Experiment with Discourse Structure Builder engine

Abstract

In this paper, some experiments with a Rule-Based Discourse Structure Builder (DSB) Engine are presented. The DSB discussed here is a Rule-Based Engine that identifies discourse relations based on centering status of entities in the text, rather than on cohesive devices. In the design of the DSB Engine some abstract notions have been defined. Addressing these notions according to a set of theories in natural language processing provides an engine for revealing discourse structure. Here in this paper, experiments with one of the possible settings are presented. The results show that the engine reveals discourse structures of the texts significantly. Although the Rule Set that was used in this experiment was a hand-tuned one, it produced Discourse Trees with 80% accuracy comparing to the ones suggested by human annotators.