Back-translation Filtering for Machine Translation Accuracy

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Automatic machine translation systems often generate incorrect or less desirable translations due to limitations in evaluating the quality of translation entries in phrase tables.

Innovation Solution

The system identifies primary phrases based on high scores or probabilities and compares them to test phrases to determine similarity, modifying the translation process to prevent or diminish the use of incorrect or less desirable translations by adjusting counts, probabilities, or flagging them for human evaluation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If automatic machine translation training procedures are used to generate translation mappings, then translation coverage and speed are improved, but translation accuracy and quality deteriorate due to generation of incorrect or less desirable translations

Engineering Contradiction:

Improvetranslation speedVSAvoidtranslation accuracy

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The patent implements a feedback mechanism where back-translations are generated and compared with original source phrases. Translation mappings that produce inconsistent back-translations are identified and filtered out, creating a quality control loop that improves translation accuracy while maintaining automated processing speed

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent performs preliminary filtering of translation mappings using back-translation validation before deploying translations. By pre-evaluating translation quality through back-translation consistency checks, the system prevents incorrect mappings from being used, thereby improving accuracy without sacrificing productivity during actual translation operations

Inventive Principle:

Principle #10Preliminary action

2Adaptability or versatility

If multiple translation alternatives are provided in the phrase table, then translation versatility and adaptability are improved, but translation quality deteriorates due to inclusion of incorrect or less desirable alternatives

Engineering Contradiction:

Improvetranslation alternativesVSAvoidtranslation quality

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent applies different quality standards to different translation mappings based on back-translation validation results. High-quality mappings that pass back-translation consistency checks are retained and prioritized, while low-quality mappings are filtered or flagged. This selective quality control maintains versatility by preserving multiple valid alternatives while eliminating incorrect ones

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS8762128B1Back-translation filtering

Publication Date: 2014.06.24 GOOGLE LLC

AI summary

A translation system receives a test pair that includes a source test phrase in a first language and a target test phrase in a second language. The test pair can be evaluated by comparing its components with phrases in primary pairs. The test source phrase can be compared to a primary source phrase that is the phrase most commonly translated by the machine translation system into the test target phrase. The test target phrase can be compared to a primary target phrase that is the phrase into which the target source phrase is most often translated. If one and/or both comparisons are sufficiently dissimilar, the machine translation system can be modified by deleting the test pair, by flagging it for human review, or in other ways.