SensEmBERT: Context-Enhanced Sense Embeddings for Multilingual Word Sense Disambiguation

Contextual representations of words derived by neural language models have proven to effectively encode the subtle distinctions that might occur between different meanings of the same word. However, these representations are not tied to a semantic network, hence they leave the word meanings implicit and thereby neglect the information that can be derived from the knowledge base itself. In this paper, we propose SENSEMBERT, a knowledge-based approach that brings together the expressive power of language modelling and the vast amount of knowledge contained in a semantic network to produce high-quality latent semantic representations of word meanings in multiple languages. Our vectors lie in a space comparable with that of contextualized word embeddings, thus allowing a word occurrence to be easily linked to its meaning by applying a simple nearest neighbour approach. We show that, whilst not relying on manual semantic annotations, SENSEMBERT is able to either achieve or surpass state-of-the-art results attained by most of the supervised neural approaches on the English Word Sense Disambiguation task. When scaling to other languages, our representations prove to be equally effective as their English counterpart and outperform the existing state of the art on all the Word Sense Disambiguation multilingual datasets. The embeddings are released in five different languages at http://sensembert.org . Introduction Word Sense Disambiguation (WSD) is the task of associating the occurrence of a word in a text with its correct meaning from a predefined inventory of senses (Navigli 2009). Over the years, two distinct lines of research have been developed to tackle this problem: supervised and knowledgebased WSD. On the one hand, supervised models rely on semantically-annotated corpora for training (Raganato, Delli Bovi, and Navigli 2017; Kumar et al. 2019; Bevilacqua and Navigli 2019) , while, on the other hand, knowledgebased systems employ graph-based algorithms on semantic networks to find the set of meanings that better disambiguate the input words (Moro, Raganato, and Navigli 2014; Agirre, de Lacalle, and Soroa 2014) . Even though supervised approaches have proved to achieve better performance,