Revisiting Machine Translation for Cross-lingual Classification
This work challenges the dominance of multilingual models in cross-lingual classification, prompting attention to MT-based baselines, which is incremental as it refines existing methods.
The paper tackled the problem of cross-lingual classification by revisiting machine translation (MT) approaches, showing that translate-test with a stronger MT system and mitigation of training-inference mismatch can perform substantially better than previously assumed, with results varying by task.
Machine Translation (MT) has been widely used for cross-lingual classification, either by translating the test set into English and running inference with a monolingual model (translate-test), or translating the training set into the target languages and finetuning a multilingual model (translate-train). However, most research in the area focuses on the multilingual models rather than the MT component. We show that, by using a stronger MT system and mitigating the mismatch between training on original text and running inference on machine translated text, translate-test can do substantially better than previously assumed. The optimal approach, however, is highly task dependent, as we identify various sources of cross-lingual transfer gap that affect different tasks and approaches differently. Our work calls into question the dominance of multilingual models for cross-lingual classification, and prompts to pay more attention to MT-based baselines.