JKirchartz
commented
5 years ago I guess others are using their issues to log what they've been doing in the project, and I'm getting a little late in the month to keep track as I go - but I've got my git commit messages to keep me warm...
As it stands, I, so let's see what I've been working on:
Overall
I've found a HUGE bug in pos2tracery that I've fixed, so it better serves you!
Also I've been experimenting with LaTeX for a while, so this year I've actually got PDF output as semi-nicely formatted books! There's a title page & table of contents and everything, so check the output directories for PDFs!
Sherlock Shuffle
This has turned out pretty well - I wish spent more time working out how to analyze the corpus, because I think getting chapter names & book titles together would be nice - I've found a single file of the complete Sherlock Holmes all together and nicely formatted, so if I ever use Holmes as a source again I'm going to use that instead of mucking with Project Gutenberg.
Winkie
I've re-implemented what pos2tracery does using winkjs, thanks to the lemmatization in wink-pos-tagger I've started grabbing stems and later transforming them with Tracery's english modifiers (and including my own ing-er). This generates text essentially the same way Sherlock Shuffle does, but there's a lot more IF/ELSE statements & NLP, so it does it somewhat better. The results are still quite nonsensical, but with proper word endings in their proper places, they feel more writerly.
Vocabularyclept
I accidentally a poetry generator. While researching lemmatization and stemming and cut-up techniques I stumbled upon the concept of a Vocabularyclept Poem - where an author alphabetized all the words in a poem, published it, and challenged other writers to come up with a new poem - essentially inventing a non-magnetic Magnetic Poetry. This project creates a set of 10 such poems, randomly choosing to leave the arrangement of poetry up to the reader, or re-arranging the poetry itself. The bit that creates actual poems only looks at syllable counts, creating a sort of syllabary grammar like tracery, then replaces the original words with random words with the same syllable count - this relies completely on word/syllable which seems to be quite correct. In a future version I'd like to get rhymes working & following the original rhyme scheme - but I'd have to develop a tagger/grammar to handle that, and I doubt I can do it by friday; fortunately NaPoGenMo is right around the corner, in 5 months... perfect timing to let this idea simmer on the backburner.
Develop a new take on the Sherlock Shuffle, using my POS Tagger that outputs Tracery grammars... and then see what else I can come up with... I really liked combining art with the text last year - so probably more of that...