Mexmix: Digital Narrative in Face of the Unreadable Machine


Lasse Scherffig

University of Bremen





Mexmix is a generative hypertext story using low-fi technology. Its central idea consists in remixing text on the basis of user behavior. As an example of generative story writing it allows us to reconsider the role of narrative as artistic form. It is argued that digital narrative as a context free endeavour does not exist while only narrative by digital means does.


1. Narrative as Artistic Form


As we are currently perceiving a return of the grand narratives (in the sense of [1]) – mainly in form of the narrative of globalization or the narrative of the clash of cultures – it might be interesting to requestion the narrative as artistic form.


The death of the latter has already been declared by Lev Manovich in  „Database as Symbolic Form“ [2]. For him „after the novel, and subsequently cinema privileged narrative as the key form of cultural expression of the modern age, the computer age introduces its correlate – database" (cited from [2]). While he did not claim the immediate death of narrative he clearly spoke in favour of a turn towards database. Even this might have been a bit hasty.


The idea of the end of narrative is closely connected to postmodern thought. When Jean François Lyotard in [1] defined the grand narratives and their role he proposed the postmodern condition to be the condition of society after the end of those grand narratives.

The existence of the grand narratives mentioned above hence puts postmodernism under pressure. As a project of revising modernism in favor of the marginal it possibly has to be declared as failed. And while this feels displeasing on the level of politics and critical theory this might help us to search for the narrative in the digital.


2. Mexmix as Example


As a piece of (already old-fashioned) hypertext, mexmix comes as a regular homepage. It shows pictures, text and hyperlinks. Every word is a hyperlink. Clicking a word yields a new page, again showing pictures, text and links. The text, however, is composed – or generated – algorithmically. The crucial element for this composition is the history of words clicked so far. This set of words determines the sentences to come. The ideas behind are simple: Automated collage writing [3] and adaptivity.


Collage writing here consists in reconfiguring the position of sentences from various sources  into a new order. These sources are personal and public, the former being my eMail inbox, the latter some texts published at indymedia (please see [4]). The reconfiguration is based on applying a text similarity measure. Sources and sentences from these are selected by comparison to another text. This other text makes up the adaptive part: It is, as mentioned, the history of words clicked so far [see figures 1 and 2].


Figure 1: The mexmix architecture. Note that it shows a closed circuit nature.


The story evolving from that is shaped by reader and program. The narration is a property of the dynamics both develop. One thus can pretend to employ what Mark Amerika already pretended to employ in Grammatron – one of the most famous pieces of  hypertext based art [5].  The „story is reading you“, Amerika says, since „neither the writer nor the reader is determining where that specific navigational route is" (as cited in [6]).


while (thereIsUserInput) {

  ui = readUserInput();


  history = readUserHistory();

  cluster = database.selectTextCluster(history);

  text[] = cluster.selectText(history);



Figure 2: Mexmix as pseudocode.


3. Digital Narratives as Unreadable Machines


The technology involved in mexmix is well known: Algorithm and its symbiont database have opened the possibility to do automatic manipulation of discrete symbols on a large scale. When Manovich introduced database as narrative's successor he also gave a very important semiotic analysis of database: French semiotics (in the tradition of Ferdinand de Saussure) orders language in a two dimensional field spanned by a paradigmatic and a syntagmatic axis. The ruling principle of the first is selection of semantically (more or less) equivalent entities. The principle of the second is combination of entities in space – ruled by syntactic rules.  In terms of French semiotics, Manovich argued, database is equivalent to the paradigmatic axis of language [2]. Algorithm, we can conclude, resembles the syntagmatic axis.


Algorithm as well as database exist in the same realm: the digital. The two dimensional field of the axes of syntagm and paradigm thus is projected onto one dimension: the computational. Moreover, this dimension leaves us with a twofold problem: Neither the syntagm nor the paradigm has been formalized yet. Approaches towards formalizing the former once looked promising. But since 1956 generative grammar is context free grammar [7]. Formalization of the latter, Umberto Eco argued, is impossible [8]. This becomes most evident in Jacques Derrida's critique of the concept of structure [9] and in his writings about grammatology.

Therefore, any kind of computational generation of sense must fail. The machines trying to do this are necessarily unreadable machines.


4. Generation as Reproduction


But we can read mexmix. Though only loosely resembling traditional (linear) narrative it might succeed in transporting atmosphere and sense.


It does so for two reasons: It reproduces and it works in a limited domain. It reproduces since the source material already is text and already is loaded with sense. It works in a limited domain since text and image data are not arbitrary but constrained: They to a large extend share the topic. They even share style.


The sources mexmix uses hence already are structured in both syntagm and paradigm. On the level of the algorithm the prestructuring is less obvious. Here the domain is the domain of formal experiments with language and (hyper-) text. Especially a tradition in which Dada or Merz are to be mentioned ([3] gives a discussion of these historic roots of hypertext) as well as contemporaries such as Mark Amerika [5] and also Lev Manovich for the language of film [10].


Sense, as an interpretation or an ascription – a result of „sense-making“ [11] – can be made of what a machine produces only since it already is there. In source material, in data structures, in decision rules and – this is very important – in the context the machine's output is placed in.


A generative narrative therefore works as good as it is written by someone (and I am not talking about a computer). Only that writing in this case might consist in writing into a database as well as inscribing structure into an algorithm or choosing the mode of presenting its results.

Digital narrative – the consequence – does not exist. Narrative in the digital, however, does.




[1] Jean François Lyotard, The Postmodern Condition: A Report on Knowledge, University of Minnesota Press, Minneapolis, 1984 (from the French original: Jean François Lyotard, La Condition postmoderne: rapport sur le savoir, Les Editions de Minuit, 1979)

[2] Lev Manovich, Database as Symbolic Form,

[3] George P. Landow, Hypertext as Collage-Writing, in The digital dialectic: new essays on new media, edited by Peter Lunenfeld (pages 151-170), MIT Press, Cambridge, 1999

[4] indymedia: Independent Media Center,

[5] Mark Amerika, Grammatron,

[6] Matthew Mirapaul, arts@large: Hypertext Fiction on the Web: Unbound from Convention, Cybertimes: The New York Times on the Web (1997-06-26),

[7] Noam Chomsky: Three models for the description of language, IRE Transactions on Information Theory, 2 (pages 113-124), 1956

[8] Umberto Eco, Einführung in die Semiotik, Wilhelm Fink Verlag, Munich 1972  (from the Italian original: Umberto Eco, La struttura Assente, Casa editrice Valentino Bompiani, Milano, 1968)

[9] Jacques Derrida, Die Struktur, das Zeichen und das Spiel im Diskurs der Wissenschaften vom Menschen, in Postmoderne und Dekonstruktion, edited by Peter Engelmann, Reclam, Stuttgart, 1990 (originally published in Jacques Derrida, Die Schrift und die Differenz, Suhrkamp, Frankfurt, 1976)

[10] Lev Manovich, Soft Cinema,

[11] Brenda Dervin, Chaos, Order and Sense-Making: A Proposed Theory of Information Design, in Information Design, edited by Robert Jacobson, MIT Press, Cambridge, 2000