johnstalcup
this is kind of a tangent but kind of related: you can use a reverse dictionary (http://www.onelook.com/reverse-dictionary.shtml) as a sort of "multi word thesaurus" when trying to condense a couple of ideas down into one word
the page is pretty, results are fast, and the letter based slots specification is nice.
but i think the results are pretty low quality (gibberish).
do this: * use an emotional connotation dictionary to classify a bunch of poems. * for each emotion category train a markov chain. * allow users to select a couple of emotions, then combine those emotion's markov chains and generate text.
for example, use tricks from poetry: assonance, consonance, alternate spellings, break the provided words into syllables and return names with multiple words that overlap on the last syllable one word and the first syllable of the next, do connotation analysis