Translation processing method and system

Abstract

Embodiments of the invention provide a translation processing method and system. The method comprises the steps of obtaining a statement of a source language; coding the statement of the source language to obtain a vector sequence, wherein the vector sequence comprises word vectors formed by converting segmented words obtained by segmenting the statement; predicting corresponding candidate words in a target language word by word according to the vector sequence; and generating a statement of the target language according to the predicted candidate words, in prediction processing of any one ofthe candidate words, obtaining a plurality of primarily selected words from a preset translation word list, and calculating translation probabilities of the primarily selected words according to a pre-trained machine translation model, thereby selecting the candidate words from the primarily selected words according to the translation probabilities. According to the translation processing method and system, a language pair of data sparseness can obtain better translation quality.

Description

technical field [0001] The present invention relates to the technical field of machine translation, and more specifically, to a translation processing method and system. Background technique [0002] With the deepening of international exchanges, people's demand for language translation is increasing day by day. However, there are many kinds of languages ​​in the world, each with its own characteristics and flexible forms, making the training of machine translation models between all language pairs an unsolved problem. [0003] In order to realize automatic machine translation, current technologies are usually based on neural network methods. Neural networks are data-driven, and for this, large-scale, high-quality parallel corpora need to be collected to obtain reliable translation models. However, high-quality parallel corpora often only exist among a small number of languages, and are often limited to certain specific fields, such as government documents, news, etc. [...

AI Technical Summary

Problems solved by technology

[0004] Therefore, machine translation based on neural networks faces a big problem: most language pairs do not have high-quality, large-scale parallel corpora

However, since each language has its own unique characteristics such as word order, vocabulary, etc., only using the "shared" neural network to train a multilingual translation model may ignore the characteristics of each language, resulting in the translation effect of the translation model changing. Difference

Images