I've got a confession to make: for the past two years, I've been trying to learn French. Having been born into a family where 70% of the people around me spoke French, the guilt of a lifetime of disinterest in the language finally caught up to me when I started dating a British girl who, despite my years of not trying, somehow spoke one of the languages I grew up hearing at family gatherings better than I did.

So I tried to replicate the method I used to learn English: Consuming Content™. In my early teens, using the basics I'd learnt in school, I essentially practiced my English by playing video games and watching hours upon hours of YouTube. That plan worked out quite well for me — I think the fact I’m writing this in English is enough proof of that — particularly because it never felt like studying. I simply wanted to play/watch/read a thing, that thing was in English, so I tried my best to understand the language it was using. Seems easy enough to replicate, I just needed to find some French things that interested me, and, little by little, I'd learn the language beyond the basic A2 level I was already at.

Movies? No problem, I already love classic French cinema, and now I get to be even more pretentious about it. Books? Plenty of great and accessible stuff to be found. Music? Easy as well, there are a trillion resources to be found, just pick your genre — plus, their music sometimes crosses over into Portuguese culture, so I already had a head start. Daily news? Pick your poison, it won't have the best politics but it’ll be good practice regardless. But YouTube videos? Social Media Content™? Well… It exists, I can find it, but the algorithmic flows and the interfaces that had helped me learn a language 10 years ago now seemed hellbent on making sure I never even had to think about language at all.

Let's start with the main culprit behind my English language skills: YouTube. My memory isn't good enough to recall exactly how I first started watching videos in English, but I do remember that subscribing to a channel I wanted to watch resulted in pretty much every subsequent video they made being shown to me on the homepage, and that's very obviously no longer the case, regardless of the video's language, so my method was broken from the get-go. Secondly, YouTube's algorithm is so strictly monolingual that even I, a Portuguese person residing in Portugal whose native language is Portuguese, don't remember ever having been recommended a video in Portuguese that wasn't from their “news” tab, despite following and watching a few Portuguese-speaking channels. It's a lost cause, and it baffles me because no other large scale recommendation algorithm I use functions this way, not even Google's own News app. Twitter, Instagram, BlueSky, TikTok, Tumblr, etc. all regularly recommend me posts in the other languages I speak (even some I don't speak, like Japanese, simply because I keep interacting with posts about anime), and had no difficulty latching on to my desire to seek out more posts in French ever since I started engaging with it.

Nevertheless, I perservered. I saved the videos I wanted to see, checked a few channels directly, and made do without the algorithm. What really got me though, was what happened when I clicked on these videos: they were automatically dubbed to English using a terrible, extremely unnatural AI voice.

This feature has been a thing since around late 2024, from what I can gather, and, despite objection from everyone, it has only gotten more pervasive. YouTube now plays the dub automatically for any language it believes you don't speak, and every single uploader has to individually opt-out of it willingly, meaning you're forced to put up with the robotic dubs unless every single channel you watch disables it.

Let me be clear: this is no “this video is in a foreign language, would you like to dub it?” pop-up (which is what Gmail does). You have to individually select the original audio track for every video you press play on. I've had professors, teaching courses in English, suddenly be jumpscared by a robotic version of Jonathan Berger speaking Portuguese, and one of my partners, who is currently teaching English in France, says she's encountered it several times during her classes. What guarantee does she have, as a teacher, that her students won’t go home and just not bother to disable the automatic dubbing of English videos, despite her whole job being to help her students understand English?

It's debatable whether the motivation behind this change is ideological or purely profit-oriented, especially in an AI-happy stock market environment, but the effect is clear: YouTube, the website that had taught me how to speak English, is now designed so that no one would ever want to go through the same thing again. This decision is especially baffling because, despite having had a perfectly competent automatic captioning and translation feature for years, the platform had made it so the choice to enable captions and machine-translate them was always a conscious one, toggled video-by-video.

Before this change, the pact between the user and the platform was apparent: by enabling automatic translation features, you understand you are undergoing a process which has inherent signal loss and no human oversight. You consented to it with every button press: first to enable the automatic captions in the original language of the audio, then to translate them into whichever language you wanted. The user interface used friction purposefully. Through rejecting the path of least resistance towards intelligibility, it achieved what other mediums, like the book publishing industry and the entertainment industry, had not: translation was not the default, it was a willing choice. You had to first recognize that the video was in a foreign language before you could opt to translate it into any other.

This design decision can obviously be attributed in large part to machine translation being, at the time, a clearly flawed and experimental technology, but it was made nonetheless. Now, with AI dubbing (also a clearly flawed and nascent technology), YouTube made the opposite choice. And every social media platform¹ is following suit.

Booting up Twitter now leads me to a timeline covered in “Translated with Grok” notices, which I have to individually click on and disable for every single language in the world, one by one. TikTok is much the same, automatically translating every video caption in a foreign language, but it at least allows you to disable the feature as whole. However, the worst offender by far is Instagram. No one at Meta seems to understand what multilingualism is, and so it asks you to choose a “Display Language”, the language every single caption is displayed in. If I want to practice my French — which, ironically, is the language a good 1/4th of the IG Reels I'm shown are in —, I have to opt to either ignore the captions completely or embrace the absurdity of browsing a feed where every English and Portuguese word I hear is transcribed to French.

These are deranged systems designed by people who seemingly have never thought about language beyond the concept of a “language barrier”, and who seem to be completely unaware of the realities of fluency and language-learning. These platforms’ misplaced blind faith in their translation algorithms is fundamentally altering the content of the things we produce and consume, and their menus and notices are hidden in ways that can often make both the producer and the consumer unaware of these alterations.

Sometimes this can result in frankly hilarious outcomes, like this post from a hardcore punk snob on Twitter correcting a Polish tweet on the correct English terms to use to describe a show:

But this Japanese post supposedly translated into “British English” showcases exactly why we previously delegated machine-translation to sub-menus: put plainly, it's not good enough, and it can never be good enough. The way these models are built inherently leads them to weird assumptions, and, as they balloon in size and scope, and absorb more information, they become prone to overreach in translation and hallucinating details that weren't there, like these stereotypically English terms:

Last issue, I talked about the overreach of the “It Just Works” mentality and how it affects AI development and AI usage. This new design overreach of social media platforms is, to me, yet another facet of this problem. I’ll freely admit the issue of computing in multiple languages is particularly dear to me, but I had been fine with machine translation implementations up until this recent flood of changes. The flaws of these systems were, up until now, (mostly) fine because they were presented in a way that made their existence obvious.

Translation was one of the few remaining processes in computing where the impact and limitations of its usage was made as clear and understandable as possible, partly because anyone who speaks a second language can understand how messy and imperfect even fully human-made translation can be. In a tech industry where the teachings of Media Studies and Critical Theory are often either discarded or co-opted, translation remained a field where the design of our tools followed the caution the social sciences preach. As our technology encroaches further into our language, letting changes like these take hold will mean relinquishing the small amount of control we still have over the words we read or hear inside the posts that are recommended to us.

Thank you for reading noReturn, a biweekly newsletter releasing roughly every other Friday. Follow for more musings on everything Media™ related. Be it tech, film, games, music… whatever really.

If you enjoyed what you read, consider donating on Ko-fi.