Why Localization Is Not Translation: The Case for Post-Editing Standards

The $49.68 Billion Question

In 2023, the global language services and technology industry generated US$49.68 billion, down 4.5% from US$52.01 billion in 2022 (CSA Research). That’s a real drop, and it’s not just a blip. The industry is consolidating, and the pressure to do more with less is real. But here’s the thing: most of that money isn’t for translation in the old sense. It’s for localization — adapting content so it feels native, not just translated. And localization, at its core, is a post-editing problem. If you’re still treating it like a translation problem, you’re bleeding money and quality.

Translation vs. Localization: A False Binary

Let’s get one thing straight: translation is the act of converting text from one language to another. Localization is that, plus adapting it to a specific culture, market, or audience. The difference isn’t just about vocabulary; it’s about context. A literal translation might be correct, but it can be wrong for the market. Think of a marketing slogan that relies on a pun — you can’t just translate it; you have to recreate it.

In the old days, localization was done by human translators who knew the target culture. They’d adjust idioms, units of measurement, even date formats. But now, with machine translation (MT) everywhere, the line has blurred. MT is the default first pass for many projects, and the human’s job becomes editing the raw output. That’s post-editing, and it’s a different skill set.

The Post-Editing Reality Check

Here’s a number that should wake you up: Google’s neural MT, when it launched in 2016, reduced translation errors by 55–85% on sampled sentences from Wikipedia and news compared to the previous phrase-based system (GNMT at production scale). That sounds great, but it doesn’t mean the output is ready for prime time. MT still struggles with idiomatic expressions, cultural nuance, and domain-specific terminology — exactly the stuff localization is about (Machine translation). So you either accept lower quality, or you have a human clean it up.

That clean-up is post-editing, and there’s a standard for it: ISO 18587:2017. It specifies requirements for full human post-editing of MT output and the competences of post-editors. If you’re doing localization and you’re not following that standard, you’re flying blind. You might think you’re saving money by skipping the editing step, but you’re just shipping half-baked content.

Why BLEU Is the Wrong Yardstick for Localization

We’ve all seen the BLEU score. It’s been the go-to MT metric since IBM introduced it in 2002 (IBM Research). But BLEU compares n-gram matches between machine output and human references. It doesn’t measure whether the translation is actually good for a specific audience. It’s a crude proxy, and it’s increasingly seen as unreliable because it correlates poorly with human judgment (IBM Research).

For localization, you need to measure post-editing effort, not just translation accuracy. That’s where metrics like Human-targeted Translation Edit Rate (HTER) come in. HTER is a variant of TER where a human creates a reference closest to the machine output, and it’s widely used as a proxy for post-editing effort (Translation Edit Rate). That’s a tool that actually tells you how much work it’ll take to get the MT output to a usable state. BLEU can’t do that.

So, if you’re a localization manager, stop obsessing over BLEU. Start measuring HTER, or better yet, use a neural metric like COMET or BLEURT, which correlate better with human judgment (COMET, BLEURT). But even those don’t capture the full localization picture. The only real test is whether a native speaker in the target market finds the content natural and effective.

The ISO 18587 Approach: A Step-by-Step Fix

So, what do you do? You adopt ISO 18587:2017 as your process framework. Here’s what that means in practice:

• Define the level of post-editing needed: full vs. light. For localization, it’s almost always full.
• Hire post-editors who have the right competences: language skills, domain knowledge, and the ability to work with MT output.
• Set up a workflow where MT is used as a starting point, but a human reviews every segment, not just the ones that look “hard.”
• Measure and track post-editing effort using tools like HTER, so you can improve your MT engine and your processes over time.

This isn’t just about compliance; it’s about quality. The European Commission’s Directorate-General for Translation, one of the largest translation services in the world, produced about 2.6 million translated pages in 2022 (European Commission). They don’t just throw MT at everything; they have a rigorous process that includes post-editing. And they offer eTranslation, a free neural MT service, but they know it’s a starting point, not the final product (eTranslation).

Quick tip: If you’re just starting with post-editing, don’t try to do it all at once. Pick one high-volume, low-stakes content type, like user-generated reviews, and pilot a full post-editing workflow there. Measure the effort and the quality, then scale up.

Bottom Line: Localization Is a Post-Editing Discipline

If you take away one thing, it’s this: stop treating localization as a translation task. Start treating it as a post-editing discipline, and use ISO 18587:2017 as your guide. That means investing in post-editing skills, measuring effort with tools like HTER, and accepting that BLEU is a relic. The industry is changing, and the money is in the editing, not the raw MT. If you don’t adapt, you’ll be left behind.

Sources

• Machine translation (Wikipedia) - https://en.wikipedia.org/wiki/Machine_translation
• IBM Research - https://research.ibm.com/blog/bleu-nlp-benchmark-anniversary
• Translation Edit Rate (ACL Anthology) - https://aclanthology.org/2006.amta-papers.25/
• ISO 18587:2017 (ISO) - https://www.iso.org/standard/62970.html
• European Commission translation department - https://commission.europa.eu/about-european-commission/departments-and-executive-agencies/translation_en
• CSA Research - https://csa-research.com/l/media/Language-Services-and-Technology-Industry-Faces-Revenue-Decline-but-Remains-Poised-for-Transformation