Disambiguating temporal connectors into TimeML relations

Abstract

The project is about learning temporal relations from unannotated text. This effort builds on the work of Lapata M. and Lascarides, A. (2006): Learning sentence-internal temporal relations, who developed a system that uses temporal connectors (after, before, while, when, as, once, until and since) in unannotated text to build a system to determine intra-sentential temporal relations. In an extension of this approach, they used their system to determine TimeML relations (before, includes, begins, ends and simultaneous) between events. Since temporal connectors do not translate one-to-one to TimeML relations, the main focus of this project is on disambiguating the temporal connectors into TimeML relations to preprocess the training data and use the system to directly learn the TimeML relations. This is done using a rule-based system and evaluated on the TimeBank corpus.