XL-WSD: An Extra-Large and Cross-Lingual Evaluation Framework for Word Sense Disambiguation

Transformer-based architectures brought a breeze of change to Word Sense Disambiguation (WSD), improving models' performances by large margin. The fast development new approaches has been further encouraged well-framed evaluation suite for English, which allowed their be kept track and compared fairly. However, other languages have remained largely unexplored, as testing data are available few only the setting is rather matted. In this paper, we untangle situation proposing XL-WSD, cross-lingual benchmark WSD task featuring sense-annotated test sets in 18 from six different linguistic families, together with language-specific silver training data. We leverage XL-WSD datasets conduct an extensive neural knowledge-based approaches, including most recent multilingual language models. Results show that zero-shot knowledge transfer across promising research direction within field, especially when considering low-resourced where pre-trained models still perform poorly. make code performing experiments at https://sapienzanlp.github.io/xl-wsd/.