In various psycholinguistic studies, attempts have been made to define what makes a text coherent. Obviously, connectives play an important role in text coherence, but without these "explicit" connectives, texts can be judged as being coherent as well.In many studies on discourse processing, a rather homogenous set of these "implicit" coherent relations can be found, consisting of referential, temporal, spatial, causal, and structural coherence relations. Each of these relations has been studied and tested thoroughly. However, the actual relations between these coherence relations have hardly been discussed.
In this paper, the links between these kinds of coherence relations will be discussed. Eleven units per proposition are defined: states of affairs (state, process, event (accomplishment), and event (achievement)), and thematic roles (agent, object, instrument, source, goal, time, and place). Connections between these units – based on the coherence relations mentioned earlier – are calculated by a computational, connectionist model. The model is "supervised" by the results of a summarization experiment. The connectionist model will be explained. Based on the results of the model and the results of several experiments in the field of discourse processing, this paper will discuss which coherence relations play a major role in text coherence and which links can be found between these relations.
To download this paper, please return to Proceedings of the 1998 Postgraduate Conference