Machine Translation Quality Estimation (MTQE)

Check out our blog post: Quality Estimation in Machine Translation: A Comprehensive Analysis.

Click here to view the API documentation.

What is it?

Pangeanic's MTQE (Machine Translation Quality Estimation) system is a proprietary AI-based framework designed to evaluate the quality of Machine Translation (MT) output without requiring reference translations. MTQE relies on an exhaustive set of linguistic and statistical features, that are detailed below. It uses pre-trained and finetuned models, along with internal mathematical operations, to generate quantitative quality predictions. It evaluates translation quality segment by segment, providing a normalized score from 0 to 100, identifying translation errors, and determining whether a segment can be accepted or requires Automatic Post-Editing (APE).

The system can be integrated directly into automated translation pipelines, enabling organizations to continuously assess, correct, and validate machine-translated content at production scale.

Reference-Free Quality Estimation: Evaluates source and target segments without requiring human reference translations.
Segment-Level Scoring: Assigns a quality score from 0–100 based on the accuracy and quality of the translation.
Error Explanation: Identifies and explains translation issues to provide transparency into the quality assessment.
Automated Quality Decision: Determines whether a translation can be accepted or should undergo Automatic Post-Editing (APE).
Low Latency: Designed for subsecond processing, enabling quality assessment in real-time and high-volume workflows.
API and UI Access: Available through both API and user interface access for flexible deployment.
Production Integration: Supports integration with production environments and localization workflows.

Quantitative Scoring

Pangeanic's MTQE generates a normalized quality score in the [0, 100] range, providing a quantitative assessment of translation quality at the individual segment level.

The system evaluates the relationship between the source segment and its machine-translated target without requiring a reference translation. Its scoring framework analyzes translation quality and identifies potential issues such as semantic inaccuracies, omissions, additions, fluency problems, and other translation defects.

In addition to the numerical score, MTQE provides an explanation of identified translation errors, allowing users to understand why a segment received a particular quality assessment.

Based on the resulting score and Pangeanic's quality framework, each segment is classified as either:

Accepted: The translation meets the required quality threshold and can proceed without further automated correction.
APE Required: The translation requires Automatic Post-Editing to address identified quality issues.

This approach enables organizations to move from simple translation scoring to an automated quality-control decision process.

Automatic Post-Editing (APE)

For segments identified as requiring correction, Pangeanic's MTQE can trigger Automatic Post-Editing (APE).

APE is executed using Pangeanic's in-house DAAIT (Deep Adaptive AI Translation) technology. The system automatically post-edits the machine-translated target to address the quality issues identified during the initial assessment.

Following APE, the resulting translation is rescored by MTQE. This second evaluation provides a new quality score, identifies any remaining translation errors, and determines whether the corrected segment requires Human Post-Editing (HPE).

The resulting workflow is:

Machine Translation → MTQE Scoring → APE → MTQE Rescoring → Human Post-Editing Decision

This closed-loop process enables automated correction and validation before content reaches human reviewers or downstream production systems.

Coverage and Adaptation

Multilingual Support: Supports 70+ languages, enabling quality estimation across multilingual translation and localization workflows.
Translation Memory Support: Translation memories can be incorporated into the scoring process to improve quality assessment within established translation environments, both on project-level and segment-level.
Glossary Support: Glossaries can be used while scoring segments to verify adherence to approved terminology, both on project-level and segment-level.
Custom Instructions: Custom instructions can be provided to adapt the quality assessment to specific requirements and workflows.
Domain Adaptation: The ability to incorporate domain-specific resources supports quality evaluation aligned with specialized terminology and content requirements.
Scalable Architecture: Designed to process large volumes of translated content while maintaining low-latency performance.

Results and Operational Applications

Automated Translation Quality Control: Automatically evaluates every translated segment and identifies content that meets quality requirements versus content requiring correction.
Post-editing Flow Optimization: Automatically routes problematic translations through APE and identifies segments that still require human intervention, allowing human linguists to focus on content that genuinely needs review.
Production Risk Mitigation: Provides a quality gate before translated content is delivered, helping organizations identify potentially defective translations before publication or downstream use.
Translation Engine Evaluation: Enables organizations to compare the quality of translations produced by different MT systems using a consistent, reference-free scoring framework.
Terminology Compliance: Supports the use of glossaries and translation memories to assess whether machine-translated content follows established terminology and translation requirements.

Use Cases

Enterprise Localization Quality Control: Organizations translating large volumes of product documentation, websites, software interfaces, and customer-facing content can use MTQE as an automated quality gate. Each translated segment can be scored without requiring a reference translation. Segments that meet the required quality threshold can proceed automatically, while lower-quality segments can be sent through APE and subsequently escalated to human post-editors only when necessary.
High-Volume Content Translation and Post-Editing: Localization providers and enterprises processing millions of translated segments can use MTQE and APE to optimize the post-editing workflow. Instead of sending every machine-translated segment to a human linguist, the system automatically identifies problematic translations, performs APE using DAAIT, and rescores the result. Human reviewers can then concentrate on segments where automated correction has not achieved the required quality level.
Translation Engine and Workflow Optimization: Organizations using multiple MT engines or translation workflows can use MTQE to continuously evaluate translation quality across languages, domains, and content types. The segment-level scores and error explanations can help identify which engines and workflows perform best for specific requirements, while translation memories, glossaries, and custom instructions can be incorporated into the evaluation process.

Technological Basis

Pangeanic's MTQE system is built on proprietary AI and language-processing technologies designed to estimate translation quality without relying on reference translations.

The system combines quantitative scoring with linguistic error analysis to produce a normalized 0–100 quality score, an explanation of detected translation issues, and an automated decision regarding the next stage of the translation workflow.

For segments requiring correction, Pangeanic's DAAIT technology performs Automatic Post-Editing. The corrected output is then passed back through MTQE for a second quality assessment.

This creates an automated quality-assurance loop in which translations can be evaluated, corrected, rescored, and routed for human review when required, providing a scalable approach to translation quality management across multilingual production environments.