Cross-lingual Retrieval for Iterative Self-Supervised training
Year: 2,020
Journal: Conference on Neural Information Processing Systems
Languages: Arabic, Burmese, Chinese (simplified), Czech, Dutch, English, Estonian, Finnish, French, German, Gujarati, Hindi, Italian, Japanese, Kazakh, Korean, Latvian, Lithuanian, Nepali, Romanian, Russian, Sinhala, Spanish, Turkish, Vietnamese
Programming languages: Python
Input data:
sentences
CRISS is a multilingual sequence-to-sequnce pretraining method where mining and training processes are applied iteratively, improving cross-lingual alignment and translation ability at the same time.