Limit languages in play to maintain statistical significant samples

[rfinean]

With 20 questions and 4 languages we got about 5 questions per language, enough to for a RAG pedagogy to distinguish between

• Green = "knows this" (>90%),
• Red = "no idea" (<50%) and
• Amber = "some understanding".

With 10 languages and only 20 questions the chances are that some languages will only have 1 or 2 questions asked, which tells the user nothing statistically significant. Need to aim for 4 or 5 questions per language.

• Should we select just 4 (random?) languages for 1st 20 questions? Then add one more language to each new "continue playing" round of 10 questions, with 50% of questions on the new language and 50% on the existing set?
• Maybe we should we stop asking questions on the "knows this" and "no idea" languages? So stop asking questions in a language after scoring 4/4, 0/4 or 1/4?

Don't want to become "hard fun", where assessment is killing the fun in the game. Evidence Centred Design (ECD) means thinking about:

1. Content model = the languages skills we are exercising = RAG comprehension of each foreign spoken language
2. Evidence model = observations and behaviours that will show 1 = the scores histogram
3. Task model = tools to engage in content and elicit 2 = the questions themselves