I've thought for a while that one could (1) take a list of the most common 10,000 words in a language (2) ignore any that are three letters or fewer (3) any that are four letters or more, assign them a three-letter code then (presto) there's a way to convey the most commonly used 10,000 words in that language using a fewer characters. Useful for something like Twitter, or for narrow bandwidth connections. Is not cryptography. Converting human-written text to and from that code might be trivial.