Contextual Post-Editing Model for Sentence-Level Translation Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sentence-by-sentence machine translation systems fail to account for contextual information, leading to errors in translation, particularly with languages that use genderless pronouns, homonyms, proper nouns, and languages that omit subjects or objects, and are computationally infeasible when attempting to translate entire language sources simultaneously.
Innovation Solution
A computing system using a machine-learned contextual post-editing model that refines preliminary translations by processing source context, including surrounding sentences, to generate refined translations, employing a two-stage translation process with a sentence-level translation system and a sequence-to-sequence architecture.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If sentence-by-sentence translation is used, then translation speed and computational efficiency are improved, but translation accuracy deteriorates due to inability to handle context
Solution Approach 1:
The translation process is segmented into two independent stages: a sentence-level translation stage that processes each sentence separately for speed, and a contextual post-editing stage that refines translations using context from surrounding sentences. This segmentation allows the system to maintain high translation speed while improving accuracy through contextual refinement.
Solution Approach 2:
The system performs preliminary sentence-level translation first to generate initial translations quickly, then applies contextual post-editing as a subsequent refinement step. This preliminary action ensures that the bulk translation is completed efficiently before contextual corrections are applied.
2Reliability
If entire language source is translated at once, then translation accuracy is improved through context awareness, but computational feasibility deteriorates
Solution Approach 1:
The translation task is divided into manageable segments (individual sentences) that can be processed independently and quickly, rather than attempting to process the entire language source at once. This segmentation makes the system computationally feasible while still enabling contextual awareness through the post-editing stage.
Solution Approach 2:
The system applies contextual post-editing to refine translations, effectively applying contextual action only where needed rather than attempting to process the entire context simultaneously. This partial application of contextual refinement maintains computational feasibility while improving accuracy.
3Reliability
If contextual information is considered in translation, then translation accuracy is improved, but processing time increases
Solution Approach 1:
The processing is segmented into a fast sentence-level translation pass followed by a more time-consuming but targeted contextual post-editing pass. This segmentation allows the system to minimize overall processing time by performing quick initial translations before applying contextual refinements.
Solution Approach 2:
The system performs preliminary sentence-level translation quickly before applying contextual post-editing. This preliminary translation step minimizes initial processing time, and the contextual refinement is applied subsequently to improve accuracy without significantly increasing total processing time.
Data Source
AI summary
Generally, the present disclosure is directed to systems and methods that leverage machine learning to perform post-editing of sentence-level translations that takes into account contextual information from the language source. As an example, the proposed post-editing system can run as a second pass to a sentence-level translation system and the goal of the post-editing system may be to refine translations which are affected by the larger context.