For my main auto-roman, I had to hack torch-rnn so that it only output words from a vocabulary. To start with, I was just playing with modifying its probability matrix (without worrying about all that vocabulary stuff) and an obvious first trick was to oulipianly ban that common glyph. Its output was so good that I spun it off into a sub-auto-roman, a simulation of Dumas’ rollicking story. Happily, many of its protagonists - Athos, Porthos, Aramis, d’Artagnan, a cardinal, Buckingham, and that notorious spy Milady - lack that glyph.
Frustratingly, but amusingly, in its original form, it always got d’Artagnan wrong:
d’Angou
d’Art--Bons
d’Art--timid
d’Art-y
d’Arto
d’Arto-diamond
d’Artois
d’Art’glanc
d’Art’gli-and
d’Art’gna
d’Art
d’Ass
Lua is an oldish programming lingo which torch-rnn runs on, and in its basic form, it trips up on non-ASCII glyphs. So that comma thingumajig which marks a missing glyph, ironically, was making it hiccup.
Although I finally got it to work (and now know a lot about manipulating utf-8 strings in Lua, hoo boy) I still think it’s good in its original, buggy form. It suits d’Artagnan’s initial position as a young guy who just wants to hang with his tough old gun bros, who in turn just try to annoy him by calling him all sorts of silly things. Ah, who’s this kid? What, that guy? Him? Oh, him? Hanging around us all day? You know, d’Arto-diamond!
So, two simulations of Dumas, both with and without d’at guy.