Automated Post-Editing of Machine Translated Content Using Generative AI

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current machine translation technologies face challenges in producing high-quality translations efficiently, as they often require human post-editing, which is costly and time-consuming. Additionally, machine translations may lack contextual understanding and domain-specific knowledge, leading to unsatisfactory quality.

Innovation Solution

The implementation of an automated post-editing system that utilizes a generative AI model, integrated with machine translation quality estimation models and contextual information, to iteratively refine machine translations. This system presents users with quality estimation scores for machine translated segments, allowing for manual intervention and feedback to enhance translation quality.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If machine translation is used to translate documents, then translation speed is improved, but translation quality deteriorates

Engineering Contradiction:

Improvetranslation speedVSAvoidtranslation quality

Core Design Contradiction:

ProductivityVSManufacturing precision

Solution Approach 1:

The system performs preliminary quality estimation of machine translated segments before human review, identifying which segments need post-editing. This preliminary action allows efficient allocation of human resources to only the problematic segments, maintaining high translation speed while improving overall quality through targeted human intervention.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system incorporates quality estimation feedback loops that continuously monitor and evaluate machine translation output. The quality estimation model provides feedback on translation quality scores, enabling the system to automatically identify and prioritize segments requiring human post-editing, thus balancing speed and quality.

Inventive Principle:

Principle #23Feedback

2Manufacturing precision

If human post-editing is performed on machine translated content, then translation quality is improved, but time consumption and cost increase

Engineering Contradiction:

Improvetranslation qualityVSAvoidtime consumption

Core Design Contradiction:

Manufacturing precisionVSLoss of time

Solution Approach 1:

Instead of requiring human post-editing of all translated content, the system applies local quality control by identifying and prioritizing only the lowest quality segments for human review. This localized approach ensures high translation quality where needed while minimizing time consumption on segments that are already satisfactory.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The quality estimation model performs self-service by automatically evaluating machine translation quality and generating prioritization lists without requiring human intervention. This automated self-assessment reduces the time and resources needed for quality control while maintaining high translation standards.

Inventive Principle:

Principle #25Self-service

3Manufacturing precision

If automated post-editing using generative AI is implemented, then translation quality is improved, but system complexity increases

Engineering Contradiction:

Improvetranslation qualityVSAvoidsystem complexity

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

The system introduces a quality estimation model as an intermediary component that bridges machine translation and human review. This intermediary layer automatically assesses translation quality and prioritizes segments for post-editing, simplifying the overall system architecture by providing a clear decision-making framework without requiring complex integration of multiple systems.

Inventive Principle:

Principle #24Intermediary (Mediator)

4Manufacturing precision

If iterative refinement with varying input is applied, then translation quality is improved, but processing time increases

Engineering Contradiction:

Improvetranslation qualityVSAvoidprocessing time

Core Design Contradiction:

Manufacturing precisionVSDuration of action of moving object

Solution Approach 1:

The system applies partial iterative refinement by performing multiple iterations only on the lowest quality segments identified by the quality estimation model, rather than applying iterative refinement uniformly to all segments. This selective approach improves translation quality for problematic areas while minimizing overall processing time by avoiding unnecessary iterations on already satisfactory segments.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS20250173524A1Systems and Methods of Automatic Post-Editing of Machine Translated Content Using a Generative AI Model

Publication Date: 2025.05.29 SDL INC

AI summary

Automatic post-editing of machine translated content using generative AI models is disclosed herein. An example method includes presenting machine translated segments of a document and their associated quality estimation scores, invoking an automated post-editing system for segments with unsatisfactory translation quality, inputting the segments into a generative AI model alongside contextual information, the contextual information comprising a variable window of text adjacent to each of the machine translated segments with unsatisfactory quality estimation scores; producing a revised translation of the segment using the generative AI model and iterating the generative AI process with varying input until a final translation is achieved or a predetermined number of attempts are reached.