A Computational Approach for Sentence Planning

Abstract

Sentence planning is a critical processing closely related to the coherence and fluency of a natural language system. After a discourse is planned, there are still several open issues before surface sentence generation. By considering the great variety of human languages, a discourse should be realized by alternative sentential representations according to its context. This research was motivated by this concern and the result successfully justified the necessity of sentence planning. In this paper, a computational approach is presented to coherently realize a discourse topic by alternative sentence representations according to its discourse context. This is a critical quality issue from the design perspective of natural language systems.