my plan today was to work on an abstract for that Philosophizing Our WWW conf but mistakingly wound up with brow furrows, frowning at LSTM programs and tutorials (working on a way of computationally making odd random words with phonological constraints)
@aparrish that's fascinating that an LSTM can (or cannot) find phonotactic constraints! I'm curious about its input data, did it contain syllabic annotations or was it a plain straightforward phonological transcription?
@ojahnn sorry not phonotactic constraints—i'm using an lstm for translation from orthography to 2d array of phonological distinctions (+bilabial, +nasal, ...), plus a 2nd lstm to go opposing way. i'm trying to put a variational autocoding that links lstm 1 and lstm 2 so i can pump out two words with similar sounds by sampling from two proximal points in a gaussian distribution (my "phonological constraint" = constraining random word making with an array summarizing phonological information)
@ojahnn though usually words from this translation do follow inglish phonotactic constraints!
@aparrish Ah, gotcha! thanks for sharing!