Multiple Alignment of Citation Sentences with Conditional Random Fields and Posterior Decoding

Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pp. 847–857, Prague, June 2007. c⃝2007 Association for Computational Linguistics Multiple Alignment of Citation Sentences with Conditional Random Fields and Posterior Decoding Ariel S. Schwartz∗ EECS, Computer Science Division UC Berkeley Berkeley, CA 94720-1776 sariel@cs.berkeley.edu Anna Divoli, Marti A. Hearst School of Information UC Berkeley Berkeley, CA 94720-4600 {hearst,divoli}@ischool.berkeley.edu Abstract In scientific literature, sentences that cite re- lated work can be a valuable resource for applications such as summarization, syn- onym identification, and entity extraction. In order to determine which equivalent en- tities are discussed in the various citation sentences, we propose aligning the words within these sentences according to semantic similarity. This problem is partly analogous to the problem of multiple sequence align- ment in the biosciences, and is also closely related to the word alignment problem in sta- tistical machine translation. In this paper we address the problem of multiple citation concept alignment by combining and mod- ifying the CRF based pairwise word align- ment system of Blunsom & Cohn (2006) and a posterior decoding based multiple se- quence alignment algorithm of Schwartz & Pachter (2007). We evaluate the algorithm on hand-labeled data, achieving results that improve on a baseline. 1 Introduction The scientific literature of biomedicine, genomics, and other biosciences is a rich, complex, and con- tinually growing resource. With appropriate infor- mation extraction and retrieval tools, bioscience re- searchers can use the contents of the literature to further their research goals. With online full text ∗Current address: Department of Bioengineering, Univer- sity of California, San Diego, La Jolla, CA 92093-0412. Email: sariel@ucsd.edu. of journal articles finally becoming the norm, new forms of citation analysis become possible. Nearly every statement in biology articles is backed up by at least one citation, and, conversely, it is quite common for papers in the bioscience do- main to be cited by 30–100 other papers. The cited facts are typically stated in a more concise way in the citing papers than in the original papers. Since the same facts are repeatedly stated in different ways in different papers, statistical models can be trained on existing citation sentences to identify similar facts in unseen text. Citation sentences also have the poten- tial to be useful for text summarization and database curation. Figure 1 shows an example of three differ- ent citation sentences to the same target paper. Most citation analysis work focuses on the cita- tion network structure, to determine which papers are most central, or uses co-citation analysis to de- termine which papers are similar to one another in content (White, 2004; Liu, 1993; Garfield, 1955; Lipetz, 1965; Giles et al., 1998). In this paper we focus instead on analyzing the sentences that sur- round the citations to related work, which we termed citances in Nakov et al. (2004). In that paper we note that one subproblem for using citances for au- tomated analysis is to identify the different concepts mentioned; a given …

Content truncated to 500 words