Treating the Internet as a database for generating new writings, the series of works titled The Readers Project (2009-2015), by John Cayley and Daniel C. Howe, draws attention to the politics of language contained in the electronic infrastructure and in its algorithmic processes for datafying our uses of language. In turn, PoemAds (2012), by Rui Torres, uses 22 advertisement slogans as syntactic templates for building a poetic engine. Both works explore the generative properties of language and the networked processability of the digital medium to interrogate the discursive and material conditions of meaning production in our pervasive algorithmic media culture. As examples of database poetics, The Readers Project and PoemAds offer powerful reflections on language as a mode of social action and on the politics of language processing in the age of big data.
electronic literature; programmed readers; textual generators; database poetics; “détournement”.
1. Writing (with) the Internet
Recent scholarship on electronic literature has tried to account for the Internet itself as a dataspace for literary interventions. 1 Through analysis of works that make use of the network as a means of poetical production, these critical discourses have directed our attention to the ways in which the Internet has become constitutive of our linguistic and social practices (see, for instance, Hayles 2012; Funkhouser 2012; Portela 2013; Emerson 2014; Pressman 2014; and Baldwin 2015). Some of the digital creations analyzed in those critical works address the new conditions of production of robotic writing at two levels: first, by adopting algorithms, protocols, resources and network interfaces as part of their material poetics; secondly, by undermining the communicational transparency of digital media, directing attention to source-code, to the specific rhetoric of software tools and to the economic and political relations built into the technology. 2
Lori Emerson’s Reading Writing Interfaces, for example, is concerned with the awareness of the interface in various writing processes and across media. Its archaeological approach to media poetics is crucial for its deconstruction of the invisibility of the interface and of the multi-touch “user-friendly” ideology that has become dominant in our current forms of computer-mediated communication and in the engineering paradigms of human-computer interaction. Through literary analysis of a series of works that include concrete typewriter poetry, Emily Dickinson’s handwritten fascicles, and several works of electronic literature – from bpNichol’s First Screening to Cayley and Howe’s algorithmic harvesting of network language –, Emerson shows the material friction of the interface as a necessary condition for expression. The retroaction suggested in “from the digital to the bookbound” offers a dynamic and layered sense of intermediation that contributes to a critical redefinition of both the “electronic” and the “interface”.
Of particular significance for the programmed works that are the subject of this article is The Internet Unconscious, by Sandy Baldwin. Baldwin articulates a literary reading of the Internet as an ungraspable writing and reading space, describing the Internet itself as a distributed ensemble of practices (from e-mail to social networks to ASCII to CAPTCHA to spam) that are constitutive of both the net as “electronic literature” and the subject of the Internet. This attempt at naming the Internet unconscious shifts our focus of attention from a strictly functional or aesthetical account of writing and reading processes taking place in and through digital media, to a self-reflexive examination of the Internet as not so much a writing space as a new form of writing, that is, as a series of difficult-to-imagine writing protocols projecting our writing acts and our bodies in the “beyondness” of the screen and its graphical user interface. Self-projection into net-writing thus becomes the introjection of its apparatus, or, in other words, the unconscious condition of our telecommunications situation. Rhetorically foregrounding the writing self by playing with the split between writing self and written self – the recurrent “I” of its text performatively enacted as a self-conscious embodiment of the writing self of net writing –The Internet Uncounscious is also a self-exemplary instantiation of the power of discourse: “Let me state: I write away, as in wear away, one drifting piece of our inconceivable inhabitation of the net. Writing away through myself and toward myself and in lieu of myself. Writing as weapon and spew and nothing.” (107)
It is as if they were mechanisms for observing this “writing with” and “being written by” the algorithms and automatic and unconscious processes of the network that The Readers Project and PoemAds become relevant, since they allow us to rethink generative and database poetics as an expression of the condition of human language in the age of automatic processing. Algorithmic literary creation becomes centred on the growing datafication of everyday uses of language, which privatizes the word and turns it into a device for surveillance and control, but also on the possibility of revealing language structures as modes of action through automation. If, in case of The Readers The Project, the generation of new textual instances is made on the basis of formalized reading protocols that re-associate elements of particular of textual corpora (which can extend from self-contained texts to open-ended n-gram searches over the endless textual database of the Internet); in the case of PoemAds, the random generation of new textual instances on the basis of a closed set of permutations dramatically re-signifies the phrases and sentences away from the original discursive contexts, giving readers the possibility of human reinscription in the midst of robotic writing.
2. Reading (with) the Internet
Writing only exists as it is read; or as function of its virtual, potential, and intermittent readability; or as function of memory, which is simply a special type of transcription within human readers. (Cayley 2013: n.p.)
John Cayley and Daniel C. Howe’s The Readers Project (2009-2015) is the general title of a series of programmed conceptual and procedural works concerned with the dynamics of writing and reading within networked databases. 3 The act of finding the text in the database is their way of locating the literary on both sides of the literary performance: as an act of writing and as an act of reading. In fact, the interweaving of writing and reading that takes place in this series of works, performances and installations suggests that the literary may be a form of reading that writes and a form of writing that reads. The reflexiveness characteristic of literary practice could then be described as the algorithmic feedback loop between reading writing and writing reading (Cayley 2011; Howe & Cayley, 2011).
The title of this article – “scripts for infinite readings” – could be rephrased as “scripts for infinite writings”, in so far as each reading is iterated as new act of writing. It is in this force field of transactions and interactions that we invent ourselves as human subjects through literary performances, now encompassing the ubiquitous and robotic processing of natural language. Yet, what is most significant about Cayley and Howe’s programmed interventions on the databases of language is not the permutational and poetic productivity of their algorithms per se, but their political and cognitive implications, one of which is the awareness of the algorithmic nature of writing and reading practices in networked programmable media. In their latest works, their characteristic algorithmic writing-reading and reading-writing reflexiveness has been programmatically developed as an aesthetical and critical intervention on the Internet itself as writing-reading infrastructure (Howe & Cayley, 2013; Cayley 2013).
The ongoing massive digitization of writing that is taking place within a technological and economic infrastructure that has been described by Howe and Cayley as an “enclosure”and “vectoralization” of natural language and symbolic practice (Howe & Cayley, 2013: 2). One of their major points, in harvesting this English-language database with the unwitting help of Google Search algorithms, is to reclaim the commons of written language from the proprietary enclosure of discourse. The fine print of the terms of service of the “big data” and “cloud computing” companies is naturalizing an unfair distribution of resources in the domain of writing and reading practices (Cayley 2013).
In How It is in Common Tongues (2012) – one of their most significant interventions on the Internet as linguistic infrastructure –, the copyright rationale of our legal system is challenged not only as a legacy from print and earlier media culture but also as the already foreseeable copyrighted and surveilled future that is in the process of being reconstituted and reinforced in electronic culture. In this work, Samuel Beckett’s How It Is is reproduced through a collage of quotes from electronic sources other than Beckett’s text, suggesting that authorial and institutional appropriation can always revert to the common flux of language. This process of collage in turn is achieved by the détournement and refunctionalizing the “algorithmic, compositional, and configurative agents of big software’s network services” (Howe and Cayley, 2013: 1). In this instance, Cayley and Howe use a Phrase-Finding Reader that moves according the Longest Common Phrase algorithm. 4
The programmed reader’s strategy is to treat the database as an archive of language where it is possible to retrieve occurrences of phrases and collocations of words that match a copyrighted source. The function of the highjacked search engine is to mediate between the work’s literary algorithm and the general database of written English language. This unconscious but structured collaboration between big data search algorithms and a readerly algorithm becomes the compositional principle of Cayley and Howe’s work, which looks for its text within the database of networked verbal production in English. The phrase-mining process culminates in the production of a printed and bound book where Beckett’s text is no longer Beckett’s text, its authorial voice having been returned to the stochastic and probabilistic stream of language as robotically assembled from many Internet servers. The definition of copyright in the eighteenth and nineteenth centuries resulted in the historical reification of originality as a legal marker for assigning property in a given organization of printed language. Cultural capitalism developed on the basis of the recognition of writing as form of property and a form of capital, and on the control and licensing of its reproduction.
Because Beckett’s text has been assembled from the stream of writings on the Internet, How It Is in Common Tongues (Figure 1) suggests that the text exists as a potentiality in the permutations of language (as instantiated by this self-exemplary networked collage), and that it is language itself and our own uses of language that have become means of production of the Internet. Instead of using their algorithmic demonstration of Ted Nelson’s notions of transclusion and transliterature to reinforce the principle of copyright, How It Is in Common Tongues draws our attention to the ways in which the Internet itself, through its big data companies and their terms of service, is determining networked writing practices and enclosing the commons of language.
This framing of the network itself as a language-based construct is a very powerful way of sustaining an argument against the privatization of language and against normalizing human desires as transaction opportunities. Language is revealed as the major invisible capta of the network – the ensemble of data provided by our daily writing and reading interactions – which becomes part of the general profit-driven advertising logic and keyword control and surveillance system. Cayley and Howe’s work is meant as an aesthetical intervention on the constraints inherent in non-reciprocal terms of service and search algorithms, which they have diverted for their literary transactions:
Both the tools we use to read and write, and the material traces of our textual practice come to be stored on systems that are removed from us as readers and writers. We are increasingly dependent on self-regulating, proprietary services without which we cannot gain access to our reading or our writing, and whenever we do gain access we do so on terms. These circumstances have momentous consequences for textual practice and their careful consideration is crucial. (Cayley 2013: n.p.)
One of the ways in which digital mediation appears to be historically unprecedented is that it offers itself as a service or facility or catalog of affordances (such as word processing for writing), but it quickly goes on to establish itself as essential infrastructure. Thus, it becomes remarkably determinative of practice and ideological framework while nonetheless continuing to be managed and developed as if it remained a service. (Cayley 2013: n.p.)
As writers and readers we are forced consider that our relationship with language and literature will never be the same. If the medium of literary art has significantly migrated to the network, where it is gathered, channeled, and filtered by big software on a massive scale, daily touching the linguistic lives of huge populations, then new practices for reading and writing with and against such services must surely arise and go beyond any uses that are constrained by the terms of service or use now made unilaterally explicit by contemporary service providers. (Cayley 2013: n.p.)
3. Language as infrastructure
While the libertarian principles of How It is in Common Tongues embody a new reading and writing practice that addresses the algorithmic condition of language as commodified data in the age of the Internet, The Readers Project has other important consequences for the production of the literary in networked programmable media and for an understanding of the performativity of the act of reading in the production of the literary. A few of their installations and performances show programmed agents creating reading paths on textual fields according to specific strategies. Although they are not attempting to simulate a particular reading practice or the actual ocular processing of text as the eyes move from point to point, its various reading algorithms give the viewer a material perception of the dynamic nature of reading. Reading traverses a field of signs, and this traversing is creating its own particular linear and multilinear constellations. Each reading vector performs an iteration of a particular reading behavior or strategy and thus becomes scripted in the textual surface. Each of these inscriptions is the result of the machine’s execution of the scripts that constitute a given machinic reader’s code (Portela 2013: 345-347).
Modes of reading that inscribe themselves in a given textual field according to a programmed reading strategy are formal experiments with far-reaching implications for locating the literary. In the various iterations of The Readers Project, the “literary” could be located at the following levels:
1. as the codependence of writing and reading: writing exists through a reading intervention which scripts itself back in the textual field according to an embodied interaction, and this means that literary writing is coproduced by literary reading – the “performative materiality” (Drucker 2013) of the literary as a particular mode of reading;
2. as a reflexive instantiation of the algorithmic processing of writing and reading in programmed networked environments: the work’s generative algorithm sustains a reading-writing experience by feeding off the network’s databases and algorithms – the literary as a particular mode rewriting the network, and making explicit the unequal contractual relations that regulate the cybernetic processing of language in today’s new media systems;
3. as a series of scripts, i.e. as a series of rules that can be formally scripted in a program, to generate new writings from linguistic permutations of the archive of written language – the literary as the open exploration of the combinatorics of language structures mediated by the memory of writing;
4. as “text generators of different scales” (Montfort 2012) through which reading algorithms harvest, highlight, underline, deform, remix, and cite various textual systems (from a particular text or work to the entire corpus of searchable English-language texts) according to a formal rule that generates writing;
5. as a “cognitive fiction” (Tabbi 2002), i.e., as a self-organizing system that represents its own workings, in this case the workings of the strange loops that connect linguistic mediations with programmed networked electronic mediations. The otherness of the Internet as a purely medial system is reframed as a language system dependent on natural language. Thus language would be one of the infrastructural elements constitutive of the network unconscious.
Since these readings of the machine are offered as writing to human readers, the writing of reading and the reading as writing contained in The Readers Project turn readers into metareaders who are made to read their own act of reading the program reading. The motion of reading, which is rhetorically enacted within the textual surface through textual animation, is also a perceptual reminder of the kineticism contained in the act of moving the eyes from letter to letter and from word to word. Because the programmed readers within the text are offering their particular mode of reading as yet another instance of writing to be read once again by a human reader, actual readings cannot be exactly mirrored or captured by the programmed readers, only reengaged as renewed encounters with signs. The particular mode of reading scripted in the reading machine and the human act of reading that particular mode reading create a cognitive position. Automated generative writing is presented as an act of reading that rewrites the text and makes it available for literary reading. Cayley and Howe’s algorithms for defining the various reading vectors manage to avoid the entropy of entirely random word selection while still making room for unexpected and improbable associations. The poetic effect is the product of the tension between the entire corpus of words in the chosen text as the work’s database and the neighborhood rules that govern particular readings. Set up as an installation to be read by human readers, the strictly rule-based semiotic processing of word association executed by the machine enters the human processes of symbolic substitution and affective apprehension. The textual motion happening in the self-reading text conflicts with the actual reading motion of parsing the text at a human reader’s pace. Because every machine reading is also a new instance of writing available for a human reader, the act of reading the machine reading becomes a model for the infinite iterability of writing as actualized by each reading act. The iterability of code is used for stressing the act of reading as a field of possibilities grounded on the iterability and citability of writing.
Cayley and Howe’s works are based on an autopoietic interaction with the network as a large-scale system of notations where natural language itself – and not just the communications’ engineering infrastructure (its hardware or its software programs) – is a major part of the means of production that sustain the Internet as the discourse network and notation system of the 21st century. Their work points to this almost invisible coupling between the network of natural language character strings and the network of encoded data. The Readers Project’s engagement with the programmable network is an engagement with the database of networked language. The reflexiveness of its reading-writing protocols creates a meta-representation of the Internet as a linguistic network. This sort of electronic equivalent of literary metalepsis enables a moving back and forth between language as meaning and language as data. Machinic iteration of reading behaviors is a way of giving literary form to our database of digital writings, an attempt at finding the text in the automated workings of our current media networks and in symptomatic manifestations of the Internet unconscious.
4. Discourse generator
Text generation is based on a formal description of the syntactic structures of language. Each phrase or sentence is a grammatically well-formed sentence if the co-occurrence of different classes of words conforms to the subject-verb-object-complement sequence of a particular language. The exact form of each collocation is further determined by the semantic probability of co-occurrence of given strings of words and by the morphological variations in the flexional word elements. Syntactic articulation of language elements works on the basis of nestedness and recursion, two properties that enable extremely complex hierarchical relations. All language production works according to this combinatorial and generative mechanism. Speakers produce new utterances through a process of selection and combination of lexical items according to syntactic, semantic, and morphological constraints.
Although the corpus of possible utterances is virtually infinite, our uses of language are not determined exclusively by the genetics of grammar. Considered as a form of social action, language is produced according to discursive practices that favor a limited number of utterances. This discursive repertoire enacts patterns of culture and structures of ideology that sustain forms of representation. Discursive practices shape our uses of language and establish the limits of what is acceptable or possible to say within any given context. Language use thus helps to define the social semiotics of those contexts and becomes one of the ways of structuring the social and political space of human interactions.
The act of writing may be described as a reflexive engagement with the properties and constraints of language, which allows the self to experience the tension between convention and invention that is at the core of any verbal act. Literary writing, in particular, can be described as form of using language in ways that make it possible to explore in a sustained manner the generative mechanism of language within and beyond those discursive conventions. 5 The domains of commercial advertisement and political propaganda are two very significant discursive contexts for observing the productivity of language as a form of social and political action in mass societies. Advertisement and propaganda attempt to create symbolic identities for commercial products or political programs that make them intelligible and referable in language through massive repetition of messages in multiple communication channels.
Those messages generally draw on stock phrases or inventive metaphors whose poetic power derives from sound and sense patterns that produce memorable associations. This social memory of language explains the resonance of particular word associations, which may result from common ways of reference, including aphorisms and idioms, but also from unexpected and original forms of reference. Once a commercial or political message is repeated in a massive scale it becomes part of that social memory of language, either as a re-signification of an already existing stock phrase or a common metaphor, or as new addition to that verbal thesaurus of a given language community. Both practices are strongly performative in the sense that they have perlocutionary effects that can be observed not only in the beliefs of the addressees but in their buying or voting behaviors.
PoemAds, by Rui Torres, gives readers access both to the corpus of 22 slogan phrases that constitute the syntactic template for lexical permutations, and also to the poematic construction, i.e, to the rules of permutation.  The discursive structure is automatically fed with lexical items collected from more than 10,000 Portuguese commercial advertisements and political slogans produced during the last decades (Torres 2012: 273). 7 Words chosen for permutation within each phrase belong to the same grammatical category and have the same flexional properties, so that every variation is syntactically well-formed (Figure 2).
5. Deprogramming meaning
A collection of slogans, spanning a period of several decades, may be treated as a historical document for analyzing evolving social representations. Their vocabulary could be explored as a probe into the changes in social values and political priorities, particularly when analyzed in conjunction with other semiotic elements such as graphic design, images, film and audio editing, and narrative content. A decontextualized appropriation of those linguistic remains of the advertisements and propaganda of the past, as happens in PoemAds, works in a different way. The defamiliarizing effect of the slogan-generating software agent points to the function of the commercial and political ad as a mechanism for the production of meaning belief-production.
Because permutations make the slogans interchangeable, they dissolve their intended meaning (and the memory of that meaning) and return them back to the generative chains of language. Message decoding cannot happen within the preferred intended meaning or even within a negotiated mode (in which receivers reinterpret the intended message with the inflections provided by their own social situadedness), but only through an oppositional strategy that allows readers to experience the arbitrariness of semantic associations between nouns, adjectives, and verbs. Remixing contaminates the political with the commercial, and vice versa; dissolves the apparent logical necessity of the meaning contained in the syntactical co-presence of a certain number of elements; and, through the randomness of the possible associations, suggests that the arbitrariness of language is a property that can extend to the semantic effects of syntactic associations (Figures 3 and 4).
PoemAds (2012) offers an epitome of the experience of the injunctions contained in messages, and a sense of the social self as a space for continuous reproduction of those messages. By suggesting the interchangeability of slogans and the equivalence between political and commercial slogans, the work offers both an x-ray of a consumption-obsessed society and a critique of current modes of political imagination and participation. Redefined by the poem as a consumer of messages, the reading self is asked to select instantiations of the automated permutations and publish them in a blog. In order to record these selected instantiations of automated permutations readers are also asked to self-identify by giving themselves a consumer reference. Choosing among linguistic combinations, within the constraints of the generated text, emulates the act of choosing among commodities and among political programs.
Yet, the critique of commercial and political marketing obtained through such a mechanism is not the main dimension of this work. In my view, through its permutational and generative rhetoric this poetic artifact offers the perception of language itself as mode of social action. As the work runs its course and readers become aware of the constraints of the interaction and the rules of engagement that are being offered, they experience how the production of meanings, and thus the production of belief in those meanings, is a form of social action. As a textual instrument, PoemAds uses text generation to show the performative nature of language as way of doing things with words.
By doing algorithmic things with ad words, mass mediated commercial and political meanings are materially embodied as perlocutionary effects of certain forms of language use. The reframing of verbal elements as a result of their arbitrary association highlights the role of those meaning structures in constituting consumer and citizen subjectivity as reproduced through media discourse. Consumer self and citizen self can be perceived as discursive functions that take a particular language form. This discursive form can be critiqued with the mechanisms of language itself.
As the work unfolds, slogan combinations result in absurd or ominous or lyric associations: “Life touches us”; “The war that searches”; “Spared by history”; “Absolute smile”; “The Internet uplifts us”. Liberated from the self-identity and stability that comes from intense repetition and the ensuing habituation, permutated slogans also suggest that the discursive structures of language are always open to appropriation and transformation. The robotic poem becomes a machine for deprogramming discursive meaning, and for reimagining a non-commodified relation to language and to the world of social and political action. By framing the generative textual experience as a supermarket of poems where each permutation suggests an alternate version of an actual (past or present) instance, PoemAds submits its own textual form to the general turbulence of semiosis.
Cayley and Howe’s project of reading and writing the political structures of the Internet, and Torres’ opening up of commodified language to the poetic power of combination are aesthetical interventions that address the coupling of language and algorithm as the new means of symbolic production. They reappropriate the commons of language as an act of resistance to the datafication of reading and writing, and as decommodification of desire and imagination. The robotic poetics of The Readers Project and PoemAds gives us a scripted experience of the infrastructure of language as means of production in the electronic network.
1. This text contains a revised and remixed version of two papers, originally written for the conferences “E-Poetry 2013: New Works, New Frames” (organized by the Electronic Poetry Center, SUNY, Buffalo, which took place at Kingston University, London, June 17-20, 2013) and “Chercher le Texte: Locating Text in the Electronic Literature” (organized by the Electronic Literature Organization, held at the University of Paris VIII, September 24-27, 2013). I express my gratitude to the organizers of both conferences, and particularly to María Mencia and Philippe Bootz. My participation in those conferences was sponsored by the research project PTDC/CLE-LLI/118713/2010, funded by FCT (Foundation for Science and Technology) and FEDER (European Regional Development Fund) through Axis 1 of the Operational Competitiveness Program (POFC) of the National Strategic Framework (QREN). COMPETE: FCOMP-01- 0124-FEDER-019715.back
2. Among the most significant examples that, over the past decade, thematize their socio-electronic conditions of production are works by Mary-Anne Breeze (MEZ), John Cayley and Daniel C. Howe, Jim Andrews, Angela Ferraiolo, Eugenio Tisselli, and Jody Zeller. back
3. Detailed documentation about the various iterations of the project – including performances, installations and publications – can be found at http://thereadersproject.org/ .back
4. Authors’ description: “A Longest-Common Phrase (or LCP) is the longest sequence of words, beginning from a specific point in a text, that can be found on the web, not written by the author or about the text in question. The LCP algorithm is the procedure, generally employed by Phrase-Finding Readers, to locate such phrases via queries to public search engines like Google and Bing. The algorithm begins by doing a search for the first K words of a text, as an exact match (i.e. as a double-quoted string), with the addition of the author's name and title words of the text excluded. If the search returns no results, one word is trimmed from the end of the string and the search is retried (if a K value of 10 was initially selected, then we search next for an exact match on the first K minus 1, or 9, words). If once again there are no results, the phrase is shortened again by one (to 8 in our example), and the search is repeated. When finally there is a match, the number of results and the list of matching URLs are stored. If for example, the search returns one or more matches for the first 6 words of the text, the next iteration of the algorithm will begin on the seventh word and proceed similarly. The algorithm terminates when the end of the text is reached and all words have thus been included in a matching phrase.” (Howe and Cayley, 2013: 2-3).back
5. Rui Torres has developed a “plagiotropic” (Torres 2012) digital poetics that uses poems and lexicon by other authors as the syntactic matrix for permutations, thus opening up poetic discourses to automated textual proliferation. For a detailed analysis of his textual motors see Portela 2012. back
6. The work is available here: http://telepoesis.net/poemads/poemads.htmlback
7. The following slogans (most of which will resonate immediately with Portuguese readers) are used (Torres 2012: 372):
Com Flora, tudo melhora – Flora
Sabor Autêntico - Super Bock
Não é Água. É Castello - Água Castello
O poder da natureza é infinito - Pedras Salgadas
A pausa que refresca - Coca-cola
Compal é mesmo natural - Compal
Puro Prazer - Iogurte Adagio
Brincadeiras de bom gosto - Óleo Fula
O Bom, Sai Bem - Pescada Pescanova
Pelo prazer de conduzir - BMW
Movidos pela paixão - Fiat
A vida inspira-nos - Millennium BCP
Valores que crescem consigo - Montepio
Há coisas que o dinheiro não compra. Para todas as outras existe MasterCard - MasterCard
Ter é poder - Unibanco
Chega de lágrimas - Champô Johnson & Johnson
Campeão de preços - Continente
Sabe bem pagar tão pouco - Pingo Doce
A vida em Movimento - Metro do Porto
A criar excêntricos todas as semanas - Euromilhões
Aposte nos seus sonhos - Lotaria Clássica
Porque a vida é agora - Visa.back