I fixed the problem with my language apps, and I should have done so sooner

One of the more practical applications of AI I’ve seen is in education. Writing flashcards, organizing notes, and collating sources take time that could be better used studying.

While AI cannot substitute for the rigors of personal research and writing, it’s fantastic for taking the work out of preparation.

Specifically, I’ve found use in using Google Translate’s practice tool to prepare for holidays. The app draws up personalized learning plans and provides ample practice of the fundamentals, without wasting my time on irrelevant topics.

However, excellent as it is, Google Translate cannot substitute for a human teacher and real-world experiences.

The subtle problem at the heart of AI-powered language apps is the same one faced by translators worldwide. AI cannot replace human translations, but that fact is not acknowledged by everyone.

Related

I'm officially learning French on the go, thanks to this free Google tool

Tiny Lessons helps you learn what to say in when going wine tasting or asking for a receipt in German

Posts 3
By  Jenny McGrath

AI lacks the necessary context for translation

It struggles to understand stylistic choices

A perfect example of the problem at the heart of AI-powered translation is a test run by literary translator Yoann Gentric (via The Guardian).

He entered the same sentence, “Bright, sharp night air, bracing,” into the AI translation engine DeepL twice, once in 2022 and once in 2026. Both times, he asked the AI to translate it into French.

The first time, the final translation: “L’air de la nuit, vif et vif, était vivifiant,” read as “The night air, lively and lively” when translated back into English.

A six-year-old could identify why the translation was so ridiculous, but in 2026, the service had dramatically improved. Now, the translation read: “L’air nocturne était vif, pur et vivifiant” (The night air was crisp, clean, and invigorating).

The result still lacked the style of the original text, but it was nevertheless a dramatic improvement over the 2022 translation.

The implications for translation jobs are clear, but these machine-driven translations also have implications for anyone learning a new language or consuming translated media.

We miss context when learning from a machine

AI cannot understand subtlety

The problem with the DeepL translation above is that it couldn’t represent the style of the original sentence.

While the translation was technically accurate, it was wholly inadequate for publishing. Yes, context is always muddied in translation, but AI is making it into a serious problem.

Learning a new language isn’t just about memorizing words and verb conjugations. Understanding the subtleties in a language comes from experiencing human speech and media.

Communication is far more than listing words; it’s about how we arrange and use them.

AI can’t understand context. While the technology has advanced sufficiently to create technically accurate translations, it is wholly insufficient for creating a full understanding of the context.

Why apps like Duolingo can do more harm than good

We learn words, not a language

Duolingo received massive backlash in April 2025 when CEO Luis von Ahn announced that it would become an “AI-first” company.

The plan was to phase out contract workers and pause hiring, diverting as much work as possible through AI.

The result was that Duolingo was able to introduce 148 new language courses, double the amount it offered before the shift to AI.

Many users have complained about the perceived drop in quality of Duolingo since this shift.

A common complaint has been that the app constantly revisits topics the user has learned while avoiding topics that need work. However, the real problem is more insidious.

Duolingo’s employees cannot vet every single part of these AI-generated courses. 148 new language courses are a colossal amount, and much of what users will see will be entirely AI-generated.

The app’s approach to teaching languages has been criticized in the past, but now it is almost actively harmful.

The AI-generated examples provided by Duolingo will never capture the nuance of a real speaker.

Subscribe for clearer takes on AI and language learning

Want smarter use of AI language tools? Subscribe to our newsletter for expert analysis, real-world examples, and actionable guidance on translations, learning apps, and preserving linguistic nuance.

Get Updates

By subscribing, you agree to receive newsletter and marketing emails, and accept our Terms of Use and Privacy Policy. You can unsubscribe anytime.

While the app is effective for basic words and phrases, in its current stage, it will do more harm for long-term learning than good.

The obvious alternative is to switch to another language app, but most use AI in some form or another.

Even my favorites, Babbel and Memrise, use AI, although they do not abuse the technology like Duolingo. Ling, another excellent app, limits AI to a chatbot in the app, which can be easily avoided.

Be mindful of all translations

The sad truth is that translators have to work just as hard to edit AI-generated translations as to create translations from scratch.

There’s a perception that editing is easier or faster than writing, which has caused rates to plummet. The result is usually effective, but it’s thanks to the human, not the machine.

The best way to ensure that you won’t be impacted by these soulless translations is to absorb as much content in the native language as possible.

Watch TV shows or movies to experience the language as it’s meant to be heard. That has always been great advice for decades, but with the advent of AI, it’s now a necessary part of learning a language.