I have been thinking about this myself. I'm working on some custom dictionaries for words I discover from my corpus of movie subtitles. Which I'm sure is not a new idea, but it's fun, because it gives me a dictionary that only contains the words that people "actually use", and with "real" example sentences. (words in quotes because movie dialogue isn't 100% as real as I'd like.)

I'm sure this is not a remotely new idea, but I'm having fun with it. I also like that I can see how common every form of every word is. I was surprised to learn that almost none of the most common words are nouns. And in my internal tools I can filter by movies released a certain date to track changes, which is neat.