Translation Verification Using Context Similarity Scores
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current translation technologies fail to consider contextual relationships within structured documents, leading to incorrect translations of single words and inconsistencies in integrated translations, which are time-consuming and costly to manually correct.
Innovation Solution
A system that calculates similarity scores between single words and their contextual phrases in both source and target documents using cosine similarity, flagging discrepancies to automate or manually correct inappropriate translations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If NLP translation is used to translate source text to target text, then translation speed is improved, but translation accuracy deteriorates due to failure to consider contextual relationships
Solution Approach 1:
The system calculates similarity scores between source text pairs and their corresponding target text pairs, then uses this feedback to identify and flag inconsistent translations. This feedback mechanism allows the system to automatically detect translation errors without sacrificing the speed of automated translation, resolving the contradiction between translation speed and accuracy.
Solution Approach 2:
The patent introduces similarity scores as an intermediary metric to evaluate translation quality. Instead of directly comparing translations for accuracy, the system uses similarity scores between contextual phrases as a mediator to indirectly assess whether translations maintain proper contextual relationships, thereby improving accuracy while maintaining automated translation speed.
2Manufacturing precision
If manual review is performed to ensure translation adequacy, then translation accuracy is improved, but time consumption increases
Solution Approach 1:
Instead of manually reviewing all translations, the system performs partial action by calculating similarity scores only for specific text pairs and flagging only those with inconsistent translations. This selective approach maintains high translation adequacy for flagged items while dramatically reducing the time consumption compared to complete manual review of all translations.
Solution Approach 2:
The system enables self-service by automatically detecting translation inconsistencies through similarity score calculations and flagging problematic translations for review. This self-detection mechanism reduces the burden on human reviewers, allowing them to focus only on flagged items rather than performing exhaustive manual reviews, thereby improving adequacy while minimizing time loss.
3Manufacturing precision
If contextual analysis is performed on all text pairs, then translation quality is improved, but computational complexity increases
Solution Approach 1:
The patent segments the translation verification process by focusing computational resources on specific text pairs that are likely to contain inconsistencies, rather than performing exhaustive contextual analysis on all translations. This segmentation approach maintains high translation quality for critical areas while reducing overall computational complexity through selective analysis.
Data Source
AI summary
Provided are a computer program product, system, and method for verifying translations of text in a source language to target text in a target language. A source similarity score is calculated indicating a similarity between a first text and second text in a source language. A target similarity score is calculated between translated first text, comprising a translation of the first text in a target language, and translated second text, comprising a translation of the second text in the target language. A determination is made whether the translation of the first text to the translated first text is of an inappropriate translation quality based on the source similarity score and the target similarity score. Indication is made of indicating the inappropriate translation quality in response to determining the inappropriate translation quality.