January 2007

How It Works

Jon Trowbridge

When given a text, Gnoetry 0.2, an ongoing experiment in computational poetry, analyzes how words are used in an extant text and tries to discern patterns. It looks at which words appear next to one another, and how often various combinations of words appear. It is important to note that Gnoetry’s approach is purely statistical. The software does not contain any a priori knowledge about grammar, and the computer has no idea about parts of speech. All it knows, and all it cares about, is how the author of the original source text put words together. From the computer’s point of view, the text could just as well be in Chinese or Swedish or some Ancient Mayan dialect.

Gnoetry0.2 also allows for the human end-user to facilitate “conversations” between disparate authors and epochs; a conversation enhanced by Gnoetry’s ability to statistically weight the texts during composition. That is, the end-user may “ask” that 23% of the time, solutions to the problem of “haiku,” for example, be found in Emma; 21.7% in The Custom of the Country; and so on up to 100%. This function allows the “voices” of the texts to be raised and lowered throughout the composition, much like a do-wop group trading solos and singing in different harmonies.

The texts statistically analyzed by the software for the gnoems reprinted as companions to this essay are The Custom of the Country by Edith Wharton; Emma by Jane Austen; Eighty Years And More: Reminiscences 1815-1897 by Elizabeth Cady Stanton; Sex and Common-Sense by A. Maude Royden (1876–1956, English preacher and social worker); Wuthering Heights by Emily Bronte; Up from Slavery by Booker T. Washington; and The Hotel Boy by Horatio Alger, Jr. (Thus representing both 19th- and early-20th-century texts, all of them about race, class and/or the sexes, and so all of them about power). Each gnoem consists of five tercets, using Gnoetry0.2’s nonce syllabic form function: the end user, Eric Elshtain, requested that every line consist of between nine and 15 syllables.

Once the computer has assembled all of the necessary information about how words appear relative to one another in the source texts, it can use that information randomly to generate new sequences of words subject to the rules and patterns that it observed. The frequency with which a word appears in the original text is an important part of the model. When generating these new sequences of words, the computer will use common words more often than rare ones, and likewise will use common combinations of words more often than rare combinations. If the language that is produced in this way is grammatical, it is only a happy accident resulting from the computer’s meticulous aping of the original authors’ patterns of word choices.

For example: the computer will never put three verbs in a row. Not because it has been explicitly programmed to avoid doing so, but because three verbs in a row is usually not a word-combination that will appear in the original text. And if this construction never appears in the original text, there will be no potential for that sort of a construction in the statistical model Gnoetry builds up—it will be assigned a probability of zero, and thus will never appear when generating the random stream of language.

Once this mechanism for generating new language is devised, the final step is to take into account the extra constraints imposed by the type of poem the user is trying to produce. So not only do users choose words subject to the original author’s usage patterns, they further limit themselves to words that fit the desired poetic form and meter.

The Implications

Eric Elshtain

Because Gnoetry is a machine/human poetic collaboration that uses extant texts as the word-level “alphabet” for its linguistic output, questions of authorship naturally arise. That is: who is exactly writing these poems? Can we even say that the computer is “writing”? If we take writing to be the ordering of words into meaningful, that is, decipherable, strings, then yes, the computer is writing. Especially in light of the human production of arguably “meaningless” or “selfless” texts (as in the work of some Modernists, Dadaists, Surrealists, Oulipeans and L=A=N=G=U=A=G=E poets) we have to claim the computer at least as co-author, though without the usual intents we ascribe to a human author, even though Gnoetry might mimic in certain respects the human poetic process since no human-derived poem is ever written in complete psychic isolation. Any poem is the product of the re-arrangement of pre-extant human utterances; in short, all poets are always borrowing from and differentiating from other humans’ texts and other humans’ speech, and Gnoetry is no different.

However: Gnoetry produces poetic language that is at a remove from its typical human sources: pre-conscious governance, psycho-historical flux, conscious-mind narration. The human mind automatically imposes a semi-discrete, hierarchical order onto the world, and we have to regard everything we think as an aspect of our ability to self-reflect. Thus, the only way towards linguistic and poetic novelty is to remove self-reflection from the primary compositional process via a non-thinking machine. So we must consider Gnoetry as an author or co-author of a different order – one without an intent or an unconscious—one without a meaning or un-meaning to convey – one that is truly and fantastically free within its constraints.

Questions of authorship will invariably lead to questions about meaning, be it semantic or semiotic meaning. If there is no authorial “self” to be relied upon when reading a poem composed with Gnoetry 0.2, then how is a reader supposed to “get” the poem? If the majority of the language of the poem is the output of the absolutely self-less computer then, ultimately, the reader is the final determiner of what meaning there might be. Because the gnoetic process can bank on the survival of grammatical and syntactical integrity, the reader’s job is made easier. The reader will have these linguistic markers, and even gestalt narrative markers such as well-known fictional place names (e.g. Baker Street), or well-known characters’ names (e.g. Doctor Moreau), to help her decide what any given gnoetic utterance might signify.

That said, poetry written with Gnoetry 0.2 is a truly open-ended poetry; not ambiguous in meaning, but open-ended with meaning. Ambiguity always leads to passivity: “Well,” a reader will say, “it can mean anything, so why bother with it?” An open-ended poetry is a poetry that demands activity on the part of the reader: humans can write open-ended poetry, but Gnoetry is always and unintentionally open-ended, because the majority of the language is “written” by a machine. Now a reader will say something like, “Well, it means everything, so I have to choose what I want it to mean,” because language is moved away from being subject to human cognizance where language can reach a meaning almost entirely outside of scripted cultural discourse – remember, Gnoetry does not know what language it is speaking. Gnoetry unwraps words from their usual human contexts until words no longer have direct instrumental relationships to the world—in fact, the very question of instrumentality is rendered moot.[1] The significance of meaning is utterly altered insomuch as each poem consists of a string of purely grammatical (or slightly un-grammatical) propositions and the reader ultimately decides what they are propositions of and/or pataphysical solutions to.[2]

Poetic language is given its own epistemic dignity. The human author is no longer accountable for meaning, or thinking about meaning, since Gnoetry renders language into almost pure technique. “Diction itself is its own image” (Forrest-Thompson 109) and Gnoetry renders language into ‘meaningless’ diction: it is the ultimate beautiful automation of poetry. The reader is “[forced]...to note that formal pattern and conventional disposition of phrases are more important than ‘ideas’ or ‘themes’ or ‘message,’ more important even than ‘reality’” and thus, through Gnoetry, the reader is “re-initiat[ed] into the rites of mediation” via poetic language speaking as much as possible through pure artifice (Forrest-Thompson 109).

The reader, then, is put in a unique position in relation to a poem composed with Gnoetry; she need not bother with wondering what is “machine” and what is “human,” since there can be no such distinguishing. Thus free from the usual limits and habits of signification – free from having to conclude that the poem contains a human message or bot-nonsense, the reader becomes the final author, the final collaborator – hers is the mind that makes meaningful, because the “cyborgian” poet that preceded her interaction with the language of the poem prefigured no such meaning.

Because of the neutrality gnoetic production produces (the question “who is writing” is meaningless in the face of gnoetry), the fact that any data can be placed into its algorithm, the gender, for example, of the writer is rendered non-representative; authority must come from something else. The past is not corrected – for example, the feminism inherited from, say, a Stanton text does not “correct for” the patriarchy affiliated with a Milton text. The texts themselves are rendered into statistical equals – whether or not that means political or linguistic equals, we leave up to the reader. Since the statistical process works on the word level, the software disassociates words from their valued context (even though the words come from texts that contain or are written with historical, cultural, racial and gender biases), and this radical decontextualization brings the words near a zero-degree of the values we typically attribute to language: that is, all the semantic and semiotic baggage of an utterance is unhinged when it has been processed by the software.[3]

The software statistically analyzes the texts based upon word adjacencies—that is, according to what words tend to be next to other words. For example, if I’ve created a poem using the statistical analysis of a dozen texts and the sentence “It’s a disappointment to my purpose, woman” appears, the reader has no idea which words came from which texts and also has to choose how to read the line. Is the “speaker” addressing a woman and mentioning a disappointment to her, or is “woman” herself the disappointment, as in “Woman is a disappointment to my purpose,” with its post-lapsarian undertones? Did “woman” come from that Nietzsche text? From the A. Maud Royden text? Was is originally used in a sentence denouncing women? Calling for their equality? The signifier has been radically unhinged from its original signified and it is up to the reader to supply a context for the poetic utterances, as there is no single or singular self/author that can be unwrapped from the gnoetic text.

Gnoetry—and other computer-generated poetry programs—allow for the liberation of a poem from authorial intent and hence gives the poem an integrity that has been hinted at but never fulfilled. Gnoetry does this not by killing the author, but by multiplying the author. While the end-user may exert her control over the gnoem by regenerating text to create, perhaps, a cogent narrative, since the building blocks she uses are not subject to her subjectivity the final product ultimately eludes her efforts to make the text mean what she intends. In this regard, Gnoetry is the fulfillment of some aspects of New Criticism and reader-response theory in that the reader is the ultimate authority of what a gnoetic text signifies. The reader is ontologically de-natured from a typical lyric understanding of poetic process through gnoetry because the end-user is as well, given her use of the machine as a prosthetic mind and hand and breath; hence, the reader is not asked, and indeed can not be asked, to fill in indeterminacies to complete the harmony of an organic whole. She is asked to leave indeterminacy in place and willfully to seek meaning elsewhere: to seek meaning in the poem itself, not worrying about finding connections to the “outside world” or seeing through the language to authorial intention or psychology. Simply put, there is no meaning behind the words—we have only the meaning of the words.

In this regard, Gnoetry can be said to lift poetic language out of the psycho-drama of the lyric, but not to do away with the echoes of the Bacchanalia, because the “voice” of Gnoetry is wild, creating a “continuity between inside and outside” which is, according to poet and Classicist Anne Carson, “abhorrent to the male nature” of purifying order (129). One paradox of the program is that while on the one hand it is an arbiter of order via the code and statistical analysis, on the other hand what it produces via the building blocks of poetic utterance is a purifying disorder. Through Gnoetry’s multiple orders of constraints, paradoxically, the poet, via the software, is free to help compose something like

In this unnerved, in this unnerved, in this
unnerved, in this unnerved, in silence, through
the pauses of the vault! In this unnerved,
in this unnerved, in this unnerved, in this

unnerved, in this unnerved, in words. The door!
The brother, and a victim to the base.
The brother, but a harmless, and the walls.
The storm. In this unnerved, in this unnerved,

in silence, through the pauses of the earth.
The brother, but in their results. The storm,
the rending of her heart? The vault! The door!
The storm, the hero of the veil. The house.[4]
(March 20, 2004, 21:32:36)

(Gnoetry0.2, et al. 17)

which is a kind-of ecstatic breathless ode to the moment of becoming unhinged--where the adjective phrase “this unnerved” can not find one thing to modify and hence becomes a place in and of itself ‘prepositioned’ “in silence” and “in words” because why: her heart has been rent by the storm? the hero is behind a veil? Is the brother a result of the “pauses of the earth” while the “she” of the poem metaphorically becomes a heroic storm, generating the only noise in the unnerving silence? The above gnoem is an open system of breathless utterance—one can imagine it panted out, noun phrase by noun phrase, prepositional phrase by prepositional phrase, in a kind of pin-point aphasia that cannot stand to say the verbs, the action too horrible to say.

As the software performs its analysis on the word level, each word is transplanted from one system of utterance to another, e.g., from novel or short story to poetic form. The poetic form, like iambic pentameter, acts as a representation of poetic breath, but since each transplanted word fully participates in the meaningful (or un-meaningful) utterance it represents, a word in a gnoem is inspired into the gnoem. It is filled with potential breath, as we read in the above poem—it is filled with breath and breathlessness, through the simulacra of breath afforded by the meter and by the fact of breath breathed into it by the reader, whoever she may be. In this way, too, Gnoetry forces us to reconsider the authorial origin(s) of a poem.

By all means, human poets—like the L=A=N=G=U=A=G=E poets—have consciously disrupted the chains of conventional signification; have questioned the stability of the self and of authorship; have removed speech acts from the contexts in which they “naturally” exist. But, though L=A=N=G=U=A=G=E Poetry is a critical precursor to appreciating Gnoetry’s fractured poetry as legitimate poetry, the L=A=N=G=U=A=G=E authors were still invested in a conservative notion of the Self; that is, there needed to be coherent political and social meaning behind all the linguistic unmeaning, and therefore there needed to be a grounded, egotistical and organic self deciding how to make the poem mean through its un-meaning. Isn’t it curious that the foremost critical allies of the L=A=N=G=U=A=G=E Movement often “decode” plots and narratives from the poems? That the foremost defenders of experimental verse do so by making claims for its reference to “the world”? See, for example, anything that Marjorie Perloff has written. Here’s a single example: After making much of Lyn Hejinian’s claim that there is no “core reality at the heart of our sense of being” and that a poem is not uttered from “an identity which is unique and seperable from all other identities” (Hejinian qtd. in Perloff, 213), Perloff states that Hejinian’s poetic utterances “represent her world” (216, italics hers). Even though there may be no authorial “‘self” at center stage to which all else is “subordinated” (Perloff 217), there is still a construction of the world taking place by someone who is in the world. Poetic language still pays, it seems, lip service to the self and to reality even if the “Self” and “Reality” are “constructed.” L=A=N=G=U=A=G=E Poetry must make reference to Reality in order to undermine it and call it merely “a” reality; as Perloff puts it, the “task of the reader is to construct a context that might make sense of [the] riddling references” (215) of a poetry like Ron Silliman’s, Rosmarie Waldrop’s or Lyn Hejinian’s. For Perloff, these “references” must refer to an outside world, and hence, must be referents produced by a self-in-the-world who has a one-to-one correspondence in mind that the reader will decode as if the poem were a crime scene and the reader a detective. On the other hand, a gnoem is the detective investigating the reader.

Unlike the political poetics of the L=A=N=G=U=A=G=E poets, however, we would also hesitate to say that the computer could make a political statement, and so whatever political or historical meaning to be gleaned from the sentence/utterance must be supplied by the reader’s spot of time, not the authors’.[5] The texts are not so much “updated” into the present moment as they are recombined into a new historical possibility: gnoems don’t say of the texts they’re derived from, “Boy, this text can still mean, even now!” which prioritizes inception, but, “This time [or any part of the intervening time between the inception of the source text and its statistical analysis by the Gnoetry program] can be meaningful and relevant to the texts.”

The latter statement highlights the efforts of the historically situated reader, not any effort to find present relevancy from within or taken out of the texts themselves. Take this excerpt from “[A few decades ago, because the whole...]”:

...We were not a lady who
escorted us, sometimes alone, sometimes
alone, sometimes alone, sometimes alone,
sometimes together and the nobles held
the lands. The coloured people who were not
the primal step in Tuskegee, in mind.
The news. The papers, to protect the year

were not a preacher, child, the mother is,
perhaps in compensation for the child.
Toward the white. The value of the white...

...Through the winter, not

the white. Besides, in order to secure
the civil war in this direction, is
in fact, the slave. The people had addressed
the men, in this description of the most
effective writing. I repeat, the whole
republic was a mere commercial gain.
[Tuesday, January 11, 2005, 15:41:20]

(Elshtain and Gnoetry0.2 10)

The language corpus for this gnoem was comprised of the statistical analysis of Elizabeth Cady Stanton’s Eighty Years And More; Reminiscences 1815-1897 and Booker T. Washington’s Up From Slavery. The continued historical relevance of these texts is not to be disputed—it doesn’t take a poem, written by human or machine, to keep these texts meaningful within the context of America’s political and social history regarding race and gender. The software dislocates words from their linguistic and hence historical location in the text, by which I mean the history of where and when the word was placed into the text by the author—every sentence houses a history in its grammar as they “[elaborate] an implicit order” and each sentence “carries within it the blueprint for the whole subsequent work” (Dewdney 73) and hence work teleologically. Out of the hands of the author and rendered into chance operations, each word and software-created triadic language token (see appendix) breed a unique utterance, a singular new thing for the text to say. By placing these words into a new poetic and historical context the words are made literally to time-travel[6] as the words are written into the “now” in order for the final arbiter of meaning—the reader—to create a meaningful context in which the gnoem may be understood. As each reader is her own individual, reading the gnoem at a different moment in time, a new historical context is written in the gnoem each time it is read. The text is ever-evolving (remembering that evolution is a theory of change through time, not a theory of progress), and the words from “old” texts are given new currency because each time they are made to mean for the text, not the text to mean in the time.

The “use” of the phrase “coloured people” and the “allusion” to “Tuskegee” give us pause. The phrase “coloured people” does, indeed, occur many times in Washington’s autobiography, and since we cannot (and do not want to) say that the gnoem quotes the phrase (at least in the usual way we understand “quoting” or “citing”), what possible semantic or semiotic currency is being spent in the gnoem with that phrase? We note that it is “used in reference” (again, the scare quotes are meant to belie any sense of intent) to “the primal step in Tuskegee.” We realize that this might be a “reference” to the tragic and criminal syphilis experiments on close to 400 African-American men between 1932 and 1972 conducted by the US Public Health Service, for the men were not infected with syphilis; they were closely watched as they died from the disease. Washington’s text was published in 1901, Stanton’s in 1898. Some aspect of the statistics of word-adjacencies in Washington’s autobiography, in conjunction with the same analysis of Stanton’s autobiography, “predicts” something that occurred in 1932, a “future,” through an analysis done in 2005 which becomes “past.” The time of gnoetic composition is made concurrent with the pasts of the texts used--not by an act of authorial superimposition or deconstruction, but because of the very non-linear nature of space-time. Just as the Gnoetry software disarticulates our notion of a single consciousness authoring a poem, the software also “rewrites” history by articulating past texts into the present. It is as if the Stanton and Washington texts were being written right now at the same time as having been written when the authors sat down to write them. A gnoem is a multi-temporal, multi-self “spot of time” through which we can understand the development of selves within histories. For example, the “primal step...in mind” is also (or is instead, depending on who’s reading the gnoem) Washington’s founding of the Tuskegee Normal and Industrial Institute in 1881. The history that is developed in the gnoem acts according to the historical moment that the reader places into the gnoem. In turn, the reader’s own development may be measured by the spot of time[7] in which the gnoem is read—the reader authors the poem’s meaning(s) as she imposes her history into the poem and makes it relevant to the connotative language and gestalt markers in the gnoem.[8] The reader exists within her own present moment and the historical moment of the text simultaneously. The texts, whether fictional or non-fictional, don’t just seem as if they speak things relevant to the present—they become the very meaning of the present moment because they fully participate in the reality they represent through the meaning-making of the reader.

By loosening texts from the grip of the fiction of an autonomous, coherent subject (which is a conception of absolute authority), Gnoetry rejects, among other things, the privileged position of gender. Words and phrases that might be considered “gendered” in a particular way become truly autonomous and can be ascribed values apart from any given societal standard. There is no “either/or” in a gnoetic text. Each gnoem is constructed under the rubric of “but/and;” the paradox here is that by mitigating the human element in poetic production, Gnoetry creates poems that are more truly Human.

[1] Gnoetry acts as a proof to Wittgenstein’s dictum that “a poem, even though it is composed of the language of information is not used in the language-game of giving information” (Zettel §160).
[2] Alfred Jarry: “Pataphysics is the science of imaginary solutions, which symbolically attributes the properties of objects, described by their virtuality, to their lineaments” (22).
[3] The question of assumptions behind the code itself the authors believe to be moot: the code “asks” nothing more or less of language than do poets across cultural, temporal and gendered borders. Like a mathematical equation, the code can be used to create a rational linguistic order or to create an irrational linguistic order. In this way, Gnoetry is as “masculine” or a “feminine” as the human end-user is “masculine” and/or “feminine.”
[4] This gnoem is based upon the statistical analysis of Edgar Allen Poe’s The Fall of the House of Usher and “The Masque of the Red Death.” Any number of texts can be used as sources for the production of ‘gnoems.’ One advantage to using few texts is that the “voices” of the authors of the source texts can show through the statistical analysis; three, nine, twenty texts leads to poetry that creates and does not “borrow” “voice.”
[5] We, of course, do not want to claim that a gnoem can mean anything to anybody; the very strictures of syntax and grammar, much of which survives the stochastic process of the program, belie this idea (which is, indeed, bogus from the outset, since it refuses all sense of materiality; even an abstract painting is limited in what it can mean because it is made from paint; media has meaning, too). What’s extra slippery about Gnoetry, though, is that it doesn’t matter what language is used in the program: “From the computer’s point of view, the text could just as well be in Chinese or Swedish or some Ancient Mayan dialect” (see Trowbridge, “How It Works” above).
[6] Vide the preface to Her Social Frame, a chapbook available through Beard of Bees Press in which the human end user implies that he went back in time and asked Edith Wharton to write some renga.
[7] By which I mean a sort of transitional object that makes the self aware of a world external to the self.
[8] Another marker is the irregular blank verse; how would a formal gnoem fit into the history of a poetic form? At the least, the ticking out of words in time is here a hallucination of poetic intent; without the human intention (if we don’t count the code as intent) to fit words into form, is the poem part of a tradition of formal verse?

Carson, Anne. Glass, Irony, and God. New York: New Directions, 1995.

Dewdney, Christopher. Immaculate Perception Toronto: Anansi Press, 1986.

Elshtain, Eric and Gnoetry0.2. The Most Effective Dress. Online. Beard of Bees Press, 2005. http://www.beardofbees.com.

Forrest-Thompson, Veronica. Poetic Artifice: A Theory of Twentieth-Century Poetry.

Gnoetry0.2, et al. Field Test Two. Beard of Bees Press, 2004. http://www.beardofbees.com

Jarry, Alfred. Exploits and Opinions of Dr. Faustroll, Pataphysician. Trans. S.W. Taylor. Boston: Exact Change, 1996.

Perloff, Marjorie. Wittgenstein’s Ladder: Poetic Language and the Strangeness of the Ordinary. Chicago and London: The University of Chicago Press, 1996.

Wittgenstein, Ludwig. Zettel. Eds. G.E.M. Anscombe and G.H. von Wright. Trans. G.E.M. Anscombe. 2nd edition. Oxford: Blackwell, 1981.