Scale Effects of Pretrained Multilingual Models on Zero-Shot XTREME-R Performance with Intermediate-Task Training

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 scale of pretrained multilingual models (e.g., 100M vs. 1B parameters) interact with cross-lingual intermediate-task training to influence zero-shot performance on XTREME-R, as measured by composite benchmark scores?

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