BabelBERT: Massively Multilingual Transformers Meet a Massively Multilingual Lexical Resource

While pretrained language models (PLMs) primarily serve as general purpose text encoders that can be fine-tuned for a wide variety of downstream tasks, recent work has shown that they can also be rewired to produce high-quality word representations (i.e., static word embeddings) and yield good performance in type-level lexical tasks. While existing work primarily focused on lexical specialization of PLMs in monolingual and bilingual settings, in this work we expose massively multilingual transformers (MMTs, e.g., mBERT or XLM-R) to multilingual lexical knowledge at scale, leveraging BabelNet as the readily available rich source of multilingual and cross-lingual type-level lexical knowledge. Concretely, we leverage BabelNet's multilingual synsets to create synonym pairs across $50$ languages and then subject the MMTs (mBERT and XLM-R) to a lexical specialization procedure guided by a contrastive objective. We show that such massively multilingual lexical specialization brings massive gains in two standard cross-lingual lexical tasks, bilingual lexicon induction and cross-lingual word similarity, as well as in cross-lingual sentence retrieval. Crucially, we observe gains for languages unseen in specialization, indicating that the multilingual lexical specialization enables generalization to languages with no lexical constraints. In a series of subsequent controlled experiments, we demonstrate that the pretraining quality of word representations in the MMT for languages involved in specialization has a much larger effect on performance than the linguistic diversity of the set of constraints. Encouragingly, this suggests that lexical tasks involving low-resource languages benefit the most from lexical knowledge of resource-rich languages, generally much more available.