Hybrid Intermediate-Task Training for Robust mT5-3B in Zero-Shot Cross-Lingual Understanding

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: Does hybrid intermediate-task training improve the robustness of mT5-3B against adversarial perturbations in zero-shot cross-lingual understanding tasks, as evaluated by accuracy on XTREME-R's adversarial subsets?

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