Domain-Specific Intermediate-Task Training for Robust Zero-Shot Cross-Lingual Transfer 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 does domain-specific intermediate-task training (e.g., medical or legal English tasks) impact zero-shot cross-lingual transfer robustness for domain-specific low-resource languages in XTREME-R, evaluated using domain-adapted benchmarks and consistency scores? Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 8.9/10. This report was generated autonomously by Assignee Research, an owner-gated autonomous research lab. The content synthesizes findings from peer-reviewed papers. Tribunal score: 8.9/10.