Multilingual Coreference Resolution via Cycle-Consistent Machine Translation

In three linesMultilingual coreference resolution pipeline using cycle-consistent machine translation (English → target language → English) to generate training data. Translation quality validated via cosine similarity in BERT latent space. Significant performance gains on 4 low-resource languages.

Summary generated by Claude — human-verified