Neural Machine Translation Accuracy via Iterative Error Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine translation methods, particularly neural machine translation, often result in mistranslation due to limitations in learning data, leading to inaccurate and faithless translations, especially with low-frequency words and uncommon word orders.
Innovation Solution
A machine translation method that involves translating an original document with a neural network, analyzing the translated document for mistranslation, modifying target words or phrases with high-frequency words from the learning data, and iteratively refining the translation until a termination condition is met, ensuring higher accuracy and faithfulness to the original document.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If neural machine translation is used to translate the original document, then translation speed and ease of operation are improved, but translation accuracy and faithfulness to the original document deteriorate due to mistranslation of low-frequency words and uncommon word orders
Solution Approach 1:
The system performs error checking on the translated document and uses the error check results to determine modification targets. This feedback loop allows the system to identify mistranslations and automatically modify the original document to improve translation accuracy while maintaining automation.
Solution Approach 2:
The system modifies the original document before re-translation by replacing low-frequency words with high-frequency synonyms. This preliminary modification prepares the document for better translation by the neural network, addressing potential mistranslation issues before they occur.
2Measurement precision
If the original document is modified by replacing words with high-frequency words to improve translation accuracy, then translation accuracy is improved, but the faithfulness to the original document may deteriorate due to excessive modification
Solution Approach 1:
The system uses error check results to guide modifications, only replacing words that are identified as causing mistranslation. This feedback-based approach ensures modifications are made only where necessary, maintaining faithfulness to the original document while improving accuracy.
Solution Approach 2:
The system applies modifications locally only to specific words or phrases identified as modification targets based on error analysis, rather than uniformly modifying the entire document. This preserves the original document's integrity in areas that do not require modification.
3Measurement precision
If iterative modification and re-translation is performed to improve translation accuracy, then translation accuracy is improved, but processing time increases
Solution Approach 1:
The system performs partial modifications only on identified error portions rather than re-translating the entire document iteratively. This reduces processing time while still improving translation accuracy by focusing efforts on problematic areas.
Solution Approach 2:
The system automatically performs error checking, identifies modification targets, and modifies the document without requiring manual intervention for each iteration. This automation reduces the time loss associated with iterative processing.
Data Source
AI summary
The accuracy of machine translation is increased. A translated document with high translation accuracy is obtained. An original document is faithfully translated. An original document is translated with a neural network to generate a first translated document; a modification-target word or phrase is determined from words and phrases contained in the original document on the basis of an analysis result for the first translated document; the modification-target word or phrase is replaced with a high frequency word in learning data used for learning in the neural network to modify the original document; and the modified original document is translated with the neural network to generate a second translated document.