Pronominal anaphora resolution for text summarisation

The assumption of term-based summarisation method is that the importance of a sentence can be determined by the importance of the words it contains. One drawback of these methods is that they usually consider the words in isolation, ignoring relations such as anaphoric links between them. This paper investigates to what extent the integration of pronominal anaphora resolution into the summarisation process can improve the informativeness of the automatically produced summaries. Evaluation of three anaphora resolution methods plus three baselines on a corpus of journal articles shows that anaphora resolution can have a beneficial effect on informativeness. In addition, one experiment which uses a simulated anaphora resolver with predefined accuracy is performed in order to demonstrate that term-based summarisation can benefit from anaphora resolution, but that high accuracy methods are necessary.