Intermediate Task Difficulty and Zero-Shot Cross-Lingual Transfer in Low-Resource Languages on XTREME-R

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 intermediate task difficulty affect the zero-shot cross-lingual transfer performance on XTREME-R when the target language is low-resource, measured by F1-score improvement per parameter size increase?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 9.3/10.