Extraction and generalisation of variables from scientific publications

Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 505–511, Lisbon, Portugal, 17-21 September 2015. c⃝2015 Association for Computational Linguistics. Extraction and generalisation of variables from scientific publications Erwin Marsi, Pinar ¨Ozt¨urk Department of Computer and Information Science Norwegian University of Science and Technology (NTNU) {emarsi,pinar}@idi.ntnu.no Abstract Scientific theories and models in Earth sci- ence typically involve changing variables and their complex interactions, including correlations, causal relations and chains of positive/negative feedback loops. Vari- ables tend to be complex rather than atomic entities and expressed as noun phrases containing multiple modifiers, e.g. oxygen depletion in the upper 500 m of the ocean or timing and magnitude of sur- face temperature evolution in the Southern Hemisphere in deglacial proxy records. Text mining from Earth science literature is therefore significantly different from biomedical text mining and requires dif- ferent approaches and methods. Our ap- proach aims at automatically locating and extracting variables and their direction of variation: increasing, decreasing or just changing. Variables are initially extracted by matching tree patterns onto the syntax trees of the source texts. Next, variables are generalised in order to enhance their similarity, facilitating hierarchical search and inference. This generalisation is ac- complished by progressive pruning of syn- tax trees using a set of tree transformation operations. Text mining results are pre- sented as a browsable variable hierarchy which allows users to inspect all mentions of a particular variable type in the text as well as any generalisations or specialisa- tions. The approach is demonstrated on a corpus of 10k abstracts of Nature publica- tions in the field of Marine science. We discuss experiences with this early proto- type and outline a number of possible im- provements and directions for future re- search. 1 Introduction Text mining of scientific literature originates from efforts to cope with the ever growing flood of pub- lications in biomedicine (Swanson, 1986; Swan- son, 1988; Swanson and Smalheiser, 1997; Hearst, 1999; Ananiadou et al., 2006; Zweigenbaum et al., 2007; Cohen and Hersh, 2005; Krallinger et al., 2008; Rodriguez-Esteban, 2009; Zweigen- baum and Demner-Fushman, 2009; Ananiadou et al., 2010; Simpson and Demner-Fushman, 2012; Ananiadou et al., 2014). Consequently the re- sulting approaches, methods, tools and applica- tions – as well as data, corpora and evaluation tasks – are rooted in the paradigm of biomedi- cal research and its conceptual framework. Typ- ical source text consists of abstracts from PubMed or full-text articles from PubMed Central. Stan- dard tasks include recognition, normalisation and mapping of biological entities (e.g., genes, pro- teins, drugs, symptoms and diseases), extraction of biological relations (e.g., protein-protein inter- action, disease-gene associations or drug-drug in- teraction) or bio-event extraction (e.g., regulation or inhibition events and their participants). There are extensive ontologies like the Gene Ontology (Consortium, 2001), annotated corpora like the GENIA (Kim et al., 2003) and BioInfer (Pyysalo et al., 2007) corpora and dedicated shared tasks in- cluding BioCreative (Hirschman et al., 2005) and BioNLP (Pyysalo et al., 2012). In short, there is a whole infrastructure …

Content truncated to 500 words