Whose Idea Was This, and Why Does it Matter? Attributing Scientific Work to Citations

Proceedings of NAACL HLT 2007, pages 316–323, Rochester, NY, April 2007. c⃝2007 Association for Computational Linguistics Whose idea was this, and why does it matter? Attributing scientific work to citations Advaith Siddharthan & Simone Teufel Natural Language and Information Processing Group University of Cambridge Computer Laboratory {as372,sht25}@cl.cam.ac.uk Abstract Scientific papers revolve around cita- tions, and for many discourse level tasks one needs to know whose work is being talked about at any point in the discourse. In this paper, we in- troduce the scientific attribution task, which links different linguistic expres- sions to citations. We discuss the suitability of different evaluation met- rics and evaluate our classification ap- proach to deciding attribution both in- trinsically and in an extrinsic evalua- tion where information about scientific attribution is shown to improve per- formance on Argumentative Zoning, a rhetorical classification task. 1 Introduction In the recent past, there has been a focus on information management from scientific litera- ture. In the genetics domain, for instance, in- formation extraction of genes and gene–protein interactions helps geneticists scan large amounts of information (e.g., as explored in the TREC Genomics track (Hersh et al., 2004)). Elsewhere, citation indexes (Garfield, 1979) provide biblio- metric data about the frequency with which par- ticular papers are cited. The success of citation indexers such as CiteSeer (Giles et al., 1998) and Google Scholar relies on the robust detection of formal citations in arbitrary text. In bibli- ographic information retrieval, anchor text, i.e., the context of a citation can be used to charac- terise (index) the cited paper using terms out- side of that paper (Bradshaw, 2003); O’Connor (1982) presents an approach for identifying the area around citations where the text focuses on that citation. And automatic citation classifi- cation (Nanba and Okumura, 1999; Teufel et al., 2006) determines the function that a cita- tion plays in the discourse. For such information access and retrieval pur- poses, the relevance of a citation within a paper is often crucial. One can estimate how impor- tant a citation is by simply counting how often it occurs in the paper. But as Kim and Webber (2006) argue, this ignores many expressions in text which refer to the cited author’s work but which are not as easy to recognise as citations. They address the resolution of instances of the third person personal pronoun “they” in astron- omy papers: it can either refer to a citation or to some entities that are part of research within the paper (e.g., planets or galaxies). Several appli- cations should profit in principle from detecting connections between referring expressions and citations. For instance, in citation function clas- sification, the task is to find out if a citation is described as flawed or as useful. Consider: Most computational models of discourse are based primarily on an analysis of the intentions of the speakers [Cohen and Perrault, 1979][Allen and Perrault, 1980][Grosz and Sidner, 1986]WEAK. The speaker will form intentions based on his goals and then act on these intentions, pro- ducing …

Content truncated to 500 words