Translation in the Age of AI

There is a new anxiety taking hold in the world of translation, and I know it because I feel it too. This time, the anxiety is not caused by another translator, another difficult language, or another impossible text, but by something that does not have a voice, a childhood, a culture, or a lived experience of language. It is coming from a machine that can process thousands of sentences in the time it takes a human translator to carefully consider one paragraph. We encounter these systems almost everywhere now, from our phones and search engines to email applications, classrooms, tourism platforms, government websites, publishing workflows, and everyday conversations. Type a paragraph, select a language, press a button, and within seconds the machine gives us something that looks like a translation, sometimes surprisingly good and sometimes embarrassingly wrong, but increasingly difficult to distinguish from competent human work.

Generative artificial intelligence is no longer knocking on the door of the translation profession because it is already inside, changing the way translation is produced, evaluated, purchased, and even imagined. It can translate emails, documents, subtitles, websites, public information, and literary texts at a speed no human translator can match, while producing polished sentences in seconds that might take a human hours to research, revise, and contextualize. I have watched this transformation with a mixture of curiosity and discomfort because I know that the technology can genuinely help us while also threatening the economic and professional structures that translators have depended on for years. The question is therefore no longer whether AI will change translation because it already has, and pretending otherwise only prevents translators from responding intelligently to what is happening. The harder question is what happens to the people whose work has always been to make one language speak meaningfully to another when a machine can generate plausible language almost instantly.

For translators, writers, teachers, publishers, linguists, and institutions working with Philippine languages, this is not some distant technological debate happening somewhere in Silicon Valley because the consequences are already beginning to reach our classrooms, publishing houses, cultural institutions, and communities. It is a professional problem because AI is changing how translation is produced and valued, a cultural problem because the languages being translated carry histories and identities, and a political problem because languages do not enter the digital world with equal resources or equal power. I am particularly interested in this question because working with Philippine languages means confronting the uneven visibility of our languages long before artificial intelligence became fashionable. When we discuss AI translation, therefore, we should not only ask whether translators should be afraid of machines but also ask which languages machines have been given enough opportunity to learn. The more important question for me is what kind of translation we want in a world where a machine can produce language almost instantly but may not understand what that language means to the people who speak it.

Artificial intelligence is often discussed as if it were an artificial person living inside a computer and understanding language in the same way that human beings do, but that description can be deeply misleading. AI systems are computational systems that process information, identify patterns, and generate outputs based on enormous quantities of data, allowing them to produce remarkably sophisticated language without possessing human cultural memory or lived experience. Their ability to recognize relationships among words, phrases, sentences, and larger units of discourse is impressive, but linguistic fluency should not be confused with understanding. A machine can produce a grammatically elegant sentence without knowing the historical circumstances that produced it, the social relationships surrounding it, or the emotional significance it may have for a particular community. For translators, this distinction matters because language is never merely an arrangement of words but a living system shaped by people, places, histories, memories, relationships, and experiences.

Translation itself is frequently misunderstood because it is reduced to the simple idea of transferring words from a source language into a target language, as though every word had one stable equivalent waiting somewhere in another dictionary. If translation were merely a matter of finding equivalent words, dictionaries and automated systems would have solved the problem long ago, and translators would have little reason to exist beyond checking grammatical errors. But anyone who has actually translated knows that meaning is shaped by culture, history, social relationships, tone, genre, audience, intention, and circumstance, and that the same expression can carry completely different meanings depending on where and how it is used. A joke can lose its humor, an insult can acquire a different social weight, a cultural expression can have no direct equivalent, and a literary image can depend on historical associations invisible to someone who knows only the dictionary definition. This is why I see translation not as moving words between containers but as negotiating between worlds, with every decision involving interpretation, responsibility, and sometimes invention.

This is precisely where the arrival of AI becomes both exciting and unsettling because the technology can now participate in almost every stage of the translation workflow. It can generate preliminary translations, suggest alternative expressions, identify terminology, revise grammar, adjust register, summarize source materials, compare different versions, produce subtitles, transcribe speech, and assist with post-editing, allowing translators to complete some routine tasks far more quickly. A translator can give a long document to an AI system and obtain a preliminary version in minutes, while researchers, teachers, travelers, businesses, and small organizations can use similar tools for different practical purposes. I would be intellectually dishonest if I pretended that these advantages do not exist simply because I am concerned about the profession. AI can dramatically increase speed and productivity, particularly when dealing with repetitive, predictable, and relatively low-risk texts, and translators need to acknowledge that reality rather than defend their profession through nostalgia. The real question is not whether productivity gains exist but whether we will use those gains to deepen the quality of translation or simply use them to demand that translators work faster for less.

The greatest advantage of AI in translation may therefore be not that it can replace translators but that it can change what translators spend their time doing. If a machine can produce a rough first draft in seconds, the translator can potentially devote more time to evaluating meaning, researching context, checking terminology, examining cultural references, improving style, and making difficult decisions that automated systems cannot reliably make on their own. This possibility is important because translation has always contained both mechanical and intellectual components, and technology can reduce the amount of time spent on some repetitive processes while making the interpretive responsibilities of the translator more visible. But this benefit depends entirely on how the technology is used because a machine-generated draft still requires someone who knows enough about the source and target languages to recognize what is wrong. I would rather imagine AI as a powerful assistant that makes translators more intellectually ambitious than as a replacement that encourages institutions to believe that fluent output is all that translation requires.

The most dangerous misconception about AI translation is the belief that fluent output is necessarily accurate output, because a translation can sound beautiful while quietly betraying the source. An AI system may add information, change details, omit important material, misunderstand a phrase, resolve an intentional ambiguity, or produce an interpretation that is unsupported by the original, and it may present the result with complete confidence. This becomes particularly dangerous when the person reading the translation does not understand the source language and therefore has no practical way of detecting what has disappeared or changed. The frightening thing about machine error is not always that it looks obviously wrong because obvious errors are relatively easy to identify and correct. The greater danger is that AI can be wrong in a way that looks right, giving readers a false sense of accuracy precisely because the language is polished and convincing.

This problem becomes even more serious when we discuss the languages of the Philippines because our country may be multilingual in everyday life while remaining deeply unequal in digital representation. English possesses enormous quantities of digital data, including books, websites, academic publications, subtitles, dictionaries, parallel texts, and other linguistic resources, while Filipino and several major Philippine languages have also developed increasingly visible digital environments. But many other languages remain significantly underrepresented in the datasets and technological resources from which AI systems learn, including languages such as Hiligaynon, Kinaray-a, Aklanon, Capiznon, Waray, Tausug, Maranao, Maguindanaon, and many others. This is not evidence that these languages are less sophisticated or less capable of expressing complex ideas because technological performance depends heavily on the availability, quality, and diversity of training data. A language with fewer digitized books, fewer dictionaries, fewer parallel texts, fewer annotated corpora, and fewer speech datasets simply gives an AI system fewer resources from which to learn.

This is where the politics of translation enters the conversation, because technological systems are built from data produced within societies that already contain histories of inequality and unequal relationships among languages. If a language has abundant digital resources, it becomes easier to develop tools for that language, while languages with little digital data face greater difficulty entering the technological ecosystem on equal terms. When an AI system performs poorly in a Philippine language, therefore, I do not think our first response should be to say that the language is difficult, obscure, or unsuitable for technology. We should instead ask who has invested in making that language digitally visible, who has created the dictionaries and corpora, who has digitized the literature, who has preserved oral histories, who has funded research, and who has decided which languages deserve technological investment. These questions are not merely technical because they concern cultural priorities, linguistic justice, ownership of knowledge, and the kind of multilingual future we are willing to build.

The Philippines is already beginning to confront these questions as government institutions and researchers explore how artificial intelligence and language technologies might support multilingual communication and public services. There are initiatives involving translation technologies, language datasets, and research on Filipino and other Philippine languages, and these developments deserve attention because they could potentially expand access to information and strengthen the technological presence of languages that have historically received less support. But technological development should not be celebrated simply because a machine can now produce a sentence in another Philippine language, because the more important questions concern representation, participation, ownership, and accountability. Questions such as whose language data are being collected, who has consented to their use, who benefits from the resulting systems, and whether language communities have any meaningful role in decisions about how their linguistic resources are processed and commercialized cannot be answered by technological efficiency alone. If we ignore these questions, we may end up reproducing old linguistic inequalities through systems that appear neutral simply because they are computational.

There is also a major opportunity here, particularly for Philippine languages, because AI could potentially become a tool for linguistic democratization if it is deliberately developed to support languages that have historically received less institutional and technological attention. It could assist in documenting vocabulary, organizing linguistic resources, developing educational materials, supporting multilingual public information, improving accessibility, creating preliminary translations, preserving cultural texts, and making local literature available to wider audiences. I can imagine a student accessing educational materials in a local language, a community producing public information in its own language, or a reader encountering literature from another Philippine language without requiring an enormous translation budget. These possibilities are significant, especially for communities whose languages have often been treated as secondary in formal institutions, but technology must be developed with communities rather than simply extracting language data from them. Otherwise, we risk reproducing an old colonial pattern in a new technological form in which the language of the community becomes the raw material while the power to control the technology remains somewhere else.

There is another issue that translators need to take seriously, and that is confidentiality, because the convenience of instant translation can make us forget that uploading information to a digital platform is itself an act involving questions of data governance. Legal documents, unpublished manuscripts, personal information, proprietary materials, research data, government records, and confidential institutional documents may contain information that should never be casually transferred into an AI system simply because doing so saves a few minutes of work. Professional translators already understand that confidentiality is part of their ethical responsibility, but AI makes this responsibility more complicated because translators must now understand not only linguistic ethics but also how technological systems process, retain, or otherwise handle information. The fact that a platform can translate a document quickly does not mean that the document should automatically be uploaded to it. Convenience should never become an excuse for abandoning professional judgment.

Another challenge that concerns me deeply is cultural flattening, particularly because I work with literature and languages in which the smallest linguistic choices can carry enormous cultural weight. AI systems work through patterns found in existing data, but those patterns can reproduce dominant linguistic conventions and favor expressions that are statistically familiar over those that are culturally specific. A distinctive local expression may become a familiar global phrase, a culturally specific metaphor may be replaced by something more conventional, a local form of politeness may disappear, and a word carrying historical or emotional significance may be reduced to its nearest dictionary equivalent. In literary translation, these changes can become especially damaging because literature often depends on ambiguity, rhythm, strangeness, silence, and forms of expression that cannot be reduced to straightforward equivalence. What a machine identifies as awkwardness may sometimes be precisely the quality that gives a writer a distinctive voice.

This is why I become suspicious whenever someone says that AI will simply make translation better, because the word better conceals a set of assumptions that we rarely bother to examine. Better according to whom, faster according to whom, more natural according to whose language, and more readable according to whose cultural standards are questions that should accompany every discussion about automated translation. When an AI system transforms a local expression into a globally familiar phrase, has it genuinely improved the translation, or has it simply made the local culture more comfortable for a dominant readership? When a regional literary voice is made to sound like standardized English, Filipino, or another dominant language, have we made the text clearer, or have we erased part of what made the text culturally distinctive? These questions are especially important in the Philippines because our linguistic history already contains unequal relationships among English, Filipino, and the many regional languages that continue to carry community identities and literary traditions.

So should translators be afraid? My answer is yes, but perhaps not of the machine itself because machines are only as powerful as the uses we permit institutions and individuals to make of them. Translators should be afraid of becoming professionals who refuse to understand the technology transforming their own field, and we should be afraid of believing that the old workflow will remain unchanged simply because it has worked for decades. We should also be afraid of defining translation so narrowly that the easiest part of our work becomes the only part we know how to perform, because that is precisely the part machines are becoming increasingly capable of doing. If AI can produce a basic first draft, then a translator who spends an entire career doing only basic first drafts is understandably vulnerable, while a translator who can evaluate, revise, research, contextualize, localize, fact-check, and defend a translation possesses a different kind of professional value.

The arrival of AI may therefore force translators to become more intellectually ambitious rather than simply more technologically efficient. The translator of the future may need to be part linguist, part editor, part researcher, part cultural consultant, part technology user, and part critic, with the ability to recognize when a machine translation is wrong becoming as important as the ability to produce a translation from scratch. This is a profound shift because translators have traditionally been trained primarily to create translations, while the age of AI requires them also to interrogate translations and examine the decisions hidden inside machine-generated language. We will need to ask why a particular word was chosen, what information was lost, what cultural assumptions entered the text, what ambiguity disappeared, what terminology changed, and whether the resulting text still belongs to the intellectual and cultural world of the source. The translator’s value may increasingly reside not in being the fastest person who can produce a sentence but in being the person who understands why that sentence should or should not exist.

This is also why translation education needs to change, because teaching future translators to use AI should never mean teaching them how to press a button and copy whatever appears on the screen. Students should learn how machine translation works at a basic level, how prompts can influence outputs, why hallucinations occur, why low-resource languages present particular difficulties, how to evaluate machine-generated translations, how to protect confidential materials, how to identify cultural distortion, and how to conduct systematic post-editing. More importantly, they should learn that technological competence does not replace linguistic competence because a person who does not know a language cannot suddenly become a reliable translator simply because an AI system produces fluent sentences in that language. If anything, the emergence of AI should make rigorous language education more important because someone must possess enough linguistic and cultural knowledge to challenge the machine when the machine is wrong.

Perhaps the most provocative question we should ask in translation education today is whether a person who cannot evaluate a translation should be allowed to call himself or herself a translator simply because an AI system produced the text. I would argue that generating a translation and taking responsibility for a translation are fundamentally different acts, because the first can be automated while the second requires knowledge, judgment, accountability, and the willingness to defend a decision. AI can generate, but the translator must evaluate, while AI can suggest alternatives, but the translator must understand their implications and decide which one belongs in the particular context. A machine can process millions of sentences, but it cannot take responsibility for the consequences of choosing one sentence over another in a specific human situation. The question is therefore not whether a machine can produce language but whether the person presenting that language as a translation actually understands and can defend what has been produced.

There is also a danger in becoming too romantic about human translators because we have to acknowledge our own failures if we want the argument for human responsibility to be credible. Human translators make mistakes, misunderstand context, introduce bias, produce awkward sentences, work carelessly under pressure, and sometimes allow ideology or insufficient training to influence their decisions, so there is nothing inherently virtuous about a translation simply because a human produced it. The stronger argument is not that humans are always better than machines but that responsible translation requires capacities such as contextual judgment, cultural knowledge, ethical accountability, intentionality, and the ability to negotiate meanings that cannot be reduced to statistical similarity. We should not defend human translation by pretending that machines are incapable of useful work, because that argument is already outdated, but by insisting that translation remains a field in which someone must be accountable for what the words ultimately mean.

The future of translation in the Philippines should therefore not be framed as a war between translators and machines because that framing distracts us from the larger struggle over language, technology, labor, and cultural power. We should instead ask whether AI will strengthen multilingualism or accelerate linguistic homogenization, whether it will help preserve Philippine languages or make them increasingly dependent on systems trained primarily in dominant languages, and whether communities will have meaningful participation in the development of language datasets. We should ask whether translators will be treated as obsolete labor or as essential evaluators of AI-generated language, and whether universities will teach future translators how to use these technologies responsibly or continue preparing students for a translation industry that no longer exists in the form we once knew. These questions are not simply about employment because they concern who will have the authority to shape the languages through which Filipinos understand themselves and one another.

These questions matter because language is not merely a communication tool, and I think anyone who has spent enough time listening to a community speak its own language understands this intuitively. Language contains memory, history, relationships, humor, identity, grief, affection, and ways of seeing the world that cannot always be separated from the words themselves. When we translate a language, we are not simply transferring information from one linguistic system to another but moving something from one cultural space into another, knowing that something may inevitably change along the way. Sometimes what is lost is obvious, while at other times the loss is so subtle that only a speaker of the original language will notice it. A machine may produce a grammatically flawless sentence while quietly removing the cultural texture that made the original worth translating in the first place.

This is why I do not believe translators should spend their energy trying to defeat AI because we will not win that battle and, more importantly, because it is the wrong battle to fight. The more urgent task is to make sure that AI becomes accountable to language rather than language becoming subordinate to AI, and that requires translators to learn the technology, linguists to participate in its development, writers to understand how their works may enter training datasets, universities to build stronger language resources, and government institutions to establish clear standards for AI-assisted translation. Communities must also participate in decisions concerning their languages, particularly when linguistic data are collected, digitized, transformed into technological resources, or potentially commercialized by institutions and private companies. And the public needs to understand one basic principle that should never become negotiable, which is that an AI-generated translation should never be treated as a verified translation simply because it sounds good.

The real danger is not that AI will make translators unnecessary because the profession has always changed whenever new technologies have altered how language is produced, reproduced, and distributed. The greater danger is that society will become satisfied with translation that is merely fast, cheap, fluent, and superficially convincing, gradually lowering its standards because convenience makes critical evaluation feel unnecessary. We may begin accepting translations that sound smooth but erase cultural differences, translations that are readable but conceptually wrong, and translations that make local expressions disappear because the machine cannot recognize their cultural value. We may eventually inhabit a linguistic environment in which every language begins to sound increasingly like every other language because technological systems have learned to prioritize what is common over what is particular. If that happens, it will not simply be a technological failure because it will represent a cultural failure that we allowed to happen in the name of efficiency.

So, will AI replace translators? I think we should answer that question without comforting ourselves with slogans about the timeless superiority of human beings. AI will certainly replace some translation tasks, automate repetitive work, reduce the time required for certain forms of translation, and change pricing models, workflows, training, and expectations within the profession, while some translators may lose particular kinds of work when clients prioritize speed and cost over quality. We should not hide this reality because technological disruption is real, and professionals who refuse to understand it may find themselves increasingly vulnerable. But acknowledging that AI will replace some tasks does not require us to conclude that translation itself has become unnecessary, because the deepest work of translation has never been reducible to producing equivalent strings of words. The question is not whether the machine will do some of our work because it already does, but whether we will allow the easiest parts of translation to define the entire profession.

AI will not replace the human translator in the deepest sense of translation because translation is ultimately about making judgments concerning meaning, context, voice, culture, intention, and consequence. It is about deciding what should remain, what can change, what requires explanation, what must be preserved, and what cannot be translated without losing something important, while also recognizing the historical weight of a concept and the emotional force of a phrase. It is about understanding the voice of a writer and the worldview of a community, but it is also about accepting responsibility when a particular linguistic decision has consequences for the people who will read or hear it. When a translation matters, someone must be willing to stand behind it and say that this was the decision I made and that I can explain why I made it. That responsibility may be assisted by machines, but it cannot simply be transferred to them.

This may be where the future of translation begins, not with the question of whether AI can translate because we already know that it can, but with the harder question of whether we understand what must not be lost. I do not think the future belongs to translators who simply reject AI, just as I do not think it belongs to those who blindly surrender translation to machines, because both positions avoid the difficult intellectual work that the technology now demands from us. The future belongs to a more demanding understanding of translation in which technology can assist production while human beings remain responsible for judgment, context, culture, and meaning. The machine may become the fastest assistant a translator has ever had and perhaps the most disruptive technology the profession has ever encountered, but the sentence is never the whole translation. The machine can generate the sentence, but the human must still decide what the sentence means.

***

Noel Galon de Leon is a writer and educator at the University of the Philippines Visayas. He teaches in the Division of Professional Education and at UP High School in Iloilo. He currently serves as faculty-in-charge of the UP Visayas Language Program Office and as secretary of the National Committee on Literary Arts of the National Commission for Culture and the Arts (NCCA). His poetry has been recognized by the Don Carlos Palanca Memorial Awards for Literature.