Earlier this year, the Journal of Experimental & Theoretical Artificial Intelligence published a paper by Messrs Manurung, Ritchie and Thompson entitled Using genetic algorithms to create meaningful poetic text. In it they describe an AI system built to produce computer-generated poetry, which they’ve named “McGONAGALL”. According to the authors:
McGONAGALL, applies the genetic algorithm to construct [poems]. It uses a sophisticated linguistic formalism to represent its genomic information, from which can be computed the phenotypic information of both semantic representations and patterns of stress.
So how does this methodology compare with reading the papers, finding a disaster, and shoehorning in as many “November/remember”, “cried/died” and “seen/green/Her Majesty the Queen” rhymes as you can? Here’s an example of the bionic bard’s handiwork when seeded with a couple of lines of Hilaire Belloc:
They play. An expense is a waist.
A lion, he dwells in a dish.
He dwells in a skin.
A sensitive child,
he dwells in a child with a fish.
Apparently, like it’s namesake, McGONAGALL is able to generate relatively meaningful poems if the constraints of metre are relaxed, or metrical poems which don’t make much sense, but struggles to combine sense and metre in a single work. Its reaction to railway bridges is unrecorded.
Perhaps predictably, I’m happy to stick with the old flesh-and-blood version of poetic imperfection. Thanks to Improbable Research for bringing this paper to my attention.