Sense-Annotated Corpora for Word Sense Disambiguation in Multiple Languages and Domains

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 5905–5911 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 5905 Sense-Annotated Corpora for Word Sense Disambiguation in Multiple Languages and Domains Bianca Scarlini, Tommaso Pasini, Roberto Navigli Sapienza NLP Group Department of Computer Science Sapienza University of Rome {scarlini, pasini, navigli}@di.uniroma1.it Abstract The knowledge acquisition bottleneck problem dramatically hampers the creation of sense-annotated data for Word Sense Disambiguation (WSD). Sense-annotated data are scarce for English and almost absent for other languages. This limits the range of action of deep- learning approaches, which today are at the base of any NLP task and are hungry for data. We mitigate this issue and encourage further research in multilingual WSD by releasing to the NLP community five large datasets annotated with word senses in five different languages, namely, English, French, Italian, German and Spanish, and 5 distinct datasets in English, each for a different semantic domain. We show that supervised WSD models trained on our data attain higher performance than when trained on other automatically- created corpora. We release all our data containing more than 15 million annotated instances in 5 different languages at http:// trainomatic.org/onesec. Keywords: Word Sense Disambiguation, Multilinguality, Semantics 1. Introduction Nowadays, deep neural models use data as their “fuel”, and have proved to attain better performance the larger the train- ing corpora they are provided with. Unfortunately, pro- ducing large manually-curated corpora is an expensive and time-consuming task. This is especially an issue for the field of Natural Language Processing (NLP), where data is usually tied to a specific language, thereby further in- creasing the time and cost of producing large-scale anno- tated corpora. An area of NLP that particularly suffers from such a problem is Word Sense Disambiguation (WSD), that is, the task of assigning a meaning to a word in a given context (Navigli, 2009). It is not by chance, in fact, that corpora annotated with senses are limited for English and almost absent for other languages. Indeed, while various sense-annotated corpora do exist for English, most of these are limited in terms of words and senses covered: MASC (Passonneau et al., 2012), a corpus providing 1M instances manually-annotated with senses, covers only 100 distinct words, OntoNotes (Hovy et al., 2006), instead, provides an- notated examples for only 6500 concepts of less than 3500 words. SemCor (Miller et al., 1993) is the largest manually- annotated English corpus currently available. It contains roughly 40K sentences and 200K content words annotated with their corresponding meanings. Despite being the stan- dard training corpus for WSD, SemCor’s main limitation resides in its low coverage of the English vocabulary in terms of both words and meanings. In fact, it provides an- notations for only 22K unique lexemes, i.e., lemma and POS tag pairs, which cover not even 15% of the words comprised in WordNet (Miller et al., 1990), i.e., the largest and most-used electronic dictionary for English. Moreover, as we shift our focus towards …

Content truncated to 500 words