“The whole world is faking it”: Computer-Generated Poetry as Linguistic Evidence

The following is a short review I wrote of discourse.cpp (pdf available here) by O.S. le Si, ed. Aurélie Herbelot, published by the Berlin-based Peer Press in 2011. The review was just published in the December issue of Computational Linguistics.


discourse.cpp (Peer Press, 2011) is a short collection of computer-generated poetry edited by computational linguistics scholar Aurélie Herbelot, produced by a computer called O.S. le Si mainly used for natural language processing, and named after a program which tries to identify the meanings of words based on their context. In this case, Herbelot inputted 200,000 pages from Wikipedia for the program to then parse and output lists of items whose context is similar to words such as “gender,” “love,” “family,” and “illness;” for example, Herbelot explains that content in the opening piece titled “the creation” was “selected out of a list of 10,000 entries. Each entry was produced by automatically looking for taxonomic relationships in Wikipedia”; and, for the piece titled “gender,” she chose the “twenty-five best contexts for man and woman in original order. No further changes.” (47) The collection is, then, as we are told on the back-cover, “about things that people say about things. It was written by a computer.”

Poets – or, for the sake of those still attached to the notion of an author who intentionally delivers well-crafted, expressive writing, “so-called poets” – have been experimenting with producing writing with the aid of digital computer algorithms since Max Bense and Theo Lutz first experimented with computer-generated writing in 1959. The most well-known English-language example is the 1984 collection of poems The Policeman’s Beard is Half-Constructed by the Artificial Intelligence program Racter (a collection which was, it was later discovered, heavily edited by Racter creators William Chamberlain and Thomas Etter). discourse.cpp is yet another experiment in testing the capabilities of the computer and computer-programmer to create not so much “good” poetry as revealing poetry – poetry that is not meant to be close-read (most often to discover underlying authorial intent) but rather read as a collection of a kind of linguistic evidence. In this case, the collection provides evidence of the computer program’s probings of trends in online human language usage which in turn, not surprisingly, provides evidence of certain prevailing cultural norms; for example, we can see quite clearly our culture’s continued attachment to heteronormative gender roles in “Gender”:

Woman                        Man
man love —                    — win title
— marry man                — love woman
— give birth                   — claim be (18)

More, this linguistic evidence also draws attention to the ever-increasing intertwinement of human and digital computer and the resulting displacement of the human as sole reader-writer now that the computer is also a reader-writer alongside (and often in collaboration with) the human.

As Herbelot rightly points out in the “Editor’s Foreword,” to a large extent this experimentation with the computer as reader-writer also comes out of early twentieth century, avant-garde writing that similarly sought to undermine, if not displace, the individual intending author. Dadaist Tristan Tzara, for instance, infamously wrote “TO MAKE A DADAIST POEM” in 1920 in which he advocates writing poetry by cutting out words from a newspaper article, randomly choosing these words from a hat, and then appropriating these randomly chosen words to create a poem by “an infinitely original author of charming sensibility.” Tzara was, of course, being typically Dadaist in his tongue-in-cheek attitude; but he was also, I believe, serious in his belief that the combination of appropriation and chance-generated methods of producing text could produce original writing that simultaneously undermined the egotism of the author. However, insofar as discourse.cpp comes out of a lineage of experimental writing invested in chance-generated writing and, later, in exploiting computer technology as the latest means by which to produce such writing, it also comes out of a certain tradition of disingenuousness that comes along with this lineage. No matter how much Tzara and later authors of computer-generated writing sought to remove the human-as-author, there was and still is no getting around the fact that humans are in fact deeply involved in the creation process – whether as cutters-and-pasters, computer programmers, inputters, or editors. The collection, then, is a much more complex amalgam than even Herbelot seems willing to acknowledge as discourse.cpp is evidence of the evenly distributed reading and writing that took place between Herbelot and the computer/program itself.


One thought on ““The whole world is faking it”: Computer-Generated Poetry as Linguistic Evidence

  1. Hack writing by scholiasts in Byzantium has many of these characteristics you attribute to machine-man production. Take a few texts, lift a bit here a bit there. make a few mistakes and a couple of howlers then add a few fantasies and then call it history. The similarities lie in the boilerplate results which pose as original single author works.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s