If you work in HR or L&D, someone' most likely asked you the question: what are we doing about AI and language training?
It's a fair one. AI is changing almost everything, and language learning isn't getting a free pass. It's worth knowing what AI actually does well in this space, what it doesn't, and where the interesting work is happening.
It's more nuanced than the headlines, and, believe it or not, a lot more interesting.
Most language apps use AI in some form. Speech recognition that catches your pronunciation. Vocabulary that adapts to what you already know. Feedback that responds to what you actually wrote, not a fixed script. None of that is new or revolutionary. It's just really good product design.
Does the app use AI? Wrong question.
Do the people who built it actually understand how language learning works? That's the one.
Passive and active language use are completely different things. Reading a sentence is passive. Translating it is passive. Typing a response while a chatbot waits patiently for you is also passive. Active use, the kind that lets you hold your own in a fast meeting in your second language without quietly translating in your head, is something else entirely. It needs automatic reflexes. And those only come from a very specific kind of practice.
Most AI tools, even the impressive ones, don't know the difference.
You ask an AI tool to help you practise your French. It answers fluently, confidently, and incorrectly. Wrong article. A sentence no French person would actually say. It doesn't correct you when you make the same mistake three times because it isn't tracking your errors. It's just generating the next plausible answer.
In AI terms, hallucinating means producing something wrong with full confidence. Not maliciously. Not lazily. The system is built to sound convincing, not to be accurate. For language learning, that's a real problem. You can spend an hour practising and walk away more fluent in your mistakes than when you started.
But that's not a feature.
At Semantics Belgium, we've been using AI inside our e-learning platform Mr Mentor for years. The difference is that Mr Mentor is built on a teaching method first. The AI works in service of the method, not the other way around.
In practice, students move through a constant flow of questions and answers, brought to life with animation. Short answers aren't allowed. You build full sentences every time. Answers are evaluated by AI on precision and accuracy, not just whether you used a keyword.
The point is to build automatic reflexes in the target language. Same goal as a classroom session, just carried into your day.
That's where AI earns its place: accuracy, repetition, the unglamorous practice that has to happen before fluency can even start to show up.
One of the more interesting things we're doing right now is using AI to help design training programmes. We recently worked with a senior manager at a major insurance company who needed to negotiate in Dutch at a high level, with the right industry vocabulary built into complex grammar, all matched to his actual job.
AI helped us pull the framework together quickly. Our pedagogical experience turned it into a programme that actually worked for him in his role.
That combination, AI doing the heavy lifting on structure and humans doing the judgement work, is where the value is right now.
Some people look at simultaneous translation and ask whether learning a language will eventually be unnecessary. If your phone translates in real time, why bother?
It won't replace it. Not in a professional context.
Sure, you can imagine sitting in a meeting with a translation device between you and everyone else, and technically people understand each other. But understanding isn't the same as connecting. Being translated isn't the same as being heard.
Fluency is confidence. It's register. It's knowing when to speak and when to listen. It's hearing the hesitation in someone's voice, noticing when the tone changes, and answering reflexively, with the right words, in the right tone, without the need to translate.
That's what good language coaching builds, and AI doesn’t do that yet.
Our edge has always been our teaching method. AI doesn't change that. It just gives us more places to apply it.
We use AI where it works. Structured practice, repetition, consolidation outside the classroom, programme design. We trust our coaches where it doesn't. Anything high-stakes, anything that needs nuance, anything where the difference between a good and a bad conversation comes down to a person reading the room.
Method first. Technology in support of it.
At Semantics Belgium, we've been building language programmes for professional teams since 2010. If you want to think through what role AI should actually play in your organisation's language strategy: