How Translation Quality Is Assured — and How You Can Check It Yourself

There is a quiet but decisive difference between inspecting a translation and assuring it. Inspection happens once, at the end, when someone glances over a finished file and hopes nothing was missed. Assurance is a system built so that errors are caught — or never made — long before delivery, by people other than the one who did the work. In legal translation the distinction is not academic. The person who eventually finds an uncaught error is the person you least want to find it: a judge reading a memorandum, a registrar at a service counter, a bank officer comparing a name against a passport. By then the cost of the error is measured in a rejected filing or a lost afternoon, not a corrected line.

This page keeps three kinds of statement carefully apart, because blurring them is how buyers get misled. Some of what follows is general professional information — how the translation industry has agreed a quality service should work. Some is an official requirement set by a UAE authority, and where that is the case we name the authority. And some is simply how we run our own review. When we describe a published model, we say it is a model; when we describe our own practice, we say it is ours; and we never dress our practice up as somebody else's rule.

It helps to say plainly what quality means here, because the word is used loosely. In legal translation quality is not elegance and it is not a pleasing turn of phrase. It is a set of testable properties: the meaning of the source rendered faithfully with nothing added or dropped; the same term rendered the same way every time; every number, date and reference transcribed precisely; every name preserved exactly as the identity documents spell it; the layout mirrored so the deliverable can be checked against the original; and a target text that reads naturally to a native professional. Each of those is checkable. Assurance simply means checking each one deliberately, by a second pair of eyes, rather than trusting that the first pass got everything right.

The rest of this page shows the mechanisms that do that work — independent revision, terminology control, format fidelity and a final verification pass — and then it does something most providers never do: it hands the tools back to you. If you do not read the target language you are not helpless, and you should not have to rely on faith. There are things you can check yourself, questions you can ask that reveal whether a real quality process exists, and warning signs that tell you when it does not. We would rather you knew them.

There is a published international answer to the question of what a quality translation service should look like. ISO 17100, issued in 2015 as the successor to the earlier European standard EN 15038, was developed by the international standards body's technical committee on terminology and language resources. It does not treat translation as one undifferentiated job; it describes a staged process with named human roles — translator, reviser, reviewer, proofreader, project manager — and it sets out requirements for the workflow, for the competences of the people in it, and for how a project is managed from the client agreement to archiving.

We describe our workflow in the vocabulary of this model because it is the model our process follows. And here we are deliberate about a line that many websites blur: we make no claim to hold any certificate. Describing the model you work to and holding an audited certificate issued by an accredited body are two different things, and we keep them apart. A provider that has genuinely been certified can say so and show the certificate; a provider that merely follows the process should say exactly that, which is what we do. If a certificate number is ever shown to you by anyone, it should be verifiable — and a description of a model is not a certificate, however professionally it is written.

The model is useful beyond its list of roles, because it also requires the things that turn good intentions into a repeatable process. Client requirements are documented at the start of a project rather than assumed. Quality-control procedures are defined in advance. And there is a feedback and corrective-action loop, so that when something does go wrong the process is adjusted to stop it recurring, instead of the same mistake being made again on the next file. One boundary is worth stating plainly: the model covers human translation. Raw machine translation followed by a post-editing pass sits outside its scope — a separate framework, ISO 18587, exists for that work — so when a provider tells you a document was translated and revised, machine output tidied up afterwards is not what those words are entitled to mean.

Finally, the model answers only one of two questions you should keep separate in Dubai. Whether a translation is admissible before a UAE court, notary or government body is a licensing question, answered by whether the translator is registered on the roll the Ministry of Justice maintains. Whether the translation was produced by a disciplined, error-resistant process is a quality question, answered by how the provider actually works. A serious provider must be able to answer both, and neither answer excuses the absence of the other. This page is about the second question. It does not, and cannot, promise how any particular authority will decide, because that decision belongs to the authority and follows its own rules.

Of all the dimensions of quality, consistency is the one a non-linguist can partly assess without help. Pick a defined term, a party name or a recurring clause, and look at whether it appears the same way each time it recurs — across the pages of one document, and across the several documents of a single matter. If the same registered company name is spelled two ways, or a defined term arrives worded differently in the second agreement than in the first, you are looking at a visible symptom of weak assurance. Professionals call the problem drift, and it does more damage than most single errors, because it makes a reader doubt the whole set rather than one line.

The mechanisms that hold terminology steady are unglamorous and effective. A matter glossary, or termbase, records how each defined term is to be rendered, which entity names must appear exactly as their official registration prints them, and which elements are not to be translated at all — trademarks, statutory titles with fixed forms, names that must match an identity document letter for letter. A translation memory sits alongside it, storing previously approved sentence pairs from your own earlier files, so that the wording your counterparties accepted before is the wording they meet again. Its value in legal work is continuity, not a discount: the fifth document in a series speaks the same language as the first.

But the tools do not assure consistency by themselves — the revision stage is where consistency is actually verified. The reviser does not only compare the translation against the source; where a matter glossary and earlier files exist, they confirm that defined terms match the glossary and match what was approved before. This is the difference between a memory that suggests a rendering and a person who confirms it survived every page. Consistency, in a real quality process, is checked. It is never simply hoped for because the software was switched on.

This gives you a concrete thing to ask any provider, and a concrete thing to watch. Ask whether there is a glossary for your matter and whether a translation memory carries your earlier work forward. Watch for the opposite: a different freelancer each time with nothing binding them to a shared list, and terminology that quietly changes from one delivery to the next. A provider that cannot tell you how it keeps the same term the same is a provider that is not, in any reliable way, keeping it the same.