How to automatically determine good Charades words?

Despite the comments, this is actually a very interesting applied question that linguistics should be concerned with. It has to do with both syntax and semantics but also paralinguistic isues such as gesture.

The answer to your question is that there's no easy way to generate such a list in a way that would replicate a human's instinct for what is easy to represent gesturally. However, you could make certain assumptions about length, frequency and abstractness. You will also want to exclude all function words and words that can be used as auxilliary. As to frequency, you'd probably want to avoid the 500 most frequent words and not venture much past words ranked about 10,000. You would also want to explore the position of the word in WordNet's synset.

There are several lexical databases that may be helpful for this but you will ultimately have to combine them for best results.

• WordNet is one place to go for semantic connections but even better might be FrameNet which has information about semantic frames which will give you an idea about concreteness

• The MRC Psycholinguistic Database has information about frequency and abstractness (but the frequency data is very out of date)

• The English Lexicon Project has more uptodate data on frequency and other useful characteristics

• WordFrequency.info has very good data on word frequency and parts of speech (some free some paid).

I think there is a great research project hiding in this.