The interesting pattern I noticed in the output is just lack of context in word roots. So you get a word that's clearly a portmanteau getting a completely unre

The interesting pattern I noticed in the output is just lack of context in word roots. So you get a word that's clearly a portmanteau getting a completely unrelated definition. Fortunately, breaking a word into parts is exactly the kind of thing a convolutional network should be great at...