A Short Survey on Sense-Annotated Corpora

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 5759–5765 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 5759 A Short Survey on Sense-Annotated Corpora Tommaso Pasini1, Jose Camacho-Collados2 1SapienzaNLP Group, Department of Computer Science, Sapienza University of Rome, Italy 2School of Computer Science and Informatics, Cardiff University, United Kingdom 1pasini@di.uniroma1.it, 2camachocolladosj@cardiff.ac.uk Abstract Large sense-annotated datasets are increasingly necessary for training deep supervised systems in Word Sense Disambiguation. However, gathering high-quality sense-annotated data for as many instances as possible is a laborious and expensive task. This has led to the proliferation of automatic and semi-automatic methods for overcoming the so-called knowledge-acquisition bottleneck. In this short survey we present an overview of sense-annotated corpora, annotated either manually- or (semi)automatically, that are currently available for different languages and featuring distinct lexical resources as inventory of senses, i.e. WordNet, Wikipedia, BabelNet. Furthermore, we provide the reader with general statistics of each dataset and an analysis of their specific features. Keywords: Word Sense Disambiguation, Semantics, Corpus Creation, Multilinguality 1. Introduction Word Sense Disambiguation (WSD) is a key task in Nat- ural Language Understanding. It consists in assigning the appropriate meaning from a pre-defined sense inventory to a word in context(Navigli, 2009). While knowledge-based approaches to this task have been proposed (Agirre et al., 2014; Moro et al., 2014; Butnaru et al., 2017; Chaplot and Salakhutdinov, 2018), supervised approaches (Zhong and Ng, 2010; Melamud et al., 2016; Iacobacci et al., 2016; K˚ageb¨ack and Salomonsson, 2016; Luo et al., 2018) have been more effective in terms of performance when sense- annotated corpora are available (Raganato et al., 2017a; Huang et al., 2019). Unfortunately, obtaining such data is heavily time-consuming and expensive (Schubert, 2006), and a reasonable amount of manually-annotated instances are available for English only (Miller et al., 1993a). One of the first attempts towards building large sense- annotated corpora was SemCor (Miller et al., 1993a) which contains instances annotated with senses from WordNet (Fellbaum, 1998). Since then, several semi-automatic and automatic approaches have also been proposed (Taghipour and Ng, 2015a; Delli Bovi et al., 2017; Pasini and Nav- igli, 2018). These automatic efforts tend to produce nois- ier annotations, but their coverage has been shown to lead to better supervised and semi-supervised WSD systems (Taghipour and Ng, 2015b; Otegi et al., 2016; Raganato et al., 2016; Yuan et al., 2016; Delli Bovi et al., 2017; Pasini and Navigli, 2017), as well as to learn effective embedded representations for senses (Iacobacci et al., 2015; Flekova and Gurevych, 2016; Mancini et al., 2017). Nevertheless, each of the aforementioned datasets come with its own for- mat, hence making it complicated to merge them or mov- ing from one to another. Vial et al. (2018) tackled this specific problem and, following Raganato et al. (2017a), proposed UFSAC, a unified repository of several sense- annotated corpora all with the same format, thus making it easier develop and test WSD models on different datasets. In this survey we present the main approaches …

Content truncated to 500 words