A Weakly-supervised Approach to Argumentative Zoning of Scientific Documents

Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 273–283, Edinburgh, Scotland, UK, July 27–31, 2011. c⃝2011 Association for Computational Linguistics A Weakly-supervised Approach to Argumentative Zoning of Scientific Documents Yufan Guo Computer Laboratory University of Cambridge, UK yg244@cam.ac.uk Anna Korhonen Computer Laboratory University of Cambridge, UK alk23@cam.ac.uk Thierry Poibeau LaTTiCe, UMR8094 CNRS & ENS, France thierry.poibeau@ens.fr Abstract Argumentative Zoning (AZ) – analysis of the argumentative structure of a scientific paper – has proved useful for a number of informa- tion access tasks. Current approaches to AZ rely on supervised machine learning (ML). Requiring large amounts of annotated data, these approaches are expensive to develop and port to different domains and tasks. A poten- tial solution to this problem is to use weakly- supervised ML instead. We investigate the performance of four weakly-supervised clas- sifiers on scientific abstract data annotated for multiple AZ classes. Our best classifier based on the combination of active learning and self- training outperforms our best supervised clas- sifier, yielding a high accuracy of 81% when using just 10% of the labeled data. This re- sult suggests that weakly-supervised learning could be employed to improve the practical applicability and portability of AZ across dif- ferent information access tasks. 1 Introduction Many practical tasks require accessing specific types of information in scientific literature. For example, a reader of scientific literature may be looking for information about the objective of the study in ques- tion, the methods used in the study, the results ob- tained, or the conclusions drawn by authors. Sim- ilarly, many Natural Language Processing (NLP) tasks focus on the extraction of specific types of in- formation in documents only. To date, a number of approaches have been pro- posed for sentence-based classification of scien- tific literature according to categories of information structure (or discourse, rhetorical, argumentative or conceptual structure, depending on the framework in question). Some of these classify sentences ac- cording to typical section names seen in scientific documents (Lin et al., 2006; Hirohata et al., 2008), while others are based e.g. on argumentative zones (Teufel and Moens, 2002; Mizuta et al., 2006; Teufel et al., 2009), qualitative dimensions (Shatkay et al., 2008) or conceptual structure (Liakata et al., 2010) of documents. The best of current approaches have yielded promising results and proved useful for information retrieval, information extraction and summarization tasks (Teufel and Moens, 2002; Mizuta et al., 2006; Tbahriti et al., 2006; Ruch et al., 2007). How- ever, relying on fully supervised machine learning (ML) and a large body of annotated data, existing approaches are expensive to develop and port to dif- ferent scientific domains and tasks. A potential solution to this bottleneck is to de- velop techniques based on weakly-supervised ML. Relying on a small amount of labeled data and a large pool of unlabeled data, weakly-supervised techniques (e.g. semi-supervision, active learning, co/tri-training, self-training) aim to keep the advan- tages of fully supervised approaches. They have been applied to a wide range of NLP tasks, includ- ing …

Content truncated to 500 words