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

Forest-based system combination method for counting machine translation

A technology of statistical machine translation and fusion methods, applied in the fields of instruments, calculations, special data processing applications, etc., can solve problems such as disordered word order, and achieve the effect of improving performance

Inactive Publication Date: 2011-08-17
HARBIN INST OF TECH
View PDF0 Cites 4 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0008] In order to solve the problem of pruning the decoding space existing in the existing system integration, the present invention prunes off candidate translations that may be good in the future, and after the construction of the confusion network, all translations except the skeleton translation will be pruned. Assumed word order disorder problem, the present invention proposes a forest-based system fusion method in statistical machine translation

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
  • Forest-based system combination method for counting machine translation
  • Forest-based system combination method for counting machine translation
  • Forest-based system combination method for counting machine translation

Examples

Experimental program
Comparison scheme
Effect test

specific Embodiment approach 1

[0019] Specific implementation manner 1: The process of the forest-based system fusion method in statistical machine translation described in this implementation manner is:

[0020] 1. Construct a confusion network based on multiple machine translation results;

[0021] 2. Construct a forest based on the information in the constructed confusion network;

[0022] 3. Add the phrase and dependent syntax information to the forest in the form of feature values, reorder all the edges of each node in the forest, and obtain a new forest;

[0023] 4. Perform statistical decoding on the forest obtained in step 3 to obtain the final translation information.

[0024] The so-called forest-based decoding method is a decoding method used in the field of machine translation and syntactic analysis in the prior art, and the decoding method in the system fusion task is similar to the decoding method of machine translation / syntax analysis. To further improve the performance of the word-level system fusion...

specific Embodiment approach 2

[0025] Specific embodiment 2: This embodiment further defines step 1 in the forest-based system fusion method in statistical machine translation described in specific embodiment 1. In step 1, the process of constructing a confusion network is:

[0026] A1. The determination of the skeleton translation. According to the results of multiple machine translations, the skeleton translation is selected through the smallest Bayesian risk E b ,

[0027] E b = arg min E ′ A E X E A E TER ( E ′ , E ) - - - ( 1 )

[0028] In the formula, E′ represents any translation result, and E represents a collection of multiple machine translation results;

[0029] A2. The translation hypothesis is aligned and normalized. The alignment algorithm is used to establish the alignment between the skeleton translation and the hypothetical translation, and the two translation results are stretched by inserting "NULL" in the skeleton transla...

specific Embodiment approach 3

[0043] Specific embodiment three: This embodiment further defines step two in the forest-based system fusion method in statistical machine translation described in specific embodiment one. In this embodiment, the method of constructing the forest described in step two The process is: according to the confusion network, a bottom-up construction method is adopted to generate a forest through nodes and edges.

[0044] The specific process of generating the forest is: according to the leaf nodes, one or more intermediate nodes are generated from the bottom to the top using edges, until the root node is generated. All nodes and edges form the forest. When generating intermediate nodes and root nodes , Calculate the score of the language model and the score of the translation model of the corresponding node, and the sum of the two scores is used as the final score of the node. See figure 2 In the forest shown in the figure, the child nodes that generate the same translation result are...

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

The invention provides a forest-based system combination method for counting machine translation, and relates to the technical field of machine translation. By the method, the problems that future possibly good candidate translations may be pruned during space pruning decoding in the system combination at the present, and the word sequences supposed by all the translations besides a skeleton translation after constructing a confusing network are disordered are solved. The forest-based system combination method comprises the following steps of: constructing the confusing network; 2, constructing a forest according to information of the constructed confusing network; 3, adding phrases and sequentially stored sentence structure information into the forest in the form of characteristic values, and re-sequencing all edges of each node in the forest so as to obtain a new forest; and 4, statistically decoding the new obtained forest so as to obtain final translation information. On the basis of the forest, twice decoding is provided, so that a better English translation is generated. In the method, a forest technology is introduced into the system combination, so the performance of the system combination is improved.

Description

Technical field [0001] The present invention relates to the field of machine translation technology, in particular to a system fusion technology. Background technique [0002] The so-called statistical machine translation is the use of statistical knowledge for translation. The source language can be translated by using a machine translation model to obtain the target language. Statistical machine translation can be divided into word-based, phrase-based and syntax-based translation systems according to whether it contains syntactic information; for syntactic translation systems, it can be divided into tree-to-string, string-to-tree and tree-to-tree translation systems. . Statistical machine translation has always been the focus of natural language research. [0003] The so-called system fusion is to fuse multiple translation results in the decoding stage or in the post-processing stage, see figure 1 Shown. The advantage is that since each translation system has corresponding cha...

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
IPC IPC(8): G06F17/28
Inventor 赵铁军刘宇鹏
Owner HARBIN INST OF TECH
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