The Role of Semantics in Automatic Summarization: A Feasibility Study

State-of-the-art methods in automatic summarization rely almost exclusively on extracting salient sentences from input texts. Such extractive methods succeed in producing summaries which capture salient information but fail to produce fluent and coherent summaries. Recent progress in robust semantic analysis makes the application of semantic techniques to summarization relevant. We review in this paper areas in the field of summarization that can benefit from the introduction of the type of semantic analysis that has become available. The main pain points that semantic information can alleviate are: aiming for more fluent summaries by exploiting logical form representation of the source text; identifying salient information and avoiding redundancy by relying on textual entailment and paraphrase identification; and generating a coherent summary while relying on rhetorical structure and discourse structure information extracted from the source documents. In addition, we review the possibility to perform automatic Pyramid evaluation of summarization quality that relies on robust semantic similarity measures.