Annotation of Temporal Relations Using Markov Logic Networks and Temporal Centering

Abstract

In recent years, annotating temporal relations has been a focus for many researchers in the area of Temporal Information Processing. By labelling these relations within text, we can identify the event occurrences and determine the order between them as well as other temporal information. Methods such as Markov Logic Networks have achieved good results when applied to this task. In this thesis we propose a combined approach of Markov Logic Networks with suggestions provided by Temporal Centering. We show the bene ts of both the probabilistic nature of Markov Logic Networks and the context based knowledge of Temporal Centering. By combining these two approaches, we allow the Markov Logic Networks to achieve a better understanding of the text, and therefore improve the results of its labelling for temporal relations. Experiments on the TempEval data showed that our approaches are competitive with the state-of-the-art systems submitted to the TempEval Evaluations.