Olga Beregovaya argues companies should generate AI content directly in target languages rather than translating from English, which risks hallucination and cultural bias. She introduces terms like "translatees" and "modelis" to describe different types of AI output, and warns that AI's polished text may reshape how humans write.
Chapters
Let's take a large enterprise with presence in 20 countries. All the plethora of content is developed in English. The traditional approach is you take all the assets from the headquarters and then you translate them into remaining 19 geo languages, just taking source and mapping it to target. Imagine a byte text. Source, target, source, target, source, target. That's the paradigm is not gonna live for long. I have 2 degrees in linguistics and I've never translated anything as a translator. You don't wait for the source to be created. You can just take the inputs like SKU tables, any kind of brand book guidelines, and you can actually jump straight to creation.
So you're jumping the step of let's create something in English first.
It might be more native to the target language or to the language of generation than that translation process. Translation process is still much more predictable. When you generate, you really need to trust the model to not dream things up. My favorite meme is a person asking a large language model, "Is this mushroom poisonous?" The model confidently says, "No." And then fast forward, there is a tombstone on this person's grave and the model goes, "Sorry, this mushroom was poisonous." There is a term of linguistic colonization. It speaks to me and translates into my, say, language that my parents spoke to me in Russian. Your parents spoke to you Mandarin. So we still want to hear our mother tongue. I still want to learn foreign languages. I mean, I'd still love to learn Mandarin as much as it is my fingertips.
I want to learn other languages too. I feel this is like when I visit Japan, I feel like if I can speak their language, it's not the same as I just chat with them using ChatGPT. It's not the same. I feel the effort level and then the level of, you know, building relationship and, you know, enhancing like mutual standing, it's It's not just about the tools. It's like I want to make a friend.
My parents do not speak to me in language model language. My parents spoke to me in Russian. Your parents spoke to you in Mandarin. So we still want to hear our mother tongue. That's why I'm saying that interpretation probably has a shelf life than translation as we know it.
I think it's an interesting future to think about. If I don't need to speak your language and you don't need to learn English, if neither of us speak English, we can still communicate very smoothly. Do you think it will work? Like for human-to-human communication, the models and the tools will so good that you won't feel the gap.
I can say that it already works when I take a cab in Beijing, but then we have a whole other set of issues, right? Which is, do we want to stay in our monolingual space? Do we want to never have to learn another language? There is a term of linguistic colonization, right? So if, for instance, again, if we talk about the models are still predominantly trained on English, even if I speak and it speaks to me and translates into my language that's much less represented, am I going to learn more about the phenomena of English-centric world? And is it eventually going to skew my cultural perception of the world?
Are people doing anything in that regard, or you think this is just happening because people want more information?
There are a lot of initiatives around collecting training data for languages that are under-resourced. If a language is under-resourced, make it fully resourced. The way we engage with models is much more personal. Models learn from you, right? They are becoming much more there's a lot of, for the lack of a better word, maybe intimacy and relationship that you build with a model and then translates for you. The way you engage with models, like chat environments, for instance, now is becoming intimate enough for you to still be able to keep that level of personal relationship, even if you talk to a person from another culture.
I almost feel that's a mindset difference. It's like a philosophical question. It's not just about using models.
It's a huge philosophical question. It's not about— no, no, the whole how do we engage with software these days? How do we engage with tools at our disposal these days, right? It's dramatically I mean, the whole paradigm is completely different. We actually— I was on a panel with a professor in ethical AI, and apparently there is a whole discipline now treating model and chatbot addictions. Suddenly you're given a friend, and again, it can be multilingual friend, it can be a friend from Japan, but suddenly you're giving a friend who's there to please you, is tailored to please you, is tailored to give you a positive answer. And suddenly, well, I was arguing until a girlfriend of mine and I spent about 4 hours trying to generate different poetry styles from the model. And like, okay, 4 hours later, okay, who am I to talk about addictions? Here I am, here I am perfectly addicted. But it's indeed, it's a philosophical paradigm. How much do I want to keep human interaction as opposed to a friend that conveniently sits in my phone and is there to please? Are we philosophically, are we now more ready for an engagement where we'll always be pleased and never contradicted? Oh, right, which is exactly, yeah, where the model can dream things up, be there to please you, say what you want to hear, adjust to your tone of voice as you like it. There is a term, another term that I love, which is modelis.
What is modelis?
In the old days of machine translation, or in general, actually, even before machine translation, human translation, there was a term 'translatees' coined by linguists, which is basically when you can absolutely spot that this text was translated and there are a lot of rabbit ears sticking right and left where you can see, okay, this was not created in the target language. Then it became machine translatees where you could actually spot that it was post-edited from machine-generated text. And modelis Mollees, as you know, you're in machine learning and data science, there are a lot of tools out there that capture content that are specifically designed to capture content generated by AI.
You're talking about more than just, you know, AI-generated content, but also translated content. So that's like double Mollees, right?
Indeed. It's Translatees meeting Mollees. So A, it's translated, and B, it carries whatever the stylistic rules you gave to the model and whatever you asked of the model. So it's very interesting because The false fluency is there. Do you want— is it even factually accurate?
I've done generating my English blog posts into Chinese and just reads— something is off, right? It's a pure translation, but it just doesn't work.
And I would say what's usually off, like sometimes one thing that models obviously saved us from is staring at a blank page where you need to write a white paper. But when you just trust the model, and I know whenever I receive text that was translated by a model or generated by a model, usually there are so many repetitive patterns. The text is overly polished, the patterns are overly predictable, and I'm sorry to say it, but the concepts are often very trivial. The way I would work with model-generated text, I was like, okay, give me 10 ideas. And I was like, okay, if the model could produce these 10 ideas, I'm going to go with the 11th because the model already generated the 10 most obvious ones.
You as a linguist, how do you feel about this? Do you think this is just that information is going to just be widely spread in a way that looks obvious and generated versus using model will get to the level that actually can fix this.
If it looks human-like, then it's treated as human content. So I think being exposed to so much automated, automatically generated content is definitely going to tweak the way we ourselves speak and write. When was the last time you used proper punctuation in your text message?
Never.
Long, long while for me. I stopped putting commas and question marks because they will people understand anyways, right? So that's just one of the simplest examples of how technology influences our speech. And I think the more model-ese we see of this more predictable text, more polished text, I think at some point in time it will also influence how we speak and write ourselves. So I think it will be this cross-pollination. It's not necessarily a bad thing. The language is here to evolve, but there will definitely be this human-machine cross-pollination.
