Mastodawn

Michel Bozgounov 🇺🇦 🇪🇺Mar 11

洪民憙 (Hong Minhee)

I'm writing this in English.

Not because English is my first language—it isn't. I'm writing this in English because if I wrote it in Korean, the people I'm addressing would run it through an outdated translator, misread it, and respond to something I never said. The responsibility for that mistranslation would fall on me. It always does.

This is the thing Eugen Rochko's post misses, despite its good intentions.

@Gargron argues that LLMs are no substitute for human translators, and that people who think otherwise don't actually rely on translation. He's right about some of this. A machine-translated novel is not the same as one rendered by a skilled human translator. But the argument rests on a premise that only makes sense from a certain position: that translation is primarily about quality, about the aesthetic experience of reading literature in another language.

For many of us, translation is first about access.

The professional translation market doesn't scale to cover everything. It never has. What gets translated—and into which languages—follows the logic of cultural hegemony. Works from dominant Western languages flow outward, translated into everything. Works from East Asian languages trickle in, selectively, slowly, on someone else's schedule. The asymmetry isn't incidental; it's structural.

@Gargron notes, fairly, that machine translation existed decades before LLMs. But this is only half the story, and which half matters depends entirely on which languages you're talking about. European language pairs were reasonably serviceable with older tools. Korean–English, Japanese–English, Chinese–English? Genuinely usable translation for these pairs arrived with the LLM era. Treating “machine translation” as a monolithic technology with a uniform history erases the experience of everyone whose language sits far from the Indo-European center.

There's also something uncomfortable in the framing of the button-press thought experiment: “I would erase LLMs even if it took machine translation with it.” For someone whose language has always been peripheral, that button looks very different. It's not an abstract philosophical position; it's a statement about whose access to information is expendable.

I want to be clear: none of this is an argument that LLMs are good, or that the harms @Gargron describes aren't real. They are. But a critique of AI doesn't become more universal by ignoring whose languages have always been on the margins. If anything, a serious critique of AI's political economy should be more attentive to those asymmetries, not less.

The fact that I'm writing this in English, carefully, so it won't be misread—that's not incidental to my argument. That is my argument.

洪民憙 (Hong Minhee)

From back when I used to flip through a thick dictionary, taking a minute for each sentence, to now, when I use an LLM to write in English, people don't realize that I'm personally speaking to them in English for their sake. Honestly, I just want to speak in Korean. I hate English.

洪民憙 (Hong Minhee)

To be precise, it's not English as a language that I hate, but English as a form of power.

Hi! It happens the same to me.

I am quite proficient with English myself and I use it to drive my points across so people can understand me.

I do hate the implications of needing to use English as a vehicular language to get understood myself.

Ian Wagner Mar 11

@hongminhee 인정

cuan_knaggs Mar 11

@hongminhee as a first language english speaker: feel free to hate english as a language if you want. it's a very silly language that's taken far too seriously

🇨🇦🇩🇪🇨🇳张殿李🇨🇳🇩🇪🇨🇦Mar 11

@hongminhee I hate both aspects of it. A ludicrous grammar, an incoherent orthography, and a vocabulary that is unnecessarily bloated with gratuitous synonyms, and a penchant for stealing random pieces of grammar and vocabulary from all over just to make the language even bigger with every passing year.

And the fact this language is the "international language" just ticks me off.

brendan (ジャンク品)Mar 11

@hongminhee i think you make some great points about how power structures impact all of this (and online i recognize i am privileged to be an english speaker, with all the access that entails).

access is important but not at the expense of clarity. it's not about grammar—the gulf between cultures can be so hard to bridge. i don't trust an (implicitly biased, probably american) llm to foster understanding. isn't that reinforcing cultural hegemony and upholding the same power structures?

Evan Prodromou Mar 11

@hongminhee 저는 당신이 영어를 싫어하는지 몰랐습니다. 영어가 모국어라서 사람들이 제 필요에 맞춰주는 것에 익숙해져 있었습니다. 앞으로는 당신의 글을 한국어(아니면 일본어?)로 읽고, 저도 한국어로 답해 보겠습니다.

Oliver D. Reithmaier Mar 11

@hongminhee this is the thing many people don't understand about far-apart languages. It's infinitesimally harder to use and speak them than, say, French would be for an English speaking person. It's like changing your way of thinking entirely. That's hard for anybody. It's the same with Japanese for me. Love the language, but its hard af to get right, and I would rather speak English or German.

Peter Brett Mar 11

@hongminhee Criticisms of Anglophonic hegemony are similar to (and inseparable from) criticisms of capitalism.

In many cases, systems people put in place to mitigate capitalism's harms inadvertently strengthen capitalism's grip (e.g. tax credits for low paid workers allow them to be paid even less).

Similarly, lowering the bar for communication in English, as you describe, makes it ever less likely that we'll ever start treating non English speakers as first class citizens.

洪民憙 (Hong Minhee)

@krans The analogy is structurally interesting, but I think it breaks down at a crucial point.

With tax credits, the argument is that the subsidy lets employers off the hook—pressure that would otherwise force wages up gets absorbed by the state instead. The discomfort falls on capital, or at least that's the intent. But when you apply the same logic to language access, the discomfort doesn't fall on the Anglophone center. It falls on the people who were already excluded. The implicit suggestion becomes: non-English speakers should communicate less fluently, so that English speakers are eventually pressured into… what, exactly? Learning Korean? There's no mechanism there.

The deeper problem is that “lowering the bar for communication in English” is not the same thing as accepting English hegemony as permanent. I use these tools to participate in a conversation that would otherwise exclude me. That's not capitulation—it's the same logic as using a wheelchair ramp. You don't refuse the ramp because its existence lets architects keep building stairs.

The structural critique of hegemony is real and I share it. But it shouldn't cash out as advice to the marginalized to make themselves less legible. That's a cost I'm not willing to ask people to pay on behalf of a structural shift that may never come.

Peter Brett Mar 11

@hongminhee My employer is perfectly capable of affording to translate our product manuals into Korean professionally — I'm here in Korea right now supporting a *massive* customer.

Instead our official policy is that Korean users will be fed LLM slop that our tech writers won't even attempt to read and validate.

That's exclusion and condescension packaged as providing access.

洪民憙 (Hong Minhee)

@krans That's a real and legitimate grievance, but it's a different argument from the one we were having.

Your employer using LLM translation to cut costs on documentation for a massive Korean customer (while having the resources to do it properly) is a decision made by someone with power, to save money, at the expense of Korean users. That's worth being angry about.

But I'm an individual trying to participate in a public conversation. I can't hire a personal interpreter every time I want to respond to a post. The choice I actually face is: use available tools, or stay silent. Those aren't the same situation, and the same tool can mean very different things depending on who's holding it and why.

If anything, your example reinforces the point. The problem isn't the tool, but it's who gets to decide when it's “good enough.”

Piranhallama Mar 11

@hongminhee @krans "good enough" is one of those phrases that always hits me as problematic.

Sometimes you can get away with cable ties and tape. Other times you'll need a weld. "Good enough" is the former to me.

Mordoukna Mar 12

@securedllama Good enough is usable, but not perfect aka Pareto principle.
Sometimes you can get away with a weld, Other times you need to exchange the part..

@hongminhee @krans

dr2chase Mar 11

@hongminhee @krans would the personal tool exist without the cost-cutting market? To make a (very stereotyped) analogy w/ transportation, if someone says "people should ride bikes" a common response is "what about the disabled?". Standard reply is "well of course they can drive, everyone else should bike/transit". But if there is no whole-population market for cars, will any get built for the disabled? What will they cost?

(auto-translation follows, and I translated back to check)

dr2chase Mar 11

@hongminhee @krans 비용 절감 시장이 없다면 개인용 도구가 존재할 수 있을까요? 교통수단을 예로 들자면, 누군가 "사람들은 자전거를 타야 한다"라고 말하면 흔히 "장애인은 어떻게 하죠?"라는 반응이 나옵니다. 이에 대한 일반적인 대답은 "물론 그들은 운전하면 되고, 다른 사람들은 자전거를 타거나 대중교통을 이용해야 한다"입니다. 하지만 자동차에 대한 전체 인구 시장이 없다면 장애인을 위한 자동차가 만들어질까요? 가격은 얼마나 될까요?

Ariadne Conill 🐰

@hongminhee @krans @dr2chase I would argue that personal-scale SLMs would already be dominating if capital did not hoard all the hardware.

I recently trained what I call a small language model (which does not try to mimic intelligence but rather convert structured data into language and back) at home and it only took a week on an RTX 6000 Blackwell.

When I bought the Blackwell GPU (~$3500) it was still expensive (as any workstation class hardware would be) but now it is the price of a recent vintage used car to buy one (~$14500).

If those compute resources had not been hoarded we would probably see more ethical hobbyist-driven models by now rather than models that require large amounts of capital to train due to the hoarding of resources.

So in other words I would say it is complicated: capital made people aware of the technology, but the same players have also restricted access to the means to make competitive implementations at the hobbyist level.

There is nothing technically blocking the creation of community-based models, the problem is the resource hoarding enabled by capital.

And I can prove that: a hobbyist released the world's first publicly accessible image generation model with an embedded LLM, Craiyon. And that model was and remains libre. We all remember Craiyon right?

But that was in 2022, before the AI frenzy drove up the price of professional GPUs by 500+%.

The problem with that argument is it assumes markets will determine what we do. If there is no whole-population market for cars, we will build motorized vehicles for the disabled. The market doesn't matter.

CC: @hongminhee @[email protected]

daughter of lilith

@dr2chase as someone who lives in the Netherlands this transportation analogy is such a painfully american perspective, non-car infrastructure is better for disabled people too, better than car infra. Personal mobility devices (outdoor electric wheelchairs et al) are very common here and use bike lanes, and if someone needs a car it's still perfectly viable on the service/access roads that are still here and can be used as both a path for people and bikes as well as a road for the rare car

The Crafty Miss 6d ago

@marta I hear the same arguement outside of the USA, both from people with disabilities that mean they need to use a car (a good faith argument) and from people who aren't currently disabled re: cycling but just don't want to drive less and use accessibility as a sort of gotcha (not a good faith argument)

500 Internal Server Error Mar 12

@hongminhee @krans in addition to this, an interpreter who is a real human could judge their employer and be a safety threat if their employer asks them to translate eg. lgbtq-related contents, especially in a socially conservative country like China

Jeff C. 🇺🇦🇬🇱Mar 11

@hongminhee @Gargron Coincidentally, I've been enjoying a YouTube series by a couple of women who escaped North Korea and settled in South Korea in their late teens.

Their English is pretty rudimentary, so they use AI translation (presumably LLM-based, written and vocal) to communicate with their (mostly American) audience. The degree of cultural interchange has been wonderful and illuminating and my world as a monolingual person would be smaller and less-informed without it.

Keira (She/Her)Mar 11

@hongminhee I had engaged with the thread you link to on a very superficial level, only regarding the poetry and cultural nuance that is missed (and I think I must have logged off before it was finished).

I appreciate the time and care you've taken to make the case for the equity and access problems that machine translation might ammend. You're right, of course. If LLMs stick around (and don't bring about a world-ending situation), I hope their use is focused on extending access and increasing cross-cultural and cross-language information exchange. Whether they can reverse the flow of information so that it's not just English-->others, relies on us Anglophones doing a better job than we're doing right now at making space for the rest of the world.

It is absolutely a bias and equity issue that I "got to" think about it so superficially in the first place.

⏚ Antoine Chambert-Loir Mar 11

@hongminhee @Gargron
I would like to add one point that I rarely see addressed: a translation is also a responsability, that the text provided is faithful to the original, possibly imperfect, but faithful.

If it is provided to you, at your request, them it's a contract between you and your provider. You might say that automatic translation is better than nothing.

But if it is imposed on you, as in subtitles, then you have no other choice but them and the responsability is on the provider. In this case, automatic translation will essentially not be admissible.

洪民憙 (Hong Minhee)

@antoinechambertloir The distinction you're drawing is a useful one, and I largely agree with it. When translation is imposed on you with no alternative, the responsibility for its faithfulness lies with whoever imposed it.

But it makes me ask: which category does my situation fall into?

I'm writing this reply in English. Not because I chose English as my preferred language of expression, but because if I wrote it in Korean, it would either be ignored or filtered through whatever tool my interlocutor happens to have on hand—and any resulting misreading would be treated as my problem. So I write in English, carefully, to preempt that. Is that a voluntary choice? Formally, yes. Practically, it sits much closer to the “imposed” end of your spectrum than it might appear.

The contract model works cleanly when both parties have genuinely equivalent alternatives. When one party's only real options are “use this tool” or “don’t participate,” the contract framing starts to obscure more than it reveals.

⏚ Antoine Chambert-Loir Mar 11

@hongminhee @Gargron on this network, I would find rude that you answer me in Korean as you would find rude that I answer you in French. But the software is connected with automatic translation system that may allow you to start some interaction, in a language or another, possibly falling out onto English. We both have this possibility, which allows us to share the responsability.
(And if you didn't speak a single word out of Korean, and I didn't speak a word out of French, we wouldn't have much choice then, but trust that system!)

⏚ Antoine Chambert-Loir Mar 11

@hongminhee @Gargron Imagine we start following each other. Then you will see the posts I send in French, and I will see yours in Korean, as I see many on German or Spanish (although I speak German but not Spanish). In any case, some freedom emerges.

洪民憙 (Hong Minhee)

@antoinechambertloir I appreciate the spirit of what you're describing, and the fediverse's approach to translation is genuinely one of its nicer features.

I'd just gently push back on one assumption: that the automatic translation available to both of us is roughly equivalent. French–English and Korean–English are not the same problem for these systems. The linguistic distance is much greater, the training data has historically been thinner, and the results have reflected that gap for a long time. It's only fairly recently that Korean–English machine translation has become reliable enough to carry a real conversation without significant loss.

So when you write in French and I write in Korean, we're not quite sharing the responsibility equally—at least not yet. Though I do think we're getting closer, which is perhaps the most honest thing I can say about where I stand on all of this.

⏚ Antoine Chambert-Loir Mar 11

@hongminhee thanks for this remark!

Maria Langer | 📝💎🌵🛥️Mar 11

@hongminhee @grumpy4n6 @Gargron Attn @jfk . This might interest you.

silverpill Mar 11

I'm writing this in English.

Do you, though? Your writing style was different in the past, so I am pretty sure that you now machine-translate, or perhaps use an LLM writing assistant.

To be honest, the non-slop version of you was much better.

洪民憙 (Hong Minhee)

@silverpill Yes, I used an LLM to help write it. I wrote my thoughts in Korean first, then had it translated. That's kind of the whole point I was making.

I'm not a native English speaker. When I write long-form English on my own, it's slow and the result is often not what I actually meant. Using a tool to bridge that gap doesn't make the thoughts less mine. It makes them more accurately mine, not less. A non-native speaker hiring a copy editor wouldn't get this reaction.

I'll grant you that “the non-slop version of you” stings a little. But I'd rather be legible and called slop than be authentic and misread.

silverpill Mar 11

@hongminhee Authenticity matters. When I see slop I usually just ignore it, because reading it is like watching paint dry, and I think I am not alone in that.
You're basically the only person with whom I continue to communicate despite all of this.

Damien de Lemeny Mar 11

@hongminhee @silverpill Hi. I'm curious (as a non-native english speaker on the other side of the argument), what gives you the confidence that machine translation won't be misread ?
I'd be way less secure about my criticism of MT if the tools were able to probe the author for meaning but we're not quite there, and I think that MT in the hands of a polyglot-ish author has better chances of being somewhat useful (at least it's a huge difference from unedited/unverified client-side translations).

Damien de Lemeny Mar 11

@hongminhee @silverpill I really think there could be a lot to do in terms of bridging the fluency gap in terms of UX. You refer to your experience flipping pages of dictionaries, and I relate to that quite hard : that's where I'd like to see effort and change in software.
However, I feel comfortable bearing the responsibility of making my speech accessible to an English or Spanish speaker that doesn't speak French, and any failure would be mine.

洪民憙 (Hong Minhee)

@ddelemeny @silverpill The confidence comes from an asymmetry I suspect many non-native speakers will recognize: I can read English much better than I can write it.

When I write in English on my own, I often know, as I'm writing, that something is off—that the sentence doesn't carry the weight I intended, or that the nuance I wanted is somewhere between the words I've chosen. I just don't always know how to fix it. When I write in Korean first and then work with an LLM, I can read the result and check it against what I meant. Sometimes I'll see a phrase and think: yes, exactly that, I didn't know how to get there myself. That moment of recognition is the verification step.

So I'm not trusting the machine blindly. I'm using my reading ability—which is reasonably good—to audit an output that my writing ability couldn't have produced alone. It's an imperfect process, but it's not as unmoored as handing a text to a system and walking away.

Your point about polyglot authors is well taken. The tool works better when the person using it can actually evaluate what it produces. I'd agree that's a meaningful distinction.

Damien de Lemeny Mar 11

@hongminhee @silverpill I see. One personal reason I don't want to rely on translators and prefer the "hard" way, is that I believe my reading and understanding is sharpened by my attempts at writing. That's the essence of the "immersion in a language" argument for me, and I have experienced it several times (positively by being immersed in English and Spanish speaking cultures, and negatively by lack of it in German and Korean). Do you relate to that ?

Damien de Lemeny Mar 11

@hongminhee @silverpill How do you think translators shape or maintain your abilities in a foreign language, as opposed to research and experimentation ?

洪民憙 (Hong Minhee)

@ddelemeny @silverpill I relate to the immersion argument, and I think it's part of why I avoided machine translation for so long—not out of principle, but because the output wasn't worth learning from. Older MT between Korean and English produced something closer to a word-by-word skeleton than actual language. You couldn't look at it and think: oh, that's how a native speaker would put it. It was more like a scaffold you had to tear down before building anything.

LLMs are different enough that I've had to revise that instinct. The output is often genuinely idiomatic, and when I read a phrase that lands exactly right, there's a recognition that functions a lot like learning—the same feeling as encountering a sentence in a book and thinking: I'll remember that. I do find myself absorbing expressions that way, probably more than I would have expected.

That said, I think your point holds at the edges. For shorter writing I still work without assistance, partly for practical reasons and partly because I notice the difference when I don't. So I suspect I'm arriving at something similar to what you're describing, just from the other direction—using the tool for longer texts while trying to keep the muscle from atrophying entirely on shorter ones.

The dynamic you mention with German and Korean is interesting too. Korean was my concern about English; I imagine the lack of immersion shapes the experience in ways that are hard to compensate for with tools alone.

Damien de Lemeny Mar 11

@hongminhee @silverpill Thank you for replying with care, your POV is really interesting.
Have you read the Reg's article about semantic ablation that was shared around some time ago ?

https://www.theregister.com/2026/02/16/semantic_ablation_ai_writing/

This may be part of why other commenters allegedly found a negative difference in your writing using MT, and a concern I have about building fluent writing skills. MT (and autocomplete) collapse possibilities into an average, that's probably correct enough but also probably low entropy.

Why AI writing is so generic, boring, and dangerous: Semantic ablation

opinion: The subtractive bias we're ignoring

The Register

Damien de Lemeny Mar 11

@hongminhee @silverpill your writing short form on your own gives you a regular workout, that's something I think is very necessary when machine assistance is involved.
On semantic collapse, I am not sure there's an equivalent practice that preserves an author from clearing a lower bar than intended and settling for a "good enough". "Good enough" is a complex object here, and while I don't challenge your ability to keep a higher standard, I fear the subjective self-indulgent bias in general.

洪民憙 (Hong Minhee)

@ddelemeny That's a useful framing, and the article is worth reading. The concern about entropy collapse is real—I've seen it happen when native speakers run their own writing through a model and get something smoother but somehow emptier back.

My situation is a bit different, though. The high-entropy original is in Korean. The LLM's job is to carry that across, not to sand it down. Whether it succeeds is a fair question, but the direction of the process matters. I'm not polishing a draft into blandness; I'm trying to get something that exists in one language to exist in another without losing its shape.

Anyway, this has been a genuinely interesting exchange. Thank you for the link.

Damien de Lemeny Mar 11

@hongminhee oh I'm glad I didn't waste your time! Thanks for engaging in the conversation!

@[email protected] do you think your writing skills will improve with continued reading of the LLM-reflected translations, to the point where you may no longer need it?

洪民憙 (Hong Minhee)

@julian I actually just addressed something close to this in a reply up the thread—might be worth a read!

@[email protected] thanks, good answer

I would wonder then that maybe you might end up sounding like an LLM, then. Best interject some of your own style later on

洪民憙 (Hong Minhee)

@julian Yeah, that's why I'm still writing short words myself. My accent won't go anywhere!

David Mankins Mar 11

@silverpill @hongminhee

I don’t interject this as an attack, but please realize that when you say “AI slop” you say “sloppy person who uses AI”.

@hongminhee very clearly is not such a person, so please don’t imply they are, even if they chose an assistant you disapprove of to help them communicate.

I am irritated by the term “AI slop” because it shifts the responsibility from the user to their tool, from the way they use the tool to something that’s inevitable.

silverpill Mar 11

@lain_7 @hongminhee I focus on the tool because it seems that a lot of people who use this tool remain unaware of how it affects them. Maybe they are aware and just don't care, maybe even majority of them don't care, but in this particular case I was assuming the former.

David Mankins Mar 11

@silverpill @hongminhee

I think it’s wrong to focus on the tool — since it shifts the “blame”. A tool can be used well, or it can be used carelessly. It’s the person that decides how the tool is used.

I have to admit that, at least in coding, AI can overwhelm a person trying to use it carefully, but that doesn’t excuse *that person* submitting a careless, sloppy pull-request.

CaohuaK ❄️Mar 12

@silverpill @hongminhee DeepL or some LLM translation assisants often use hyphens to replace commas or semicolons, so I usually use this feature to confirm if a text is machine-translate.

Cleverson Mar 11

@fastfinge @hongminhee This cultural selectiveness is why I am for a constructed auxiliary language as lingua franca. The most popular is Esperanto, but there are technically better projects nowadays.

patricus Mar 11

@clv1 @fastfinge @hongminhee well, that isn't possible as such a language would need to be usable *somewhere* on it's start.

🇨🇦Samuel Proulx🇨🇦Mar 11

@patricus @hongminhee @clv1 It also doesn't solve any of the problems of "just learn English": can you afford the lessons? Do you have the cognitive ability to learn two languages? Do you have the time for the lessons? Are teachers and materials available? Are they accessible? Do you have a place to practice outside of the classroom? Also, "let's erase everyone's culture" doesn't sound, to me, like any better of an answer than "let's make one culture king".

I speak as someone who studied French for eight years straight, an hour a day, and never managed to pass a single course. There are people who just literally can't, when it comes to language learning. I'm one of them. Though to be fair, it's almost certainly a combination of the environment, the instruction, and other factors, rather than some flaw innate to me. But either way, I've never found a method that works.

I actually toyed with learning Korean, thinking that maybe it was the gendered nature of French, as well as the spelling, that was the problem. Plus I thought a more regular alphabet might help me. But after eight years of bashing my head against the French wall, I just...couldn't. Picking up a new language course felt like going back to hell, and I couldn't make myself stick with it for more than a week.

Cleverson Mar 11

@fastfinge @patricus @hongminhee Auxiliary languages are constructed to be easier on purpose. A regular alphabet is only one of many features that must be easy on a lingua franca, and there isn't a single natural language whose all features are easy. As for cognitive ability, as far as I've observed, it seems that the teaching method counts a lot. Have you tried different methods?