undyingking: (Default)
undyingking ([personal profile] undyingking) wrote2007-10-23 09:40 am
Entry tags:

Free rice game

If you haven't already seen this, you might like to give it a go -- not just because of the free rice aspect, but because it's a fun and well-designed game.

Get three in a row right and you go up a level -- get one wrong and you go down a level. At higher levels you get asked more difficult words, difficulty being assessed by the proportion of previous contestants who've got them wrong.

50 is the highest level, so if you can stay there for any length of time, you're doing pretty well. My longest run at 50 was 4 words. Beat that if you dare!

And you'll learn lots of interesting (but, for the most part, practically useless) words.

(Thanks to [livejournal.com profile] watervole for pointing me at it!)

[identity profile] bateleur.livejournal.com 2007-10-23 09:46 am (UTC)(link)
How about:

* A wrong answer is considered "good" if it receives a high proportion of votes.
* If the spread of quality amongst wrong answers to a question is wide enough, swap out the worst after some fixed number of uses.
* If the spread is narrow, randomly swap in any other dictionary word for the worst option, again after some fixed number of uses.

This will converge all questions to a situation where all but one of the false options are maximally confusing.

The tricky part is that you'd have to have a hand-coded list of all words which could be considered correct answers so that you never present one of those intending it as a false option!

[identity profile] undyingking.livejournal.com 2007-10-23 10:20 am (UTC)(link)
That sounds good. I guess ideally each word would have a bank of wrong answers classified by their quality from which it could select randomly according to a defined desired spread of quality (maybe something like 50%, 33%, 17% of people who get it wrong chose that one), with the quality values being adjusted over time according to "performance during most recent n exposures to players" sort of thing.

It'd take a lot of trials to converge on good values, but I guess they have plenty of trials going on.

(I suspect the hand-coding of synonyms is something that'll be difficult to do without in any automated system. It wouldn't surprise me if they had it already.)