A framework, not one universal score
The name MQM does not by itself tell a reader how a result was produced. A useful evaluation description needs the selected categories, severity rules, scoring model and acceptance conditions. The content and audience also matter.
In our editorial view, the practical benefit is explainability. “There is an omission in this instruction” can be examined more directly than “The translation feels weak.” The finding still needs evidence and a decision about whether it is actually an error.
An illustrative error discussion
“You can cancel until Friday.”
“You must cancel on Friday.”
The example changes permission, timing and reader expectations. A reviewer can point to that change and explain why it conflicts with the source and project brief. Another reviewer should be able to inspect the same evidence.
Now compare a translation that preserves the meaning but sounds more formal than the requester wanted. Whether that difference is an error depends on the agreed requirements. The exercise demonstrates why semantic accuracy and preference should not be merged without explanation.
These examples are original illustrations, not official MQM training cases.
Relationship to LISA and DQF
MQM's documented history identifies the LISA QA Model as an influence and describes a more modular approach. It also records the harmonized DQF-MQM typology and its replacement by MQM Core. This is a documented conceptual relationship, not organizational succession. MQM history.
Using AI without hiding the evaluation setup
Our recommendation is to keep automated findings provisional until their reliability is understood. Preserve the source, target, instructions and evaluator configuration. Have humans adjudicate a representative set of findings and missed errors before using a score for consequential decisions.
Deterministic calculation of an approved scorecard is different from asking a model to decide which errors exist. The first can be repeatable while the second remains uncertain. A model's confidence should not silently become error severity.
Reading results
Ask what the result supports: a correction, a diagnostic profile, an acceptance decision or a comparison. Do not assume that two reports using the MQM name are directly comparable.
Source record
MQM Council, scoring guidance and history. This article describes the framework and offers editorial advice; it does not claim endorsement.
MQM source material is credited to the MQM Council under CC BY 4.0; see its licensing notice. The explanations here are paraphrased and abridged; examples and implementation advice are original additions. No endorsement is implied.