Cross-lingual Transfer Performance in Multilingual vs. English-Only Intermediate-Task Training for 1B and 10B Parameter Models

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 intermediate-task training on multilingual benchmarks like XTREME-R compare to English-only training in zero-shot cross-lingual transfer performance for models with 1B vs. 10B parameters, measured by average F1 scores across XTREME-MR tasks?

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