Daniel Escandell-Montiel (in English)

Introduction: The Emergent Question

The increasing importance of user agency in engagement with hypermedia (that is, interactive media, of which video games are the supreme example; Nelson) has provoked the question of designing narrative structures as a space that reacts to the input a receiver provides to a synthetic world. LeBlanc outlined the concept of emergent narrative at the end of the last century, and its use was taken up again in Jenkins’s work, though significantly reduced and restricted to ambient narrative. According to LeBlanc, the emergent narrative is a way of constructing and telling stories that do not rely on blocks of narrative written by a scriptwriter (which he views as an embedded narrative within the hypermedium). In other words, there is no predefined narrative sequence (linear or otherwise), but instead, what takes place in the narrative is an organic reaction to how the user interacts with the world.

LeBlanc’s proposition does not unseat the author, but it does redefine their work, no longer positioning them in the role of a deterministic god and transforming them into a god that accepts free will. This author sets out rules and parameters that establish what the world is like, what sorts of interactions can take place, what consequences each action can produce, and what interactions influence others, but from there the world is no longer governed by determinism (or at least does not seem to be so to the receiver). The role of the receiver/user, too, is altered: previously confronted with a scripted space (maybe even one that allowed a genuine possibility of making moral decisions, but within a predesigned narrative sequence) in which they remained a kind of puppet moving through the constrained terrain laid out for it, now they have the ability to act on their own story and on the rest of their environment. Within limits, of course: though it is clear that there is no set sequence of events and the illusion of free will is carefully crafted, this is possible only insofar as the preestablished and designed mechanisms of interaction allow for it. In fact, these very mechanisms can also be used to force the user to carry out a particular action or sequence, meaning that narrative construction is also present. Nevertheless, under this paradigm, narration and interaction are ideally one and the same, as a result of which the line between narrative and ludological elements blurs, thus transcending classic questions such as that posed by Adams, for whom interaction and narration were practically opposing concepts.

It is doubtlessly significant that, decades later, the most strictly conceptual debate on this topic remains unresolved (Suárez) and literary analysis has failed to take it very far, which has left the discursive framework for this question centered in gamestudies. Another issue, already raised by LeBlanc himself, is that it can be quite complicated to generate interesting, meaningful, or emotionally satisfying stories using only mechanisms of interaction without straightforward scripting. And, of course, this poses another subordinate, but fundamental, question: where does it leave the traditional script and, therefore, the focus of literary study that makes it relevant as a medium for the discipline?

Video games’ ability as a medium to develop relevant stories has advanced as technological capacity, which originally even limited the amount of text a program could contain, has increased (Escandell Montiel). As such, we can consider a hypermedium to be a sort of metamedium—that is, a general foundation (effected electronically) able to encompass multimodalities to which the receiver relates through a series of interactions. It is in this broad—but necessary—conception that we findoutcomes that tip the balance toward one type of medium or another. For example, the first adventure video games were purely text-based (they had no—or only negligible—graphical capacity), more akin to choose-your-own-adventure books than to our current notion of video games.

That may be why Device 6 was such a surprise: an interactive textual-visual story (with a substantial textual component) with audiovisual and 3D photographic elements that presents readers with integrated puzzles that they must solve to advance in their reading. The creators promoted it as a video game, but it is not unique; the Electronic Literature Organization’s various compilations of international digital literature have included works such as Dwarf Fortress (fantasy adventure with textual graphics), Kentucky Route Zero (a magical realist story with an Americana-inflected aesthetic), 80 Days (a decolonial rewriting of Verne’s novel), and A Slow Year (visual poems with an eight-bit aesthetic), which have already explored the gray areas between the textual/literary and the ludic/interactive, sometimes using in their taxonomies the label “literary game,” and at others “interactive fiction.”

It is clear that contemporary criticism accepts and understands the narrative and intersectional capacity of the video game; nevertheless, LeBlanc’s question remains relevant. The experience of a predesigned (embedded) narrative with an (emergent) derivative of user actions produces internarrative dissonance. That is, the illusion of free will fostered by the emergent narrative collides with the narrative touchpoints and developments designed to build a story and a world. The emergent narrative, likewise, can only reach a particular point through the actions-reactions (explicit and implicit) generated by users regarding themselves and the world, by the world regarding the users, and by the world regarding itself. Taken to the extreme, emergent narratives seem to be produced, above all, in pure play—that is, in the stories that users can tell themselves in variations of the Lego formula (such as Minecraft); in titles where a kind of divine role is adopted to alter the characters’ lives and world (such as The Sims); or in games in which the narrative is secondary and the social component assumes greater importance, such as in the massively multiplayer online role-playing genre (e.g., Worldof Warcraft), or even in the amalgamation of social network and synthetic world (e.g.,Second Life). In these game concepts, the narrative experience is a secondary aspect at best.

Procedural Narratives: The Prism of Machine Content Generation

Certainly, these models do not entirely match the experience of a free or truly emergent narrative. That is achieved with procedural narrative as one of the strategies of procedural content generation (PCG). PCG is a growing part of software design, as when using trained artificial intelligences for aspects such as character or location design (for example, ensuring that each person has a face and body of their own, or that the vegetation in a forest is unique) in synthetic worlds. This leads us to procedural narrative generation (PNG)—in other words, where an artificial intelligence does not merely randomize which script preestablished by scriptwriters and narrative architectsis followed in response to an input, but in fact itself creates characters, contexts, determinants, events, etc., presenting the user with different situations and stories. As a narrative component, it should also be accompanied by an advanced capacity for natural language generation (NLG) with which to supply illocutive acts to the characters, narrators, and contents to ensure satisfactory communication between software and receiver.

NLG has been employed in virtual mobile assistants (such as Apple devices’ Siri) and in a variety of chat systems that continue to make strides in generating complex, reactive, and simulatedly human texts. The PNG arena, however, has drawn less attention. Akimoto and Ogata, jumping off from the theories of Barthes, Propp, and Genette, have designed programs capable of presenting random narratives that also react to the receiver’s input (Akimoto and Ogata). Unsurprisingly, their work serves as a relevant example of the application of digital humanities by combining established narratological theories (especially those that have sought to atomize and systematize narrative and narrator functions) with advanced computer programming.

Following Akimoto and Ogata’s work, we can establish three well-defined phases of PNG that also parallel the human creative process. At first, the system builds a general framework (the story) with which it designs an initial sequence of basic markers to which it associates content in a treelike structure, predicting possible outcomes and consequences. In the second phase, the discourse is established, which involves shaping the narrative by selecting the elements to be presented to the receiver/user; this includes defining which aspects can be altered through input generated by the receiver’s interaction. The third phase is the textual-audiovisual expression of the previous one—that is, the execution of the narrative’s surface (what the receiver will see, hear, and be able to interact with) in the medium and format in which it will be narrated.

As noted above, this system of creation is essentially identical to that used by a human when devising the predefined narratives mentioned in the previous pages; the difference resides in that artificial intelligence can perform it systematically, automatically, and as an ongoing reaction, especially if the interactions deviate from what had been predicted, constructing new narratives in a world that never has to be considered finished because it can potentially present an unlimited number of permutations. The result is true emergent narrative, in that it can transcend the limitations of the general framework itself but implements the structures that continue to be considered vital to a coherent, efficient narrative. It is unquestionably harder to predict emotional impact, though in reality the world is full of failed stories in every medium and format, so a failure to evoke satisfaction or an emotional reaction in the receiver does not necessarily mean that said experience should be dismissed. In any case, we must not forget that digital services companies are fully informed on our tastes and interests thanks to their algorithms and their analysis of all our data (Graefe et al.), which they use to offer us more content that they predict we will like (and to give instructions—or not-so-subtle advice—to their creatives when designing new products).

The work of any artificial intelligence, as currently conceived, involves training—that is, exposing the system to a substantial, properly prepared corpus to be analyzed and patterns extracted to reproduce and debug, until the software is able to generate its own content. This explanation, though tremendously simplified, allows people without computer expertise to understand how AIs such as the DALL-E image creator and the ChatGPT conversationalist, to cite some examples popular with the general public, have achieved their results. It was years ago now that an artificial intelligence succeeded in learning to play a video game on its own (Jaderberg et al.), a major technical milestone since it means the program was able to learn to move and act both actively and reactively to play against human users and achieve results that allowed it to defeat professional players. The present-day computing capacity, number of variables per second, and algorithmic complexity are greater than what was needed for Deep Blue to defeat Kasparov.

Several years earlier, a novel generated by an artificial intelligence (though revised by a human) designed at Future University Hakodate was named a finalist for a literary prize (Shoemaker). The project lead explained that whereas AIs had so far been used to solve problems, the new challenge was to attempt to emulate human creativity, as has in fact been the case ever since.