Building Event Threads out of Multiple News Articles

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 958–967, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Building Event Threads out of Multiple News Articles Xavier Tannier LIMSI-CNRS Univ. Paris-Sud Orsay, France xavier.tannier@limsi.fr V´eronique Moriceau LIMSI-CNRS Univ. Paris-Sud Orsay, France moriceau@limsi.fr Abstract We present an approach for building multi- document event threads from a large corpus of newswire articles. An event thread is basi- cally a succession of events belonging to the same story. It helps the reader to contextual- ize the information contained in a single arti- cle, by navigating backward or forward in the thread from this article. A specific effort is also made on the detection of reactions to a particular event. In order to build these event threads, we use a cascade of classifiers and other modules, tak- ing advantage of the redundancy of informa- tion in the newswire corpus. We also share interesting comments con- cerning our manual annotation procedure for building a training and testing set1. 1 Introduction In this paper, we explore a new way of dealing with temporal relations between events. Our task is somewhat between multidocument summariza- tion and classification of temporal relations between events. We work with a large collection of En- glish newswire articles, where each article relates an event: the main topic of the article is a specific event, and other older events are mentioned in order to put it into perspective. Thus, we consider that an event is associated with an article and that defining temporal relations between articles is a way to define temporal relations between events. 1This work has been partially funded by French National Research Agency (ANR) under project Chronolines (ANR-10- CORD-010). We would like to thank the French News Agency (AFP) for providing us with the corpus. The task is to build a temporal graph of arti- cles, linked between each other by the following re- lations: • Same event, when two documents relate the same event, or when a document is an update of another one. • Continuation, when an event is the continua- tion or the consequence of a previous one. We also define a subset of continuation, called reaction, concerning a document relating the reac- tion of someone to another event. Some examples of these three classes will be given in Section 3. These relations can be represented by a directed graph where documents are vertices and relations are edges (as illustrated in all figures of this article). Figure 1 shows an example of such a graph. Press articles, and especially newswire articles, are characterized by an important redundancy of re- lated events. An important event2 is likely to be treated by several successive articles, which will give more and more details and update some num- bers (mainly, tragedy casualty updates, as shown in Figure 2). On the one hand, this redundancy is an issue since a system must not show duplicate infor- mation to the user; on the other hand, we …

Content truncated to 500 words