Comparative Scaling of High-Resource Non-English and English Intermediate Tasks for XTREME-R Under Fixed Compute

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 high-resource non-English intermediate tasks compare to scaling the number of English intermediate tasks in terms of XTREME-R benchmark performance when controlling for total training compute?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 8.3/10.