A Mongolian-Chinese machine translation method for placeholder disambiguation based on pointer generation network

Abstract

A method for Mongolian-Chinese machine translation based on pointer generation network to realize placeholder disambiguation, based on encoder-decoder architecture, is characterized in that it also includes an auxiliary network and a backbone network, and the auxiliary network is a lexical position for each input source Generate a binary gate to dynamically select the vocabulary to be focused on. The backbone network is a pointer generation network with a gating mechanism for attention; in the encoding stage, the source text is encoded into a hidden layer through the word embedding layer of the encoder State, and then the gating mechanism determines whether the information from the current state flows in or is replaced by placeholders; in the decoding stage, the ability to copy source text and generate new vocabulary is used to generate network pointers, and the placeholder context is used for decoding. Before the final data output, additional modules are used to perform linguistic checks to detect possible translation anomalies and mark them, and adjust relevant parameters to achieve the best translation effect.

Description

technical field [0001] The invention belongs to the technical field of machine translation, in particular to a Mongolian-Chinese machine translation method for realizing placeholder disambiguation based on a pointer generation network. Background technique [0002] With the rapid economic development in various regions of the world, more and more attention has been paid to the communication between different languages, and machine translation has emerged accordingly. The development of the modern Internet has driven the upsurge of machine translation research, but the current machine translation has not yet achieved the effect of human translation. [0003] After the three stages of rule-based translation, statistical machine translation, and neural network machine translation, the translation effect has become more and more remarkable, but the problems that have always existed have not been really solved, such as polysemy, grammatical problems, etc., so for improving the ma...

AI Technical Summary

Problems solved by technology

[0006] With the emergence of the Transformer model, BERT also appeared. Compared with the previous word embedding method represented by word2vec, the BERT model further increases the generalization ability of the word vector model, fully describing the character level, word level, sentence level and even between sentences. Relational features can model polysemy to a certain extent, but it requires a large amount of data sets, and it has a great impact on Mongolian-Chinese translation, a language with a small corpus

Images