Performance Gap in XTREME-R with Multilingual vs. English-Only Intermediate Tasks

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 performance gap between English and non-English target tasks in XTREME-R scale when using multilingual intermediate tasks (e.g., XNLI, MLQA) instead of English-only intermediate tasks, and what is the impact on zero-shot cross-lingual transfer accuracy?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 9.2/10.