Translation Workflows: From Sausage Making to Smart Collaboration

This article was originally published as part of a series of articles on our LinkedIn page:

From Craft to Segments

Traditionally, translators worked holistically: reading an article, story, or manual, researching terminology, and creating a unique, coherent text in the target language. For marketing copy and prose, that approach is still gold.

But when it comes to large volumes of repetitive technical content, few people enjoy translating the same “Plug A into B” sentence hundreds of times – and even fewer want to pay for it.

This is why translation memory systems became the backbone of the industry. By breaking content into segments and storing them for reuse, companies saved enormous amounts of time and money. But there’s a downside: Most machine translation engines see only isolated sentences. They don’t know what the “it” refers to, and in languages like German, where objects can be masculine, feminine, or neuter, this can create embarrassing mistakes. Human translators step in to restore fluency, but increasingly they’re polishing machine output rather than translating from scratch. Unsurprisingly, many people dislike the work: it often pays less and feels less creative.

Enter Large Language Models

Recently, large language models (LLMs) have shown that translation is just another form of “text transformation.” Answering “What color is the sky?” isn’t fundamentally different from translating it into Italian. Both involve mapping input to expected output.

The catch? LLMs sometimes “wander off,” adding or omitting details. That’s unacceptable when manuals need to remain aligned with a single source of truth.

A Third Way: Hybrid Workflows

So, where does this leave us?

Imagine your source content – manuals, quick guides, marketing copy – starts in English. Professional translators or agencies deliver reliable one-to-one translations. These are accurate, but they can lack a certain polish, that elusive je ne sais quoi.

Traditionally, brands with large budgets would send the translation for secondary review by another human expert. Today, LLMs can take on that “last mile.” A brand-side native speaker, even if not a full-time translator, can use an LLM to refine tone, terminology, and flow – guided by a style guide, glossary, or even just a well-crafted prompt. The result: content that is factually correct and aligned with the brand voice, without tying up scarce human resources.

What This Means for Organizations

Large corporations may hesitate to trust such “downstream optimization,” but for smaller teams, the approach can be powerful. Human translators ensure accuracy, machines handle volume, and LLMs bring in the finishing touches.

It’s not about choosing human or machine, NMT or LLM. It’s about orchestrating all three – so that every person, and every machine, contributes where they’re strongest.

Over to You

How does your organization approach translation? Do you rely entirely on your language service provider to polish documents, or do you bring in your own teams for the last mile? I’d love to hear your experiences.

↻ 2025-11-07