Zero-Shot Cross-Lingual Transfer via English 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 robust is zero-shot cross-lingual transfer via English intermediate-task training to domain shifts in the target task, evaluated using accuracy on XNLI and other benchmarks with different domain distributions?

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