Mining Scientific Terms and their Definitions: A Study of the ACL Anthology

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 780–790, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Mining Scientific Terms and their Definitions: A Study of the ACL Anthology Yiping Jin1 Min-Yen Kan1,2 Jun-Ping Ng1 Xiangnan He1 ∗ 1Department of Computer Science 2Interactive and Digital Media Institute National University of Singapore 13 Computing Drive, Singapore 117417 {yiping, kanmy, junping, xiangnan}@comp.nus.edu.sg Abstract This paper presents DefMiner, a supervised sequence labeling system that identifies scien- tific terms and their accompanying definitions. DefMiner achieves 85% F1 on a Wikipedia benchmark corpus, significantly improving the previous state-of-the-art by 8%. We exploit DefMiner to process the ACL An- thology Reference Corpus (ARC) – a large, real-world digital library of scientific arti- cles in computational linguistics. The re- sulting automatically-acquired glossary rep- resents the terminology defined over several thousand individual research articles. We highlight several interesting observations: more definitions are introduced for conference and workshop papers over the years and that multiword terms account for slightly less than half of all terms. Obtaining a list of popular defined terms in a corpus of computational lin- guistics papers, we find that concepts can of- ten be categorized into one of three categories: resources, methodologies and evaluation met- rics. 1 Introduction Technical terminology forms a key backbone in scientific communication. By coining formalized terminology, scholars convey technical information precisely and compactly, augmenting the dissemi- nation of scientific material. Collectively, scholarly ∗This research is supported by the Singapore National Re- search Foundation under its International Research Centre @ Singapore Funding Initiative and administered by the IDM Pro- gramme Office. compilation efforts result in reference sources such as printed dictionaries, ontologies and thesauri. While online versions are now common in many fields, these are still largely compiled manually, re- lying on costly human editorial effort. This leads to resources that are often outdated or stale with re- spect to the current state-of-the-art. Another indirect result of this leads to a second problem: lexical re- sources tend to be general, and may contain multi- ple definitions for a single term. For example, the term “CRF” connotes “Conditional Random Fields” in most modern computational linguistics literature; however, there are many definitions for this acronym in Wikipedia. Because only one correct sense ap- plies, readers may need to expend effort to identify the appropriate meaning of a term in context. We address both issues in this work by automat- ically extracting terms and definitions directly from primary sources: scientific publications. Since most new technical terms are introduced in scientific pub- lications, our extraction process addresses the bottle- neck of staleness. Second, since science is organized into disciplines and sub-disciplines, we can exploit this inherent structure to gather contextual informa- tion about a term and its definition. Aside from performance improvements, the key contributions of our work are in 1) recasting the problem as a sequence labeling task and exploring suitable learning architectures, 2) our proposal and validation of the the use of shallow parsing and …

Content truncated to 500 words