What is MQM
MQM (Multidimensional Quality Metrics) is a standardized Linguistic Quality Assurance (LQA) framework for evaluating translation quality across error categories.
Evaluating translations under MQM helps you identify strengths and weaknesses in your localization process, measure the impact of process changes on translation quality, and make data-driven decisions. For example, a rise in terminology errors may indicate your glossary needs updating; a rise in style errors may point to gaps in your style guide.
MQM LQA can be applied to both human and machine translations.
MQM Schemas
MQM evaluates translation quality by categorizing errors and assigning each a severity weight. The number and severity of errors across categories produce an MQM score for a body of content.
Smartling offers several schema templates when setting up LQA:
- Smartling LQA Schema (Simplified MQM)
- Smartling LQA Schema (Full MQM)
- Smartling LQA Schema (MQM with Repeated Error Types)
- Smartling LQA Agent MQM Schema
Smartling LQA Schema (Simplified MQM)
A streamlined version of the full MQM schema with fewer categories and error types, designed to make evaluation faster and more accessible for evaluators.
| Category | Errors | Description |
| Accuracy |
|
Errors in this category occur when the target text does not accurately correspond to the propositional content of the source text. |
| Linguistic Conventions |
|
Errors that are related to the linguistic well-formedness of the translated text. |
| Style |
|
Errors in text that are grammatically acceptable but deviate from organizational style guides or exhibit inappropriate language style. |
| Technical & Locale |
|
Errors occur when the translation product violates locale-specific content or formatting requirements for data elements. |
| Other |
|
Smartling LQA Schema (Full MQM)
The full MQM schema follows the industry-standard MQM framework and includes the complete catalog of error types.
| Category | Errors | Description |
| Terminology |
|
Errors arise when a term does not conform to normative domain or organizational terminology standards or when a term in the target text is not the correct, normative equivalent of the corresponding term in the source text. |
| Accuracy |
|
Errors in this category occur when the target text does not accurately correspond to the propositional content of the source text. |
| Linguistic Conventions |
|
Errors that are related to the linguistic well-formedness of the translated text. |
| Style |
|
Errors in text that are grammatically acceptable but deviate from organizational style guides or exhibit inappropriate language style. |
| Locale Conventions |
|
Errors occur when the translation product violates locale-specific content or formatting requirements for data elements. |
| Audience Appropriateness |
|
Errors arising from the use of content in the translation product that is invalid or inappropriate for the target locale or target audience |
| Design & Markup |
|
Errors related to the physical design or presentation of a translation product, including character, paragraph, and UI element formatting and markup, integration of text with graphical elements, and overall page or window layout. |
| Other |
|
Smartling LQA Schema (MQM with Repeated Error Types)
This schema separates initial and repeated errors. Repeated error types have a Neutral severity by default and are non-modifiable. Linguists cannot change their severity unless a content owner updates the error type's severity setting.
| Category | Errors | Description |
| Accuracy |
|
Errors in this category occur when the target text does not accurately correspond to the propositional content of the source text. |
| Audience Appropriateness |
|
Errors arising from the use of content in the translation product that is invalid or inappropriate for the target locale or target audience |
| Custom |
|
|
| Linguistic Conventions |
|
Errors that are related to the linguistic well-formedness of the translated text. |
| Style |
|
Errors in text that are grammatically acceptable but deviate from organizational style guides or exhibit inappropriate language style. |
| Technical & Locale |
|
Errors occur when the translation product violates locale-specific content or formatting requirements for data elements. |
Smartling LQA Agent MQM Schema
The LQA Agent schema is customized to match the error types that LQA Agent can identify and record.
| Category | Errors | Description |
| Accuracy |
|
Errors occurring when the target text does not accurately correspond to the propositional content of the source text, introduced by distorting, omitting, or adding to the message. |
| Audience Appropriateness |
|
Errors arising from the use of content in the translation product that is invalid or inappropriate for the target locale or target audience. |
| Linguistic Conventions |
|
Errors related to the linguistic well-formedness of the text, including problems with grammaticality, spelling, punctuation, and mechanical correctness. |
| Style |
|
Errors occurring in a text that are grammatically acceptable but are inappropriate because they deviate from organizational style guides or exhibit inappropriate language style. |
| Technical & Locale |
|
Errors occurring when the translation product violates locale-specific content or formatting requirements for data elements. |
Error Severity
Each error is assigned a severity level (Critical, Major, Minor, or Neutral) with a default weight. These weights determine how much each error affects the overall MQM score.
| Severity | Description | Default Weight |
| Critical | Severe errors with potential legal or commercial consequences. Examples: omitting negation in a terms and conditions text, or mixing up the two parties in a legal contract. | 25 |
| Major | Errors that may confuse readers, sound unnatural, introduce inconsistencies, or deviate from established linguistic assets (style guide, TM, glossary, instructions). | 5 |
| Minor | Objective errors, including typos and punctuation issues, that do not affect the reader's understanding. | 1 |
| Neutral | Errors that should not affect the final score but still need to be documented. Use for preferential suggestions, feedback, or kudos to translators. | 0 |
For repeated and preferential errors, recording severity can still be useful. Depending on your LQA program, however, you may want to exclude these errors from the overall MQM score. When setting up an MQM-compatible schema, you can optionally neutralize specific error types so they do not affect the MQM score. Neutralized errors still appear in the LQA Errors & Arbitration report with their recorded severity.
Customization
You can customize the name, severity level weights, and Acceptable Penalty Points on any unpublished schema. Severity levels (Critical, Major, Minor, Neutral) are fixed and cannot be changed.
- Go to Account Settings > Linguistic Quality Assurance.
- On any draft schema, click the ellipsis (…) > Edit.
- In the Edit Schema modal, update any of the following:
- Name: rename the schema.
- Severity Level Weights: adjust the weight for Critical, Major, Minor, or Neutral errors. The sum of all weights cannot exceed 100.
- Acceptable Penalty Points: set the pass/fail threshold. The Raw Quality Score and Calibrated Quality Score passing thresholds update automatically as you adjust this value.
- Click Save.
Acceptable Penalty Points
Acceptable Penalty Points (APP) define the pass/fail threshold for your MQM evaluation, giving you a consistent, objective way to determine whether translated content meets your quality bar.
When editing a draft schema, you can set the APP value in the Edit Schema modal. The default is 20, which produces a Raw Quality Score passing threshold of 98 and a Calibrated Quality Score passing threshold of 80. Both thresholds update dynamically as you adjust the APP value.