training an LSTM to generate noun/noun juxtapositions (kennings) from word vectors (input A: hyphen-concatenated text of two words; input B: mean of vectors for those words; predicting on single-word vectors). some results so far...
abacus → secretary-throwing
allison → substant-constructor
cheese → construct-strangle
daring → spectrometeration-starting
mastodon → stoner-stoner
parrish → vinict-harmony
purple → barb-black
trousers → construction-steep
whoops → crest-hipphese
zoo → allego-gar
@aparrish You made up "substant-constructor" and slipped it in, right? Either that, or the LSTM has learned to troll.
@danima I mayyyy have run the generator a few times to get results that looked good :)