Translation Quality Evaluation
Does your translation work? Measure it, don't guess.
"Good" and "bad" are not useful answers when you need to approve a translation, compare vendors, or decide whether machine or AI translation is ready for your content.
We evaluate translations against clear, agreed specifications using the Multidimensional Quality Metrics (MQM) framework — so you get evidence, a score, and a decision you can defend.
What Is MQM?
MQM is an open framework for analytic translation quality evaluation. It works for human translation, machine translation, and AI-generated translation alike.
Instead of a general impression, trained evaluators mark each issue in the translation, classify it with a shared error typology, and rate how serious it is. A scoring model then turns those annotations into a quality score you can compare across projects, languages, and vendors.
MQM is stewarded by the MQM Council, building on work that began in the EU-funded QTLaunchPad and QT21 projects. It is aligned with ISO 5060:2024, the international standard for evaluating translation output.
The MQM Error Typology
MQM Core groups translation issues into seven dimensions. Each one has more specific error types beneath it, so you can use the full set or a subset that fits your content.
Terminology
Terms that break domain or company standards, or that are not the right equivalent of the source term.
Accuracy
Meaning that is distorted, left out, or added compared with the source.
Linguistic Conventions
Grammar, spelling, punctuation, and other problems with well-formed language.
Style
Text that is correct but does not follow the style guide or uses the wrong tone or register.
Locale Conventions
Dates, numbers, currencies, addresses, and other formats that do not match the target locale.
Audience Appropriateness
Content that does not suit or is not valid for the target audience or culture.
Design and Markup
Layout, formatting, tags, and other problems with how the text is presented.
From Errors to a Score
Severity matters as much as the number of errors.
Each issue gets a severity level. Penalties grow sharply with severity, so one serious error counts far more than several small ones. A common setup is:
- Neutral (0 points): a preference or a note that is not an error
- Minor (1 point): does not affect meaning or usability
- Major (5 points): affects meaning or would confuse the reader
- Critical (25 points): makes the content unfit for purpose — often an automatic fail
The penalty points are added up and normalized by word count, then converted into a quality score. Comparing the score with an agreed passing threshold tells you whether to accept, revise, or reject the translation.
MQM offers a raw score, which is best for comparing across projects and organizations, and a calibrated score, which you can tune to your own tolerance for error. Read more about the MQM scoring models.
How We Evaluate
A clear process for a defensible result.
- Agree on specifications. We define the purpose of the content, the audience, the error types that matter, severity levels, weights, and the passing threshold.
- Select the sample. We choose a representative sample — or evaluate the full text — based on your risk and budget.
- Annotate. Qualified evaluators mark every issue with its error type and severity, segment by segment.
- Score and decide. We calculate the quality score and give a clear pass or fail against your threshold.
- Report and improve. You get a report with examples, error patterns, and root causes — source text, terminology, workflow, or tools — so the next translation is better.
When Quality Evaluation Helps
Machine and AI Translation
Find out whether machine or AI output is fit for your content, compare engines and prompts, and decide where human post-editing is needed.
Vendor Selection and Management
Compare translation providers on the same objective scale and track their quality over time.
High-Risk Content
Get independent evidence of quality for health, legal, regulatory, and public-facing content before it reaches your audience.
Quality Programs
Set up your own MQM-based specifications, scorecards, and thresholds, and train your reviewers to use them consistently.
Source Content Clarity
Many translation errors start in the source. We connect evaluation results to plain language improvements upstream.
English and Spanish
Evaluation by ATA-certified translators, with partner evaluators for other language pairs.
Let's Measure Your Translation Quality
Tell us about your content, languages, and goals — a one-time evaluation, a vendor or machine translation comparison, or an MQM-based quality program. We'll get back to you within 24 hours on business days.