Split the taskDécouper la tâche

Here is a draft list using terminology that should work well in Canadian French, with a preference for terms commonly used in academic and professional contexts in Québec.

A few entries are worth flagging because usage is not fully settled. GML is a good French equivalent for LLM, although grand modèle de langage is also widespread. For prompt, requête is terminologically cleaner than the very common borrowing prompt, but depending on your audience, invite and instruction also occur. For foundation model, modèle de fondation is established, while modèle de base can be ambiguous. For RAG, génération augmentée par récupération is reasonably transparent, but several competing translations are still in circulation.

artificial intelligence
machine learning
deep learning
neural network
transformer
large language model
generative AI
foundation model
token
tokenization
prompt
prompt engineering
context window
attention
embedding
fine-tuning
retrieval-augmented generation
inference
hallucination
alignment

For a less technical audience, I would replace the architecture-level terms with concepts that occur more often in public discussion and everyday use.

For the four you flagged, my preferred substitutions would be:

• transformer → chatbot or AI assistant
• attention → AI agent
• embedding → training data
• tokenization → drop it and retain token only

Other good candidates, depending on what you want the glossary to cover, are multimodal AI, AI-generated content, synthetic data, AI bias, AI safety, AI system, and natural language processing.

So I might revise the 20 to:

artificial intelligence
machine learning
deep learning
neural network
generative AI
foundation model
large language model
AI model
chatbot
AI assistant
AI agent
prompt
prompt engineering
context window
token
training data
fine-tuning
retrieval-augmented generation
hallucination
alignment

I would consider neural network, deep learning, fine-tuning, and retrieval-augmented generation the remaining relatively technical items. If the intended readership is genuinely general-public rather than “non-specialists who work with AI”, I would probably replace some of those too.

For Canadian French equivalents

I would use three sources, in roughly this order:

1. OQLF, Grand dictionnaire terminologique / Vitrine linguistique. This is probably your best primary source if you specifically want Québec/Canadian French. The OQLF has an actively maintained AI vocabulary, L’intelligence artificielle en évolution : les termes de l’IA. Its entries distinguish termes privilégiés, acceptable variants, abbreviations, usage notes, and definitions. For example, it currently gives grand modèle de langage for large language model and jeton for token. [Vitrine linguistique +1]
2. TERMIUM Plus, from the federal Translation Bureau. This is the other source I would systematically check. It is explicitly Canadian, bilingual, and its AI records have been updated quite aggressively in 2025–2026. For instance, its current token record gives jeton / jeton textuel, and it has dedicated recent records for generative-AI terminology. [Termium Plus +2]
3. Translation Bureau’s Artificial Intelligence Terminology Concept Map. This is particularly handy for your task because it is already a curated English–French AI term list rather than a general terminological database. It includes things such as artificial intelligence → intelligence artificielle, generative AI → IA générative, deep learning → apprentissage profond, prompt engineering, and retrieval-augmented generation, with links into TERMIUM. The Bureau explicitly says its terminologists continually update TERMIUM as AI terminology evolves. [Our Languages]

For this particular glossary, I would cross-check every term in both OQLF and TERMIUM. Where they agree, the choice is easy. Where they differ, that difference is itself useful information: OQLF tends to tell you what is preferred in Québec, whereas TERMIUM tells you what the federal Translation Bureau recognizes in Canadian government usage.

I would use FranceTerme only as a secondary comparison source, not as the basis for a Canadian French glossary. The OQLF sometimes imports FranceTerme records alongside its own, so it is important to look at the author of the individual GDT record, not merely assume that anything appearing in the GDT is an OQLF recommendation. For example, there are separate OQLF and FranceTerme records for grand modèle de langage. [Vitrine linguistique +1]

If the goal is a 20-term US English → Canadian French terminology list, OQLF + TERMIUM is a very defensible methodology.

That gives a clear hierarchy:

1. OQLF
2. TERMIUM Plus
3. DataFranca as tiebreaker when OQLF and TERMIUM disagree

For each term, I’d record the preferred Canadian French equivalent, and only flag alternatives where there is a meaningful terminological difference rather than harmless variation.

I replaced AI assistant with multimodal AI, which is both more useful conceptually and has a clear OQLF/TERMIUM equivalent. The forms below prioritize OQLF, with TERMIUM cross-checking; for example, both support requête, jeton, affinage, fenêtre de contexte, and grand modèle de langage. [Vitrine linguistique +4]

Copy-paste this directly into Excel:

US English	Canadian French
artificial intelligence	intelligence artificielle
machine learning	apprentissage automatique
deep learning	apprentissage profond
neural network	réseau de neurones artificiels
generative AI	intelligence artificielle générative
foundation model	modèle de fondation
large language model	grand modèle de langage
AI model	modèle d'intelligence artificielle
chatbot	robot conversationnel
multimodal AI	intelligence artificielle multimodale
AI agent	agent d'intelligence artificielle
prompt	requête
prompt engineering	rédactique
context window	fenêtre de contexte
token	jeton
training data	données d'entraînement
fine-tuning	affinage
retrieval-augmented generation	génération augmentée par récupération
hallucination	hallucination d'IA
AI alignment	alignement de l'IA

Two choices worth flagging: rédactique is specifically the OQLF-preferred term for prompt engineering, although TERMIUM and DataFranca also recognize ingénierie de requêtes. [Vitrine linguistique +2] For retrieval-augmented generation, OQLF prefers génération augmentée par récupération, whereas TERMIUM proposes génération améliorée par récupération d’information (GARI) but also records the OQLF form; I therefore kept the OQLF form. [Vitrine linguistique +1]

Here is a draft list using terminology that should work well in Canadian French, with a preference for terms commonly used in academic and professional contexts in Québec.

A few entries are worth flagging because usage is not fully settled. GML is a good French equivalent for LLM, although grand modèle de langage is also widespread. For prompt, requête is terminologically cleaner than the very common borrowing prompt, but depending on your audience, invite and instruction also occur. For foundation model, modèle de fondation is established, while modèle de base can be ambiguous. For RAG, génération augmentée par récupération is reasonably transparent, but several competing translations are still in circulation.

artificial intelligence
machine learning
deep learning
neural network
transformer
large language model
generative AI
foundation model
token
tokenization
prompt
prompt engineering
context window
attention
embedding
fine-tuning
retrieval-augmented generation
inference
hallucination
alignment

For a less technical audience, I would replace the architecture-level terms with concepts that occur more often in public discussion and everyday use.

For the four you flagged, my preferred substitutions would be:

• transformer → chatbot or AI assistant
• attention → AI agent
• embedding → training data
• tokenization → drop it and retain token only

Other good candidates, depending on what you want the glossary to cover, are multimodal AI, AI-generated content, synthetic data, AI bias, AI safety, AI system, and natural language processing.

So I might revise the 20 to:

artificial intelligence
machine learning
deep learning
neural network
generative AI
foundation model
large language model
AI model
chatbot
AI assistant
AI agent
prompt
prompt engineering
context window
token
training data
fine-tuning
retrieval-augmented generation
hallucination
alignment

I would consider neural network, deep learning, fine-tuning, and retrieval-augmented generation the remaining relatively technical items. If the intended readership is genuinely general-public rather than “non-specialists who work with AI”, I would probably replace some of those too.

For Canadian French equivalents

I would use three sources, in roughly this order:

1. OQLF, Grand dictionnaire terminologique / Vitrine linguistique. This is probably your best primary source if you specifically want Québec/Canadian French. The OQLF has an actively maintained AI vocabulary, L’intelligence artificielle en évolution : les termes de l’IA. Its entries distinguish termes privilégiés, acceptable variants, abbreviations, usage notes, and definitions. For example, it currently gives grand modèle de langage for large language model and jeton for token. [Vitrine linguistique +1]
2. TERMIUM Plus, from the federal Translation Bureau. This is the other source I would systematically check. It is explicitly Canadian, bilingual, and its AI records have been updated quite aggressively in 2025–2026. For instance, its current token record gives jeton / jeton textuel, and it has dedicated recent records for generative-AI terminology. [Termium Plus +2]
3. Translation Bureau’s Artificial Intelligence Terminology Concept Map. This is particularly handy for your task because it is already a curated English–French AI term list rather than a general terminological database. It includes things such as artificial intelligence → intelligence artificielle, generative AI → IA générative, deep learning → apprentissage profond, prompt engineering, and retrieval-augmented generation, with links into TERMIUM. The Bureau explicitly says its terminologists continually update TERMIUM as AI terminology evolves. [Our Languages]

For this particular glossary, I would cross-check every term in both OQLF and TERMIUM. Where they agree, the choice is easy. Where they differ, that difference is itself useful information: OQLF tends to tell you what is preferred in Québec, whereas TERMIUM tells you what the federal Translation Bureau recognizes in Canadian government usage.

I would use FranceTerme only as a secondary comparison source, not as the basis for a Canadian French glossary. The OQLF sometimes imports FranceTerme records alongside its own, so it is important to look at the author of the individual GDT record, not merely assume that anything appearing in the GDT is an OQLF recommendation. For example, there are separate OQLF and FranceTerme records for grand modèle de langage. [Vitrine linguistique +1]

If the goal is a 20-term US English → Canadian French terminology list, OQLF + TERMIUM is a very defensible methodology.

That gives a clear hierarchy:

1. OQLF
2. TERMIUM Plus
3. DataFranca as tiebreaker when OQLF and TERMIUM disagree

For each term, I’d record the preferred Canadian French equivalent, and only flag alternatives where there is a meaningful terminological difference rather than harmless variation.

I replaced AI assistant with multimodal AI, which is both more useful conceptually and has a clear OQLF/TERMIUM equivalent. The forms below prioritize OQLF, with TERMIUM cross-checking; for example, both support requête, jeton, affinage, fenêtre de contexte, and grand modèle de langage. [Vitrine linguistique +4]

Copy-paste this directly into Excel:

US English	Canadian French
artificial intelligence	intelligence artificielle
machine learning	apprentissage automatique
deep learning	apprentissage profond
neural network	réseau de neurones artificiels
generative AI	intelligence artificielle générative
foundation model	modèle de fondation
large language model	grand modèle de langage
AI model	modèle d'intelligence artificielle
chatbot	robot conversationnel
multimodal AI	intelligence artificielle multimodale
AI agent	agent d'intelligence artificielle
prompt	requête
prompt engineering	rédactique
context window	fenêtre de contexte
token	jeton
training data	données d'entraînement
fine-tuning	affinage
retrieval-augmented generation	génération augmentée par récupération
hallucination	hallucination d'IA
AI alignment	alignement de l'IA

Two choices worth flagging: rédactique is specifically the OQLF-preferred term for prompt engineering, although TERMIUM and DataFranca also recognize ingénierie de requêtes. [Vitrine linguistique +2] For retrieval-augmented generation, OQLF prefers génération augmentée par récupération, whereas TERMIUM proposes génération améliorée par récupération d’information (GARI) but also records the OQLF form; I therefore kept the OQLF form. [Vitrine linguistique +1]