Transfer learning parallel statement pair extraction method and device based on language similarity

The invention relates to a transfer learning parallel statement pair extraction method and device based on language similarity, and belongs to the field of natural language processing. According to the method, firstly, corpora of Thai and Lao are preprocessed, sub-words and words in the Thai are replaced based on phonetic symbols, and unified representation of Thai and Lao statements is obtained; a Chinese-Thai parallel statement pair extraction model is migrated to a Chinese-Lao model by utilizing a data migration and model migration method based on Thai-Lao language similarity, and finally, Chinese-Lao parallel statement pairs input into the model are predicted by utilizing the pre-trained parallel statement pair extraction model. According to the method provided by the invention, the language similarity can be effectively modeled, and the Chinese-Thai statement pair extraction model with rich resources is migrated to a Chinese-Lao statement pair extraction model with scarce resources, so that the purpose o