Meeting TempEval-2

Abstract

Temporal expressions are one of the important structures in natural language. In order to understand text, temporal expressions have to be identified and normalized by providing ISObased values. In this paper we present a shallow approach for automatic recognition of temporal expressions based on a supervised machine learning approach trained on an annotated corpus for temporal information, namely TimeBank. Our experiments demonstrate a performance level comparable to a rule-based implementation and achieve the scores of 0.872, 0.836 and 0.852 for precision, recall and F1-measure for the detection task respectively, and 0.866, 0.796, 0.828 when an exact match is required.