Articulations

by Allison Parrish

Poetry. Fiction. Literary Nonfiction. Artificial Intelligence. Digital Humanities. The poems in ARTICULATIONS are the output of a computer program that extracts linguistic features from over two million lines of public domain poetry, then traces fluid paths between the lines based on their similarities. By turns propulsive and meditative, the poems demonstrate an intuitive coherence found outside the bounds of intentional semantic constraints.

Genres: Poetry

120 pages, Paperback

Published January 15, 2018

Reviews

[Ally]

A super-cento sorted by sound. Parrish harvests all of the lines of poetry on Project Gutenberg, assigns them a sound vector, and performs a random walk through sound space, picking the nearest line of poetry at each step, so that the only thing connecting each line to the next is similarity in sound.

The result lives interestingly close to the border of sense - it benefits greatly from the fact that the majority of poetry in English is in blank verse, and from the selection biases of Project Gutenberg (where the mass of the material clusters in a relatively narrow historical range - luckily the range closest to the modern day).

It's a neat idea and fun to read aloud, though no particular moment struck me in 100+ pages. (This is, for me, a general problem with longer generative works using simple rules, in both poetry and music - it's easier for the meatiness of the idea and the duration of the realization to drift apart than it is for a through-composed piece.) The lines themselves, being taken from other poems both verbatim and out of context, are low in surprise. Given the nature of the procedure, it's hard to tell what is happening when the lines *don't* feel connected - does the algorithm think two lines are close in sound when my ear doesn't, or did we just take a large step, or do we happen to be wandering through a big gap in soundspace without many nearby lines in the corpus?

I would probably have gotten more pure pleasure out of a few pages of shorter, curated examples, though from Parrish's body of work I expect that minimizing the impact of authorial/curatorial taste, outside of the design of the algorithm, is an important part of her intention. Much of Parrish's work is also so incredibly labor intensive, and it's a pleasure to hold a book-length thing by her to stand in for other pieces I love. It doesn't seem practical that could happen unless the robots get to run free once in a while and we get the occasional Articulations and Everyword, which aren't reallllly readable cover to cover but are still cool artifacts.