Automatic Annotation of Semantic Term Types in the Complete ACL Anthology Reference Corpus

Automatic Annotation of Semantic Term Types in the Complete ACL Anthology Reference Corpus Anne-Kathrin Schumann, Héctor Martínez Alonso ProTechnology GmbH, Dresden, Germany, Thomson Reuters Labs, Toronto, Canada annek_schumann@gmx.de, hector.martinezalonso@thomsonreuters.com Abstract In the present paper, we present an automated tagging approach aimed at enhancing a well-known resource, the ACL Anthology Reference Corpus, with semantic class labels for more than 20,000 technical terms that are relevant to the domain of computational linguistics. We use state-of-the-art classification techniques to assign semantic class labels to technical terms extracted from several reference term lists. We also sketch a set of research questions and approaches directed towards the integrated analysis of scientific corpora. To this end, we query the data set resulting from our annotation effort on both the term and the semantic class level level. Keywords: semantic labeling, terminology, history of science 1. Introduction Science changes continually: While certain research top- ics may be in a state of stagnation or decline, other re- search fronts move forward rapidly. However, even “dor- mant” (Menard, 1971) science can regain importance if new data is produced or methods are developed to tackle unresolved research problems. Scientific thought exhibits intricate evolutionary patterns (Fleck, 1980) and paradigm change (Kuhn, 1962) can affect the structure and outline of a whole discipline. The availability of large collections of digitized scientific text enables systematic studies of the processes that drive scientific development. Recent years have seen a no- table increase in quantitative studies of scientific text col- lections, e. g. Hall et al. (2008), Gupta and Manning (2011), Michaelis et al. (2013), Mariani et al. (2014), Babko-Malaya et al. (2015), Schumann and QasemiZadeh (2015b), Asooja et al. (2016), Francopoulo et al. (2016), Schumann (2016), Heyer et al. (2016). The present study is a contribution to this research strand. Our work centers on the use of semantic labeling tech- niques for the automatic enhancement of a small corpus of manual term and semantic class annotations. The ulti- mate goal of our work, however, is to use this information as one feature in the profiling of scientific papers, commu- nities, and disciplines. In using semantic class labels as one source of information, we take a macro- rather than a micro-perspective: While individual words and terms are certainly indicators of scientific trends and developments, it is necessary to relate them to more coarse-grained cate- gories for an overall view of a scientific discipline. Seman- tic class annotations allow us to answer detailed questions about the evolution of computational linguistics over time. The data set described in this paper is made available to the research community1. 2. Motivation and Related Work The present investigation is conceptually related to earlier studies dedicated to the lexical analysis of diachronic cor- 1https://github.com/anetschka/ complingterm. pora. Since the well-known work by Hall et al. (2008), topic modeling has been widely employed in the analysis of diachronic data. Topic modeling, however, has the dis- advantage that it is ignorant to the concept of domain rele- vance. Topics, therefore, have to be painstakingly inferred …

Content truncated to 500 words