Towards Multiple Antecedent Coreference Resolution in Specialized Discourse

Towards Multiple Antecedent Coreference Resolution in Specialized Discourse Alicia Burga1, Sergio Cajal1, Joan Codina-Filb`a1, Leo Wanner1,2 1NLP Group, Department of Information and Communication Technologies Pompeu Fabra University Roc Boronat, 138, 08018 Barcelona, Spain alicia.burga, sergio.cajal, joan.codina, leo.wanner@upf.edu 2Catalan Institute for Research and Advanced Studies (ICREA) Abstract Despite the popularity of coreference resolution as a research topic, the overwhelming majority of the work in this area focused so far on single antecedence coreference only. Multiple antecedent coreference (MAC) has been largely neglected. This can be explained by the scarcity of the phenomenon of MAC in generic discourse. However, in specialized discourse such as patents, MAC is very dominant. It seems thus unavoidable to address the problem of MAC resolution in the context of tasks related to automatic patent material processing, among them abstractive summarization, deep parsing of patents, construction of concept maps of the inventions, etc. We present the first version of an operational rule-based MAC resolution strategy for patent material that covers the three major types of MAC: (i) nominal MAC, (ii) MAC with personal / relative pronouns, and MAC with reflexive / reciprocal pronouns. The evaluation shows that our strategy performs well in terms of precision and recall. Keywords: Multiple antecedent coreference, split antecedent coreference, coreference evaluation, patent processing 1. Introduction Coreference resolution has been a popular research topic for a considerable time now both in theoretical and in computational studies; see, among others, (Lasnik, 1989; L.Eguren and Fern´andez Soriano, 2007; Kayne, 2005) for the first and (Mitkov, 1999; Recasens et al., 2010; Recasens and Hovy, 2010) for the latter. It is an obligatory task for any language understanding application that goes be- yond surface-oriented parsing. However, the overwhelm- ing majority of the state-of-the-art coreference resolution works focused so far exclusively on one kind of corefer- ence, namely the single antecedent coreference illustrated in (1): (1) [John]i met Mary in New York. At that time, [he]i stud- ied Computer Science. Multiple antecedent coreference (henceforth MAC), or split antecedent coreference (Lasnik, 1989; Hornstein, 1999), as illustrated in the examples (2) and (3), has been largely ne- glected. (2) [John]i convinced [Mary]j to move to Barcelona. Now, [both]i+j enjoy living close to the sea. (3) [John]i, who adores [Mary]j from the first day, thinks that [they]i+j make a great couple. Some computational proposals that deal with coreference mention the phenomenon, but do not treat it (and do not justify in a satisfactory way the decision to ignore it ei- ther); see, e.g., (Mart´ınez-Barco and Palomar, 2011; Bald- win, 1997). Others treat it, but do not present the achieved performance in detail (Aone and Bennett, 1995; Kennedy and Boguraev, 1996). But, as already pointed out, most of the existing proposals (among them, e.g., (Ge et al., 1998; Refoufi, 2007)) do not even mention the phenomenon. The exclusion of MAC from coreference resolution is also re- flected in the annotation tools that are frequently used: ei- ther they are not implemented to directly capture this phe- nomenon, as, e.g., GATE (Cunningham et al., …

Content truncated to 500 words