[She list in dignity]
She list in dignity since that last reaching
To winter’s frost. Whilst eternal I thus
Compass’d all the trauma, we reach’d the present.
The world I’ve been moving, she’s lost. Wilderness.
Feb 28, 2012, interactive bigram generation with jGnoetry. Source texts: 60% Dante Alighieri, The Vision of Hell, Translated By The Rev. H. F. Cary; 40% Joy Division, lyrics.
As I mentioned in comments, I’m more a fan of the “generate and select” approach than I am of gnoetry-style constrained interaction. But I figured I should eat my own dog food and try out jGnoetry a bit.
Note for further thought: when using multiple corpora, it’s currently impossible for jGnoetry to get “stuck” in a certain combination. For example, originally I had the phrase “Whilst eagerly I thus…”, which I didn’t like, but apparently in this corpus the only word that follows “Whilst” is “eagerly,” from Dante. However, the way I have it programmed now, when jGnoetry generates a word, it first selects which corpus to look at, and then selects a word from that corpus. To guide its selection of a word, it looks at what word its generating immediately afterwards, so it may create a bigram. But if there is no bigram in the corpus it’s selected, it just falls back to a unigram, i.e. picks a single word without reference to the word before it. So if I’ve almost finished a poem, and am only trying to replace the word “eagerly”, if I click regenerate 10 times, 6 of those times it will give me “eagerly” from Dante, and the other 4 times it will give me words from Joy Division. Is this the right way to do it? Dunno… it recalls “smoothing” from n-gram modeling. But it highlights how, even with a fairly standardize generation approach, the decisions you make during implementation will affect what you end up with. This is possibly related to the oulipian notion of clinamen.