Pre-training on diverse English tasks for multilingual adversarial robustness in zero-shot cross-lingual settings

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: Does pre-training on a diverse set of intermediate English tasks improve the robustness of multilingual models to adversarial attacks in zero-shot cross-lingual settings, as measured by accuracy drop on the XTREME-R benchmark under back-translation perturbations?

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