Extracting glossary sentences from scholarly articles: A comparative evaluation of pattern bootstrapping and deep analysis

Proceedings of the ACL-2012 Special Workshop on Rediscovering 50 Years of Discoveries, pages 55–65, Jeju, Republic of Korea, 10 July 2012. c⃝2012 Association for Computational Linguistics Extracting glossary sentences from scholarly articles: A comparative evaluation of pattern bootstrapping and deep analysis Melanie Reiplinger1 Ulrich Sch¨afer2 Magdalena Wolska1∗ 1Computational Linguistics, Saarland University, D-66041 Saarbr¨ucken, Germany 2DFKI Language Technology Lab, Campus D 3 1, D-66123 Saarbr¨ucken, Germany {mreiplin,magda}@coli.uni-saarland.de, ulrich.schaefer@dfki.de Abstract The paper reports on a comparative study of two approaches to extracting definitional sen- tences from a corpus of scholarly discourse: one based on bootstrapping lexico-syntactic patterns and another based on deep analysis. Computational Linguistics was used as the tar- get domain and the ACL Anthology as the corpus. Definitional sentences extracted for a set of well-defined concepts were rated by do- main experts. Results show that both meth- ods extract high-quality definition sentences intended for automated glossary construction. 1 Introduction Specialized glossaries serve two functions: Firstly, they are linguistic resources summarizing the ter- minological basis of a specialized domain. Sec- ondly, they are knowledge resources, in that they provide definitions of concepts which the terms de- note. Glossaries find obvious use as sources of ref- erence. A survey on the use of lexicographical aids in specialized translation showed that glossaries are among the top five resources used (Dur´an-Mu˜noz, 2010). Glossaries have also been shown to facil- itate reception of texts and acquisition of knowl- edge during study (Weiten et al., 1999), while self- explanation of reasoning by referring to definitions has been shown to promote understanding (Aleven et al., 1999). From a machine-processing point of view, glossaries may be used as input for domain ontology induction; see, e.g. (Bozzato et al., 2008). ∗Corresponding author The process of glossary creation is inherently de- pendent on expert knowledge of the given domain, its concepts and language. In case of scientific do- mains, which constantly evolve, glossaries need to be regularly maintained: updated and continually extended. Manual creation of specialized glossaries is therefore costly. As an alternative, fully- and semi-automatic methods of glossary creation have been proposed (see Section 2). This paper compares two approaches to corpus- based extraction of definitional sentences intended to serve as input for a specialized glossary of a scien- tific domain. The bootstrapping approach acquires lexico-syntactic patterns characteristic of definitions from a corpus of scholarly discourse. The deep ap- proach uses syntactic and semantic processing to build structured representations of sentences based on which ‘is-a’-type definitions are extracted. In the present study we used Computational Linguis- tics (CL) as the target domain of interest and the ACL Anthology as the corpus. Computational Linguistics, as a specialized do- main, is rich in technical terminology. As a cross- disciplinary domain at the intersection of linguistics, computer science, artificial intelligence, and mathe- matics, it is interesting as far as glossary creation is concerned in that its scholarly discourse ranges from descriptive informal to formal, including math- ematical notation. Consider the following two de- scriptions of Probabilistic Context-Free Grammar (PCFG): (1) A PCFG …

Content truncated to 500 words