The Part of Translation Nobody Talks About

Ask someone what AI has changed about translation and they'll probably mention speed, or quality, or the ongoing anxiety about whether human translators still have a future. These are the conversations that get airtime. They're real conversations. But they tend to skip over something more unglamorous, and in some ways more consequential, what happens to the document itself.

Translation isn't just moving words from one language to another. It's also moving a document. And documents are stubborn, finicky things that don't travel well.

A scanned PDF of a certified certificate. A multi column corporate brochure. An invoice with a tightly formatted table. These files have layouts, carefully constructed ones, and when you translate their contents, the layouts break. Text expands. German and French, notoriously, are wordier than English. A text box that fit perfectly in the original becomes a typographical disaster in the target language. And that mess has to be fixed, manually, by someone with design skills, before the translated document is usable.

This is Desktop Publishing work, and it has quietly consumed somewhere between 20 and 30 percent of many localization budgets for years. Not because anyone wanted it to. Because there was no better option.

The Hidden Tax on Every Document

DTP in localization is one of those problems that's invisible until you're inside it. From the outside, you commission a translation and receive a translated document. From the inside, a project manager has spent hours rebuilding a layout from scratch, an OCR tool has mangled a scanned file beyond recognition, and a designer has been brought in to fix what the software couldn't.

The traditional workflow runs something like this: extract the text, rebuild the template, translate, reformat, review. Five steps before the document is ready to deliver. For a standard Word file, that's annoying. For a scanned PDF or a complex layout, it's genuinely expensive.

The problem has always been that most translation tools were built for text. They handle strings well. They fall apart when the content is baked into an image, or when the document's structure is part of its meaning, which, for certified documents, legal contracts, and marketing collateral, it usually is.

What AI Actually Changes Here

The more interesting development isn't AI translation. It's AI reconstruction.

Tools like Cipher, built by Translayte, take a different approach to the problem. Instead of extracting text and hoping the layout survives, they use AI to analyse the document, translate the content, and rebuild the original layout simultaneously, on a blank canvas, in seconds. The output is a formatted document that's 95% ready to deliver, with a human reviewer making final tweaks rather than rebuilding from scratch.

Tayo Ademolu, Translayte's founder, built Cipher out of direct frustration with the status quo. Most OCR and machine translation tools handle clean digital files reasonably well, but struggle badly when faced with scanned documents or complex visual layouts. That's exactly the category of document that causes the most pain in professional translation workflows, and the one that AI, applied thoughtfully, turns out to be particularly good at handling.

The workflow it enables is genuinely different: upload the file, let the AI reconstruct and translate it, review the output in a browser editor, export. What used to take hours takes seconds. Not as a theoretical benchmark, but as a practical daily reality for the language service providers and project managers who are already using it.

Why This Matters Beyond Localization

It's tempting to frame this as a niche localization story. But the underlying problem, documents that need to be transformed without losing their structure, is everywhere.

Legal teams dealing with multilingual contracts. HR departments distributing policy documents across global offices. Marketing teams localising brochures for regional campaigns. Any workflow that moves formatted documents across language boundaries runs into some version of this problem. The localization industry has simply been living with it the longest, which is why the solutions are emerging there first.

The deeper shift is that AI is starting to understand documents as objects, not just as containers for text. Layout, structure, visual hierarchy, these are things that carry meaning, and that have historically been very hard to preserve automatically. The fact that they're now within reach of automation changes the economics of a lot of work that nobody was particularly excited to be doing by hand.

The Work That Disappears

There's a pattern worth noticing across how AI is changing knowledge work. The tasks that disappear first are rarely the ones that feel important. They're the ones that felt like friction, necessary, time consuming, invisible to anyone outside the process.

DTP in translation is a good example. Nobody got into localization because they loved rebuilding scanned PDFs in InDesign. It was just the cost of doing the work properly. Now it's becoming the kind of cost that, quietly, doesn't need to exist anymore.

That's not a dramatic story. It won't get written up as a revolution. But for the project manager who used to spend half their day on it, it's probably the most useful thing AI has done for them so far.

If you want to see it in action, Tayo Ademolu walked through a live demo of Cipher on a recent episode of Language Tech Unboxed, including some genuinely tricky documents, landscape transcripts, expanding invoices, scanned certificates. Worth watching if you've ever winced at a broken layout and reached for InDesign.