What is translation, really? And why does it still require human expertise in the age of AI?
The short answer: Translation is the professional practice of preserving meaning, intent, and cultural context across languages. It requires linguistic ability, cultural fluency, and strategic judgment about what matters in a given context.
The longer answer – and the one that explains why I am so excited to continue my “word work” after two decades in the Intelligence Community- is more interesting.
The Foundation: Beyond Word-for-Word
In intelligence work, you learn quickly that translation isn’t about finding equivalent words. It’s about understanding intention, recognizing subtext, and knowing what’s at stake when meaning shifts even slightly. A mistranslated phrase in a diplomatic cable isn’t just inaccurate, it can misrepresent intent, escalate tensions, or obscure critical information.
That IC background taught me something essential: context determines meaning, and missing subtext can mean missing everything.
When people ask what makes a good translation, they often expect a technical answer: accuracy, grammar, maybe vocabulary. Those things matter, but they’re just the foundation. The real work of translation lies in nuance—in what’s implied, softened, or rearranged to sound natural, credible, and true.
Accuracy gets you to correct. Nuance gets you to right.
What Linguistic Expertise Actually Looks Like
Let me show you what I mean with a few simple examples that trip up even sophisticated machine translation.
Take the Czech words stavebník and stavitel. Both can translate to “builder” in English. These words were recently used interchangeably (in Czech) in a Machine Translation Post-Editing job I was working on. Except a stavebník is the developer—the person commissioning / paying for the construction—while a stavitel is the contractor doing the physical building. That distinction determines liability, responsibility, and contractual obligations.
My guess is the Machine Translation was rendering original English content (where the meaning of the word ‘builder’ could be understood through context) into Czech and was using the words stavitel / stavebník interchangeably, ?? for variety ?? In Czech, the two words have different meanings. A professionally nuanced translation preserves the relationship between them because it understands what’s at stake.
Or consider the word pomalu. In standard written Czech it means “slowly.” But in the sentence Už bude pomalu jaro, (common in spoken Czech), it doesn’t describe pace—it expresses anticipation. The natural translation isn’t “Spring is slowly coming” but “It’s almost spring now.” The literal translation is technically correct but misses the speaker’s actual meaning entirely.
One more from some post-editing I was doing: The question (in English) had been something like “Is this a redwood or a sequoia?” And it had been machine-translated to Je to sekvoj nebo sekvoj? Even if you don’t understand a word of Czech, you can probably sound out that sekvoj sounds like “sequoia” and realize that it essentially asked, “Is this a sequoia or a sequoia?” Ummm, yes? The fact is, if you type that English question into Google Translate, you get the sekvoj or sekvoj version in Czech. Do more precise words for these trees exist in Czech? Of course. But in this case a human had to recognize they were needed: “Redwood” = sekvoj (vždyzelená); “Sequoia (giant)” = sekvojovec (obrovský).
These aren’t edge cases. They’re the everyday texture of cross-language communication.
Cultural Context and Register
Nuance isn’t just linguistic—it’s social and cultural.
Czech has formal and informal pronouns (Vy vs. ty, similar to French vous/tu), but the distinction goes beyond grammar. It signals relationships, respect, and professional boundaries. (If you’re REALLY interested, please read my 400+ page dissertation on this subject here).
A research question occurs to me: Given, say, a translation of a short story or a novel- from English into Czech (or Russian, or French, or German, or Polish) – how does the Machine Translation or Generative AI / LLM fare expressing relationships through pronouns and their related structures (corresponding verbs, etc.)? How can it know whether the translation of ‘you’ should be Ty or Vy? (or Ty or Pan/Pani in Polish, etc.)?
I’m going to hypothesize right now that a computer-generated translation will not be able to consistently get this right. From English to Czech for example it will probably get it right for true singular / plural, but not for the polite form; going Czech to English it all gets flattened to ‘you’ but that’s the problem- English might need more words to express the actual relationship conveyed by the pronouns. Which the computer will not know to generate. More to come on this.
Why This Matters: The Stakes of Getting It Wrong
In the Intelligence Community, I learned that translation failures aren’t just embarrassing—they have consequences.
A phrase that seems neutral might carry cultural or political weight. A shift in formality might signal changing relationships or priorities. The difference between “we are building” and “we will build” can indicate commitment level, timeline confidence, or negotiating position. When you’re working with communications for intelligence reporting, missing that nuance doesn’t just produce awkward phrasing; it produces misunderstanding. In national security, that can even cost lives.
The same principle applies across other high-stakes contexts:
Legal and contractual documents: A misread nuance can shift obligations, change liability, or create ambiguity where precision is required.
Marketing and brand communications: Tone mismatch undermines credibility. A campaign that works in English might feel presumptuous, cold, or culturally tone-deaf in Czech or Russian if the register isn’t adapted.
Academic and research writing: Precision affects reputation. A poorly translated methodology section can make rigorous research appear sloppy.
Business communications: Relationship-building depends on appropriate formality, warmth, and cultural awareness. Getting the register wrong can damage trust before the conversation even begins.
This is why translation is a professional discipline, not just a language skill.
Human + Machine Teaming; or What AI Does Well, and When Humans Are Essential
AI translation has improved dramatically, and it has a place in modern language workflows. For instance, for straightforward technical documentation, machine translation can be remarkably effective.
But here’s what AI still can’t do: it can’t tell you whether that Czech speaker meant “almost spring” or “slowly spring” when they said už bude pomalu jaro. It can’t distinguish whether your audience expects formal distance or collegial warmth. And it can’t flag the moment when literal accuracy produces cultural awkwardness or changes meaning in ways that matter. These aren’t failures of technology. They’re the limits of pattern matching without contextual judgment.
I’m not arguing against AI—I’m arguing for strategic use of it.
Human expertise is essential when:
Tone and relationship matter. Formal vs. informal, authoritative vs. collaborative—these aren’t just vocabulary choices; they’re relational signals.
Cultural context shapes meaning. Idioms, metaphors, humor, and directness vary dramatically. AI can identify idioms but rarely knows how to adapt them. More on the art of “transcreation” soon.
Ambiguity requires resolution. When a sentence could mean two things, AI picks the most common interpretation. A human translator asks what makes sense given the speaker, audience, and purpose.
The stakes are high. Contracts, medical documents, brand positioning: these aren’t places for approximation.
The key is knowing which tool fits which task. And that discernment? That’s also a kind of translation expertise.
For many clients, an acceptable approach combines both human and machine: AI handles first-pass translation of straightforward content, and human review ensures accuracy, tone, and cultural fit. Human-only translation is reserved for anything involving nuance, relationships, legal weight, or brand voice. But don’t be fooled: often that “human review” of an AI translation takes as much time, energy, and research as a from-scratch “human only” translation does.
How I Approach Professional Translation
At AC Stepan WordWorks, I treat every project as a judgment call: What does this text need to accomplish? Who’s the audience? What’s at stake if the tone is off or the nuance is lost?
Translation isn’t about converting words; it’s about preserving intent, building trust, and ensuring that what you meant to say is what they actually hear (read).
That means:
Reading for subtext and assumption. Every sentence carries assumptions about who’s speaking and to whom.
Considering cultural rhythm. Some languages prefer directness; others value indirection. Some expect formality; others prioritize warmth.
Asking what could go wrong. Where could ambiguity create problems? Where might tone undermine credibility?
Knowing when to push back. Sometimes the best translation advice is “this won’t work in the target language—here’s why, and here’s what would.”
Machine translation and LLMs can be tools in the process. Human expertise is what ensures the result works—not just linguistically, but strategically, culturally, and contextually.
Translation as Empathy
Good translation is like good architecture: the structure must be sound, but the light must fall right, too.
After decades of working across languages, cultures, and contexts—in academia, in intelligence, now in professional language consulting—I’ve come to see translation as a kind of empathy. It’s not about finding matching words. It’s about finding matching minds.
It’s about understanding not just what someone said, but what they meant. Not just the message, but the relationship it’s trying to build. Not just the content, but the context that makes it matter.
If you’ve ever stumbled across a phrase that almost worked but didn’t quite, you’ve already felt the gap that nuance fills. That’s where the translator lives—in the space between languages, finding the version that doesn’t just say the same thing, but does the same work.
Translation accuracy isn’t just about words—it’s about meaning, culture, and trust. At AC Stepan WordWorks, I help clients bridge the gap between human nuance and efficiency through careful review, editing, and linguistic consulting.
Explore more insights and case studies on my WordWorks Blog or reach out for a consultation on how to evaluate and elevate your multilingual content.
Anne Cofield Stepan
Founder & Principal Language Consultant
AC Stepan WordWorks
AC Stepan WordWorks Language Consulting Services 