On Virtually Disembodied Expression

Poetry in the digital age is a curious subject. Writers with aesthetic, lyrical, political, and other agendas abound, despite mass culture's minimal engagement with this literary form. The popularity of MFA Creative Writing programs is remarkable (even if the quality of poetry produced is sometimes called into question), and readings and publications featuring non-academics are abundant. Small presses thrive and diversify on the Web. One of the few basic objective observations to be made about contemporary poetry is that some poets choose to write using structures established centuries ago, and many others are writing in open, or process-based, forms more recently developed. Questions as to what poetry is, and what it accomplishes, resound, and a singular definition or purpose is practically impossible to establish authoritatively.

Beyond these somewhat ambiguous circumstances, we live in an age when, as improbable as it may seem, poetry can be written by a computer programmed to do so. This is not a new possibility, either. Since the late 1950s, and throughout subsequent decades, programmers have gone to great lengths to create digitally infused verse.

The first wide-scale notice of this artistic development dates as far back as 1962, when TIME (5/22/62) published a brief article called "The Pocketa, Pocketa School," critically profiling a computer program named Auto-Beatnik. As I observe in my book Prehistoric Digital Poetry: An Archaeology of Forms, 1959-1995, Auto-Beatnik's output does not reflect sensitivity, but poetic traits found in the program's output emulates free verse (particularly stream of consciousness) and resembles a strain of Beat poetry. Auto-Beatnik's poems, like those made with many (digital) poetry generating programs, are not without obvious flaws (overt, mundane repetition is among the primary villains) and its premise is formidable.

Yet since poetry, historically, was calculated using fixed metric structures and patterns, maybe we shouldn't be surprised to now see poetry being formulated by humans working with computers. From another angle, automatically randomizing texts with computer programs is also a logical next step in the Dada progression. Like so many who labored with the craft in previous eras, contemporary poets (who happen to work with computers) confront social and artistic fragmentation in the world around them, using technology to atomize and hybridize texts that subvert, reflect, and perhaps extend the complex of cultural information. Authors use fragmentation to legitimize fragmentation and challenge the stability of language as a point of meaning; this process of re-assembling disparate pieces via technology can be used as a means to impart a sense of coherence and transformation.

Although approaches to the task are varied, the foremost characteristic of automatically generated poems is permutation. A set of words is used again and again, sometimes slotted into templates. However, if programming instructions are complex, and the databases employed are large (or multiple), the obviousness of such traits can be diminished (while linguistic content of many works is limited, other select words from entire dictionaries). Virtual poetry (or writing machines) can be entirely original, interactive (readers set constraints, parameters, and add vocabulary), and/or seek to simulate a certain style of writing, or the tone of a particular writer. In fact, the inclination to incorporate words and verbal intonations used by historically known poets, giving them new context and visibility, is fascinating attribute that has persisted throughout the span of this programming practice.

The first poetry generator, Theo Lutz's "Stochastic Text" (1959), for example, utilized words and subjects chosen from Kafka's The Castle. One of the earliest experiments by an American, Emmett Williams's concretist poem "Music," incorporated the most popular words from Dante's Divine Comedy. Such efforts were not engineered to duplicate, but rather propel new circumstances for (and possibly new understandings of) the original texts. Lutz's choice to build the first computer poems based on Kafka's book is especially intriguing, and adds a layer of significance to the endeavor. It is possible that Lutz chose Kafka's incomplete novel as a foundation out of respect for poetry, as a way to question the communicative values of machine modulated verse. While the processes of generating or consuming the poetry do not particularly reflect or require the reader to embody the type of mysterious bureaucracy experienced by the protagonist of Kafka's novel, an alienated, barren tone pervades the output of the program (see http://www.stuttgarter-schule.de/lutz_schule_en.htm for documentation; see http://auer.netzliteratur.net/0_lutz/lutz_original.html and http://copernicus.netzliteratur.net/index1.html for emulators created by Johannes Auer). As in other successful works of this sort, the best examples of Lutz's generated poems are impressive because the reader, via the condensation and computer processing of the materials, can rediscover the essence of Kafka's story, or somehow experience new perspectives derived from the original text. Ideally, the selection of words, combined with a stimulating programming method, enables a speculative, self-reflexive, unconventional style of expression.

To say a poem is automatically generated does not mean an enormous amount of effort has not gone into its production; nor does it mean that it has reached its final form (i.e., critic/practitioner Charles O. Hartman sees his programs as "first draft" writers). Writing a sophisticated poetry generator is not easy work, and positive reviews of the effort are far from being guaranteed because the results often dissatisfy discerning readers. As early as 1967, critical articles written by serious practitioners of the form began to appear. Insiders, such as John Morris, (whose experiments with, and discussions about, computer haiku are certainly worth reading), challenged and even denounced the purposes of the endeavor. Pioneers in the field, such as Nanni Balestrini, never returned to the task after early investigations. Nonetheless, experimentation has continued, and the results are impressive and intriguing for various reasons.

For instance, a few years ago, Jim Carpenter (a programmer affiliated with the Wharton Business School), had an idea that he could write a program that would output publishable poems, and he has succeeded. In the "Directed Poetics" section of his generator Erika (see http://etc.wharton.upenn.edu:8080/Etc3beta/), interactive pull-down menus enable the reader/user to establish a poem's content. In addition selecting the type of stanza, the "word pool" (i.e., topic context, topic only, topic synonyms, topic antonyms, and alliteration), and subject/object preferences, the reader also prescribes "grammars" and "context sources" for the poems. These latter components purport to allow the poem to embody and grammatical structures appropriated from known foundations. Besides offering grammars labeled "Mimetic," "nominatives," "titles," "Subordinate clauses," "fragments," "hinge clause," "lyric," "nominatives," "questions," and "common," the option to incorporate grammatical styles by Sylvia Plath, Rachel Blau DuPlessis, Frank O'Hara, and Gary Snyder is also offered. While I am not sure that the "voices" of these writers always emerge from the noise, the words of the two writers who are always at the core of Erika's poetry, Emily Dickinson and Joseph Conrad, are effectively repurposed by the program.

Millie Niss (with Martha Deed) has produced another generator that uses the styles of contemporary and canonical writers. Niss's "The Electronic Muse" (http://collection.eliterature.org/1/works/niss__oulipoems.html), which appears in the Electronic Literature Collection, Vol. 1 (a publication reflecting the wide variety of techniques writers have deployed using computers), creates and accumulates lines written in the manner of John Hollander ("Since broken and indivisible poverty rages, that works raging possible telephones on dense leaves"), Shakespeare ("We with speed crept around pale dispriz'd vows"), Anne Sexton ("If you urinate, then pine-paneled and well-born kisses menstruate"), Harriet Mullen ("Recyclable, we deliver with in the family way lights", Robert Browning ("Have vulgarest and apprised word like this natural true ear!"), or "Dick and Jane" ("Don't play blue ball as good as those green and green dogs now"). While generally less complex than Carpenter's work in terms of variety, users can add vocabulary to the poem identify its part of speech (i.e., noun, adjective, adverb), and edit the generated lines they have created. Quotes attributed to the styles of the authors (as seen above) are not entirely implausible. Another Web-based program, The Shannonizer (http://www.nightgarden.com/shannon.htm), is "a web toy with delusions of literacy" that gives a user the opportunity to "rewrite" input texts. Built on the premises of Claude Shannon's information theory, the program supplies "editors" named Lewis Carroll, Raymond Chandler, God, Miss Manners, Edgar Allan Poe, Dr. Seuss, Hunter S. Thompson, and Mark Twain. Here the results of the programmatic processing are not only effective, but humorous. A software program called Gnoetry (http://www.beardofbees.com/gnoetry.html) has also produced a number of compelling poems by completing a statistical analysis of pre-existing texts, although the program is not yet publically available, as are the others mentioned above.

While I have a special appreciation for the playful and often serious ancient generators discussed in my book, these recent generators are amongst the best yet produced. They indicate a desire for the virtual embodiment of known forms, styles, even authors. Are we facing a scenario where the dead poets might somehow, at least textually, come back to life, bringing new verses to wanting readers? What an odd, but yet not so far-fetched a possibility. In 2006, while living in Malaysia, I was approached by an Iranian professor who had heard one of my lectures on digital poetry. His attention to my research involved our common interest in generated poetry. He proposed a surprising collaboration: that we use AI (Artificial Intelligence) related Neural Networks information processing techniques to write new Rumi poems. What we needed to do in order to make this happen was to build a vocabulary, then teach the machine how to speak like Rumi. It was a refreshing, alternative approach to the task, and one that might work if we were crafty enough. But due to our physical separation (I returned to the US), unfortunately, nothing ever happened with the scheme we concocted.

The moment at which a poem is generated is a fusion of mind and world, or thought; often a captured, crafted observation, stated with a sense of sound or lyric. Can such occur as a result of the relationship between the software/algorithm, the interface, and the reader/participant? If so, perhaps a refined program could endlessly create unique transformative poems. Can, at one moment, the materials be-like a dormant or unknown thought-one thing (i.e., a set of words in a database), and at another they are something else-words shaped into a conceptually infinite poem? Production of serial texts in this manner, mutations and manipulations of the language of a database, opens the possibility of a continuous perpetuation of language and ideas. Emerging during a period when poets, critics, and others newly explore the relation of language to the world, this form of expression pays particular attention to language as a system with variable properties. When we encounter a computer poem, we see a representation of our highly technological world, but couldn't there be more than that? Within the myriad types of expression, artists working in this often seek to expose, and sometimes subvert, the various binary oppositions that support our dominant ways of thinking about writers, literature, and about communication in general.

Image by zen, courtesy of Creative Commons license.