Word Sense Induction

Word sense induction (WSI) is the unsupervised task of automatically identifying different meanings (senses) of ambiguous words based on their contextual usage. Current research focuses on leveraging contextualized language models, such as BERT, and employing techniques like hierarchical clustering, mutual information maximization, and substitution-based methods to improve sense induction accuracy across multiple languages, even low-resource ones. These advancements are crucial for improving various NLP applications, including lexical semantic change detection, dictionary creation, and building more robust multilingual resources like WordNets. The development of standardized benchmarks and evaluation metrics is also a key area of ongoing work, facilitating more reliable comparisons and reproducibility of results.