Aesthetic language generation (ALG) systems have long been plagued by two very difficult design problems. First, emulating originality requires that the ALG product be random to some degree. But randomization soon leads to gibberish. Second, absent authorial intent, ALG systems have generally lacked compositional cohesion. Their outputs don't have beginnings, middles, and endings.I got there from Erin O'Connor's.
Erica has two primary components: 1) a language database that models the relational semantics of English in normal discourse and 2) a collaborating set of programmed libraries that define a grammar for use with the semantic model.
Here's "my" poem:
The waxy cageThe little stutter in "waltzzes" is particularly fine.
Organelle sheltered nervous.
Growing, hits the terrorist.
Terminating the circuit in some area.
Travels the member.
Travels a time.
Travels the child.
Travels forward the holiday.
Travels a booster.
Travels helpful advice.
Like the hall, absents 18 days.
Inflaming, jitterbugs fluorescent.
Gametocyte, as practically free as the box.
Finding in a prescribed organic matter, waltzzes.
Ends mosaic, saying to a material genuine article.
2 comments:
I don't want to publish my poem in this comment, because it was really weird, but if you want to look at it, the poem ID is 1497. My seed words were penguin, frolic, and vociferous. Go figure. :)
Man, what is it with weird poetry and penguins? Somebody help me out here.
Post a Comment