AI in Quebec French: Why It’s a Fight That Matters

Think about ChatGPT. You ask it a question in French, and it answers. Seems okay, right? Except… does it answer in Quebec French or Parisian French? And is it really calibrated for your reality?

The problem: AI speaks French, but not our French

Most of the data used to train AIs comes from the English-speaking Internet. Then they add a bit of French — but mostly French from France. Why? Because Paris has more web presence than Montreal. It’s like training an AI to understand hockey by mostly showing it soccer videos.

Result: an AI can understand “je vais bien,” but it doesn’t always get “ça va-tu bien?” or “c’est-tu correct?” It doesn’t know our expressions, it doesn’t grasp the Quebec context, and it can produce content that sounds weird to us.

Why is it called a “fight”?

Because making AI in Quebec French isn’t trivial. You need:

Quality data: texts, conversations, Quebec articles. There aren’t enough online.
Local expertise: people who know what we mean when we talk the way we talk.
Cultural pride: saying our language deserves to be properly represented in technology.

And it’s a real issue: if all tech speaks mostly English and Parisian French, Quebecers are progressively being digitally marginalized.

What we can do

Quebec companies are starting to create AIs that are more respectful of our French. Even OpenAI is working on improving its Quebec French (slowly, but working on it). The real game changer? Quebec researchers and companies building their own models.

Want to dig into this? Check out Sherpa to test the same question in different languages, or explore Laeka Research to read more about local AIs.