Can Monolingual Pretrained Models Help Cross-Lingual Classification?
This addresses the issue of multilingual models lagging behind monolingual ones in cross-lingual tasks, offering an incremental improvement for natural language processing applications.
The paper tackled the problem of improving zero-shot cross-lingual classification by transferring knowledge from monolingual pretrained models to multilingual ones, resulting in methods that outperform vanilla multilingual fine-tuning on two benchmarks.
Multilingual pretrained language models (such as multilingual BERT) have achieved impressive results for cross-lingual transfer. However, due to the constant model capacity, multilingual pre-training usually lags behind the monolingual competitors. In this work, we present two approaches to improve zero-shot cross-lingual classification, by transferring the knowledge from monolingual pretrained models to multilingual ones. Experimental results on two cross-lingual classification benchmarks show that our methods outperform vanilla multilingual fine-tuning.