Enhancing Zero-Shot Cross-Lingual Code Generation via Multimodal Benchmark 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: Can intermediate-task training with English multimodal benchmarks (e.g., MMMU, SEED-Bench) enhance zero-shot cross-lingual transfer in code generation tasks involving multimodal inputs, as measured by functional correctness or latency on platforms like CodeParrot or MultiPL-E?

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