To Learn to Speak a Language, You Have to Actually Speak It

01 /The Concept

Most people who've studied a foreign language know the feeling: months of grammar drills and vocabulary flashcards, followed by the crushing realization that they can't actually hold a conversation. That's because speaking a language is a fundamentally different skill from knowing its rules – or even being able to read it.

To learn to speak, you have to speak. Not study how speaking works. Not listen to others speak.

Univerbal built an app for conversational language learning – one where users can practice speaking across 20 languages.

The key insight is that Univerbal users learn by actually talking – not with other humans, but with AI characters who are simultaneously engaging conversation partners and skilled teachers.

This sets Univerbal apart from two established alternatives: native speakers who usually have no teaching experience, and professional tutors who often stick to uninspiring, textbook-driven topics.

Because the conversation partner is an AI, it can hold a meaningful discussion on virtually anything. Users pick their own interests, and every conversation revolves around those topics.

During conversations, the AI character corrects mistakes in real time, explaining not just what was wrong but why the correct form works better.

Crucially, there's no fear of embarrassment. The student is talking to a bot, not a person – which turns out to be liberating. Students make more mistakes. And the more mistakes you make and correct, the faster you improve.

The AI isn't just chatting idly, either. It builds a structured learning arc around the goals the student sets at the start, so each conversation is effectively a lesson advancing toward a defined objective – just packaged as a natural dialogue rather than a lecture.

Beyond the core B2C model, Univerbal is also testing a B2B approach, positioning the app as a supplementary tool for language schools and private tutors.

Univerbal went through Y Combinator last winter but only published to the YC blog ten days ago, which is how this crossed the radar. By that point, the startup had already grown to over 100,000 users and raised $1.5 million in additional funding on top of $500,000 from the accelerator itself.

For good measure, Univerbal also landed on Wired's list of Europe's 100 hottest startups.

02 /Why It Matters

Y Combinator's interest here seems to be less about Univerbal specifically and more about the category – AI-powered conversational language learning. The proof: the current YC batch includes ISSEN, a startup pursuing almost exactly the same concept.

In other words, YC is confident the category will take off. It just isn't sure which specific startup will win it. Hence, multiple bets.

But the implications run even deeper. This isn't really just about language learning – it's about a paradigm shift toward what might be called dialogic learning: acquiring new knowledge not by passively receiving a teacher's explanation, but through live, back-and-forth dialogue.

The method is ancient. It's how Socrates taught his students. He famously claimed he couldn't teach anyone anything – he could only make them think, so they'd learn for themselves. His lessons were structured as dialogues full of provocative questions designed to draw out reasoning and self-discovery. That approach became known as the Socratic method.

Dialogic learning is genuinely effective – but historically terrible to scale. Once education industrialized, the individualized Socratic dialogue gave way to the one-to-many lecture: a teacher broadcasts knowledge; students memorize it. The thinking, processing, and understanding got optimized away as too slow and expensive.

Strangely, the internet made this worse. It enabled lectures to reach unlimited audiences, making one-way information delivery the de facto norm – "watch the YouTube video and stop bothering me"

AI changes the equation. Conversations can now be conducted not by scarce, expensive humans but by AI characters with deep knowledge, trainable dialogue skills, and infinite patience.

This is already happening. One example: Humy, covered last month, built a platform with AI twins of well-known figures that teachers can invite into lessons – each AI character capable of holding individual conversations with multiple students simultaneously on any topic the teacher sets.

03 /Opportunities

The most direct opportunity is building conversational AI language learning apps.

Yes, these already exist. But if Y Combinator is still placing new bets in the space, the potential clearly hasn't been tapped – it's barely been scratched. There's still a lot of room to dig and grow.

The broader opportunity is applying dialogic AI methods to other subjects and disciplines – from K-12 and universities to adult professional development.

Some of the most interesting applications will be unexpected. One example from the current YC batch: Soma Labs ([covered here](/review/dlja-masshtabirovanija-obuchenija-ne-hvataet-vot-jetogo)), a platform featuring AI patients that medical students interview to practice extracting the information needed to reach a diagnosis.

The range of possible applications is enormous. The challenge is picking the domain where dialogic AI delivers the maximum impact at the lowest cost of implementation. What would that domain be? It's a genuine question, not a rhetorical one – share your thinking in our community chat.