Syntactico-Semantic Analysis: A Hybrid Sentence Extraction Strategy for Automatic Summarization

Abstract

Automatic summarization systems often make use of sentence extraction methods to select significant content in texts. This paper presents two-sentence extraction strategies. The first one is based on a semantic analysis using word senses built by means of a network of lexical co-occurrences. The second one uses a combination of semantic and syntactical analysis to extract a set of relevant sentences. Both strategies are evaluated against a large corpus of newspaper and scientific articles, and their results are compared to Mead, a reference summarization tool. Evaluations results suggest that, inside the extractive summarization framework, the proposed hybrid strategy improves the relevance of the resulting summaries.