In recent years, discussions about artificial intelligence and translation have grown louder – especially when it comes to languages with complex morphology and deep cultural layers like Kazakh. But what does AI really get right? Where does it fail? And how are translators and editors adapting to a future where machines can draft text in seconds?
To answer these questions, we turned to four people who work with language every day and in very different ways: linguist and researcher Bibarys Seitak, editor and translator Gulnur Kuanishbekova, editor Bayan Khasanova, and literary translator Dinara Mazen. Their backgrounds range from academic linguistics to technical translation, from editorial work to the delicate craft of rendering literature from one language into another.
What unites them is a shared understanding of AI’s limits and possibilities. All four see it as a helpful assistant that can clarify meanings or speed up technical work — but never as a true translator. They point to familiar errors in Kazakh: mismatched suffixes, English-like word order, flat tone, and even invented details. And while AI can ease routine tasks, it still cannot capture subtext, intention, rhythm, or artistic voice — the core of literary translation.
They also note that AI is reshaping both workflow and the economics of the field. Post-editing is becoming standard, rates are shifting, and the boundary between professional and amateur translators is blurring. But their message is clear: skilled human judgment — especially editorial judgment — remains essential, and may become even more valuable.
Their insights reveal not only how AI assists translators, but also where its limits become impossible to ignore.
Editor Bayan Khasanova: AI text still sounds like a robot
In my editorial practice, I use AI only when I need to understand the meaning of a complicated scientific term or concept that does not yet exist in Kazakh. AI helps by offering explanations and examples, but I do not rely on it to translate or edit text. The essence of editorial work — shaping voice, maintaining clarity, and preserving stylistic consistency — must remain human.
One of the first things I notice in AI-generated text is the absence of emotion. Human writing carries rhythm, tone, and subtle variations, while AI tends to rely on repetitive structures and familiar phrasings. The result sounds mechanical. When I edit such text, my role is to restore a human tone, introduce variation, and ensure the language flows naturally. This requires experience and analytical sensitivity.
Compared to ten years ago, AI has made certain aspects of editing easier, particularly when dealing with unclear translations or concepts from source languages unfamiliar to some editors. Tools now provide quick explanations of terms that once required extended research. However, this convenience comes with a risk: people may become overly dependent on AI and lose the habit of deep textual analysis.
Despite the rise of AI, I don’t believe the essence of the editor’s role has changed. Different genres — scientific, literary, publicistic — demand different registers and stylistic choices, and maintaining these distinctions requires human judgment. AI can support, but not replace, the nuanced decisions that shape a readable and meaningful text.
Linguist and researcher Bibarys Seitak: AI mirrors language – it doesn’t shape it
I use ChatGPT quite regularly for translation: both from English to Kazakh and vice versa. It has simply become part of my workflow. Before that, I relied on Google Translate, but ChatGPT turned out to be much more accurate and flexible6 especially when I feed it shorter segments rather than long passages.
When it comes to AI’s impact on language, I don’t think AI will fundamentally alter the natural evolution of any language. At the end of the day, AI isn’t inherently creative, it simply mirrors the way people already speak or write and adapts to user input. It relies on patterns found in its training data rather than inventing something truly new. Because of that, I don’t expect it to reshape Kazakh or any other language in a radical way. Its role is more reflective than directive.
In translation, there are many types of errors AI tends to make. Morphological errors are common. For example, attaching suffixes or case endings in the wrong way. There are also phonological issues… Or perhaps system errors in text-to-speech synthesis. One of the most frequent issues I used to notice was the misreading of numbers: it would pronounce numbers in Turkish when generating a response in Kazakh or likewise randomly switch to English when generating Russian responses. This has improved a lot in recent models though. Syntactic errors also happen often: the word order can be unnatural because Kazakh and English have almost opposite syntactic structures. A sentence that is 1-2-3-4-5 in Kazakh can easily become 5-4-3-2-1 in English but the model sometimes transfers that pattern back into Kazakh. So far, ChatGPT doesn’t consistently avoid these mismatches.
Even so, the model feels much closer to “understanding” in English, simply because the training data is richer and it covers both everyday speech and high-register texts. When asked about the meaning of a word in an English sentence, it usually provides accurate interpretation. In Kazakh, however, it may give incorrect descriptions or even fabricate details. Still, even without genuine semantic understanding, it can help reveal nuance by paraphrasing an abstract idea in several different ways.
This moment in history also marks another major leap for the English language. In the AI domain, English is overtaking not only other global languages but even the major European ones. Its influence is becoming overwhelmingly strong. This wasn’t always the case, but the rise of AI has amplified its dominance. The stronger English becomes in technological ecosystems, the more its structures and conventions shape and influence literally thousands of languages across all continents.
Literary translator Dinara Mazen: AI cannot reproduce artistic style
I’ve been working as a translator for about 15 years, mostly with literary works and academic texts. I didn’t use AI tools before, but now AI and machine translation have become helpful assistants that make a translator’s work easier. I use AI tools like ChatGPT mainly for technical translations.
AI has made technical translation easier by providing consistent terminology and accelerating routine tasks. AI is an excellent assistant. Its output still needs to be edited, but when an AI model is trained to adapt to a translator’s style, it can significantly streamline the workflow. In the AI era, translators may rely less on memorizing terminology and more on refining and correcting AI-generated translations.
However, for literary translation – where creativity, nuance, and the translator’s own interpretive skill are essential, AI cannot be trusted. I never use AI for literary projects, whereas for technical content I train it on specific terminology and use it only as a supporting tool. Researchers note that AI can sometimes produce inaccurate information or add details that were not in the original text, and even AI developers warn about this. That’s why we shouldn’t depend on its output entirely. It is a tool that supports the translator, not something that can replace human judgment.
I work with both local and international companies, and many foreign clients began using AI and machine translation as support tools years ago. For example, Machine Translation Post-Editing has become a standard service – it means editing machine-translated output from engines like Google or Yandex. Now there is also a growing demand for editing AI-generated translations. These services are usually cheaper than human translation. Some publishers are experimenting with AI-generated drafts for motivational or non-fiction books, but not for true literary works, since AI cannot reproduce an author’s unique artistic style.
My prediction is that human translation will remain essential in literature and complex academic fields, and the value of such translations will likely increase. Many translators may shift toward roles focused on editing AI-generated content. One concerning issue is that people who do not actually know English now use AI and call themselves “translators,” which creates opportunities for fraud. This means editors will need to be even more skilled and deeply knowledgeable in translation to ensure quality.
Editor and translator Gulnur Kuanishbekova: AI changes the market for translators
In my work, I use AI mainly as a tool for clarifying meaning. When a sentence contains a word with several possible interpretations, or when a phrase carries a nuance that is difficult to unpack, I ask AI to explain it. It offers examples, paraphrases, and semantic shades, which helps me choose the most natural Kazakh equivalent. I do not translate entire sentences or paragraphs through AI, nor do I rely on it for editing; the responsibility for the final text remains mine. Because I primarily translate from English to Kazakh, I ask my clarifying questions in English. AI often helps illuminate concepts, especially in technical or scientific texts where a term has no direct Kazakh analogue. These explanations allow me to construct Kazakh phrasing that reflects the original meaning without distorting it.
I believe that as AI becomes more integrated into translation workflows, the translator’s role will increasingly resemble that of a strong editor. Machines can draft or suggest versions of terminology-heavy passages, but human expertise is needed to refine them, restore naturalness, and ensure that the cultural and stylistic context remains intact. The future translator must therefore possess excellent editorial skills.
At the same time, AI is changing the economics of the profession. Rates for translators have not increased in recent years; if anything, they are declining. Clients sometimes openly suggest that I can “use AI and just fix the Kazakh.” In professional chat groups, translators discuss how a page once paid at three thousand tenge is still paid at the same rate, even though the cognitive demands of the work remain the same. I have personally had clients send me machine-translated documents asking only for stylistic correction. These are signs that the market is shifting. Translators will not disappear, but their labor risks being undervalued, and their work may increasingly be framed as editing rather than translating. Translation is demanding, complex mental work, and yet the financial recognition does not reflect that. This imbalance is becoming more visible as AI tools spread more widely.
