When we test Eddy-AI, it is suppose to answer in the language the user uses. If I write in norwegian it will answer in norwegian, and if i write in english it will translate everything in the Workspace and answer in english to the user.

But we have an issue, because norwegian (from Norway) and danish from Denmark are extremely similar in many words and sentences. With short questions, then the AI struggles to know what language to answer in.

If I understand you correctly there sort of is a "fallback language" as you say it will answer in the default language of the knowledge base if there is only one or two words.

So it seems like that threshold should just be extended a bit until the AI is sure that is is a different language from the kowledge base?