AI may change translation, but human judgment still matters
As AI-powered tools become faster and more fluent, experts say translators are likely to evolve into post-editors, cultural advisers and quality controllers rather than disappear
By Akhtar Pathan
KARACHI, Pakistan (MNTV) — With artificial intelligence looming over jobs across industries, translators and editors have reason to wonder whether they could become among the last professionals in their fields.
But language and translation experts offer a more measured outlook: However sophisticated AI-powered translation tools become, they still lag behind seasoned human translators where cultural understanding, judgment and context matter most.
Rather than ending the human role in translation, AI is redefining it.
Blind expert evaluations have found that top-tier AI systems can match the performance and error rates of junior-to-midlevel freelance translators but still consistently lag behind senior professional linguists.
AI can handle massive volumes of low- and medium-stakes text with little or no human intervention. But experts in linguistics and artificial intelligence say human supervision and final revision remain essential for high-stakes, culturally sensitive and legally binding communications.
The result is less an extinction of the profession than an evolution. The traditional translator is increasingly becoming a bilingual editor, cultural consultant, terminology specialist and AI supervisor.
What AI can miss beneath words
Ghulam Ali, director of the Centre for Languages and Translation Studies at Allama Iqbal Open University, says AI has become highly proficient at what he calls “surface accuracy” — vocabulary, grammar and standard phrasing.
“What it misses is what sits beneath the surface, and that is exactly where professional fields are most sensitive,” he says.
In journalism, even a seemingly small choice of words can alter the meaning of a report.
“I am afraid if AI can be employed in journalism, its word choice may pose a problem. Whether someone is ‘killed’ or ‘martyred,’ a group is ‘militant’ or ‘freedom fighter,’ changes the meaning of the whole report. A machine picks the statistically likely word. A journalist-translator asks what the word does politically in the target language,” Ali says.
In law, concepts do not always map neatly across legal systems. A term from common law may have no exact counterpart in another legal tradition, and a small error can alter a party’s rights.
In diplomacy, the distinction between “concerned,” “deeply concerned” and “gravely concerned” can serve as a calibrated signal. Missing that distinction can turn a warning into a courtesy.
Literature presents another challenge.
A translation system may miss irony, silence or what a writer deliberately chooses not to say. Urdu provides a particularly clear example. The choice between ‘tu’, ‘tum’ and ‘aap’ is not simply a matter of grammar. It can convey intimacy, hierarchy, respect and, at times, rebuke.
“A machine often flattens these registers. A human translator hears them,” Ali says.
Hafiz Azizur Rehman, in-charge of the Department of Translation Studies at the Islamia University of Bahawalpur, agrees that AI-powered translation has improved dramatically.
Modern systems are far better than earlier machine-translation technologies at grammar, sentence structure and understanding context. But Rehman says they are not consistently equal to experienced human translators.
Recent research continues to find differences in accuracy, style and contextual appropriateness, particularly in demanding texts. A 2026 comparative study, he says, found human translators performing better overall than the AI systems tested.
Translation, Rehman argues, is about far more than replacing words in one language with words in another. It involves equivalence, semantics, culture, literature, rhetoric, terminology, press translation and religious translation.
Human advantage: judgment and responsibility
Asked what element of human translation AI cannot fully replicate, Ali identifies what he calls “judgment under responsibility.”
“If I must name one, it is judgment under responsibility. Cultural context, emotion, intention, and humour are all real, but they are aspects of one deeper thing: a human translator makes a decision and answers for it,” he says.
Translation is not simply the transfer of words, Ali explains. It is an act of interpretation.
A translator reads an author’s intention, imagines the reader’s world, weighs what may be lost and decides which loss is acceptable.
“A translator lives inside a culture and carries its memory, its grief, and its jokes in the body,” he says.
“When I translate a couplet of Ghalib, I am drawing on more than a dictionary. I bring the way that poetry was recited, quoted at funerals, and argued over.”
AI can produce a plausible translation, Ali says, but it cannot be accountable for the result.
“It does not risk anything when it gets it wrong, and translation, at its best, is an ethical act,” he says.
Rehman similarly identifies understanding the full human context behind words as the most important element AI cannot fully replicate.
A good translator asks: Why was this said? Who said it? To whom? In what situation? What does it mean culturally and emotionally?
AI can analyze language patterns with extraordinary efficiency, Rehman says. Human translators, however, bring live cultural knowledge, judgment, responsibility and an understanding of human intention.
When ‘correct’ can still be wrong
One of the more serious dangers of AI translation, Ali says, is not an obviously bad translation but one that reads fluently while quietly distorting cultural meaning.
“I would say this is the more serious danger, precisely because it is invisible. A clumsy translation announces its own failure. A fluent but misleading one persuades the reader that they have understood,” he says.
Culture, Ali argues, is not stored in individual words. It lives in historical memory, social convention and symbols.
A single term connected with Partition or religious devotion, for example, can carry centuries of associations that no literal rendering can fully convey.
An author’s voice — whether ironic, elegiac or understated — is likewise a matter of rhythm and restraint, not simply meaning.
Rehman also warns that AI can produce a translation that is grammatically correct but culturally misleading.
Expressions, humor, metaphors, historical references and symbolism often cannot be understood from the words alone. Literary translation therefore requires judgment about style, voice, imagery and emotional effect.
That does not make AI useless in literary translation. Rehman says it can generate useful drafts and alternatives, but a skilled human must determine whether the result genuinely represents the writer and the culture.
Risk for indigenous and underrepresented languages
For Ali, the challenge goes beyond accuracy.
He is particularly concerned about AI’s impact on Urdu and other Pakistani languages that remain underrepresented in the data used to train language systems.
“AI systems learn from the data they are given, and our languages are under-represented in it,” he says.
The danger, he argues, is that AI systems may quietly reflect assumptions embedded in dominant languages, causing smaller languages to be rendered through a foreign lens.
“Translation then repeats an old pattern in which the powerful language explains everyone else,” Ali says.
His response is to build indigenous language resources, including meaning corpora, terminologies and models developed by local scholars.
“Without them, we would be depending on others’ technologies to represent ourselves,” he says.
From translator to AI supervisor
Will reliance on AI weaken the translation profession or simply turn it into a more technologically advanced field?
Ali says both outcomes are possible, depending on the choices institutions and professionals make now.
The near-term reality, he says, is likely to be increased automation of routine, repetitive and high-volume work, including manuals, standard correspondence and first drafts.
Translators who use AI effectively will be able to work faster. But complex decisions involving ambiguity, tone and sensitive content will continue to require human judgment.
There are risks if institutions treat translation primarily as a cost to be minimized, Ali says. They could reduce human review, pay translators less for post-editing and accept mediocre quality.
Poorly reviewed machine translation in courts, hospitals or public policy is therefore a matter of justice, not simply style, he warns.
“So my position as a director training the next generation is that we must not teach students to compete with machines at speed. We should teach them what machines cannot do: critical judgment, cultural expertise, subject knowledge, and ethical responsibility,” Ali says.
“At CeLTS, I want our graduates to be AI-literate translators who can evaluate, correct, and guide the technology, not merely operate it.”
Rehman likewise expects some routine translation work to become faster and more automated. At the same time, translators are increasingly taking on roles as post-editors, reviewers, terminology specialists, language-technology users and quality controllers.
Rather than simply replacing translators, AI is likely to change what translators do and the skills they need.
A partnership, not a replacement
Ali does not believe AI will completely end the role of human translators, although he expects that role to change substantially.
Translation, he notes, has always adapted to new technologies, from the printing press and dictionaries to computer-assisted translation tools. Each technological shift changed the work rather than making it disappear.
What is happening now may represent a much bigger transformation, he says, but the underlying human need remains: People want to understand one another, and understanding is more than decoding.
Wherever meaning carries significant weight — whether legal, political, literary or emotional — people will want a human being to stand behind the words.
“A treaty, a verdict, a poem, or a diagnosis is not just information. It is a relationship of trust,” Ali says.
“I would put it this way: AI can translate text, but translation is about people. The future belongs not to machine or human alone, but to a partnership in which technology extends our reach and human judgment keeps it honest. Our task is to make sure that in this future, languages like Urdu are not merely translated but are heard on their own terms,” he says.
The emerging picture, Rehman says, is therefore neither “human versus AI” nor “AI replacing everyone.”
It is more likely to be humans using AI responsibly.
AI can provide speed and scale. Human translators provide judgment, cultural understanding, creativity and accountability.
We must take charge of our narrative. Research indicates that 92% of their Muslim coverage is negative. Help us change that.
Have Your Say!
Have you been affected by the issues raised in this story? Please share your experiences by emailing haveyoursay@soundvision.com.
Please include a contact number if you are willing to speak to a Muslim Network TV’s representative. You can also get in touch in the following ways:
Tweet: @MuslimNetworkTV
