Fine-tuning Pretrained Models with Multilingual Intermediate Tasks for Enhanced Cross-lingual Reasoning

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 fine-tuning a pretrained model on a mix of English and high-resource non-English intermediate tasks (e.g., 5 English + 5 non-English) improve zero-shot cross-lingual reasoning capabilities on XTREME-R, as measured by accuracy on tasks requiring logical or commonsense reasoning compared to using only English intermediate tasks?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 9.1/10.