PARADISE: Exploiting Parallel Data for Multilingual Sequence-to-Sequence Pretraining
This addresses the problem of enhancing multilingual NLP models for tasks like translation and inference by leveraging parallel data, offering a computationally efficient approach that is incremental over existing methods.
The paper tackles the underutilization of parallel data in multilingual sequence-to-sequence pretraining by introducing PARADISE, which integrates parallel corpora and dictionary-based noise into training, resulting in average improvements of 2.0 BLEU points in machine translation and 6.7 accuracy points in cross-lingual natural language inference.
Despite the success of multilingual sequence-to-sequence pretraining, most existing approaches rely on monolingual corpora, and do not make use of the strong cross-lingual signal contained in parallel data. In this paper, we present PARADISE (PARAllel & Denoising Integration in SEquence-to-sequence models), which extends the conventional denoising objective used to train these models by (i) replacing words in the noised sequence according to a multilingual dictionary, and (ii) predicting the reference translation according to a parallel corpus instead of recovering the original sequence. Our experiments on machine translation and cross-lingual natural language inference show an average improvement of 2.0 BLEU points and 6.7 accuracy points from integrating parallel data into pretraining, respectively, obtaining results that are competitive with several popular models at a fraction of their computational cost.