Using Word Embeddings for Unsupervised Acronym Disambiguation

Proceedings of the 27th International Conference on Computational Linguistics, pages 2610–2619 Santa Fe, New Mexico, USA, August 20-26, 2018. 2610 Using Word Embeddings for Unsupervised Acronym Disambiguation Jean Charbonnier Hochschule Hannover Expo Plaza 12 D-30539 Hannover jean.charbonnier @hs-hannover.de Christian Wartena Hochschule Hannover Expo Plaza 12 D-30539 Hannover christian.wartena @hs-hannover.de Abstract Scientific papers from all disciplines contain many abbreviations and acronyms. In many cases these acronyms are ambiguous. We present a method to choose the contextual correct definition of an acronym that does not require training for each acronym and thus can be applied to a large number of different acronyms with only few instances. We constructed a set of 19,954 examples of 4,365 ambiguous acronyms from image captions in scientific papers along with their contextually correct definition from different domains. We learn word embeddings for all words in the corpus and compare the averaged context vector of the words in the expansion of an acronym with the weighted average vector of the words in the context of the acronym. We show that this method clearly outperforms (classical) cosine similarity. Furthermore, we show that word embeddings learned from a 1 billion word corpus of scientific texts outperform word embeddings learned from much larger general corpora. 1 Motivation Scientific papers usually contain a high number abbreviations and acronyms that might be very general or very specific for a certain domain and that might be very common in that domain or that are constructed ad hoc. For various tasks, like retrieval or information extraction we need the expanded form of the acronyms.1 In most cases we can find the definition for an acronym in the paper. Finding this definition automatically is not as trivial as it might sound and there is a lot of literature on this topic. However, a lot of cases remain where we cannot find such a definition, because the extraction method fails or because there is no definition in the paper. In these cases we can use a definition found elsewhere. As already noted by Chang et al. (2002) in these cases acronyms often turn out to be ambiguous and the correct definition has to be chosen by considering the context of the acronym. Exactly this task is the topic of the present paper. Thus, although the goal of our work is acronym expansion, the work is more related to word sense disambiguation (WSD) than to typical work on acronym resolution. The main difference with WSD is that we do not have dictionaries with description of possible senses. Instead the possible expansions of the acronym directly represent the different senses. In the Project NOA (Reuse of Open Access Images, http://noa.wp-hs-hannover.de) we analyze captions from images in scientific papers to collect appropriate metadata for those images (Charbonnier et al., 2018). In order to get maximal information from the usually short image captions we need to expand the acronyms. Moreover, the acronyms are often very specific technical terms that give a lot of information on the displayed image. Since the focus …

Content truncated to 500 words