Scaling Effects of Model Size on Zero-Shot Cross-Lingual Robustness to Adversarial Examples

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 the scaling of model size (e.g., from mT5-base to mT5-xxl) in zero-shot cross-lingual transfer affect the robustness to adversarial examples differently with and without intermediate-task training, as measured by accuracy drop on the PAWS-X benchmark?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 9.4/10.