SVMs for the temporal expression chunking problem

Abstract

This technical report provides a description of the Temporal Expression Chunking problem, which consists in finding the mentions of temporal expressions in free text documents, and reports on the results obtained using SVM classifiers to tackle this task. We first define the task in detail, then present a brief survey of relevant literature, state our approach and finally present the results of several experiments. These include a comparison of the correct recognition rate achieved using different polynomial kernel degrees, sets of features, and sizes of the context window. We use 5-fold cross-validation on the ACE 2005 corpus for training and for testing.