Bag of Senses Versus Bag of Words: Comparing Semantic and Lexical Approaches on Sentence Extraction

Sentence extraction is a valuable technique for automatic summarization. This paper presents LIC2M’s first participation in TAC evaluation campaign (update summarization task). We describe two main extractive approaches for summarization. The semantic strategy makes use of a bag-of-senses to calculate sense concentration on each sentence. In the lexical strategy, source text sentences are ranked according to their lexical similarity against a topic statement represented as a bag-of-words. Both approaches are compared and the evaluation results analyzed. An alternative version of the semantic strategy is proposed, where sense concentration ranking takes into account syntactic dependencies.