I’ve worked in the language industry for more than 30 years. Sometimes I can feel like a dinosaur, but really, I just feel experienced.
The language industry has changed significantly, especially with new technologies. Lately, it feels like that’s all anyone talks about.
But what is being left out of the conversation are the details. And it is true about translation, perhaps more than anywhere else, that the devil is in the details.
Errors are a natural part of translation. It’s always been a human task, and people aren’t perfect. Unfortunately, AI isn’t perfect either.
So, imperfections and errors do happen. A slight shift in meaning here. An ambiguous phrase there. A missed opportunity for a culturally and linguistically appropriate adaptation. So, in almost every context, we expect them to happen, and we have (so far) been able to rely on the fact that somehow, someone will “catch them” in the end.
That approach might work in some settings, but it’s not enough for regulated, high-stakes environments. In those cases, errors often point to larger issues, such as gaps in governance.
In the past, multilingual content was usually considered late in the process. That was actually the best case, since at least someone planned for the content to be translated after it was drafted and approved.
At worst, translation was not considered at all. When that happened, there were not only timeline problems but also budget issues.
When problems came up, like a client complaining about a translation, we often blamed the usual suspects: tight deadlines, insufficient context about the audience or intent, mistakes in the original text, or errors from the client’s reviewer, who is rarely a trained linguist.
Sometimes these reasons are true, but often they’re just symptoms, not the real problem. When language is key to compliance, safety, or public trust, treating it as an afterthought doesn’t make sense.
In regulated settings, language is more than communication. Getting it right becomes as important as providing evidence.
For example, translated content explains clinical trial results to regulators, informs patients about health risks before surgery, and reminds employees about safety instructions.
In this context, an error isn’t just a quality issue to fix at the end. It’s a sign of a governance problem that raises bigger questions, such as:
• Who is accountable for the accuracy of this message across languages?
• Who validated the message, and how does the message land with different audiences?
• Who set the rules for what tools could or could not be used and at what point(s) of the translation workflow?
• Who signed off, and on what basis?
If those answers are missing or unclear, the issue is with the system, not just the language.
We see these same patterns in life sciences, education, and the public sector. The most common issues are:
Language decisions are made informally
Teams often rely on memory – “this is how we’ve always done it” - and personal judgment instead of clear rules. This works until people change roles, teams grow, or the stakes get higher.
Technology outpaces policy (every time)
Many quickly adopt machine translation and AI tools to save time or money, but they don’t set clear boundaries. No one defines when automation is okay, when human review is needed, or how to assess risk.
Ownership is scattered
Legal, compliance, operations, procurement, and vendors all handle multilingual content, but often no one owns the whole process from start to finish.
A review is a box to check
Bilingual reviewers are often asked to “take a look” without clear responsibilities. This makes reviews inconsistent and can even create new risks.
At first, these problems don’t look like governance failures. They show up as delays, extra work, and frustration - until something more serious happens.
When language is tied to regulation, safety, or rights, the stakes rise right away.
A patient could be harmed if a hospital uses a bilingual employee or family member instead of the right vendor. A policy might be translated accurately but not fit the local legal context. An employee could get hurt at work, even if the safety instructions were reviewed.
At that point, the question isn’t “Who mistranslated this?” anymore. The real question becomes, “Why did our system allow this to happen?” This is where we move from quality assurance to governance.
In a previous blog, we mentioned W. Edwards Deming’s point that you can’t “inspect quality into a product.” Instead, quality should be built into the whole process of multilingual communication. I suggest organizations treat multilingual communication as a governed system, not just a service. That system should:
• Define ownership of the individual parts of the multilingual process.
• Determine risk exposure in all workflows.
• Identify acceptable uses of Artificial Intelligence and Machine Translation
• Describe review roles and give them explicit accountability
• Produce documentation that captures not only what was done, but why quality is important. But without governance, quality alone can’t withstand the demands of growth and constant scrutiny in regulated industries.
The biggest change is in how we think about the problem. Stop asking whether your translations are accurate. Start asking whether your multilingual communication is governed.
Today, language errors are rarely just about language. They’re early warning signs that governance hasn’t kept up with real-world needs. Organizations that see this early can avoid learning the lesson the hard way.