AI-MTPE for Game Localization Based on a Customized MQM Framework

How Peoplying Delivered High-Quality AI MTPE for a Game Client Through Domain-Specific MQM Customization

AI-driven machine translation is changing the way localization teams approach post-editing. However, AI-MTPE cannot be handled in exactly the same way as conventional MTPE. While traditional MTPE often focuses on correcting machine translation output to meet linguistic accuracy and fluency standards, AI-MTPE requires a more structured understanding of AI-generated error patterns, domain-specific expectations, and quality evaluation criteria.

This case study shows how Peoplying supported a game client by performing a large-scale AI MTPE program based on a customized MQM framework, helping the client improve translation quality, train post-editors effectively, and apply consistent quality standards across high-volume game localization content.

Project Overview

The client required AI MTPE support for game localization content generated through AI-based machine translation. The content required not only linguistic correction, but also careful review of terminology, tone, context, character voice, in-game usability, and player-facing readability.

The client’s goals were to:

• establish a clear quality framework for AI MTPE,

• identify and correct AI-specific translation errors,

• customize MQM criteria to match game localization requirements,

• train post-editors based on the client’s preferred quality standards,

• process a large volume of AI MTPE work while maintaining consistent quality.

Key Challenges

1) AI MTPE Requires a Different Quality Perspective

AI-generated translation output can appear fluent on the surface while still containing serious contextual, terminology, or meaning-related errors. For game content, this creates additional risk because mistranslations can affect gameplay understanding, character tone, player immersion, and overall user experience.

2) MQM Is Not a Fixed Error Framework

MQM provides a structured basis for translation quality evaluation, but it is not a rigid, one-size-fits-all error system. Error categories, subcategories, severity levels, and weighting can be adjusted depending on the domain, content type, and client expectations.

For game localization, certain error types may require different severity treatment compared with general business, technical, or marketing content.

3) Game Content Requires Domain-Specific Error Judgment

Game localization involves unique content types such as UI strings, dialogue, item names, quest descriptions, system messages, skill descriptions, and player-facing prompts. Each type requires different judgment regarding accuracy, tone, consistency, space limitations, and in-context usability.

4) Large-Scale AI MTPE Requires Consistent Post-Editor Training

When multiple post-editors are involved, quality consistency becomes a major challenge. Without a shared understanding of customized MQM criteria, individual editors may apply different standards to similar errors.

Peoplying’s Approach

1) Customized MQM Framework for Game Localization

Peoplying worked with the game client to adapt MQM criteria to the client’s content and quality priorities. Rather than applying MQM as a fixed checklist, Peoplying treated it as a configurable quality framework.

The customization process included:

• reviewing AI MT output patterns specific to the client’s game content,

• identifying key error types relevant to game localization,

• adjusting severity levels based on client priorities,

• clarifying how each error type should be judged in actual post-editing,

• aligning evaluation criteria with gameplay context and player-facing quality expectations.

This allowed the project team to establish a practical MQM model that reflected the client’s real localization needs.

2) Severity Adjustment Based on Error Impact

Peoplying and the client discussed how different error types should be weighted depending on their impact. For example, an error affecting gameplay instructions, item effects, character identity, or user action could be treated as more serious than a minor stylistic issue.

This helped the team distinguish between:

• errors that directly affect meaning or gameplay,

• errors that reduce immersion or character consistency,

• errors that damage terminology consistency,

• errors that affect readability but do not block understanding,

• minor fluency or style issues with limited impact.

By adjusting severity levels in this way, Peoplying ensured that post-editors focused on the errors that mattered most to the client and the end user.

3) Post-Editor Training Based on Client-Specific MQM Criteria

Peoplying trained post-editors using the customized MQM framework agreed with the client. The training focused not only on general post-editing principles, but also on how to recognize and correct AI MT errors in game localization.

The training covered:

• AI MT error patterns commonly found in game content,

• client-specific MQM categories and severity rules,

• examples of acceptable and unacceptable edits,

• terminology and style consistency requirements,

• judgment standards for context-sensitive game strings,

• practical guidance for balancing correction effort and final quality.

This helped post-editors apply a consistent standard across large volumes of content.

4) Quality Control Designed for AI MTPE

Peoplying implemented a review process designed specifically for AI MTPE rather than conventional MTPE. The review focused on whether the edited output met the customized MQM standards and whether the post-editors were applying severity rules consistently.

The review process evaluated:

• accuracy of meaning,

• terminology consistency,

• tone and character voice,

• in-game usability,

• fluency and readability,

• consistency with the customized MQM framework.

This allowed Peoplying to detect recurring issues, provide feedback to post-editors, and stabilize quality throughout the project.

5) Scalable Workflow for High-Volume Game Content

Because the project involved a large volume of AI MTPE, Peoplying designed the workflow to support both speed and quality. The customized MQM framework served as a shared reference point for all linguists, reviewers, and project managers.

The workflow supported:

• consistent editor onboarding,

• efficient issue classification,

• controlled feedback loops,

• quality monitoring across batches,

• stable application of client-specific standards.

This enabled Peoplying to handle large-scale AI MTPE work without allowing quality standards to vary by editor or batch.

Results

Peoplying delivered a high-quality AI MTPE program for the game client by combining MQM customization, post-editor training, and domain-specific review.

The project achieved:

• a client-specific MQM framework tailored to game localization,

• clearer judgment standards for AI MTPE errors,

• improved consistency among post-editors,

• better handling of AI-specific translation issues,

• reduced risk of surface-fluent but contextually incorrect translations,

• scalable quality control for large-volume game localization content.

By adapting MQM to the client’s game content and training post-editors accordingly, Peoplying helped the client move beyond conventional MTPE and establish a more reliable AI MTPE workflow.

What This Shows

• AI MTPE requires a different approach from traditional MTPE because AI output can be fluent while still containing serious contextual or domain-specific errors.

• MQM is not a fixed error system, but a customizable quality framework that can be adjusted by domain, client, content type, and error impact.

• For game localization, customized severity rules are essential because some errors directly affect gameplay, immersion, and player experience.

• Large-scale AI MTPE can achieve stable quality when post-editors are trained on client-specific MQM criteria and supported by a structured review process.