Robustness of Intermediate-Task Training with English-Only vs. Multilingual Synthetic Data in Cross-Lingual XTREME 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: How does the robustness of intermediate-task training with English-only vs. multilingual synthetic data differ when evaluated on adversarial or out-of-domain cross-lingual XTREME tasks, using metrics like F1 score and accuracy?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 9.0/10.