Automatic text summarizing based on sentence extraction: A statistical approach

Abstract

The present work describes a system for automatic summarizing of texts. Rather than focusing on abstractsC a hard NLP task of not asserted effectiveness the system produces extracts through selection of most important sentences in the original text. Statistical concepts are involved in order to evaluate the degree of significance of words, groups of words and sentences. Currently both Japanese and English texts can be treated. Procedures for computing importance, information content of sentences and measures of correlation between sentences are implemented. Comments are given on the feasibility of the approach and future developments.