Cross-lingual transfer robustness of intermediate-task training against domain shifts in XTREME-R

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 robust is the cross-lingual transfer benefit of English intermediate-task training against domain shifts, as measured by performance drops on XTREME-R tasks from high-resource to low-resource languages when using models of varying sizes (e.g., 1B, 10B, 175B parameters)?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 8.7/10.