Linguistic Tuning
Making AI truly fluent in how Canadians actually speak — from Québécois French to regional Canadian English.
AI that speaks Canadian
Most AI language models are trained on data that skews heavily toward American English and European French. This means Canadians regularly encounter AI that misunderstands their context, gets idioms wrong, and defaults to cultural references that do not fit.
Québécois French is not European French. When a Québécois says « char » they mean car, not chariot. « Pantoute » means not at all. « Magasiner » is the word for shopping, not « faire du shopping. » These differences are not trivial — they are the difference between an AI that feels natural and one that feels foreign.
Canadian English also has its own character. Spelling conventions (colour, centre, defence), terminology (toque, loonie, riding), and cultural context all matter when AI is helping Canadians with everyday tasks.
Our initiatives
Québécois French Benchmark
Creating a comprehensive evaluation benchmark for how well AI models understand and generate Québécois French, covering slang, idioms, register, and cultural context.
Canadian English Fine-Tuning Datasets
Building curated datasets of Canadian English text — government documents, journalism, literature — to improve model performance on Canadian-specific tasks.
Bilingual Model Evaluation
Systematic testing of leading AI models on their ability to switch between Canadian English and Québécois French naturally, including code-switching patterns common in bilingual communities.
Regional Dialect Mapping
Documenting linguistic variations across Canadian regions — from Maritime English to Prairie French — to ensure AI models can serve all Canadians, not just those in major cities.