Pre-trained language models (LMs) encode rich information about linguistic structure but their knowledge about lexical polysemy remains unclear. We propose a novel experimental setup for analyzing this knowledge in LMs specifically trained for differ
ent languages (English, French, Spanish, and Greek) and in multilingual BERT. We perform our analysis on datasets carefully designed to reflect different sense distributions, and control for parameters that are highly correlated with polysemy such as frequency and grammatical category. We demonstrate that BERT-derived representations reflect words' polysemy level and their partitionability into senses. Polysemy-related information is more clearly present in English BERT embeddings, but models in other languages also manage to establish relevant distinctions between words at different polysemy levels. Our results contribute to a better understanding of the knowledge encoded in contextualized representations and open up new avenues for multilingual lexical semantics research.
We propose a novel method of homonymy-polysemy discrimination for three Indo-European Languages (English, Spanish and Polish). Support vector machines and LASSO logistic regression were successfully used in this task, outperforming baselines. The fea
ture set utilised lemma properties, gloss similarities, graph distances and polysemy patterns. The proposed ML models performed equally well for English and the other two languages (constituting testing data sets). The algorithms not only ruled out most cases of homonymy but also were efficacious in distinguishing between closer and indirect semantic relatedness.
Deciding whether a semantically ambiguous word is homonymous or polysemous is equivalent to establishing whether it has any pair of senses that are semantically unrelated. We present novel methods for this task that leverage information from multilin
gual lexical resources. We formally prove the theoretical properties that provide the foundation for our methods. In particular, we show how the One Homonym Per Translation hypothesis of Hauer and Kondrak (2020a) follows from the synset properties formulated by Hauer and Kondrak (2020b). Experimental evaluation shows that our approach sets a new state of the art for homonymy detection.
