Order Variation in Intermediate-Task Training for mT5-Large's Cross-Lingual Inference Accuracy

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: What is the effect of varying the order of intermediate-task training (e.g., XTREME-U followed by XTREME-R vs. XTREME-R followed by XTREME-U) on the zero-shot cross-lingual inference accuracy of mT5-Large on XTREME-R's natural language inference tasks?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 9.2/10.