Comparative Analysis of Multimodal and Text-Only Intermediate Tasks for Zero-Shot Cross-Lingual Transfer on M3C

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 the performance of multimodal intermediate-task training (e.g., image-text alignment) on the M3C benchmark compare to text-only intermediate tasks for zero-shot cross-lingual transfer in terms of retrieval accuracy and inference efficiency?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 8.9/10.