 152 Japan

We have made 6 new video-recordings (360 minutes) of post graduate seminars in the engineering department at TIT in addition to 10 video-recordings which were made last year. Through observation of the video-recordings, this paper analyses cohesion and coherence of seminar dialogues. In order to analyze spoken text, we have to treat it in a formal way because sentences are not uttered in the same way as written text due to ill-formed and imperfect sentences and so on.
We try to observe collocation of words in the spoken text focusing on paraphrased parts of the dialogues in order to find cohesion in the spoken discourse. On the other hand, we note functional words such as conjugations and demonstratives in the context of meta markers of coherence.
We define seminar dialogue as a mutual understanding process for reaching for new information or knowledge, which is different from idle chatter or strategic dialogues for sales talks.
A speaker often uses paraphrases as rhetorical strategies, when he wants to persuade the audience. He also uses them to repair communication breakdown for the following reasons:

(1) Physical reasons of both speakers and listeners (the handicapped; difficulty in hearing or speaking)
(2) Language abilities of both speakers and listeners (children vs adults, foreigners vs native speakers)
(3) Background knowledge of both speakers and listeners (experts vs nonexperts)
(4) Reasoning ability of both speakers and listeners.

In this context, a paraphrase is defined as replacing words or phrases in other words or phrases. We also define cohesion as word relationships within the text and coherence as word relationships with reference to background knowledge (the world).
We observe the following patterns of cohesion within paraphrases in dialogues of the data: (a) substitutions (b) ellipsis (c) reference (d) synonymy (e) hyponymy (f) metonymy (g) syneodoche (h) antonymy.
Those words which are paraphrased should be close to others in the sense of notional relationships. On this premise, we use a notional dictionary produced by EDR (Japan Electronic Dictionary Research Institute Ltd.) to measure the distance between words and the paraphrased words. We also try to show notional network maps.
On the other hand, we observe the structure of coherence by dividing functional words as discourse markers into additives, adversative, causal, temporal, exemplifying and summing up.
From this analysis we conclude the following:

(1) In analyzing spoken discourse, we have to use different methods from written text. It seems a useful method to measure collocations of cohesion in the text by using a notional dictionary, although we need more data for measuring the distance between words within the text.
(2) We can observe spoken discourse from different viewpoints such as cohesion and coherence. Cohesion shows collocations of words or notional relationships within a discourse: coherence shows logical structures.

The following topics form the basis of further research:

(1) We will collect more data for analyzing in order to measure notional distances with a notional dictionary.
(2) We will form a descriptive rhetorical structure model according to notional relationships using computational algorithms.
(3) We will form protocols of discourse structure at seminars (comparing with idle chat, strategic talks etc.)

Keywords: coherence, collocation, background knowledge, notional dictionary, discourse marker