Event representation across genre

Abstract

This paper describes an approach for investigating the representation of events and their distribution in a corpus. We collect and analyze statistics about subject-verb-object triplets and their content, which helps us compare corpora belonging to the same domain but to different genre/text type. We argue that event structure is strongly related to the genre of the corpus, and propose statistical properties that are able to capture these genre differences. The results obtained can be used for the improvement of Information Extraction. 1