Online Machine Translation Model Adaptation via Post-Edit Feedback

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Updating machine translation systems with post-edits from human translators is a difficult endeavor, as existing methods struggle to efficiently incorporate corrections into statistical models in real-time, leading to suboptimal translation quality.

Innovation Solution

A method that immediately updates a machine translation system by receiving post-edited sentence pairs, aligning and adding phrases to the model, adapting the language model, calculating translation statistics, adjusting weights, and retranslating sentence pairs using the updated model to match human corrections, allowing for fast online domain adaptation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If post-edits are incorporated into the machine translation system, then translation quality is improved, but the complexity of updating the system increases

Engineering Contradiction:

Improvetranslation qualityVSAvoidsystem update complexity

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

The system automatically updates itself by incorporating post-edits from human translators without requiring manual retraining or complex intervention. The machine translation system performs self-learning by integrating correction data directly into its statistical models, enabling autonomous improvement while maintaining operational simplicity

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system implements a feedback loop where post-edits from human translators are captured and fed back into the translation model. This continuous feedback mechanism allows the system to learn from corrections and automatically adjust its translation quality, resolving the contradiction between improvement and complexity

Inventive Principle:

Principle #23Feedback

2Manufacturing precision

If the machine translation system is updated with post-edits in real-time, then translation quality improves faster, but the processing time and computational resources increase

Engineering Contradiction:

Improvetranslation qualityVSAvoidprocessing time

Core Design Contradiction:

Manufacturing precisionVSLoss of time

Solution Approach 1:

The system performs partial updates by selectively incorporating only the necessary portions of post-edit data into the translation model, rather than retraining the entire system. This approach achieves quality improvement with reduced computational overhead and faster processing

Inventive Principle:

Principle #16Partial or excessive action

Solution Approach 2:

The system prepares and pre-processes post-edit data as it is received, organizing and aligning corrections before integrating them into the model. This preliminary preparation reduces the computational burden during actual translation operations, balancing quality improvement with processing efficiency

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS9213694B2Efficient online domain adaptation

Publication Date: 2015.12.15 SDL INC

AI summary

Systems and methods for efficient online domain adaptation are provided herein. Methods may include receiving a post-edited machine translated sentence pair, updating a machine translation model by adjusting translation weights for a translation memory and a language model while generating test machine translations of the machine translated sentence pair until one of the test machine translations approximately matches the post-edits for the machine translated sentence pair, and retranslating the remaining machine translation sentence pairs that have yet to be post-edited using the updated machine translation model.