Multilingual Representation Distillation with Contrastive Learning

Explore multilingual sentence representations and their application in cross-lingual information retrieval through this 23-minute conference talk by Steven Tan from Johns Hopkins University's Center for Language & Speech Processing. Delve into the integration of contrastive learning with multilingual representation distillation for quality estimation of parallel sentences. Discover how this approach enhances multilingual similarity search and corpus filtering tasks, particularly in low-resource languages. Learn about the significant performance improvements achieved over previous sentence encoders like LASER, LASER3, and LaBSE, as demonstrated through extensive experiments.