A Method for Translation of Low-Frequency Words Based on Semantic Information Fusion

Abstract

The present invention proposes a low-frequency word translation method based on semantic information fusion, which belongs to the field of machine translation. A pair of sentences is input into the translation system, wherein the source language sentence X and the target language sentence y corresponding to the source language sentence are obtained in the source language The subword sequence of the low-frequency words in the sentence, the target translation corresponding to the low-frequency words in the target language sentence is obtained, and the wildcard UNKi is used to replace the low-frequency words in the bilingual sentence pair (x, y), and the new bilingual sentence pair is obtained. The vector representations of words and / or low-frequency words in the target language are fused with the vector representations of the wildcard UNKi. The present invention closely follows the core idea of ​​semantic fusion, and proposes three specific forms of semantic fusion, including integrating low-frequency word vector representations in the source language, integrating low-frequency word vector representations in the target language, and integrating low-frequency word vector representations at both ends. Vectors in two languages ​​and two vector spaces are used to represent the semantic information of low-frequency words.

Description

technical field [0001] The present invention relates to the field of machine translation, especially for the conversion task of low-frequency words in the neural machine translation system. By making full use of the semantic vector representation of source-side and target-side low-frequency words in the process of model training and decoding, the low-frequency words and even the whole sentence are improved. translation quality. Background technique [0002] Low-frequency words refer to a class of words that are sparse in frequency or never appear in large-scale bilingual parallel corpora. Depending on the degree of frequency, it is usually called Unknown words or Out of vocabulary in natural language processing. Due to its characteristics of sparse frequency and single translation, the translation of low-frequency words has always been the focus and difficulty of machine translation research. Especially in the current mainstream neural machine translation, the word list is...

AI Technical Summary

Problems solved by technology

In addition, considering that the current neural machine translation system usually uses two methods of subword segmentation and replacement mechanism to solve the translation problem of low-frequency words, at this time, low-frequency words are usually segmented into subword sequences

Images