Dataset Size Scaling and Multimodal vs. Text-Only Transfer Performance in Zero-Shot Cross-Lingual Tasks

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: To what extent does scaling the size of the target language dataset (e.g., varying the number of samples in XQuAD) impact the relative performance gains of multimodal vs. text-only intermediate-task training in zero-shot cross-lingual transfer, as measured by EM/F1 scores?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 9.3/10.