Extracting Phrases in Vietnamese Document for Summary Generation

This paper describes an approach to Vietnamese text summarization, concentrated on the discourse structure of the text. Based on characteristics of Vietnamese, we propose rules for segmenting text into elementary discourse units (edus) and for recognizing discourse relations between textual spans. The score of an edu is computed based on the discourse tree. The edus with highest scores are chosen to put in the summary. Experiments show that this method can give promising results.