Scaling of Intermediate-Task Training in Multilingual vs. English-Only Models for Zero-Shot Cross-Lingual Performance

Intermediate-task training---fine-tuning a pretrained model on an intermediate task before fine-tuning again on the target task---often improves model performance substantially on language understanding tasks in monolingual English settings. We investigate whether English intermediate-task training is still helpful on non-English target tasks. Using nine intermediate language-understanding tasks, we evaluate intermediate-task transfer in a zero-shot cross-lingual setting on the XTREME benchmark. We see large improvements from intermediate training on the BUCC and Tatoeba sentence retrieval tas

Research goal: How does the scaling of intermediate-task training with English-only data compare to multilingual intermediate-task training when evaluated on XTREME-R across different model families (e.g., mT5, XLM-R, BLOOM) in terms of zero-shot cross-lingual performance metrics like XTREME-R score?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 7.7/10.