Machine translation method and apparatus, and machine translation system training method and apparatus

Abstract

An embodiment of the invention provides a machine translation method. The method comprises the steps of receiving an original character string of an input source language, wherein the original character string comprises a named entity character and a non-named entity character, and the named entity character has a subordinate entity class tag; translating the original character string into a middle character string, specifically, replacing the corresponding named entity character in the original character string with the entity class tag, and translating the non-named entity character in the original character string into a target language character; and translating the middle character string into a target character string, specifically, searching for a target character matched with the named entity character and the corresponding entity class tag in a preset mapping table, and correspondingly replacing the entity class tag in the middle character string with the target character. According to the method, the translation quality can be improved.

Description

technical field [0001] The present invention relates to the technical field of language processing, in particular to a machine translation method and device, a machine translation system training method and device, and a machine translation device and a machine translation system training device . Background technique [0002] At present, the development of globalization has brought about an urgent need for machine translation (Machine Translate) between multiple languages. Among them, due to the advantages of simple construction of the neural machine translation system and good translation quality, machine translation through the neural machine translation system has become the mainstream. [0003] However, the high requirements of the neural machine translation system on computing equipment and its system framework determine that the size of the translation vocabulary is inversely proportional to the efficiency of machine translation and machine training. Therefore, in o...

AI Technical Summary

Problems solved by technology

Therefore, in order to ensure the timeliness of machine translation, it is inevitable to limit the size of the translation vocabulary, which makes it difficult to cover some low-frequency but important words

Especially for some named entity characters (Named Entity, NE), because they are not included in the translation vocabulary, the corresponding translation of named entity characters cannot be performed during translation, which may cause the problem of translation loss

[0004] Therefore, the current translation mechanism is difficult to fully cover the translation of named entity characters, which seriously affects the translation quality

Images