Utente:Lydia Tuan/Generative Literature

Generative literature refers to literature that is completely or partially generated by an autonomous system, such as a computer program, that algorithmically produces generated literary texts. Closely linked to the field of generative art, generative literature is often seen as its subset, as both artistic forms rely on an autonomous system, recognized as a non-human entity that produces the literary text independently from a human author, for its literary production.

History

Generative art's increasing popularity in the late twentieth century was due, in part, to the computational possibilities offered via computers, which gave generative art a new platform. Art historian Grant D. Taylor notes that computer art’s introduction in 1963 sparked outrage, mostly from non-computer artists who feared that the written poem, representing “communication from a particular human being” and “one last refuge for human beings” would no longer serve that function in the computer age.^[1] Computer art was often seen as “another example of the vulgarization of science, where besotted artists, dallying with the latest scientific and technological media, produced what was tantamount to science as kitsch,” paralleling the fascination of computer art with modernist responses to the development of pure sciences in the early twentieth century.^[1] Prior to the mainstream acceptance of computer poetry as art in 1990s, people had hoped that machines would fail, having coveted art as a “refuge from the onslaughts of our whole machine civilization.”^[1] The stigma attached to computer art was voiced by artists such as Paul Brown, who lambasted the use of computers in art as the “kiss of death”^[1] to describe computer artists who were rejected from galleries once it was revealed to curators and directors that computers played a role in their work’s creation.

Jean-Pierre Balpe and surrealism

Unlike generative art, the introduction of generative literature did not receive such negativity. One of the first, most prominent uses of generative literature as a term can be traced to French generative writer and theorist Jean-Pierre Balpe, who in the mid-1970s, was inspired by surrealism, which fueled his exploration of automatic text generation’s artistic potential.  Balpe defines generative literature as “the production of texts that continually change since they are based on a specific dictionary, on a set of rules and the use of algorithms”^[2] and that understanding the complexities of generative literature requires awareness of its “niveaux d’engrammation” or different "levels of engrammation" that specify modes of communication between humans and machines behind the generativity.^[3] Balpe believes that all literature, to an extent, is generative.^[4]

Balpe spent the early 2000s working on several computer-generated novels online, including Fictions and Trajectoires (2001), including creating the poetry machine Babel Poésie (2004), which produced poems by generating French, Italian, and Spanish words. Poems from Babel Poésie cannot be generated more than once, and while the content of its poems has been described as “the poetry of trash language, word garbage, chaos speak,” the poems’ forms have been praised as “a new poetry which works with boundless text flow and is conceived as an associative and endless process.”^[5] According to Balpe, generative texts dismantle normative reading habits of temporally situating texts in relation to texts encountered earlier on the diegetic axis because “[t]he narrative is not totally built in advance but put together from a lot of virtualities which are — or are not — actualizing themselves in the course of reading.”^[6] In other words, readers will neither see the same texts presented to them a second time nor read the same the text as another reader.

Codework poetry

The idea that code can be read, analyzed, and written as literature is not unprecedented. Codework poetry, known as the construction and stylization of verse using a mixture of programming languages with natural languages to produce literature, is a literary treatment of data. Using programming languages like natural languages by giving them syntactical and semantic meanings produces a concrete poem-esque effect when juxtaposed together in the same context. Published anonymously in the networked discussion system Usenet, “Black Perl” (1990) serves as an example of a codework poem. Written in the programming language Perl (short for “Practical Extraction and Report Language”) as an example of Perl Poetry, “Black Perl” was intentionally written in valid Perl commands so that it could be understood by computer and human reading. The step-by-step commands listed in each line of the program transform into a narrated event when read line-by-line as a poem.  The code’s form, such as the inclusion the asterisks and parentheses, influences the readability of the code as a poem, as various punctuation marks serve different semantic purposes when read in Perl than in English, for example. However, “Black Perl” was intentionally written as a poem, meaning that this particular codework poem has more in common with practices of constraint writing than generative literature. In fact, “Black Perl” is not generative for the reason that it is not program-generated output but, is, instead, the program itself. The usefulness of this poem, however, is to demonstrate the duality of human and computer readability in “Black Perl” and how programming languages are not completely devoid of literary value.  

Controversy

Despite the loose parameters for what qualifies as art today, the debatable literary status of algorithmic outputs has been an ongoing contention even amongst new media artists.  Digital technology theorist Yuk Hui called algorithmic outputs “algorithmic catastrophes” rather than anything worth studying at all, defining outputs, or “the product of automated algorithms,” as “the failure of reason,” not even “material failure.”^[7] Portuguese experimental poet Rui Torres, whose corpus of creative works includes presenting poetry in hypermedia contexts, asserted, while fielding questions after a talk delivered at the University of California, Berkeley in April 2016,^[8] that algorithmic outputs can never transpierce the literary realm, thus barring algorithmic outputs as literature and siding with Hui’s idea that algorithmic behaviors suggest a “failure of reason.”

To address this skepticism maintained by Hui and Torres, as well as other generative art skeptics — that algorithmic output cannot qualify as art — inevitably attracts past debates on art’s definition that have been hashed and re-hashed out since the emergence of the avant-garde.  According to generative artist Philip Galanter, the oft-discussed question of “What is art?” within art theory does not go unnoticed when formulating generative art theory.  If art is to be understood as a product of expression, then generative art, Galanter notes, faces another obstacle, namely, the frequently encountered question within artificial intelligence discourse: “Can it be claimed that a computer can and will express itself? Alternatively, when the computer determines forms not anticipated by the artist, does its creation still qualify as the artist’s expression?”^[9]

Examples of generative literature

Raymond Kurzweil's "Cybernetic Poet"

First introduced sometime in the mid-1980s, Raymond Kurzweil’s Cybernetic Poet is an online program that generates poetry by reading an extensive collection of poems written by human authors. On his website, entitled “CyberArt Technologies,”^[10] Kurzweil introduces the Cybernetic Poet’s functionalities in greater detail:

RKCP [the Cybernetic Poet] uses a recursive poetry-generation algorithm to achieve the language style, rhythm patterns, and poem structure of the original authors whose poems were analyzed. There are also algorithms to maintain thematic consistency through the poem.  The poems are in a similar style to the author(s) originally analyzed but are completely original new poetry.  The system even has rules to discourage itself from plagiarizing.[2]

The Poet’s ability to produce original poetry by reading, first, an extensive selection of poems by one or several authors mimics a writing process that could very well be practiced by human poets — especially if we recall that novel literary forms and styles often emerge from the influence and desire to depart from current and preceding literary movements. Kurzweil has seemingly programmed the Cybernetic Poet to function like a human author, as its abilities to “maintain thematic consistency through the poem” and “discourage itself from plagiarizing” all suggest the development of an authorial personality.  Functioning as a “poet’s assistant authoring tool,” the Cybernetic Poet aids human authors by “assist[ing] and stimulat[ing] a (human) poet in finding the right verbal images and phrases,” which, Kurzweil notes, “are often intriguing and surprising.”[3]  The Cybernetic Poet’s participation aids the human author in a way that could potentially contribute to the authoring process as a co-author, even though it does not replace the role of human authors by writing for them, as they could always reject the bots’ suggestions. 

[1] “Kurzweil CyberArt Technologies.”

[2] Kurzweil (1999), 163.

[3] Kurzweil, “The Cybernetic Poet.”

William Chamberlain and Thomas Etter's "Racter"

In spite of its popularity, the Cybernetic Poet was not the only poetry generator from the mid-1980s. William Chamberlain and Thomas Etter’s Racter, whose namesake derives from raconteur, is a software written in the programming language BASIC that generates prose on an IMS (Information Management System) computer without prompts from a human operator.  A collection of Racter’s early fiction was published in a book entitled, The Policeman’s Beard is Half Constructed (1984), and aside from spelling mistakes corrected by Chamberlain himself, the text is completely computer-generated.  Racter writes from a database containing 2,400 words to match nouns with contextually appropriate adjectives, and it ensures continuity by tracking used phrases,[1] allowing the book to have some form of cohesion that we might call a narrative (even though there are human-drawn sketches that serve as visual aids that potentially contribute to this cohesion). Racter’s choice of words is completely random, producing senseless text that literary critic Jack Barley McGraw calls “empty text” resembling “Dadaist nonsense” that cannot be close read.  Any attempt at close reading Racter’s “disturbingly superficial” prose, according to McGraw, would be a futile exercise in “conceptual justification (seemingly out of thin air) for vaguely related strings of words.”[2]

[1] Simanowski, 96-97.

[2] Ibid. Quotes are Simanowski’s quotation of McGraw.

Prefacing the book, Chamberlain writes that Racter’s goal is to “replicate human thinking” — or, in other words, represent a utopian actualization of the vision that certain people had for computers during the mid-1980s, precisely that computers were “designed to accomplish in seconds (or microseconds) what humans would require years or centuries of concerted calculation effort to achieve,” and, in some cases, were absolutely needed, as certain tasks could not be accomplished without the use or assistance of a computer.[1]  Chamberlain’s description of Racter parallels Kurzweil’s Cybernetic Poet in the sense that the goal of both poetry generators is to make creative choices that a human might make.  This commonality between Racter and the Cybernetic Poet not only reveals the utopian appeal of computers that some people held during the late twentieth century, but also reveals the hope that people had for computers to be their friendly, helpful companions rather than representative extensions of themselves that may threaten the role of humans in the creation of humanity, as now discussed in many posthumanist discourses of the twenty-first century.[2] 

[1] Chamberlain, The Policeman’s Beard is Half Constructed, Introduction (no pagination).

[2] For literature on computers representing extensions of selfhood, refer to Sherry Turkle’s writings on the second self.

Nick Montfort's #!

More recent examples of generative literature include Nick Montfort’s book of computational poems, entitled #! (2014) but pronounced ‘she-bang,’ which, from the dictionary, means “the set of all circumstances.” Published thirty years after Racter and Kurzweil’s Cybernetic Poet, #! contains poems written by algorithms and the algorithms that generated the poems. The book is divided into sections; each section begins with the algorithm, followed by its output on the subsequent pages.  Some of the outputs end with ellipses to signify that they could not be printed due to their infinite length.  #! is also ambiguous in its intended readership, as its title, for example, is a valid Python command; the placement of a hashtag before any given text commands the computer not to read any text following the hashtag.  In this interpretation, “#!” conveys surprise — a reaction that summarizes the general sentiment when attempting to read or make sense of this book.

In a review of Montfort’s #!, Cayley writes that even though the programs are meant to read by the program producing the output, but the inclusion of both program and output in #! makes the code “a (constitutive) facet of the poem.  It is (also) the text.”[1]  In this way, both the code and its output become the text — but only if they are considered as such in relation to each other.  The effect of sharing the source code, according to Galanter, not only further creates confusion as to whether the source code is the text but also allows other artists to create variations of the output, which “breaks with the paradigm of the heroic single artist creating a ‘fixed’ masterpiece.”[2]  

[1] Ibid.

[2] Galanter (2016), 171.

Computational sublime

The “computational sublime” addresses this fascination voiced by generative art critics and generative artists respectively, that code could be programmed to produce writing that may have discernable meaning and make sense.  Termed by digital poets and critics Jon McCormack and Alan Dorin in 2009, the computational sublime borrows from the notion of sublime established in the eighteenth and nineteenth centuries: the fear of being unable to experience or quantify the totality of all that exists to be experienced in the world while also feeling overwhelmed (but potentially in a pleasurable way) while acknowledging this fact.  Per the authors’ formulation, the computational sublime is:

the instilling of simultaneous feelings of pleasure and fear in the viewer of a process realized in a computing machine.  A duality in that even though we cannot comprehend the process directly, we can experience it through the machine — hence we are forced to relinquish control. It is possible to realize processes of this kind in the computer due to the speed and scale of its internal mechanism, and because its operations occur at a rate and in a space vastly different to the realm of our direct perceptual experience.[1]

The feeling of dual pleasure and fear that programs can produce text that, from a human perspective, could pass as human-authored is exacerbated by an awareness of a computing machine potentially becoming or, at least, approaching the status of a creative equal to humans. Furthermore, the loss of control allows humans to experience the work through the machine and through an understanding of the machine’s computational abilities rather than engaging with the output directly.  The feeling of being overwhelmed by the recognition that machine operations “occur at a rate and in space vastly different to the realm of our direct perceptual experience” draws a concern echoed by generative and computer art critics, namely the possibility that computer programs, generating surprising and unexpected output, could either amount to or supersede the human capacity for literary production.

McCormack and Dorin’s computational sublime echoes generative artist Marius Watz’s notion of “genuine surprise,” defined as “a temporary loss of subjectivity, as a relinquishment of one’s subjective intention, either to another’s control or to objective forces beyond one’s control.”[2]  According to Watz, the experience of surprise is key to generative art, and one way of aesthetically judging generative art might depend on the work’s ability to induce surprise.  It is, however, arguably anxiety-inducing when program-generated outputs and human-authored literature cannot be confidently differentiated, as such was experienced in new media artists Daniel C. Howe and Braxton Soderman’s undergraduate digital writing workshops at Brown University from 2007 to 2008.  Having discussed Watz’s genuine surprise with their students, who created and analyzed generative literature, Howe and Soderman reported that program-generated texts often prompted students’ anxieties about the texts’ meaning and authorship, supported by their fear that the computer might even have an “individuality.”  As the article will later discuss, the problem of authorship is always central to debates regarding generative literature, providing interesting perspectives on questions of authorship that have been central to criticisms of existing literary forms.

[1] McCormack and Dorin, 78.

[2] Howe and Soderman, “The Aesthetics of Generative Literature: Lessons from a Digital Writing Workshop.”

References

1. (EN) Grant D. Taylor, When the Machine Made Art: The Troubled History of Computer Art, a cura di Francisco J. Ricardo, collana International Texts in Critical Media Aesthetics, vol. 8, New York, Bloomsbury, 2014, pp. 5-6.
2. ^ (EN) Peter Gendola and Jörgen Schäfer (a cura di), The Aesthetics of Net Literature: Writing, Reading and Playing in the Programmable Media, Bielefeld, Transcript Verlag, 2007, p. 13.
3. ^ (EN) Peter Gendolla and Jörgen Schäfer (a cura di), The Aesthetics of Net Literature: Writing, Reading and Playing in the Programmable Media, Bielefeld, 2007, p. 25.
4. ^ (FR) Jean-Pierre Balpe, Fiction et écriture générative (PDF).
5. ^ (EN) P0es1s.digitale Poesie, su p0es1s.net. URL consultato il 12 giugno 2016 (archiviato dall'url originale il 24 maggio 2019).
6. ^ (EN) Jean-Pierre Balpe, Jean-Pierre Balpe: Principles and Processes of Generative Literature, su dichtung-digital.de, 2005 (archiviato dall'url originale il 24 maggio 2019).
7. ^ (EN) Yuk Hui, Algorithmic Catastrophe—The Revenge of Contingency (PDF), in Parrhesia, vol. 23, 2015, p. 123.
8. ^   (EN) Rui Torres, Rui Torres – Unlocking the Secret Garden: Electronic Literature from Portugal, Institute of European Studies, UC Berkeley, 4 maggio 2016. URL consultato il 24 maggio 2019.
9. ^ (EN) Philip Galanter, Generative Art Theory, in Christiane Paul (a cura di), A Companion to Digital Art, 1ª ed., John Wiley & Sons, Inc., 2016, p. 169.
10. ^ (EN) Raymond Kurzweil, Kurzweil CyberArt Technologies Home Page, su kurzweilcyberart.com (archiviato il 24 maggio 2019).

Further reading