Sentence-level bilingual alignment method and device, and computer-readable storage medium

Abstract

The invention discloses a sentence-level bilingual alignment method and device and a computer readable storage medium, and the method comprises the steps of S1, obtaining Z trained convolution kernels, where Z is an integer greater than or equal to 1; S2, carrying out sentence segmentation processing on the two to-be-aligned texts respectively, and establishing a text similarity matrix U of the two to-be-aligned texts; S3, carrying out convolution on the text similarity matrix U by adopting each convolution in the Z trained convolution kernels to obtain Z optimized text similarity matrixes; and S4, obtaining statement alignment results of the two to-be-aligned texts by utilizing the Z optimized text similarity matrixes. The method and the device are beneficial to improving the text statement alignment efficiency.

Description

technical field [0001] The present invention relates to the technical field of natural language processing, in particular to a sentence-level bilingual alignment method and device, and a computer-readable storage medium. Background technique [0002] Parallel corpora are important data for translation algorithms based on natural language processing. Parallel / corresponding corpora are bilingual / multilingual corpora composed of source texts and parallel corresponding target language texts. The degree of alignment can be divided into word-level , sentence-level, paragraph-level, and article-level. Among them, the sentence-level parallel corpus is the most commonly used corpus. Therefore, the paragraph-level and article-level parallel corpora are often converted into sentence-level parallel corpora. However, in the corpus , the original text and the translated text do not necessarily have a one-to-one correspondence. For example, due to the difference between the structure of th...

AI Technical Summary

Problems solved by technology

[0002] Parallel corpora are important data for translation algorithms based on natural language processing. Parallel / corresponding corpora are bilingual / multilingual corpora composed of source texts and parallel corresponding target language texts. The degree of alignment can be divided into word-level , sentence-level, paragraph-level, and article-level. Among them, the sentence-level parallel corpus is the most commonly used corpus. Therefore, the paragraph-level and article-level parallel corpora are often converted into sentence-level parallel corpora. However, in the corpus , the original text and the translated text do not necessarily have a one-to-one correspondence. For example, due to the difference between the structure of the article and the author's writing habits, there may be 15 Chinese sentences corresponding to 22 English sentences, or 16 Chinese sentences corresponding to 50 English sentences. sentence, so it is necessary to consider complex and diverse sentence pairing situations. At present, the corpus of paragraphs and chapters is mainly split and grouped into one-to-one corresponding sentences manually. This method requires a lot of manpower and time, which is not conducive to sentence alignment. efficiency improvement

Images