Intermediate-Task Training for Zero-Shot Cross-Lingual Transfer to Low-Resource Languages in X-MMR

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: Can intermediate-task training on English multimodal tasks (e.g., VQA or COCO) improve zero-shot cross-lingual transfer specifically for low-resource languages in X-MMR, as measured by relative F1 score gains compared to high-resource languages?

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