Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

A method and device for training a translation model

A technology of translation model and training method, applied in the field of natural language processing, which can solve problems such as expensive manual labeling, restrictions on large-scale deployment of manual standards, and difficulty in obtaining parallel corpus

Pending Publication Date: 2019-01-25
BEIJING ZIDONG COGNITIVE TECH CO LTD
View PDF2 Cites 21 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

In the case of insufficient parallel corpus or no parallel corpus, it is difficult for current neural machine translation methods to achieve satisfactory translation performance
In practical application scenarios, it is generally difficult to obtain large-scale parallel corpus; although manual annotation is a feasible way to obtain parallel corpus, however, the cost of manual annotation is very expensive and time-consuming, which seriously limits the large-scale deployment of artificial standards.

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • A method and device for training a translation model
  • A method and device for training a translation model
  • A method and device for training a translation model

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0084] figure 1 A flow chart of a translation model training method provided by an embodiment of the present invention, the method is applied to a translation model whose network structure is a deep neural network, such as figure 1 As shown, the method includes:

[0085] S1: Obtain large-scale monolingual corpus in the source language and monolingual corpus in the target language;

[0086] The source language refers to the language to be translated in the translation task, and the target language refers to the output language after the translation task is completed. For example, in a Chinese-to-English translation task, Chinese is the source language and English is the target language.

[0087] Monolingual corpus refers to unlabeled corpus, that is, there is no one-to-one correspondence between the source language and the target language. In this embodiment, the monolingual corpus may be a text corpus obtained from the Internet and manually sorted out, or a text corpus manu...

Embodiment 2

[0100] A translation model training method provided by an embodiment of the present invention, the method is applied to a translation model whose network structure is a deep neural network, and the flow of the method is as described in Embodiment 1, and will not be repeated here;

[0101] figure 2 A flow chart of a word vector training method provided by an embodiment of the present invention, such as figure 2 As shown, in the embodiment of the present application, step S2 includes:

[0102] S21: Segment the monolingual corpus on the source language side and the target language side respectively, and obtain the word after word segmentation.

[0103] For specific word segmentation methods, reference may be made to various related technologies. Such as the traditional word segmentation method based on conditional random field and the current mainstream word segmentation method based on long short-term memory network.

[0104] It should be noted that if the current language ...

Embodiment 3

[0112] A translation model training method provided by an embodiment of the present invention, the method is applied to a translation model whose network structure is a deep neural network, the flow of the method is as described in Embodiment 1 or 2, and will not be repeated here;

[0113] image 3 A flow chart of a word vector alignment method provided by an embodiment of the present invention, such as image 3 As shown, in the embodiment of the present application, step S3 includes:

[0114] S31: Searching for common words from the monolingual corpus of words in the source language and the target language.

[0115] Specifically, the common words in the source language and the target language usually include Arabic numerals, commonly used symbols, and the like.

[0116] S32: Construct a one-to-one mapping relationship according to the shared words of the source language and the target language;

[0117] Specifically, the word vectors corresponding to the words shared in th...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

A method and a device for training a translation model are provided in the embodiment of the present invention, the translation model is trained with large-scale source language monolingual corpus andtarget language monolingual corpus, the trained translation model can simultaneously perform two-way translation from the source language to the target language and from the target language to the source language, and has the following advantages: high translation performance can be achieved without parallel corpus. Therefore, the method can greatly reduce the dependence of neural translation model on parallel corpus, and greatly reduce the cost and time of manual tagging corpus.

Description

technical field [0001] The embodiments of the present invention relate to the technical field of natural language processing, and in particular to a translation model training method and device. Background technique [0002] With the development of globalization, international exchanges and information transmission have shown explosive growth. The traditional method of relying on human translation has been unable to meet the needs of people's daily cross-language communication. As a technology that can automatically provide accurate translation results, machine translation has gradually received extensive attention and research. From traditional rule-based translation methods to statistical machine translation methods, and now mainstream neural machine translation methods, machine translation performance has made tremendous progress. [0003] The current mainstream neural machine translation methods rely heavily on large-scale high-quality parallel corpora. In the case of...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06F17/28G06F17/27
CPCG06F40/289G06F40/58G06F40/30
Inventor 王峰
Owner BEIJING ZIDONG COGNITIVE TECH CO LTD
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com