Using Tree Kernels for Classifying Temporal Relations between Events

Using Tree Kernels for Classifying Temporal Relations between Events ∗∗∗∗ Seyed Abolghasem Mirroshandel, Gholamreza Ghassem-Sani, and Mahdy Khayyamian Department of Computer Engineering, Sharif University of Technology Azadi Ave., 11155-9517 Tehran, Iran mirroshandel@ce.sharif.edu, sani@sharif.edu, khayyamian@ce.sharif.edu Abstract. The ability to accurately classify temporal relations between events is an important task in a large number of natural language processing and text mining applications such as question answering, summarization, and language specific information retrieval. In this paper, we propose an improved way of classifying temporal relations, using support vector machines (SVM). Along with gold-standard corpus features, the proposed method aims at exploiting useful syntactic features, which are automatically generated, to improve accuracy of the SVM classification method. Accordingly, a number of novel kernel functions are introduced and evaluated for temporal relation classification. Our evaluations clearly demonstrate that adding syntactic features results in a considerable performance improvement over the state of the art method, which merely employs gold-standard features. Keywords: Temporal Relations between Events, Classification, Information Retrieval, Text Mining Copyright 2009 by Seyed Abolghasem Mirroshandel, Gholamreza Ghassem-Sani, and Mahdy Khayyamian 1 Introduction In recent years, many progresses have been made in natural language processing (NLP). Combining statistical and symbolic methods plays a significant role in these advances. Tasks such as part-of-speech tagging, morphological analysis, parsing, and named entity recognition have been addressed with satisfactory results (Mani et al., 2006). Problems such as temporal information processing that requires a deep semantic analysis are yet to be addressed. Lately, the increasing attention in practical NLP applications such as question answering, information extraction, and summarization have resulted in an increasing demand for temporal information processing (Tatu and Srikanth, 2008). In question answering, one would expect the system to answer questions such as "when an event occurred", or "what is the chronological order between some desired events". In text summarization, especially in multi-document type, knowing the order of events is a useful source for merging related information correctly. It is also the case that in some information extraction applications, the temporal information between events can be very useful and effective (Alonso, 2009). Temporal information is usually encoded in the textual description of some events. For a given ordered pair of components ( x1, x2) , where 1x and 2x are times or events, a temporal information processing system tries to identify the type of relation ir that temporally links 1x to 2x . The type of relation ir can be one of the 13 types proposed by James Allen (Allen, 1984). For example, in the sentence "Ocean Drilling said (e22) it will offer (e23) 15% to 20% of the contract-drilling business through an initial public offering (e25) in the near future (t67). (wsj_313), there are some relations between pairs (e23, e25), and (e25, t67). The task is to automatically tag these pairs with relations INCLUDES and BEFORE, respectively. 355 23rd Pacific Asia Conference on Language, Information and Computation, pages 355–364 With recent construction of the Timebank corpus (Pustejovsky et al, 2003), the efficiency of different machine learning methods can …

Content truncated to 500 words