Multi-granularity Chinese-Vietnamese parallel sentence pair extraction method based on graph attention network

Abstract

The invention relates to a multi-granularity Chinese-Vietnamese parallel sentence pair extraction method based on a graph attention network, and belongs to the technical field of natural language processing. According to the method, the twinning neural network is mainly adopted as a main body frame to share Chinese and Vietnamese bilingual in the same semantic space. The method comprises the following steps: firstly, in combination with the structure of a document, dividing a Chinese-Chinese document into four different levels of granularities, namely sub-words, sentences, paragraphs and documents, constructing different levels of graph network structures, and realizing multi-granularity document modeling; then, carrying out coding by utilizing BERT, obtaining feature representations of nodes with different granularities through a graph attention network layer, and carrying out fusion representation on a graph information integration layer; and finally, training a classifier by using a full connection layer, calculating the parallel probability of the two sentences, and extracting Chinese-Vietnamese parallel sentence pairs. According to the method, the language difference of the Chinese-Vietnamese sentence pairs can be reduced, the Chinese-Vietnamese parallel sentence pairs are effectively extracted, and powerful support is provided for development of neural machine translation.

Description

technical field [0001] The invention relates to a multi-granularity Chinese-Vietnamese parallel sentence pair extraction method based on a graph attention network, and belongs to the technical field of natural language processing. Background technique [0002] Neural machine translation is data-driven and relies on massive bilingual data. Generally, the larger the data size, the better the translation model will be. However, the existing neural machine translation of low-resource languages ​​is not effective, especially in Chinese-Vietnamese neural machine translation, where parallel corpora are scarce and difficult to obtain directly on the Internet, resulting in poor model performance. Parallel sentence pair extraction is one of the important methods to improve the quality and scale of parallel corpora. Many studies have shown that large-scale and high-quality parallel corpora can effectively improve the quality of neural machine translation for low-resource languages. Th...

AI Technical Summary

Problems solved by technology

However, the effect of neural machine translation of existing low-resource languages ​​is poor, especially in Chinese-Vietnamese neural machine translation, where parallel corpus is very scarce and difficult to obtain directly on the Internet, resulting in poor model performance

Images