# Curating Simulated Storyworlds — Chapter 3: The Pleasure of Emergent Narrative

> NOTE: THIS IS A MARKDOWN CONVERSION INTENDED FOR LLM CONSUMPTION.
> IF YOU ARE A HUMAN BEING, READ THE PDF INSTEAD:
> https://jamesryan.computer/papers/2018_ryan_curating_simulated_storyworlds.pdf
>
> The PDF is the canonical document, typeset with every figure and table. This file
> is a plain-text conversion of the LaTeX source, published HERE so that software agents
> can read the thesis without parsing a PDF. It is lossy and is not meant for humans.
>
> This file is chapter 3 of 14 only: The Pleasure of Emergent Narrative (~21,000 words).
> See below for links to markdown conversions of the individual chapters.
>
> **Structure.** Sections are `##`/`###`/`####` headings. Cross-references are resolved to
> their numbers ("Section 3.1.1", "Chapter 5"), so a heading can be searched for directly.
> Citations are inline bracketed keys such as `[ryan2017grimes]`; every key used in the text
> has an entry in the bibliography, formatted as `- **[key]** Author. Title. Venue, Year.`
> Footnotes appear inline as `[Note: ...]`.
>
> **Conversion limits.** Figures and most complex tables appear as `[Figure omitted]` or
> `[Table omitted]`. Code fences carry no language tag. A few diacritics in author names
> were simplified from the Latin-1 source.
>
> **All files.**
>
> Index (this list, plus the abstract; small): https://jamesryan.computer/thesis/index.md
> Whole thesis in one file (large, ~1.6 MB): https://jamesryan.computer/thesis/full.md
>
> By chapter:
> - 01. Introduction (~4,600 words, ~30 KB) — https://jamesryan.computer/thesis/01-introduction.md
> - 02. Defining Emergent Narrative (~6,900 words, ~47 KB) — https://jamesryan.computer/thesis/02-defining-emergent-narrative.md
> - 03. The Pleasure of Emergent Narrative (~21,000 words, ~135 KB) — https://jamesryan.computer/thesis/03-the-pleasure-of-emergent-narrative.md
> - 04. The Pain of Emergent Narrative (~41,000 words, ~268 KB) — https://jamesryan.computer/thesis/04-the-pain-of-emergent-narrative.md
> - 05. Refining Emergent Narrative (~11,000 words, ~73 KB) — https://jamesryan.computer/thesis/05-refining-emergent-narrative.md
> - 06. James Ryan Generator (~2,800 words, ~16 KB) — https://jamesryan.computer/thesis/06-james-ryan-generator.md
> - 07. World (~20,000 words, ~129 KB) — https://jamesryan.computer/thesis/07-world.md
> - 08. Case Study: Diol/Diel/Dial (~9,300 words, ~59 KB) — https://jamesryan.computer/thesis/08-case-study-diol-diel-dial.md
> - 09. Talk of the Town (~25,000 words, ~162 KB) — https://jamesryan.computer/thesis/09-talk-of-the-town.md
> - 10. Case Study: Bad News (~22,000 words, ~132 KB) — https://jamesryan.computer/thesis/10-case-study-bad-news.md
> - 11. Hennepin (~32,000 words, ~209 KB) — https://jamesryan.computer/thesis/11-hennepin.md
> - 12. Looking Ahead: Sheldon County (~11,000 words, ~75 KB) — https://jamesryan.computer/thesis/12-looking-ahead-sheldon-county.md
> - 13. Conclusion (~2,700 words, ~17 KB) — https://jamesryan.computer/thesis/13-conclusion.md
> - 14. Bibliography (~29,000 words, ~233 KB) — https://jamesryan.computer/thesis/14-bibliography.md

Through its utilization in important videogames such as *Dwarf Fortress* and *The Sims*, emergent narrative represents perhaps the most successful approach to procedural narrative in the public consciousness. But why are these works beloved—why do people like them? What is the pleasure of emergent narrative? In this chapter, I attempt to answer this question by discussing a series of related phenomena with which emergent narrative shares critical features: nonfiction, stories of lived experience, worldbuilding, and art brut. Finally, I identify a set of aesthetics that are yielded by these shared features. I call these the *aesthetics of emergent narrative*, and I believe that they undergird the pleasure of the form (and thus provide an account for why people like its successful examples). Throughout the chapter, I refrain from discussing my own work, though in Part II I will apply the intellectual framework developed here in the case studies included there. Nonetheless, I should note that the ideas articulated in this chapter were developed not only through interpretation of the projects discussed herein, but also through the evolution of my own simulation and media practice. As such, my autoethnography in Part II tells a story about how I arrived at these arguments.

## Analogous Forms

In this section I will articulate a series of analogies by which we may view emergent narrative as having certain structural, functional, or superficial resemblances to other narrative and artistic forms: nonfiction, stories of lived experience, worldbuilding, and art brut. With this foundation in place—particularly the resemblances to nonfiction—it becomes feasible to identify a set of aesthetics that are distinctly associated with emergent narrative, which I provide at the end of this chapter.

### Analogy to Nonfiction

Emergent narrative works like nonfiction. Here is what I mean: events that emerge out of the interaction of processes in a computational system actually *happen*, and so stories that recount such events work like nonfiction, to the degree that nonfiction is conceived as the recounting of a series of events that have happened. Note that here I mean to call upon the *genre* of nonfiction, as opposed to the *ontology* of nonfiction, the plausibility of which has been challenged by thinkers (both ancient and contemporary) who have argued against the possibility of human objectivity. [Note: To be clear, by ‘genre of nonfiction', I mean narrative works that (ostensibly) faithfully recount a series of (ostensibly) real events, as opposed to, for example, a math textbook or a manual for operating a typewriter. I will elaborate on this notion below.] In this section, I outline an array of scholarly perspectives on nonfiction before providing a case study that applies these perspectives to an example of emergent narrative, to show that it works like nonfiction. [Note: My argument in this section is essential to the ideas promulgated in this thesis, and so I have sought to be diligent in making it. Starting now, I will depart on a rather adventurous foray into a wide array of scholarship on the ontological status of nonfiction. While some of the upcoming discussion may feel at best tangential, trust me when I say that all of it is in fact highly relevant to the core argument of this section: that emergent narrative works like nonfiction. I promise that the journey will make sense by the destination.]

#### Essential Objections to Objectivity

There are ancient arguments that question whether there is even an external world that is constituted in a reality. For example, the solipsistic position holds that only the self can be known to exist [audi2005epistemology]. A more extreme position, attributed to the Greek sophist Gorgias, contends that in fact *nothing* exists, and even if something did exist, we could not perceive it, and even if we could, we could not impart such perception [kerferd1955gorgias]. Thus, if nonfiction is the recounting of a sequence of real events, and the very notion of a real event (or realness itself) is called into question, then the ontology of nonfiction, as typically conceived, is challenged. Alternatively, if reality exists but is constituted by oneself, or if an external reality can be perceived but not recounted to others, then the very project of nonfiction is frivolous—it is just the recording of a diary for one's own consumption.

#### Contemporary Objections to Objectivity

Even if we believe in the possibility of real events existing and being recounted, contemporary scholarship (especially from postmodernism) problematizes the notion of *faithfully* recounting real events. Here, an essential critique maintains that while a true account of reality would be *objective*, humans are subjects whose experiences, understandings, and expressions are (thereby) inherently *subjective* [foucault1969archaeology,rorty1979philosophy,hutcheon1988poetics]—Satya Mohanty has called such denial of objectivity “the core thesis of postmodern literary and cultural theory” [mohanty1997literary, p. 10]. Due to its predication on human experience, understanding, and expression, the very project of nonfiction is subjective by this view. Moreover, even if the propositional content expressed in a work is somehow all true, the work will (of course) still exclude many truths about reality, and so its particular inclusions and exclusions are the byproduct of a subjective process of selection. Indeed, this selection—moreover, any mode of artistic or scientific or generally communicative production—is necessarily couched in relativistic social and cultural practices [kuhn1962structure,berger1967social,vygotsky1978mind,peshkin1988search].

Roland Barthes attacks the idea of nonfiction from a semiotical perspective: the sign system of a recount creates signifieds that may correspond to real referents, but as signifieds they are not literally those referents [barthes1987rustle]. He calls the implicit or explicit claims of objectivity in writing a *referential illusion*—the author implies that the signifiers are actually the real referents they signify—that yields a *reality effect* (it produces a style called realism). This referential illusion is typified, Barthes says, by the historian Adolphe Thiers' directive for historiographic practice: “To be simply true, to be what things are and nothing more than that, and nothing except that” [barthes1987rustle, p. 148].

#### Recuperation

So, if nonfiction is not a faithful account of an external reality, is it just fiction? This may be the case in an ontological sense, and that may be unavoidable, but nonetheless there is clearly a thing called ‘nonfiction'. Indeed, lay people seem to have little trouble delineating the forms—simply consider how bookstores are organized to feature nonfiction sections. [Note: Interestingly, Christine Gallagher argues (partly on the basis of etymology) that the idea of fiction is actually quite new (less than a few centuries old) [gallagher2006rise]. Meditating on the origins of language, the writer and thinker E. L. Doctorow has suggested (relaying the opinion of a professor from his college days [savvas2011there, p. 124]) the perhaps inevitable case of a time when fiction and nonfiction could not be distinguished because “the designative and evocative functions of language were one and the same” [doctorow1977false, p. 217]. This notion brings to mind the late-career work of Sheldon Klein, a main character in Chapter 4, on the origins of human cognition [klein1983analogy,klein2002analogical,klein2003invention]. Curiously, the word ‘nonfiction' seems to express fiction's privileged status as a default mode, as Gérard Genette notes [genette1990fictional, p. 756].] In summarizing the work of Laurence Goldman and Michael Emmison [goldman1995make,goldman1998child], Jean-Marie Schaeffer notes that such distinction appears to span all human cultures:

> But even if it may be true that fictional narrative as a socially recognized practice is not an interculturally universal fact, all human communities seem to distinguish between actions and discourses that are meant to be taken “seriously” and others whose status is different [schaeffer2009fictional, p. 99]

Moreover, there are scholars outside the postmodernist tradition who are still interested in characterizing the delineation of fiction and ‘nonfiction' (or characterizing, alternatively, a spectrum of fictionality). Such work has tended to appear in three distinct traditions, which may be called the *semantic*, *formalist*, and *pragmatist* perspectives on fictionality. I will discuss the first of these perspectives in the next section (with a particular focus on John Heintz's description of fictional worlds), and the latter two in the subsequent sections.

#### John Heintz's Fictional Worlds

The *semantic perspective* on nonfiction is characterized by logicians who evaluate the ontological status of narrated entities and the truth values of the propositional content concerning such entities. [Note: My writing of this section and the following two was greatly aided by Sebastian Deterding, who sent me a helpful array of pointers to scholarly work dissecting the notion of nonfiction.] As Jean-Marie Schaeffer notes [schaeffer2009fictional, p. 104], this tradition proceeds from the denotation of Gottlob Frege [frege1948sense] as extended by Bertrand Russell [russell1905denoting]: a statement's truth value depends on whether that statement obtains with regard to the one actual universe. It is explored primarily in the area of analytic philosophy, where it flourished especially in the 1970s. [Note: For more comprehensive reviews, consult the writings of John Woods, who has been active in this area for decades [woods1974logic,woods2007fictions,woods2018truth]. Matthieu Fontaine and Shahid Rahman also provide a recent overview [fontaine2010fiction].] In an influential paper called “Truth in Fiction” [lewis1978truth], the philosopher David Lewis wrestles with the intuition that a sentence like *Sherlock Holmes lived on Baker Street* seems to be true while *Holmes and Watson are identical* seems to be false. [Note: John Woods has called sentences like the former *bet-sensitive*: if one person bets another that Holmes lived on Baker Street, the former wins the bet even through both parties will agree that Holmes does not exist [woods1974logic].] The problem here is that ‘Sherlock Holmes' is “denotationless” (it fails to refer to an entity in the actual universe) by Russell's program [lewis1978truth, p. 38]. One potential solution is to take up the Meinongian ontology, whereby any concept that can be described, even a non-existent one, is viewed as an actual object that may be denoted (referred to) [jacquette2011meinongian]. This project has been taken up by Terence Parsons [parsons1975meinongian], but Lewis cites a number of attendant difficulties [lewis1978truth, p. 37]. [Note: Parsons's Meinongian view of fictional characters leads to some evocative writing, such as: “I suppose that all of these [fictional characters] were objects before they were written about; they were so to speak only identified by the author, and writing about them did not confer objecthood on them. Nor do I suppose that writing about them brought into existence, for they do not exist” [parsons1975meinongian, p. 79].] Instead, Lewis provides a reasonable solution by which (in part) these sentences are taken as implying a prefix, *In such-and-such fiction...*, which means the troublesome sentence is actually understood as something like *In the Sherlock Holmes stories, Sherlock Holmes lived on Baker Street*. Others have posited similar refinements [woods1973considerations,parsons1975meinongian,heintz1979reference]. By this account, we may say that nonfiction obtains when (or to the degree that) a work's propositions are true without being prefixed in this way. [Note: It is interesting to consider this approach with regard to Barthes's referential illusion, mentioned above. Barthes would argue that even if the propositional content of a narrative refers and is all true, the very procedure of denotation makes nonfiction impossible: sign systems cannot contain components of the actual world, but rather only simulacra of those signifieds.] This work, however, is concerned not with the delineation of the fictional and factual, but with building a logic that is robust enough to handle fictional worlds.

This is also the nature of John Heintz's project [heintz1979reference], but his approach is to first identify a set of peculiar logical features that characterize fictional worlds (since a robust logic for such worlds would have to handle such features). Specifically, he argues that fictional worlds are necessarily *incomplete* and are often *inconsistent*.

First, they are incomplete in the sense that they fail the law of bivalence, which maintains that, for some proposition $p$, either $p$ is true or $p$ is false. [Note: Writing later on, Daniel Dennett follows this same reasoning to deem fictional worlds *undecidable* [dennett1992self].] As Heintz states, for any given fictional world, truth values are undecided for nearly all conceivable propositions about the world:

> authors [of fiction] are not foreign correspondents, reporting, sometimes incompletely, on events they witness in some far-flung corner of the world. They create (most of) the events they write about. What they fail to tell us, either explicitly or by implication, simply does not exist. [heintz1979reference, p. 92] [Note: In the paper, Heintz addresses a clever attempt at producing an exception to this: “Dudley Shapere suggested in conversation that it might be possible to so write a story that every statement about its characters would be decided. For example one might write ‘and everything not explicitly stated or implied here is false'. I do not know of any actual cases of this. Of course, if mathematics is any part of the story, then Godel's incompleteness result, which guarantees that no consistent set of sentences will yield all the truths of arithmetic, rules out even this possibility” [heintz1979reference, p. 91].]

 The potential counterargument to Heintz's claim is that the fictional world is an actual one that exists independently of the conception of an author—that is, rather than constructing the world, the author has gained special access to it. This is an interesting idea, but it is not especially convincing, as Heintz argues:

> To believe otherwise is to posit a multiplicity of pre-existing fictional worlds to some of which authors have some kind of special access which, in principle, non-authors might come to share. A decision about the fate of a character could not then be decided by examining the text, [but] by gaining access to the relevant fictional world. We could not just take Barrie's word that Peter Pan could fly; we should be obliged to seek access to the fictional world Barrie was reporting on, and see for ourselves whether or not Peter Pan actually relied on hidden wires or jet propulsion. In the absence of well-articulated arguments for a position with such bizarre consequences, I accept the more natural view that writers make up their stories, and that what they write down is, as far as the fiction goes, the way it is. When nothing in the text supports either [$p$ or $ p$], neither is true of that fictional world. [heintz1979reference, p. 92]

Beyond being necessarily incomplete, Heintz argues that fictional worlds are frequently *inconsistent*. As an example, he states that “a careful reading of Anna Karenina reveals that it begins on a Monday and a Tuesday” [heintz1979reference, p. 92]. [Note: Apparently this claim was the subject of subsequent discussion that was predicated in part on a misreading of “Monday and Tuesday” as “Monday or Tuesday” [humberstone2011connectives, pp. 767–768]. For this reason, I quote the passage directly.] While such an inconsistency could be viewed as an authorial slip-up that is resolved in the context of the rest of the work, it could of course be possible for such context to variously rely on multiple components of the paradox. If this occurs, the context does not resolve the paradox, but in fact reinforces it. As Heintz puts it, “An author may at one place assert a sentence, at another its denial, and the point of the story turns on both” [heintz1979reference, p. 94].

Let us consider what Heintz's fictional worlds tell us about nonfiction. As noted above, Heintz discusses these characteristics of fictional worlds in pursuit of the development (or identification) of a logic that is robust to such peculiarities. In the case of nonfiction, however, such special logics are not needed, since the world of nonfiction is our external reality: if this is conceived as a sole actual world, Russell's program is robust enough; if instead it is conceived as one of a series of actual possible worlds, then a standard modal logic is robust enough. Thus, we might say that Heintz's description of fictional worlds implies a definition of nonfiction as being constituted in works for which these logics suffice—i.e., works that do not have the troublesome characteristics of fictional worlds that he identifies: incompleteness and inconsistency. Note that this does not necessitate that a nonfiction *narrative* be complete and consistent, but rather that its *subject world* be complete and consistent. Below, I will argue that the subject worlds of emergent narrative (simulated storyworlds) tend to be complete and consistent, and thus the approach itself tends to work like nonfiction.

#### Formalist Perspective

As an alternative to semantic approaches, the *formalist perspective* ignores the logical status of narrative content to instead identify the stylistic tropes that characterize works of fiction [schaeffer2009fictional,schaeffer1998fiction,schaeffer2010fiction]. For example, Kate Hamburger isolates its means of entering the subjective worlds of characters [hamburger1973logic], while Ann Banfield asserts that fiction is marked by discourse that expresses mental perspectives in a third-person voice [banfield2002grammatical]. Contra these arguments, however, are claims that such features are culturally variable and may be present in both works of fiction and nonfiction, as Henrik Nielson, James Phelan, and Richard Walsh articulate:

> No technique is found in all fiction and/or only in fiction, even though within certain cultural and historical contexts certain textual features can become strong conventional indices of a fictive communicative intent [...] Thus, while some techniques can contextually signal fictive intent, there is no necessary homology of form and function [nielsen2015ten, pp. 66–67] [Note: Indeed, as Sebastian Deterding noted in personal correspondence, the genres of *new journalism* [hollowell1977fact] and *fake documentary* [lebow2006faking] demonstrate that the stylistic tropes of fiction works may be used in factual accounts and, likewise, the stylistic tropes of factual works may be used in fictional accounts (personal communication, Apr 5, 2018).]

#### Pragmatist Perspective

For scholars taking the *pragmatist perspective*, nonfiction is all about communicative intent: fiction is produced when fiction is intended and fiction is apprehended. This pragmatic view is most associated with John Searle, [Note: Schaeffer asserts [schaeffer2009fictional, p. 110] that this view actually originates in work as early as David Hume's [hume2003treatise].] who views fiction as a string of simulated speech acts: “the author of a work of fiction pretends to perform a series of illocutionary acts” [searle1975logical, p. 325]. These speech acts depend on an interlocutor perceiving them correctly, and so Searle speaks also of the “shared pretense” by which pretend speech acts (and their referents) are understood as the author intends [searle1975logical, p. 320]. Expanding on this, Stanley Fish notes the activation in fiction (and all communication) of “a set of discourse agreements which are in effect decisions as to what can be stipulated as a fact” [fish1976things, p. 1022]. Here, Paul Grice's *cooperative principal* [grice1975logic] is called to mind [walsh2007rhetoric, p. 21–23]. [Note: R. Michael Young has written about a kind of “narrative contract” in interactive storytelling that works similarly to the cooperative principal [young1999notes, p. 164].] Of course, a shared pretense might also underpin the construal (by both author and audience) of nonfiction—this has been discussed in the context of myths and sacred texts [veyne1988did,sternberg1987poetics]. Scholars beyond the ones I have named have also taken up the project of characterizing fiction according to a notion of speech acts [iser1975reality,pratt1977toward,currie1985fiction]. In a paper preceding Searle's, Richard Ohmann highlights the reader's active role in the shared pretense:

> by providing the reader with impaired and incomplete speech acts which he completes by supplying the appropriate circumstances [...] the literary work is asking him to participate in the imaginative construction of a world—or at least as much of one as is necessary to give the speech acts an adequate setting [ohmann1971speech, p. 17]

 Along these lines, Dorrit Cohn notes how the fiction writer must do the work of laying *signposts* that call attention to the fictionality of her work [cohn2000distinction]. Relatedly, scholars have reported how similar measures may be taken to call attention to the (ostensible) objectivity of a work. In a seminal paper on the craft of newspaper journalism, Gaye Tuchman introduces the notion of *objectivity as a strategic ritual*:

> the correct handling of a story, that is, the use of certain procedures discernible to the news consumer, protects the newspaperman from the risks of his trade, including critics. [tuchman1972objectivity, p. 661]

 Of course, Tuchman's idea is not unique to news stories, and indeed we might think of all nonfiction as being constituted in a kind of strategic ritual carried out by its author according to social and cultural contexts. [Note: This notion appears to be very general. Citing Eugene Weinstein [weinstein1966toward], Tuchman writes this also: “Weinstein (1966) speaks of ‘credibility enhancing tactics' and suggests (personal communication, 1971) tactics connoting ‘objectivity,' such as quoting others or presenting alternative possibilities by citing negative evidence or conflicting opinion, may be used to enhance credibility in interpersonal communication” [tuchman1972objectivity, p. 661]. Such tactics are clearly also germane to academic writing—in fact, I am employing them right now, in this very section (and in this very footnote).]

#### A Fictionality Spectrum

To summarize, scholars have demarcated fiction from nonfiction in three distinctive ways. Caring exclusively about semantic concerns, logicians define nonfiction as having actual referents and true propositional content. Alternatively, formalists seek to identify the endemic stylistic tropes of fiction, such as direct access to character subjectivity and the communication of mental states in a third-person voice; nonfiction, then, could be characterized by the absence of such features. Finally, proponents of a pragmatist approach might argue that nonfiction is produced when nonfiction is intended and nonfiction is received. Again, I refer to these as the *semantic*, *formalist*, and *pragmatist* perspectives, respectively. In the first view, nonfiction depends on the factuality (truth value) of the story, whereas in the latter accounts nonfiction is a *genre*—with stylistic tropes and or a communicative contract between author and audience—in which case factuality may be more or less orthogonal to fictionality.

Note that these schemes do not necessarily depend on binary criteria for nonfiction, which means fictionality may be conceived as a matter of degree. Indeed, several scholars have explicitly stated this [searle1975logical,fish1976things,nielsen2015ten]. Marie-Laure Ryan describes such gradation in terms of a two-dimensional plane whose axes pertain to the frequency of referencing an extra-textual world and the implied guarantees about the truth of story content [ryan1997postmodernism, pp. 179–183]. That paper is in opposition to the postmodernist demotion of nonfiction to fiction, outlined above, which Ryan calls ‘panfictionism'. To Ryan, this “doctrine” was successful in bringing about a kind of *paradigm shift* [kuhn1962structure] in our conception of nonfiction, but as a totalizing account it does some disservice—here is her lucid articulation:

> The doctrine of panfictionality took us out of the complacency with which we have long approached nonfiction. If structure, tropes, conventions, rhetorical devices and the idiosyncrasies of the medium produce meaning in fiction, so do they in nonfiction. Until the doctrine of panfictionality came along, nonfictional genres of discourse enjoyed an illicit immunity from textual or semiotic forms of investigation. Thanks to the postmodern critique, we are now better aware that texts of nonfiction display an image distinct from their reference world, and that the construction of this image deserves consideration. [...] But the postmodern attack on the dichotomy loses most of its thrust if we give up the simplistic equation of nonfiction with truth and fiction with non-truth. [...] Under the present account, extending the definition of fiction to all discourse would have the negative effect of protecting nonfiction from what I believe are vital forms of inquiry. If fiction has a performative value, it is inappropriate to ask such questions as: “How does the author know?” “How was the data collected and what principles led to its selection?” “Is the interpretation justified by the data?” On the other hand, it is necessary to submit historiography, scientific discourse, literary criticism, or ethnographic writing to this type of scrutiny, just as it is necessary to ask of these genres the question raised by postmodernism: “What kind of paradigm and tacit assumptions led from this data to this interpretation?” [ryan1997postmodernism, p. 180]

 Indeed, it is this nuanced account of fictionality that best characterizes the departure of these scholars from the postmodernist totalizing account of all human endeavor as fiction. [Note: As some have noted, such a totalizing view can have dangerous repercussions: when everything is fiction, no one is beholden to any notion of historical truth, and this may be seen as enabling projects like Holocaust denial [lipstadt1993denying,kakutani2006bending]. Others, however, argue that postmodernism actually provides a methodology by which such efforts may be adequately confronted: by emphasizing the subjective aspects of historical work, postmodernists are uniquely equipped to expose nefarious aims in ostensibly sincere historical projects [eaglestone2001postmodernism].] In the rest of this section, I will rely on this kind of justification for the existence of a distinct thing that we call ‘nonfiction'.

#### Hayden White's Curation

Lastly, I would like to introduce some ideas from the late Hayden White that will prove foundational to my argument that emergent narrative works like nonfiction. [Note: Hayden White passed away in March 2018 here in Santa Cruz, where he was professor emeritus in the History of Consciousness department at UC Santa Cruz. I never met him, and neither did my coadvisors, which means his influence on my thinking in this dissertation is incidental to our living in the same town.] In his seminal *Metahistory* (1973), White characterizes the act of history writing as one of rhetorical curation [white2014metahistory]. While his arguments are specifically about historiography, I think it would be fair (and useful) to extend them to the broader phenomenon of nonfiction. [Note: One could argue that nonfiction is necessarily about the past. Even in cases where the present is being documented, by the time the documentation exists, that present has fallen into the past. In any event, this nuance is not important for my argument.] By emphasizing the constructive nature of historiography, White both destabilizes and augments the predominant and long-held conception of such endeavor, which White summarizes in this way:

> It is sometimes said that the aim of the historian is to explain the past by “finding,” “identifying,” or “uncovering” the “stories” that lie buried in chronicles; and that the difference between “history” and “fiction” resides in the fact that the historian “finds” his stories, whereas the fiction writer “invents” his. [white2014metahistory, p. 6]

This earlier view implies that narrative historical accounts exist in an historical record, fully intact, which makes historiography a simple matter of retrieval. To this, the scholar Arthur Danto has articulated a clever refutation in the form of what he call a *narrative sentence*, whereby a description of multiple events is loaded into a sentence about a single event. A canonical example is “The Thirty Years War began in 1618”, as Danto explains:

> “The Thirty Years War began in 1618” refers to the beginning and to the end of the war, but it is about the beginning of the war. On the assumption that the war was so-called because of its length, nobody could presumably describe it in 1618—or at any time before 1648—as the “Thirty Years War”. Of course someone might predict that the war would last just that long, and put sufficient confidence in his prediction actually to describe the war that way. But he would be making a claim on the future [danto1962narrative, p. 155]

 Danto posits this idea in the context of the *ideal chronicler*, an entity that is capable of maintaining a perfect historical record called the *ideal chronicle*:

> I now want to insert an Ideal Chronicler into my picture. He knows whatever happens the moment it happens, even in other minds. And he is to have the gift of instantaneous transcription: everything that happens across the whole forward rim of the Past is set down by him, as it happens, the way it happens. The resultant running account I shall term the Ideal Chronicle [danto1962narrative, p. 152]

 As Danto shows, the ideal chronicler cannot construct narrative sentences, because doing so requires foreknowledge of the future, but the entity is omniscient only with regard to the present. Given this, historical records can never contain narrative sentences—but narratologists from Roland Barthes [barthes1987rustle] to William Labov [labov1997narrative] show that storytellers tend to cue the causal relations between events. As such, the ramification of Danto's argument is that historical records cannot contain stories. Intriguingly, this suggests a purest distillation of historiography, and all nonfiction, as the matter of crafting narrative sentences.

Hayden White, partly through the influence of Danto [white2014metahistory, p. 275], does not distill, but rather expands these ideas to describe a more constructive procedure by which subject phenomena are captured in a raw historical record that is then *curated* to produce a narrative historical account with particular rhetorical features. [Note: To be clear, White does not himself use the term ‘curation' (or any other variant), but the procedure that he describes is in line with the sense of the term at work in this dissertation.] As White explains, this procedure is as much about sensemaking as it is about argumentation:

> [the procedure entails] processes of selection and arrangement of data from the *unprocessed historical record* in the interest of rendering that record more comprehensible to an *audience* of a particular kind. [white2014metahistory, p. 5]

 At the beginning of this procedure, the subject phenomena (constituting an *historical field*) are captured in a *chronicle* that is further curated into a *story*:

> First the elements in the historical field are organized into a chronicle by the arrangement of the events to be dealt with in the temporal order of their occurrence; then the chronicle is organized into a story by the further arrangement of the events into the components of a “spectacle” or process of happening, which is thought to possess a discernible beginning, middle, and end. [white2014metahistory, p. 5]

 In addition to rearranging and filtering the events in a chronicle, constructing a story requires the historian to charge that selected material with rhetorical dynamic. White describes such functionality in terms of *inaugural motifs*, *transitional motifs*, and *terminating motifs*:

> This *transformation of chronicle into story* is effected by the characterization of some events in the chronicle in terms of inaugural motifs, of others in terms of terminating motifs, and of yet others in terms of transitional motifs. An event which is simply reported as having happened at a certain time and place is transformed into an inaugurating event by its characterization as such: “The king went to Westminster on June 3, 1321. There the fateful meeting occurred between the king and the man who was ultimately to challenge him for his throne, though at the time the two men appeared to be destined to become the best of friends....” A transitional motif, on the other hand, signals to the reader to hold his expectations about the significance of the events contained in it in abeyance until some terminating motif has been provided: “While the king was journeying to Westminster, he was informed by his advisers that his enemies awaited him there, and that the prospects of a settlement advantageous to the crown were meager.” A terminating motif indicates the apparent end or resolution of a process or situation of tension: “On April 6, 1333, the Battle of Balybourne was fought. The forces of the king were victorious, the rebels routed.” [white2014metahistory, p. 5]

Thus, it is the distinctly narrative quality of the story—marked by the intentional selection of material and the attachment of rhetorical functionality to that selected material—that distinguishes it from the chronicle:

> When a given set of events has been motifically encoded, the reader has been provided with a story; [...] Historical *stories* trace the sequence of events that lead from inaugurations to (provisional) terminations of social and cultural processes in a way that *chronicles* are not required to do. Chronicles are, strictly speaking, open-ended. In principle they have no *inaugurations*; they simply “begin” when the chronicler starts recording events. And they have no culminations or resolutions; they can go on indefinitely. Stories, however, have a discernible form (even when that form is an image of a state of chaos) which marks off the events contained in them from the other events that might appear in a comprehensive chronicle of the years covered in their unfoldings. [...] In the chronicle, [an] event is simply “there” as an element of a series; it does not “function” as a story element. The historian arranges the events in the chronicle into a hierarchy of significance by assigning events different functions as story elements in such a way as to disclose the formal coherence of a whole set of events considered as a comprehensible process with a discernible beginning, middle, and end. [white2014metahistory, pp. 5–6]

White's curation procedure entails three additional steps, which correspond to increasingly higher-order modalities of explanation. In the first, which White calls *emplotment*, a story is cast in the mode (‘emplotment') of an archetypical genre (romance, tragedy, comedy or satire). The choice of emplotment affords a particular kind of argumentation, and the afforded argumentation supports a particular ideological position. Finally, by taking this ideological stance, the historian, through the curated history, prescribes a social praxis that the reader may adopt to navigate her present epoch. These final steps in the derivation of an historical account, though interesting, are not particularly relevant to my argument, as I will explain at the end of this section.

To summarize, White describes a procedure by which subject phenomena (an historical field) are captured in a chronicle (an exhaustive historical record) which is then curated into a story that is finally embedded into an historical account with particular rhetorical features. [Note: In a later book, *The Content of the Form*, White distinguishes between the chronicle and another primitive to the historical account, the *annal* [white2009content, p. 16]. Relative to the chronicle, the annal has even less narrative structure, and tends to be sparer in its event recording.] As White states, this curatorial process is precisely what separates historical writing (all nonfiction) from fiction:

> Unlike literary fictions, such as the novel, historical works are made up of events that exist outside the consciousness of the writer. The events reported in a novel can be invented in a way that they cannot be (or are not supposed to be) in a history. This makes it difficult to distinguish between the chronicle of events and the story being told in a literary fiction. In a sense, the “story” being told in a novel such as Mann's *Buddenbrooks* is indistinguishable from the “chronicle” of events reported in the work, even tough we can distinguish between the “chronicle-story” and the “plot” (which is that of an Ironic Tragedy). Unlike the novelist, the historian confronts a veritable chaos of events *already constituted*, out of which he must choose the elements of the story he would tell. [white2014metahistory, p. 6]

By now the connection of White's ideas to my claim in this section, and to the larger framework of this dissertation, should be clear: curationist emergent narrative is constituted in the same curatorial process that White attributes to historiography, but instead of taking real-world phenomena as its subject material, it takes simulated phenomena.

Before moving on, I should mention that others have made the connection between Hayden White and procedural narrative [anderson2007past,boluk2013dwarven,sarnek2015chaptertwo,czarnota2016eve]. Here, of particular relevance is a fabulous paper (that I will reference several more times) in which Stephanie Boluk and Patrick LeMieux argue that some of *Dwarf Fortress*'s [dwarf] generated material resembles a Whitean chronicle:

> *Dwarf Fortress*'s linguistic forms of historical inscription in Legends mode bear a striking formal and thematic resemblance to early forms of writing, such as the medieval annal and chronicle. [boluk2013dwarven, p. 144]

I will now proceed to argue that (curationist) emergent narrative works like nonfiction by discussing an example of the form in terms of the definitions of nonfiction that I have provided so far.

#### Case Study: *Oilfurnace *
Let us finally return to the thesis of this section: emergent narrative works like nonfiction. [Note: My argument in this section was bolstered by conversations with Max Kreminski and Jason Grinblat, who each take a similar stance on these issues.] To ground this discussion, we will consider the case of *Oilfurnace* (2010), a narrative comic by Tim Denee that recounts the emergent events of a *Dwarf Fortress* [dwarf] gameplay session. The comic was commissioned by the Australian magazine *PC PowerPlay* and appeared in its June 2010 issue [denee2010oilfurnace]; an excerpt is shown in Figure 3.1. [Note: The entire comic is freely available online at https://www.timdenee.com/oilfurnace [denee2010oilfurnaceweb]. Prior to *Oilfurnace*, Denee produced a similar comic called *Bronzemurdered* [denee2010bronzemurder], and he is also the illustrator of *Getting Started With Dwarf Fortress* [tyson2012getting]. Many more stories recounting *Dwarf Fortress* gameplay are available on the website *Dwarf Fortress Stories* [dfstories], which is curated by Taran and Trevor Van Hemert.] *Oilfurnace* recounts a *Dwarf Fortress* gameplay session that Denee carried out over the course of two or three months [mastrapa2010kiwi], and as such it is clearly a work of emergent narrative—and more particularly of curationist emergent narrative, as I will emphasize in Chapter 5.

[Figure omitted]

*Oilfurnace* begins by explaining that the titular Oilfurnace is a dwarven colony established on a place called Dread Island, a barren isle in the region of The Fatal Forests, by the decree of a dwarven king in the (dwarven) year 1050. The colonists are The Fatal Forests Trading Company, a troupe of seven individuals of varying specializations—four are miners, the others are multiskilled—who are led by a character named Ast Akrulikal. Prior to Oilfurnace, four earlier colonies had been attempted at the site, but each ended in disaster: zombie gorillas, thirst and starvation, zombie elephants, and flood. [Note: This narration is an allusion to the emergent narrative of earlier gameplay sessions carried out by Denee in the same simulated storyworld.] From this background, the comic proceeds to tell the story of the colony in terms of four distinct periods in its history (each taking up its own page): The Foundation, The Rise, The Golden Age, and The Fall.

In Denee's narration of the fist period, we learn of the foundation of the colony through an ordered series of discrete events that are recounted in comic panels that each couple an illustration and a sentence or two of descriptive natural language. These pertain primarily to the successful construction of an elaborate shaft allowing access to the rock layer below an aquifer that had brought the demise of two of the earlier colonies (in one it could not be traversed, in the other it flooded). The narration of this period culminates in a miner discovering platinum in the rock layer below the aquifer.

At the beginning of the next period, the dwarves dig into the dry stone beneath the aquifer and carve out a subterranean fortress. From here, industry proceeds—trees are harvested, ore is mined, coal is burnt, metal is smelted—and new migrants arrive at the island fortress by boat. An invasion of undead slugmen and snailmen challenges the fortress, but a militia comprising Ast Akrulikal and a dwarf dubbed ‘Batdwarf' (he wears a robe made of bat leather) swiftly disposes of them. More migrants come and so do more invaders, but the militia grows in turn and these challenges are also overcome. A period of starvation ensues, but this is solved by the development of a series of cave farms in which mushrooms are cultivated. At this point, precious metals abound in the fortress and a first mayor is elected. As a final hardship in this period, troglodytes begin to emerge out of the fortress mine shafts, but a series of traps are constructed accordingly. Finally, a shaft is dug deep, into a magma layer in “the bowels of the earth”, and the dwarves use this as a garbage disposal. The period culminates, as before, in the discovery of a precious mineral—this time, miners encounter adamantine at the bottom of the magma trash chute.

From here, Denee proceeds to narrate The Golden Age of Oilfurnace. On this page, the explanation is more visual than narrational, illustrating the layout of the fortress and concerns such as the processes that characterize its steel and adamantine industries (which emerge from the game's *crafting system* [grow2017crafting]). Additionally, we learn that the six surviving members of the previous period's militia have each attained the status of ‘legendary killer'—this cohort is now dubbed the ‘zombie slayers', as they deal swiftly with a series of zombie hordes. A new generation of dwarves is growing up in the fortress, the population is booming, and an arena for entertainment and training has even been constructed. A foreboding panel at the bottom of the page, however, notes the presence of five *forgotten beasts* [tyson2012getting, p. 172] in the magma layer below the fortress.

Finally, Denee narrates the closing period in the history of the colony: The Fall of Oilfurnace. Here, he returns to a predominantly narrational style, with ordered event panels; this page is shown in Figure 3.1. [Note: Note that the temporal sequence is made explicit by numbering each panel; this device was also used on the second page.] The period begins with the fortress mayor demanding, hubristically, that coins be minted out of pure adamantine, which requires miners to dig near the deepest parts of the magma sea below the fortress. Unfortunately, this opens a gate to the *underworld* [dfunderworld]. From here, the narration tightly reconstructs a rapid series of events that begins with a group of demons pouring out of the gateway to meet the zombie slayers. The demons, Denee explains, are too powerful and too numerous: members of the militia perish, and Batdwarf is forced to sacrifice himself to allow Ast Akrulikal the time and space to retreat from the depths. Miners are ordered to breach the aquifer (to drown the demons), but they panic or die before this can be done. An alarm bell sounds to call for evacuation, and Ast (with no weapon and a broken arm) and six other survivors manage to escape and close the fortress gate behind them. With the dwarves now outside and helpless, a band of zombie slugmen emerge from the dead woods of Dread Island to descend upon them. Only Ast survives the onslaught—“he punches a slugman through the head”—and he retreats to a secure building in the woods. In this secret keep, Ast is shown reading a letter from the dwarven king that reads *Remember: losing is fun* (panels 17–18)—this is the *Dwarf Fortress* community's favorite slogan [tyson2012getting, p. 2]—and finally he pulls a “self-destruct lever”, thereby locking himself safely (but permanently) inside. In a final panel, Ast is shown rotted to bones in the keep, and Denee concludes the tale:

> The entrance to the fortress is blocked by pulling the self-destruct lever. With nothing waiting for him outside the lever-keep, Ast waits for death inside, away from the harsh glare of the overbright. If a dwarf cannot die underground, at least let him die with a roof over his head. A noble end to an ignoble fortress; it is all any dwarf can hope for.

#### *Oilfurnace as Nonfiction*
I contend that *Oilfurnace* is a work of nonfiction (or at least works like one). To explore this claim, let us treat the descriptions of nonfiction given above as classification schemes: if they can be applied to *Oilfurnace*, then its status as nonfiction is supported, and from here it is reasonable to extrapolate this characterization to other works of emergent narrative.

First, it is prudent to consider the objections to the very notion of nonfiction that I have outlined above. If Gorgias, taken seriously, is correct, then nothing exists: there is no *Oilfurnace*, no dissertation, no me, no you. If one allows for solipsism, then still I cannot be sure that *Oilfurnace* exists and neither can you. Moving beyond these troubles, we encounter the postmodernist assassination of objectivity. In the case of the artifact at hand, this perspective might lead us to maintain that, in constructing a comic that recounts the emergent events of his *Dwarf Fortress* gameplay session, Tim Denee has produced a work of fiction, because it impossible to produce an objective account of anything. [Note: Barthes, in a similar vein, might specifically argue that the sign system of the comic has as its signifieds not the actual characters and material of the Oilfurnace colony and its larger simulated world, but rather simulacra of those entities. That is, the denotational procedure constituted in the composition of any human communication, including a comic, makes objectivity impossible.] This is a reasonable view that I accept, but like the various other scholars who were discussed above, I think it is also reasonable to say that objectivity is not a necessary feature of nonfiction. That is, there is an identifiable genre called nonfiction, and it persists in spite of the impossibility of human objectivity. This, again, is the sense of ‘nonfiction' that I appeal to in making this argument, and so the postmodern critique is not a showstopper here. [Note: To be clear, this is what I mean when I say that emergent narrative ‘works like' nonfiction.]

Before moving on, let us consider the postmodernist critique as it would apply to emergent narrative as it more commonly appears: an unfiltered stream of phenomena that transpires over the course of an interaction with a simulation. In Chapter 4, I will build on Espen Aarseth's critique of this approach to emergent narrative (my view: narrative only obtains when the raw stream is curated), but nonetheless it is currently the predominant conception of the form. As such, it is interesting to ponder: are these streams of raw emergent narrative objective? It turns out, I think, that this consideration does not really compute, since it is like asking whether the raw stream of reality is objective—we can only know it through a subjective experience of it, and the artifact of our encounter (whether that be a mental understanding or a work of media) is thus always subjective. In the case of a raw stream of simulated material, we might say that the narrative obtains in the mind of the interactor as the stream transpires, but then the result is clearly subjective. Really, this consideration does not make much sense because it requires us to ignore that such streams lack the narrative structure that characterizes the fiction and ‘nonfiction' artifacts that are the subject of the postmodernist critique (which means we cannot proceed in good faith). This conceptual trouble is an expression of what I view as the fundamental issue with emergent narrative: raw event streams are not narrative, but for some reason they have been viewed as such in the particular case of simulation.

#### Semantic Perspective: *Oilfurnance 's World*
Next, let us consider the *semantic perspective* on fictionality. According to the view articulated by David Lewis, nonfiction obtains when (or to the degree that) a narrative asserts true propositions about real referents. Who are the referents of *Oilfurnace*'s narration? They include characters such as Ast Akrulikal, Batdwarf, and other named dwarves; unnamed characters including the various zombies and forgotten beasts; locations such as The Fatal Forests, Dread Island, the Oilfurnace colony and fortress, and components of the fortress; the named earlier attempts at colonizing Dread Island; and many more. Logical propositions about these entities, then, are asserted through the various storytelling mechanisms employed in the comic: visual images, natural language prose, panel arrangements, and any other technique one might encounter, for instance, in the poetics of sequential art proffered by Will Eisner [eisner1985comics] or Scott McCloud [mccloud1993understanding] or others [jhala2017discourse]. So, are these real referents and true propositions?

This line of thinking ultimately converges on a discussion of the ontology of simulation, and in particular whether simulated entities and events are realer than their counterparts that obtain in conventional fiction. One might contend that a simulated entity is not real because it is virtual, not physical. To this, another might counter that the simulated entity *does* indeed have a physical constitution, one that obtains in the material of the computing machine. This is probably not a great argument, however, because that physical constitution will vary across runtime instances, even if the virtual entity remains identical (by some policy for determining isomoprhism).

Alternatively, one could say a simulated entity is not real because it is a simulacrum. But if Ast Akrulikal is a simulacrum, what is the original entity that it represents? Due to the use of world generation in *Dwarf Fortress*—i.e., due to its *roguelikeness* [garda2013neo]—Ast Akrulikal only exists in the storyworld that is associated with the particular world seed and software version that Tim Denee used in his gameplay session. Thus, Ast Akrulikal is no simulacrum: he is not a human being, or even a being of any sort, but he is something of a particular kind—a *Dwarf Fortress* dwarf—and he exists in his world at the full fidelity of existing as that kind. [Note: More specifically, this kind is a *Dwarf Fortress* dwarf in the particular software version that Denee executed. When Tarn Adams changes game code that affects the representation or behavior of dwarves (or of any part of the gameworld that affects dwarves in any way), then the ontology of the *Dwarf Fortress* dwarf changes. This means that Ast Akrulikal and a dwarf of another version of the game may be of different kinds. As Ian Horswill noted in his feedback on an earlier draft of this thesis, the treatment of simulated entities as being real raises peculiar ethical questions: “If DF dwarves are real, what are the ethics of killing them?” (personal communication, July 28, 2018). I will not dive into this issue here, but for discussion of it with regard to *The Sims* [maxis2000sims], for instance, see this paper by Juyun Kim and Stephen Petrina [kim2006artificial].] That is, he could not be more perfectly modeled, just like you could not be more perfectly modeled—you are the perfect representation of yourself, because you are yourself, and the same is true of this character.

One might argue that this does not hold for characters in conventional fiction: the Sherlock Holmes that one encounters in print, for example, is a lossy representation of the pure Sherlock Holmes that existed in the mind of Sir Arthur Conan Doyle, making the former a simulacrum of the latter. [Note: Interestingly, since Doyle is dead, one could state that the original is gone, and all that is left is the copy. This calls to mind Jean Baudrillard's writing, in *Simulacra and Simulation* [baudrillard1994simulacra], about copies with no (extant) originals. As Noah Wardrip-Fruin noted in his feedback on an earlier draft of this thesis, this argument is made tenuous by the shift in literary studies away from the author (e.g., [barthes1967death]). Indeed, the argument will fall apart by the end of this paragraph, but I think the idea is still worth mentioning.] Intriguingly, this situation only obtains in roguelike procedural narrative—an approach often taken in works of emergent narrative—since authored characters still appear in other kinds of computational media. For instance, my claim about Sherlock Holmes could be applied to the *Facade* character Trip [mateas2005structuring], though now the original(s) would exist in two minds (belonging respectively to co-creators Michael Mateas and Andrew Stern). In this way, the argument begins to fall apart. Moreover, even in the case of Ast Akrulikal, another player could access the same storyworld by using the same seed and software version (or Denee himself could, for a second time), and assuming divergent gameplay inputs, the world would proceed differently and Ast Akrulikal himself would likely change as well. Generally, I think it is probably futile to make ontological claims about the realness of simulated entities (and even if realness could be established, issues with objectivity could prevent direct access to the true forms). [Note: Again, Barthes's reference illusion comes to mind: perhaps Ast Akrulikal is merely a kind of signified with no referent. I will note also that others have made ontological claims about expressive simulations. Cameron Kunzelman postulates, evocatively, that the system of a videogame constitutes a real, living body that humans live both with and within [kunzelman2014nonhuman]. Several thinkers have suggested that the gameworlds of massively multiplayer games like *Eve Online* are real worlds [carter2016internet], though this idea is bolstered by fact that many human interactors pilot characters in its multiplayer networked storyworld simulation. In a distinct but related vein, Stephanie Boluk and Patrick LeMieux describe simulated storyworlds as monuments to the real mechanisms that produce them: “Despite its minimal textual interface, the process of generating this history weighs heavily on the central processing units of most computers. The millions of events logged during world generation are granular enough not only to correspond to the history of the gamespace represented on-screen but also to ultimately historicize the processor cycles of the computer itself” [boluk2013dwarven, p. 126].]

Instead of considering the ontology of simulated entities, one might explore an intuition: simulated events seem to *happen* in a way that authored fictional events do not. That is, the stuff of emergent narrative is material that is not authored, but instead emerges out of the complex interaction of authored processes. I am not sure how to construct a formal semantic account of this intuition, since the representation of true propositions referring to nonexistent entities would require a peculiar logic. Instead, I will contend that this intuition, and the feeling that procedurally generated entities are somehow realer than authored fictional characters, supports a loose idea that emergent narrative works like nonfiction. Whether this is true or not, the notion, when held by an author or reader, supports a particular set of aesthetics—indeed, ontology may be orthogonal to aesthetics, since the aesthetics of a work for an individual depends on her conception of that work. [Note: In my project *Sheldon County*, discussed in Chapter 12, I specifically frame the media artifact as a work of nonfiction (both in the work itself and in supporting materials). Whether that is true or not, it is how I conceive of the work, and by framing it accordingly, listeners will be led to also view it in this way.] In Section 3.2, I discuss the aesthetics supported by this intuition.

While I find this intuition alone to be powerful, I will now provide a concise logical account of *Oilfurnace* as nonfiction that is rooted in John Heintz's identified features of fictional worlds. As outlined above, Heintz describes fictional worlds as being logically incomplete and logically inconsistent, and thereby he implies a definition of nonfiction as narrative that recounts complete and consistent worlds.

Due to the nature of computer simulation, the simulated storyworlds of emergent narrative are in fact complete and consistent. As an example, let us consider the case of the *Dwarf Fortress* world that *Oilfurnace* recounts—specifically, let us discuss logical *decidability* (whether and how one can identify truths and nontruths) with regard to this world. As the world obtains through the execution of the game's software, a set of facts about the world will be asserted as an unavoidable byproduct of that execution. These facts will pertain to concerns such as entity attributes, the order of simulated events, details of the physical world model, and much more. Some of these facts will be stored in computer memory, which means propositions about the storyworld may be evaluated by querying against this data. As such, the notion of a truth (pertaining to the simulated storyworld) may be operationalized as a fact that can be queried during execution of the game's software (of course, with the seed and software version that indexes that world being used). [Note: As Ian Horswill noted in his feedback on an earlier draft of this thesis, in the *Versu* project Richard Evans and Emily Short very deliberately model the storyworld as a set of facts (asserted in a modal logic). As they argue in a paper on the project, this design method produces a number of authorial and architectural affordances, namely the ability to “find out what is true” [evans2014versu, p. 118].] Note that some truths about the world will not be recorded as persistent facts that are stored in data, because it is not feasible to record all data about the gameworld—we might call these *ephemeral facts*. Nonetheless, even an ephemeral fact qualifies as constituting a truth under this operationalization scheme, because the fact could have been queried in the brief moment that the pertinent data existed in memory. As such, all truths about the storyworld will be rendered in asserted facts as the simulation proceeds. That is, with regard to a simulated timestep $t$, all truths about the world that hold through $t$ will have been asserted in the execution to that point, and no truth that holds at $t$ will not have been asserted.

Still, there is an outstanding issue that pertains to the notion of decidability in the *Oilfurnace* storyworld: we may wish to evaluate certain propositions for which corresponding queries cannot be formed. This would occur when an attempt is made to evaluate a proposition that depends on something that is not modeled in the storyworld. For example, to adapt a predicate that Daniel Dennett has applied to Sherlock Holmes [dennett1992self], we might wish to ask whether Ast Akrulikal has a mole on his left shoulder blade. While *Dwarf Fortress* is famous for its detailed modeling (e.g., of organ tissue [dftissue]), I do not believe such a query could be formed, since the simulation does not represent moles and thus the software would not “understand” the query. More precisely, ‘false‘ would not be returned—as it would be if a query asked whether Ast's name is ‘Batdwarf', for instance—but rather some kind of error would occur, since the query would not be well-formed with regard to the game's data representation. One might call the propositions corresponding to such queries *undecided*, which would mean that the *law of bivalence* (all propositions are either true or false) does not hold for *Dwarf Fortress* storyworlds, which would mean that they are incomplete. [Note: To be fair, it is not clear that this law holds for the real world either [van1973semantic,misak2004peirce], as Ian Horswill noted in his thesis feedback.]

Alternatively, however, it is tempting to say that in such cases Ast does *not* have a mole, because such a mole is not modeled—that is, if a query cannot be formed, the negation of the corresponding proposition is implied. Is this fair, though? One could just as well say that anything about Sherlock Holmes that is not explicitly stated in the text of Doyle's stories is false. To me, however, it feels more intuitive to allow this in the case of a simulated world whose ontology is explicitly represented as structured data. Additionally, while Doyle could have written a new installment about the mystery of the mole on Holmes's left shoulder blade, thereby rendering the proposition's truth value decided, this cannot be done in the case of *Dwarf Fortress*: when the code that produces a given storyworld is changed, the world becomes inaccessible. [Note: I admit that it would be possible for Tarn Adams to alter the code such that the storyworld remains accessible and intact, except for the advent of moles, through a clever engineering of the world-seeding functionality. Indeed, Adams spends considerable effort on ensuring backward compatibility in the game, which means storyworlds from old versions (stored as save files) can be used in newer versions [dfsaves]. That being said, what I am attempting to elicit here is a particular intuition: simulated storyworlds feel like closed systems, and fictional worlds less so.] Thus, I contend that the world of *Oilfurnace* is complete, or at least that it *feels* complete: even if a detailed consideration reveals that the law of bivalence does not hold in computer simulation, it *feels* like it does. This feeling, which I will continue to validate throughout the rest of the section, is the bedrock that supports the notion that emergent narrative works like nonfiction. Even if one takes the ontological position that simulated storyworlds are not consistent, they still feel consistent, and this supports a particular set of aesthetics—these are the aesthetics of emergent narrative, which I outline in Section 3.2. [Note: As I have already explained above, my argument here is fundamentally aesthetical in nature, rather than ontological or even philosophical: I am interested, foremost, in how emergent narrative feels how it feels, and why humans like that feeling.]

Due to the nature of simulation and of computation more broadly, the simulated storyworld of *Oilfurnace* is also *consistent*. Recall that Heintz characterizes fictional worlds as being inconsistent when incompatible propositions are asserted—for instance, *Anna Karenina* implies both that the events of its story begin on a Monday and on a Tuesday. At a certain level of detail, this kind of inconsistency is impossible in computing, since it may be seen as requiring incompatible data to be stored in the same memory locations. [Note: This kind of *superposition* is actually possible in *quantum computing* [gruska1999quantum], but the ontological implications of that paradigm extend far beyond our concerns here.] This points to a fundamental difference between a fictional world and a simulated world, which is that the latter emerges from a system of laws (the mechanics of the simulation), whereas the former obtains through a kind of human artifice. Of course, in simulation a human may still author the system of laws, but upon being authored those laws constitute a closed system that is consistent. It may not be consistent with the real world or whatever it may be modeling, but it is nonetheless an internally consistent system that is complete with regard to itself. [Note: I should mention a counterexample here. In his independent and collaborative doctoral work [robertson2014finding,robertson2016model,robertson2017directing,robertson2018perceptual], Justus Robertson has explored how incompatible assertions about a storyworld can be maintained as equally valid, using a metaphor of *superposition*. Specifically, a superposition (set of competing assertions) may collapse (onto a sole assertion) in accordance with a human observer's experience of the world. For example, an observer might encounter two doors, one on the left and the other on the right, whose respective connecting rooms have not been decided by the system. As the observer enters the left door, a connecting room is decided for it, and from now on the door on the left will always connect to that room, while the door on the right will not. Alternatively, had the observer entered the right door, *it* would have connected to that room and the left would not have. The ontological and aesthetic ramifications of Robertson's approach are fascinating and worthy of extensive discussion, though unfortunately that is also beyond the scope of this study.] Thus, when it comes to consistency, human invention is mutable, while simulation tends not to be. [Note: In his notes on an earlier draft of this thesis, Ian Horswill identified an important exception here: “there are lots of cases of games in which there is a detailed simulation, but there's something that appears in the game that's referred to, either in the dialog or in the visuals, that isn't modeled in the world” (personal communication, July 28, 2018). This is a great point, and one that further erodes the basis for any argument that simulated storyworlds are real in an ontological sense. As I will explain more at various points below, my ultimate aim here is not to make ontological claims, but rather to argue in support of an artistic position: regardless of their true ontology, simulated storyworlds *feel* real, and this produces a distinct pleasure.]

It is worth addressing the potential counterclaim that these properties of completeness and consistency, to the degree that they are based in ontology, in fact apply to all works of computational narrative (i.e., even ones outside the paradigm of emergent narrative). Indeed, in each case a storyworld will obtain through the execution of a computer program. However, the primary difference, I contend, originates in the hallmark of emergent narrative: it is driven bottom-up by simulation, not top-down by narrative. In emergent narrative, the rules of the simulation are the laws that govern the storyworld itself, while in conventional computational narrative the rules of the simulation govern the narrative, the instantiation of which may *suggest* a larger storyworld in the same way that the prose of print fiction suggests a larger storyworld. For instance, consider the case of *Facade* [mateas2003faccade]. There is a simulation at work—it models the physical environment of the apartment, the hallway outside it, and apparently the balcony outside [old2017facade]—but it does not model the entirety of the storyworld that the unfolding narrative implies. For instance, past events (such as a vacation in Italy) are implied, but these events are not actually simulated. The history of the world of *Oilfurnace*, on the other hand, was actually simulated prior to the beginning of gameplay—this is *Dwarf Fortress*'s famous *world generation* procedure [hall2014dwarf,martin2012landscape]. While all simulations have gaps (see Section 4.1.5) and imply extraneous material, this occurs less frequently as the level of simulation increases, and the hallmark of emergent narrative is intensive world simulation. As such, the storyworlds of emergent narrative feel more complete and more consistent, and so emergent narrative works more like nonfiction.

#### Formalist Perspective: *Oilfurnace 's Tropes*
As I outlined above, the *formalist perspective* on fictionality ignores the logical status of narrative content to instead focus on the stylistic tropes that characterize works of fiction. Identified tropes include a means of entering the subjective worlds of characters and surfacing the internal concerns found therein, often in a third-person voice. Intriguingly, the simulation of characters in a storyworld may easily support this phenomena: a system containing such a simulation will be omniscient with regard to it, in the sense that any truth about the simulation may be accessed through querying (as discussed above). Indeed, in my own work, I am deeply concerned with modeling of the internal lives of synthetic characters, and ultimately my goal is to surface interesting internal phenomena that emerges through such modeling. Does this make emergent narrative more like fiction, then? Not necessarily, I contend. First of all, nonfiction, as a genre of writing, is rarely divorced from the internal worlds of its characters (real individuals). Even beyond *new journalism* [wolfe1973new], [Note: In the midst of writing this paragraph, I learned that Tom Wolfe, the figurehead of the new journalism movement, died today. Rest in peace.] classical ‘objective' nonfiction writes about the internal worlds of individuals, sometimes as a byproduct of indirect access (e.g., through someone's diary writings), but often through a rational postulation. In any event, the only impediment to this direct accessing for the nonfiction writer is her lack of omniscience. But if an omniscient being wrote nonfiction that delved with perfect accuracy into the mind of a real individual, we would not say that such access turns that account into fiction.

As such, though the tropes of fiction identified by this formalist view may apply to works of emergent narrative, this does not mean they do not work like nonfiction. Moreover, through its characteristic omniscient access to a storyworld, emergent narrative enables a unique brand of nonfiction that is in fact only made possible through something like computer simulation. Indeed, the very possibility of omniscience in emergent narrative may be viewed as an argument for why it works like nonfiction. Again, however, this is a feature that is also present in other works of computational narrative, though again to a lesser degree corresponding to the reduction in explicit modeling of a storyworld.

#### Pragmatist Perspective: *Oilfurnace 's Contract*
According to the *pragmatist perspective*, fictionality depends on the *communicative intent* of the author and a kind of social contract that is consummated between author and audience. That is, nonfiction obtains when an author intends to produce nonfiction and the audience agrees that the artifact composed thereby is a work of nonfiction. Thus, nonfiction is made possible by a *shared pretense* [tomasello2003makes]. Here, we might consider a pivotal factor in the interpretation of *Oilfurnace*: whether the audience is aware of *Dwarf Fortress* and the comic's relation to it. To the uninitiated, the comic *constructs* a narrative set arbitrarily in a fantasy world, while to the initiated it *recounts* a narrative that emerged out of the game's famous simulation. Indeed, while I have not encountered the comic except as someone who was already aware of *Dwarf Fortress*, I suggest that the comic, as a standalone artifact divorced from any understanding of the game, is a strange artifact. Its narrational style is peculiar, and the quality of the narrative is perhaps even dubious. If Tim Denee were to write a story from whole cloth, he might produce something more extravagant—the quality of this story, however, is dependent on an understanding that the events that it recounts *actually happened*, in the sense that they emerged out of the complex interaction of processes in a computer simulation. As I will express more thoroughly in the later sections of this chapter, a story that recounts actual events will tend to be inherently more interesting than a hypothetical fictional counterpoint with the same content. This is the basis for ‘based on a true story'—it is what produces the *pleasure of nonfiction*.

In the case of emergent narrative, this means that the stories generated by this approach receive a boost in interestingness *when the audience is made aware of the simulational origins of the content*. To the degree that the simulation is inhibited through interventionist techniques, such as drama management or narrative planning, this interestingness boost dissipates. [Note: I will expand on this argument in Chapter 5.] One might even say that procedural narrative works like nonfiction to a degree that is commensurate to this interestingness boost. It is critical, however, to realize that the boost is not actuated merely by the simulational origins of a generated story, but by a *contract* between author and audience that transacts interestingness in exchange for guarantees about the ‘actuality' of the recounted events—this might be called the *contract of emergent narrative*.

The primary takeaway here is that works of emergent narrative must do some work to emphasize the actuality of the recounted simulated events so that a contract, transacting such actuality in exchange for a boost in perceived interestingness, may be consummated between the author and the audience. This emphasis may be carried out in the content of the work itself, or in the *paratext* [genette1991introduction], meaning any external materials that frame interpretation of a work. In Chapters 10 and Chapter 12, I discuss how this contract is consummated in *Bad News* (primarily through its preliminaries and epilogue) and in *Sheldon County* (primarily through its paratext and the opening segment of each generated pilot episode).

#### Whitean Perspective: *Oilfurnace 's Curation*
Finally, let us view *Oilfurnace* through the lens of Hayden White's model of historiography as a procedure of curation. By this procedure, which entails three distinct phases, subject phenomena of interest to historians are eventually recounted in historical accounts with particular rhetorical features.

{First, the subject phenomena are captured, as they are transpiring, in a *chronicle* that records them by some method of inscription. While the chronicle itself is the result of a kind of curation, since it will never perfectly capture the subject phenomena (inscription is lossy), it is, for the historian's purposes, the raw historical record. That is, because it is the only documentation of the subject phenomena, it contains the only material that the historian may use to construct her account of that phenomena (though she of course may augment this, or adulterate it, with extraneous material). The chronicle is open-ended, with no narrative beginning (but rather the unceremonious onset of a process of recording that produced the record) and no narrative end (but instead an unceremonious termination of that recording process). From this chronicle, the historian crafts a *story*. This process entails the selection of a subset of the chronicled events, which may then be used to construct a discernible narrative structure in which some of selected events actuate *motifs*: *inaugural motifs* cue meaningful causal sequences that will culminate eventually with *terminating motifs*, and along the way *transitional motifs* signal abeyance in the causal sequence. Finally, the story is embedded in a particular *emplotment*—e.g., tragedy or comedy—which unlocks a set of rhetorical effects that support a targeted ideological stance.} [Note: As I have noted already, while White delves into considerably more detail as to the latter stages of the historiographic process [white2014metahistory], it is the initial steps that are of particular concern to our purposes here.]

Let us now consider analogues between the components of emergent narrative and White's notions of an historical field, chronicle, story, and emplotment. Again, we will consider emergent narrative through the concrete example of *Oilfurnace*.

First, what I have called the ‘subject phenomena' of a story, White terms the *historical field*. In *Oilfurnace*, this corresponds to the unfiltered stream of simulated phenomena that transpired during Denee's gameplay session. For the most part, this phenomena is ephemeral, since it would be infeasible (from the standpoint of computer memory) for the system to record the whole plenitude.

As for White's ‘chronicle', we might identify two analogues in *Oilfurnace*. First, whatever record keeping the system *does* manage to do (in terms of recording simulated phenomena) will result in a kind of chronicle (in the form of stored data). Second, Denee himself likely maintained a kind of chronicle in the form of notes or other inscriptions (whether physical or digital) recorded during the course of gameplay. [Note: We might also posit that Denee's own memory of the gameplay session is a kind of chronicle. I am not sure whether this idea would be in accord with White's thinking, but this may be due to the fact that he describes the particular case of historiography, and typically the historian is personally (and temporally) removed from her subject phenomena. As I articulate below, while I am interested in human-curated works like *Oilfurnace*, I am especially interested in the prospect of automated systems that procedurally curate simulated histories. This is what I am exploring now with my project *Sheldon County*, and as I explain in Chapter 12, I have found that such automatic curation is greatly assisted by the simulation maintaining extensive records of its generated phenomena. In this configuration, where one computer program is curating the material inscribed in a record kept by another computer program, we find an uncanny analogy to White's notion of the historian who constructs a story by curating a chronicle.] Note that the existence of multiple such records poses no trouble to this analogy, since for many historical concerns the notion of a chronicle takes the form of a composite record that combines information from multiple sources. Thus, we may say that in the case of *Oilfurnace*, these records combine together to form a composite chronicle.

Next, we may identify as an analogue to White's ‘story' the subset of material from this chronicle that Denee selected for expression in the comic. Here, we even find White's ‘motifs': an early panel reading “Strike the earth!” inaugurates the start of an extensive dig that terminates finally in the fatal demon invasion that was caused by the dwarves (hubristically) digging too deep. Along the way, a series of transitional motifs reference the discovery (through increased digging) of increasingly appealing minerals.

Lastly, White's ‘emplotment' mechanism appears in *Oilfurnace* through its evocation of the tragedy, which is signaled frequently: a sense of doom is instilled immediately through the initial reference to four failed earlier attempts at colonizing Dread Island; new challenges are encountered and overcome, but victory is always tenuous; ultimately a final challenge (the demon invasion) cannot be overcome, and everyone perishes; the king's emergency letter, opened by Ast as a last resort, reminds the reader that “losing is fun”. [Note: Indeed, this slogan and the tragic fatalism that it celebrates suggests that the emplotment of *Dwarf Fortress* gameplay itself is the tragedy.] Because there are different stakes to the rhetoric of historiography and to that of emergent narrative, White's articulated concerns at the level of emplotment and beyond become less relevant to our concerns here. [Note: This is not to say that it would be impossible to curate emergent narrative to particular rhetorical effect. Indeed, my initial dissertation topic [ryan2016situated] was a kind of story generation by which characters situated in a storyworld would curate their accumulated knowledge of that world (a kind of chronicle) to tell stories to one another to particular rhetorical effects. That is, it would be *pragmatic* story generation, whereby characters storytell as a way to change the world. Moreover, it is also certainly possible to do curationist emergent narrative in a way that would target a particular ideological stance (with regard to the real world, not just a simulated one). This is essentially one of the aims of my colleague Melanie Dickinson, who is exploring of use of social simulation for social justice [dickinson2017social]. Though not a work of emergent narrative, *Terminal Time* [mateas1999terminal,mateas2000generation,domike2003recombinant] is a remarkable example of procedural nonfiction that explicitly targets particular rhetorical effects. In the project, an AI system works to generate a documentary film that exaggerates an audience's collective stance on an issue to the level of a lampoon. Michael Mateas, who is one of its cocreators (along with Steffi Domike and Paul Vanouse) and also my coadvisor, recently told me that several individuals who experienced the piece remarked that it was essentially operationalizing the ideas of Hayden White.]

Clearly, analogies abound between White's model of historiography and the process by which *Oilfurnace* would have been created. While White's greater aim is to show that historiography is primarily a rhetorical practice, the process of curation that he describes signals a critical distinction between fiction and nonfiction. To emphasize this point, let us again consider White's articulation on the matter (which was already quoted earlier):

> Unlike literary fictions, such as the novel, historical works are made up of events that exist outside the consciousness of the writer. The events reported in a novel can be invented in a way that they cannot be (or are not supposed to be) in a history. This makes it difficult to distinguish between the chronicle of events and the story being told in a literary fiction. In a sense, the “story” being told in a novel such as Mann's *Buddenbrooks* is indistinguishable from the “chronicle” of events reported in the work, even tough we can distinguish between the “chronicle-story” and the “plot” (which is that of an Ironic Tragedy). Unlike the novelist, the historian confronts a veritable chaos of events *already constituted*, out of which he must choose the elements of the story he would tell. [white2014metahistory, p. 6]

As I have noted above, I think it is fair (and valuable) to consider White's model as a description not just of historiography, but of nonfiction more generally. In taking this broader view, we find an even stronger account of how emergent narrative works like nonfiction: in fiction and other forms of computational narrative, the subject matter of a story is instantiated *through the process of narration*, whereas in emergent narrative these events are *already constituted at the time of narration*. [Note: In his comments on an earlier draft of this thesis, Ian Horswill wrote: “This is a great model for Oilfurnace, but DF itself doesn't have a curator. So would you then say that a person just ‘playing' DF, and not curating a story about it for a third person, was not experiencing emergent narrative?” I address this matter extensively below, in multiple sections, so if you are wondering the same thing, hold tight or else jump to Section 5.3.2.] Note that this is especially the case in curationist emergent narrative, where effort is dedicated to the construction, through curation, of an actual narrative artifact. Indeed, I contend that this curationist approach does better than any other method for emergent narrative to unlock the pleasures of nonfiction and the aesthetics that result thereby. Let us conclude this section with a quote from Tim Denee, in which he himself describes the creative process that produced *Oilfurnace* as one that resembles the curation procedure described by Hayden White:

> I think Dwarf Fortress generates such complexity and depth of information, that the task of the storyteller is to simply sift through and find the narrative [mastrapa2010kiwi, n.p]

### Analogy to Stories of Lived Experience

In the fall of 2002, I was fifteen years old and deeply immersed in *Grand Theft Auto: Vice City* [vicecity]. The game was enjoyable to me not for its notorious content and violent gameplay, but for its simulation of a vibrant bustling city. Earlier open-world simulation games had typically been set in fantasy or science-fiction environments, but this game was set in a version of the real world, which appealed more to my tastes. [Note: For this reason, I often feel like an outcast among videogame players, developers, and scholars, but I have a comrade in Gonzala Frasca: “I find most fantasy-related videogames quite boring. I have always preferred stories about human affairs and social issues to magic spells and mean dragons. This is why I always salute any attempt to bring human characters to videogames” [frasca2001sims].] Most importantly, the game's emergent possibility space seemed marvelously vast. I felt that I could navigate this space in a constructive manner—like I was crafting a story through the way that I played. [Note: This idea of gameplay as a constructive act that produces a kind narrative is the crux of Aaron Reed's notion of *sculptural fiction* [reed2017changeful]. To the degree that such an act works like a collaboration between system and player, the work affording such interaction falls in the purview of Ben Samuel's *shared authorship* [samuel2016crafting]. Later on, I will discuss these ideas in more depth, since they relate to aspects of my framework for curationist emergent narrative. Interestingly, much of the earliest writing on interactive fiction viewed the reader as co-author [jennings1984word,armour1986interactive,costanzo1986reading,campbell1987interactive,andersen1989interactive,ziegfeld1989interactive,sloane1991interactive,graves1991automatic,graves1991plot] or “co-actor” [andersen1992narrative, p. 61] or “prosumer” (portmanteau of ‘producer' and ‘consumer') [schwartz1988student, p. 60]; these terms evoke Bertold Brecht's “spect-actor” [boal2000theater].]

Eventually, I became disenchanted by the hollow modeling of non-player characters (they do not afford meaningful interactions beyond different types of collision) and particularly by the lack of world persistence: due to memory constraints, only the player character's immediate vicinity is simulated, which means that characters and objects leave memory (and disappear forever) as soon as one turns a corner. This does not reflect a design flaw, but rather a difference in emphasis. The *Grand Theft Auto* series is all about the near term: gratification is disbursed continuously as gameplay moves from visceral moment to visceral moment, and there is no reason to maintain dependency between such moments. What I personally wanted, however, was something longer term—I wanted to carry out an extended project of my own design.

#### A Thrilling Emergent Experience

Before putting down the game, however, I played through its linear mission structure, and it was in this unlikely context that I had perhaps the most thrilling emergent experience in all of my history with computational media. In a mission called “Psycho Killer”, shown in Figure 3.2), the player is tasked with apprehending an unhinged character who seeks to kill the members of a band, Love Fist, with whom the player character is associated. The mission is designed to play out in a short car chase that culminates in the killing of the character being chased, but in my playthrough things went off the rails.

[Figure omitted]

Over the course of the car chase, both of our vehicles became damaged to the point of starting on fire—in the *Grand Theft Auto* series, this indicates that the vehicle is about to explode, which means that its passengers must promptly exit or else die in the explosion. As such, both the killer and I were forced to leave our vehicles, but because this occurred in a relatively empty neighborhood near the edge of the island city, there were no nearby vehicles for either of us to hijack. Having no other choice, the killer proceeded to run from me on foot, and I chased after by the same means. This is where the situation took on an almost magical quality. After a few blocks, the killer took a turn toward the water, walked out onto a marina, and hopped into a boat that was docked there. I followed after and commandeered another boat, and from here the chase transpired on water, through the beautiful ocean passages of Vice City. By this point, I was mesmerized and had become more interested in prolonging the chase than in executing the stipulated task of killing the killer. We cruised in our boats for what felt like a long time, but eventually the killer approached a dock somewhere else in the city, disembarked, and again proceeded on foot. I did the same, and as we emerged back onto the street, the killer stole a car and I did the same. At this point I began debating whether it would be more poetic for the killer to get away—after taking such creative means, this seemed to perhaps be the just outcome—or to succumb finally to the mission's tragic design. I chose the latter, and this culmination terminated an emergent experience that I recall lasting upward of an hour.

#### The Personal Dimension

Why was this experience so powerful for me? Why do I still remember it fifteen years later, while I can recall essentially nothing from the game's scripted narrative? When the authored story of the mission (a brief car chase) veered off the rails into an emergent sequence that perhaps no other player ever experienced, something more akin to nonfiction obtained. This strange chase by car and foot and boat actually *happened*, and this makes it inherently more interesting—if the designers of the game had scripted a mission designed to transpire like my emergent experience did, the experience of that mission would not be as interesting.

While the actuality of the experience is important, I contend that the primary source of intrigue for this emergent sequence is something else: *I personally experienced it*. Because I was an active participant in this story, for me it does not merely work like nonfiction, but moreover like *lived experience*. Thus, my account of this emergent experience, as it has appeared on these pages, is a story of lived experience. While a particular pleasure of nonfiction is rooted in the ostensible actuality of the recounted events, an additional pleasure obtains when the recounted events were personally experienced. This is intuitive, since we encounter this phenomenon all the time. Consider the kinds of stories that we tell one another about our daily experiences—these are not inherently riveting (they would not make good novels or good screenplays), but instead are interesting because they recount actual events, and moreover lived experience. Often, such stories are more interesting, or are only interesting, to the ones who tell them, because they actually participated in the recounted events. These are the kinds of stories that are told after a day at the workplace—they may require indulgent audiences, but they are interesting enough to the teller to merit the telling.

#### Personal Relatedness and Story Interestingness

Here, we do not have rely completely on intuition, since a number of researchers have corroborated the argument that I am making here. Since at least the late 1970s, cognitive scientists exploring story understanding (and discourse processing, more broadly) have investigated the notion of *story interestingness*. While I discuss this work in more depth in Section 4.1.1, I will mention here that several researchers have pointed out the obvious: stories about ourselves or those we know are inherently more interesting. In a seminal paper in this tradition, Roger Schank cites “personal relatedness” as an operator that may work in tandem with other interestingness factors to boost story intrigue [schank1979interestingness, p. 15]. Since that paper, many others have agreed with Schank or taken the notion for granted [kintsch1980learning,graves1983writing,smith1985recall,hidi1986interestingness,ram1990knowledge,hidi1991motivational,alexander1994subject,benton1995knowledge,young1996some]. In Section 3.1.2, I discussed personal relatedness at length by articulating a connection between emergent narrative and stories of lived experience. Interestingly, a recent paper by Petri Lankoski on story intrigue in videogames reports positive correlation between interestingness judgments and characteristics that are more typical of emergent narrative, namely that a game “support[s] different play styles” [lankoski2013models].

#### A Lack of Curation

While the pleasure of nonfiction obtains in emergent narrative, the pleasure of stories of lived experience obtains in the more specific case of *interactive* emergent narrative. However, as I will argue in Chapter 4 as an extension of Espen Aarseth's critique of emergent narrative, few interactive works in this area actually produce narrative artifacts. Consider the case of my *Grand Theft Auto: Vice City* experience. The events that I have recounted did not constitute a narrative artifact merely by their transpiring, but instead the story came to exist *through* my recounting of those events. Thus, my account is a story of lived experience—one that I can look back on fondly, reaping the benefits of both nonfiction and personal narrative—while the raw experience recounted thereby is merely the subject of that story.

So while a core appeal of interactive emergent narrative is that it supports such stories of lived experience, these works tend to require that the interactor do the actual curation work to produce such stories. Typically, this curation process occurs in the mind of the interactor—or, more specifically, it may be enacted in the course of a conversational recounting to someone else—and indeed this is how stories of lived experience tend to originate. In other cases, however, curation by the interactor may result in a work of media that recounts the lived experience. In fact, this is the pattern of curation that characterizes *Oilfurnace*, the case study that appeared in the last section: for Tim Denee, the gameplay that he recounts was a personal experience, and so for him the comic is a story of lived experience. Because *Dwarf Fortress* is a *god game* [burns2002civilization] (the player does not control a situated avatar), Denee's sense of that experience on Dread Island being one that was lived by him may be reduced relative to my own experience in Vice City. Here, there appears to be an interesting spectrum, where certain interaction configurations are better poised to support stories of lived experience.

To return to the question of who is doing the curation: I am excited by the prospect of systems that would automatically curate gameplay to recount, to a player, her own lived experience that has transpired thereby. Aside from a particular variant of *Diol/Diel/Dial*, a project that I discuss in Chapter 8, my work that I report in this dissertation does not really venture in this direction. While both *Bad News* and *Sheldon County*—the subjects of Chapters 10 and Chapter 12, respectively—curate emergent material to form narrative artifacts, this emergent material does not (typically) result from player activity.
Nonetheless, this is something that I am interested in, and a broader way of describing the task is one of *gameplay summarization*, which is an area in which I have done some work. [Note: Of course, not all gameplay works like emergent narrative, so this category is broader than what I have described so far in this section.]

#### Summarizing Interactive Emergent Narrative

On an earlier project, collaborators and I sought to summarize the emergent material produced by *Prom Week* gameplay [antoun2015generating]. To my knowledge, this notion of procedurally generating a post-mortem after gameplay was first explored by Anthony Davey's *Proteus* system, reported in his 1974 dissertation [davey1974formalisation,davey1978discourse], which could play tic-tac-toe and generate game summaries afterward. [Note: David McDonald's 1981 dissertation reported on an application of his system *Mumble* to the tic-tac-toe summarization task that Davey had explored [mcdonald1980natural]. In 1984, Graeme Ritchie reported on a rational reconstruction of *Proteus* [ritchie1984rational].] In his 1996 masters thesis, Paul Bailey demonstrates the generation of stories summarizing tic-tac-toe, card games, and the *blocks world* [bailey1996generating]. [Note: For his PhD work, which he appears to have abandoned, Bailey proposed reader-centric story generation [bailey1997reader,bailey1999searching,bailey1999reader], as I will discuss later on.] Ian Frank also proposed the generation of gameplay postmortems in an obscure 1999 paper [frank1999explanations]. In our project with *Prom Week*, this meant generating narrative accounts from *playtrace data* recorded by the game. While other projects have used such *Prom Week* data for authoring support [samuel2014introducing,osborn2014evaluating] or interpretation [samuel2015playing], this was a first attempt at producing narrative accounts.

Along with my advisors Michael Mateas and Noah Wardrip-Fruin, I call this task of producing narrative accounts of raw data *narrativization*. In an earlier paper, we proposed that a tool like *Playspecs* [osborn2015playspecs], which supports automated playtrace analysis, could be used to recognize interesting emergent sequences for the purposes of narrativization [ryan2015open, p. 21]. Interestingly, this process does not necessarily have to occur after the emergent sequence has terminated—indeed, the narrativization process could occur in an online manner, narrating the story of lived experience as it is unfolding. Along these lines, in his *Writing Buddy* project, Ben Samuel is currently exploring the use of *Playspecs* to recognize unfolding emergent sequences and also to consider prospective completions to them [samuel2016design, p. 393]. In another recent project, Camile Barot and collaborators have produced *Bardic*, a system that narrativizes *Dota 2* gameplay [barot2017bardic]. Other workers in this area have explored the summarization [levy1996icca,bjornsson1998fritz,marsland1999fritz,mallett2010herschberg] or narrativization of chess gameplay [gervas2014composing,gervas2014metrics], procedural commentary or narrativization for sporting events [tanaka1998mike,zheng2010automated,allen2010statsmonkey,bouayad2012perspective,lee2014automated], and the narrativization of logs from other videogames [cheong2006framework,cheong2008automatically,hoek2014generating,jacko2018generating]. Here, the most extensive exploration has probably come in the development of procedural commentary for commercial sports videogames [rhodes2010enhancing,lee2012automated,baerg2013digital,stein2013televisual,adams2014fundamentals]. [Note: An interesting variation on this notion comes in the form of a recent patent on technology for procedural commentary of a videogame itself, in the style of film commentary [walker2014enhanced]. This would take the form of a gameplay mode that procedurally triggers the playback of audio segments that provide insight about what is currently being experienced in gameplay (for example, how a level was developed).]

#### Personal Relatedness Without Interaction

It is worth considering how important the dimension of interactivity is to the phenomenon I have described in this section. While a feeling of lived experience will be far more activated in the case of interactive emergent narrative, it is still possible in non-interactive works. For instance, if a person encounters a story that has been generated by a computer program just for her, she may feel a personal attachment to it, since no one else has encountered it and no one else will. Thus, the kind of ephemerality that characterizes works of emergent narrative, interactive or not, may work to engender a sense that generated narrative outputs are personalized, which makes encountering such outputs a kind of personal experience. It would thus not be unreasonable to say that such cases work (to some degree) like stories of lived experience.

#### Computational Personal Narrative

Lastly, I would like to note that there is an interesting line of work that explores the intersection of computational methods and personal narrative. Andrew Gordon and Reid Swanson (and at times other collaborators) have developed techniques for automatically recognizing personal narratives found in blog entries, which they used to build a corpus comprising millions of examples [gordon2005automated,gordon2007automated,gordon2008storyupgrade,gordon2009identifying]. This corpus was then harnessed to drive Swanson's dissertation project, the aforementioned story generator *Say Anything* [swanson2008say,swanson2010enabling,swanson2012say], and others have used automatic methods to study features of these stories [sagae2013data,rahimtoroghi2013evaluation,rahimtoroghi2014minimal,swanson2014identifying,ryan2016recognizing,rahimtoroghi2017modelling,reed2017learning]. Other work out of Marilyn Walker's lab (led by students Chao Hu, Stephanie Lukin, and Kevin Bowden) has demonstrated approaches for stylistic (and even multimodal) *regeneration* of such personal narratives [lukin2015generating,hu2016corpus,bowden2016m2d,lukin2017personabank]. It would be interesting to apply these techniques to the task of narrativizing event sequences from interactive emergent narrative.

### Analogy to Worldbuilding

Due to its emphasis on simulating the background and larger context of a storyworld, emergent narrative may be viewed as employing a kind of *worldbuilding* [roine2016imaginative,pacat2017fantasy]. In worldbuilding, an author endeavors to construct a larger world in which the characters and events of a story may be situated.

#### Tolkien's Aristotelian Inversion

While this is arguably an inevitable feature of all fiction [ryan2001narrative], in worldbuilding the world is both *preliminary* and *primary*. As the writer China Mieville states, the practice of worldbuilding curiously inverts the typical procedure of fiction authoring: “The order is reverse: the world comes first, and then, and only then, things happen—stories occur—within it” [mieville2009there]. This is in a sense anti-Aristotelian, as Hanna-Riikka Roine notes in her dissertation on worldbuilding: Aristotle contends that action and plot give rise to the other fictional elements, which may be viewed as constituting the fictional world, but in worldbuilding the world gives rise to the action [roine2016imaginative, p. 17]. J. R. R. Tolkien, who is likely the individual most associated with the technique, called it “sub-creation”: a mode of fiction crafting that entails a subprocess through which the author “makes a Secondary World which your mind can enter” [tolkien1947fairy, p. 12]. As Roine notes [roine2016imaginative, p. 17], while Tolkien of course had his own influences, the mode of worldbuilding (or ‘sub-creation') that he employed in his work (particular the *Lord of the Rings* series) established the fantasy genre and inspired projects outside of print fiction, including *Dungeons & Dragons* [gygax1974dungeons] (and thereby *interactive fiction* [montfort2005twisty]) and the practice of *transmedia* [jenkins2009transmedia,dena2009transmedia]. While worldbuilding may viewed as a reformulation of fiction's authorial process (world precedes plot), it may also be viewed as a conceptual orientation (world above plot) or a communicative orientation (world concretizes a thought experiment) [roine2016imaginative].

#### Emphasizing the Simulated Storyworld

The analogy between worldbuilding and emergent narrative is this: just as the former inverts the process of fiction authoring to proceed from world to action, the latter inverts the orientation of computational narrative such that a simulated world generates action (and not vice versa). By the interventionist approaches that characterize conventional computational narrative, namely *drama management* [laurel1986toward,weyhrauch1997guiding,mateas1997oz,roberts2008survey] and *narrative planning* [young1999notes,riedl2010narrative], the conceptual center of gravity is the plot. Indeed, it should not be surprising then to discover that these approaches are rooted intellectually in the Aristotelian poetics that worldbuilding inverts [laurel1986toward,laurel1991computers,kelso1993dramatic,mateas1997oz,mateas2000neo,mateas2001preliminary]. [Note: Hartmut Koenitz has recently articulated an evocative account of the intellectual history of the drama management approach (citations inserted for clarity): “instead of emphasizing the connection to ancient drama, Mateas' and Stern's Facade can be more aptly described as based on the theoretical framework developed by Mateas himself (2001) [mateas2001preliminary], modifying Laurel's (1991) [laurel1991computers] reinterpretation of Neo-Aristotelian concepts by implementing Janet Murray's (1997) [murray1997hamlet] analytical and phenomenological categories. The reference to the rather distant—several times removed—original notion is more distracting than helpful. Indeed, Aristotle's unmodified concept is inappropriate for Interactive Storytelling in its basic demand of a plot that is perfect if nothing needs to be added or can be removed, a perspective which is in fundamental disagreement with the idea of interactive narrative that needs to offer excess material for alternative paths and optional elements” [koenitz2017interactive, pp. 8–9]. As for narrative planning, its figurehead is R. Michael Young, whose earliest writings on the topic also make the Aristotelian connection [young1999notes,saretto2001mediation]. Interestingly, in his dissertation, Mark Riedl (who was a student of Young) cites Lajos Egri's *premise*-oriented process of dramatic writing [egri1960art] as the primary intellectual influence on his own approach to narrative planning  [riedl2004narrative, p. 44].] While this perspective may be viewed as plot-centric (Aristotelian) and others as character-centric [Note: Mark Riedl [riedl2004narrative, p. 44] associates the character-centric orientation with the more contemporary craft advice of Lajos Egri [egri1960art] and Robert McKee [mckee1997substance].] or author-centric, the brand of emergent narrative promulgated in this dissertation might be called *world-centric*. [Note: The tripartite model of plot-centric, character-centric, or author-centric story generation appears to originate with Paul Bailey [bailey1999searching,mateas2003narrative], whereas an earlier character vs. author duality was first articulated by Natalie Dehn in 1981 [dehn1981story]. (The difference between the author-centric and plot-centric approach is that the former attempts to model the creative process of a human author, while the latter operates more directly at the level of plot manipulation.) Beginning with Masoud Yazdani's initial reporting of *Roald* in 1982 [yazdani1982write], and more famously with Michael Lebowitz's first *Universe* paper of 1983 [lebowitz1983creating], practitioners have shown that systems may bridge such conceptual divides. Furthermore, others have introduced alternative approaches to these three, such as Chris Crawford with his “verb-based” *Erasmatron* [crawford1999assumptions, p. 122] and Bailey with his reader-centric approach [bailey1997reader,bailey1999searching,bailey1999reader]. Bailey's project was proposed while he was a graduate student at the University of Edinburgh [ross1997prospectus,bailey1997reader] and did not reach fruition, but I would like to see greater exploration of the reader-centric approach to story generation. Inspired by Bailey, Peter Mawhorter originally proposed to move in this direction with his dissertation project [mawhorter2013reader], which after some reformulation resulted in the introduction of *choice poetics*, an intellectual and technical framework that considers reader response in the context of choice-based interactive narrative [mawhorter2014towards,mateas2015intentionally,mawhorter2018choice]. Stacey Mason has also produced a descriptive model that incorporates reader concerns [mason2013games]. Eventually, Mawhorter operationalized his framework in a system called *Dunyazad* [mawhorter2015generating,mawhorter2016artificial,mawhorter2016dunyazad], which generates choice-based interactive narrative that targets particular reader responses. Along related lines, an earlier system (1998) by Boris Galitsky attempted to generate *persuasive* stories for internet ads (to persuade potential buyers) [galitsky1998formal,galitsky1999narrative], and Kim Binsted and Graeme Ritchie proposed generating *story puns* that target a reader response across generation instances [binsted1996speculations,binsted2001towards]. It even appears that Jim Meehan was beginning to explore reader-oriented story generation just before he left UC Irvine [meehan1982stories, p. 459]—unfortunately, I have not been able to obtain his paper on this new direction [meehan1981boy]. Finally, Pablo Gervás and Carlos León have recently argued in favor of this direction [gervas2014reading].] While emergent narrative has typically been called character-centric [dehn1981story,riedl2003character,louchart2004theoretical], the world-centric approach that I advocate here extends character simulation to a *massive* scale: instead of simulating the activity of characters with regard to a specific narrative premise, I am interested in simulating entire storyworlds containing hundreds or thousands of characters who live out lives that extend beyond the context of a given narrative premise. Thus, while I still view classic systems like *Tale-Spin* [meehan1976metanovel] and *FearNot!* [aylett2005fearnot] and *The Sims* [maxis2000sims] as exemplars of emergent narrative, I am especially interested in the kind of grandiose world simulation that is typified in *Dwarf Fortress* [dwarf]. As such, the fundamental difference between a character-centric system and a world-centric system is the increased magnitude and granularity at which the latter simulates a larger world as a backdrop for potential emergent stories.

Thus, while the hallmark of emergent narrative is its character-centric orientation, I aim to promote an increased breadth and depth of simulation such that the result might be called world-centric. When this approach is taken, the pleasure of worldbuilding obtains, and in particular this orientation undergirds the *aesthetics of a larger context*, as I outline in Section 3.2.

#### Curating the Simulated Storyworld

That being said, the presence of a massive world simulation constitutes only the first concern underpinning my framework of curationist emergent narrative. My tagline for this approach is *overgenerate and curate*, and while a world simulation satisfies the ‘overgenerate' imperative, I also believe that a simulation does not on its own constitute a work of emergent narrative. In my view, it is only when the material produced by such a simulation is curated that a narrative artifact obtains. When this is done, the narrative artifact is imbued with greater intrigue because it stands against the backdrop of a larger context that is constituted in the simulated storyworld. This context on its own, however, is not narrative, and moreover it has no backdrop to stand against, since as a raw and unprocessed storyworld it *is* the entire context. [Note: To return to Tolkien, we might characterize the negative reception of *The Silmarillion* (as opposed to *The Hobbit*) [foot1977middle,adams1977hobbit] as being rooted in the failure of the former to curate a distinct narrative artifact from the congeries of the storyworld's raw history. Writing about *Dwarf Fortress*'s Legends mode [dflegendsmode], Stephanie Boluk and Patrick LeMieux make the same connection: “The result is a textual archive that reads more like *The Silmarillion* than *Lord of the Rings*” [boluk2013dwarven, p. 143]. This analogy does not perfectly apply, however, since the book is still a narrative account, not literally a raw storyworld (in the sense that a computer simulation may be one).] Thus: overgenerate *and curate*. In Chapter 5, I will return to this argument to articulate and advance my framework for curationist emergent narrative.

### Analogy to Art Brut

[Figure omitted]

I begin here with my favorite computer-generated story:

> Henry Ant was thirsty. He walked over to the river bank where his good friend Bill Bird was sitting. Henry slipped and fell in the river. He was unable to call for help. He drowned. [meehan1976metanovel, p. 128]

 This is one of *Tale-Spin*'s famous *mis-spun tales* [meehan1976metanovel, p. 126], a collection of outputs reported by its creator Jim Meehan as illustrative examples of the story generator going awry (usually due to gaps in Meehan's knowledge engineering). [Note: An aesthetically kindred example, produced by a story generator based on *Tale-Spin*, appeared on the ‘comp.ai‘ newsgroup in 1993: “Once upon a time there was a bear named Jack. Jack was dead. Jack was still dead. The end.” [hearn1993computer, n.p.].] When things went “right”, the resulting stories were regenerated Aesopian fables with dubious prose. I am not alone in appraising the mis-spun tales as far more evocative than their correctly spun counterparts. Indeed, Noah Wardrip-Fruin has noted that these curious artifacts are the system's most famous (and most frequently reprinted) outputs [wardrip2009expressive, p. 130]. [Note: As Wardrip-Fruin explains, this may be due in part to the prose of the mis-spun tales being hand-authored by Meehan (from the underlying computational representation), whereas the prose of the successful outputs was computer-generated.] Wardrip-Fruin cites the “amusement value” of the tales, and others might characterize them as a kind of *glitch art* [menkman2011glitch], but I think there is something more here.

Certainly the story above expresses a kind of glitch (though one rooted in human authoring), and I do agree that the story is humorous—but its comedy is dark and tragical and strange and alien. I contend that there is an *uncanny* quality of the tale that yields its peculiar aesthetic impact. It does not work like a human story—that is, it does not seem to be a story that a human would be compelled to tell. Below, in Section 4.1.1, I discuss the notion of *tellability*: a narratological (and sociolinguistic) concept that pertains to whether a story is even tellable (where the stakes are losing face due to the wasting of a recipient's time and cognition). Put differently, a story is a recounting of a tellable sequence of events. It is tempting to say that this story is intriguing because in human culture it is not tellable, but in the mind of this computer, it is. I do not think this is the case, though: it is easy to generate untellable stories that are not intriguing, but such tales are simply *boring*. This story is not boring—there is something strangely enchanting about it—and in this section I intend to characterize this mysterious quality so that it may be explicitly targeted (or at least celebrated) in future generative works.

#### A Computer Art Brut

I view this mis-spun tale and related examples as representing a kind of computer *art brut*. Sometimes called ‘low art' or ‘folk art' or ‘naive art' or ‘outsider art', [Note: I agree with others who have deemed ‘outsider art' a poor synonym, since that phrase is often used to refer to a broader category of works [maclagan2010outsider,clarke2014fi].] art brut (literally ‘raw art') is a term that was coined by artist and thinker Jean Dubuffet to refer to artistic practices that exist outside of the art establishment and that are (usually incidentally) in opposition to its aesthetic hegemony [peiry2001art]. This art—“unscathed by artist culture” [dubuffet1988art, p. 31]—has often been identified with children [cardinal2009outsider], mental illness [macgregor1989discovery], and developing nations [minturn2004dubuffet], but such factors are not necessary or sufficient conditions for art brut, but rather recurring circumstances. [Note: In attempting to celebrate the aesthetic sensibilities of artists who are deemed to be ‘outside' of some boundary, there is clearly a risk that such demarcation will result in an othering, or in a patronizing fetishization, of the characteristics that are viewed as contributing to one's outsider status. This has been articulated before [wojcik2016outsider,frank2016should,evans2016many], and though the stakes are different when considering the otherness of a computer, I think these considerations are still worth acknowledging.] While art of various kinds may evoke the ‘art brut' classification, certain aesthetic qualities underpin the idea: a sense of the crude, uncanny, alien, eccentric, deranged, marginalized, pure [dubuffet1988art,whiteley1990banham,weiss1992shattered,minturn2004dubuffet]. As such, the form is bound primarily by a set of *family resemblances* [wittgenstein2009philosophical], rather than any concrete concerns pertaining to medium, content, and so forth.

I am interested in asking *how would a computer art brut work?*, and I think the uncanny mis-spun tale reported above is a potential answer. It is art by a computer that is distinctly removed from the aesthetic sensibilities of human art. Here is another generated story that I love for the same reason:

> The Paranovs live in a distant province. The father is Emelya. The mother is Maria. The oldest son is Foma. The younger son is Baldak. Vasilisa is the only daughter. Emelya says Baldak, do not go to the house. Emelya dies horribly. Baldak goes to the house. A bear flys [sic] into the distant province. The bear asks Marco where is Baldak. Marco says that Baldak is in the house. The bear orders the murder of Baldak. Baldak is secretly freed. Baldak is wandering. Baldek meets a forest knight along the way. The forest knight brawls in a forest hut with Baldak. The forest knight twice repels Baldak. They fight for the third time. Baldak defeats the forest knight. A magic carpet is given to Baldak. Baldak starts back home. Baldak returns home. [klein1974modelling, p. 64]

 This eerie tale was produced by a system developed in the mid-1970s by Sheldon Klein and his students to operationalize Vladimir Propp's famous model of Russian folk tales [propp2010morphology]. [Note: As Klein explained, he did not set out to create strange stories, and in fact he viewed deficiencies in the generated outputs as proof that Propp's model was itself deficient (citations inserted for clarity): “That the results are bad Russian fairytales is a major theoretical significance, for they confirm that Propp's model is for the syntax of fairytales, and that there must also exist a higher level semantics of fairytales outside the description of Propp. Our experiment provided a concrete verification of the points about Propp made by Levi-Strauss (1960) [levi1960analyse] and Meletinski *et al*, 1971 [meletinsky1974problems])” [klein1977meehan, p. 2]. I am, however, interested in celebrating these “deficiencies”, especially when they are rooted in the alien peculiarities of a distinctly computational sensibility. Note that this is different from celebrating the mere bad. An earlier project in Proppian story generation that produced more mundane stories was abandoned because its practitioner, Joseph E. Grimes, found the outputs to be boring [ryan2017grimes]. I do not advocate boringness, and that is exactly why I am not intrigued by *Tale-Spin*'s “successful” outputs.] In an improbably *more* surreal output, one character keeps acting even after death (though perhaps in a liminal hereafter):

> The Popoviches live in a distant province. The father is Vladimir. The mother is Dunia. The oldest is Boris. The younger son is Erema. The youngest son is Foma. Vasilisa is the only daughter. Erema asked Dunia where is your wisdom. Dunia says that my wisdom is in a magic egg. Dunia murders Erema. A lament is sung for Erema. Erema is wandering. Erema meets a stove along the way. The stove proposes that Erema eat the meal which it has prepared. Erema responds by partaking of the meal. A magic potion is consumed by Erema. Erema obtains incredible speed. Erema starts back home. Erema returns home. [klein1974modelling, p. 65]

 The evocative *computer art brut* aesthetic of these generated tales is further revealed in their original form, as presented in a technical report via printout by a drum printer, which produces a distinctly computer-generated look; this is shown in Figure 3.3.

#### Uncanny Emergent Narrative

While Klein's system is not an example of emergent narrative, *Tale-Spin* is, and in any event we might consider how projects taking this approach could work like art brut to yield a striking idiosyncratic aesthetics. Due to the nature of emergent narrative, such qualities would likely be based in the peculiarities of computer simulation—perhaps the gaps or exaggerations or other alien qualities of a simulated storyworld could be emphasized to evocative effect. Such qualities are probably unavoidable in computer simulation, but I would like to see projects in emergent narrative that embrace them. The stories recounted in this way would be openly and proudly about the surreal ontology of the storyworlds from which they emerge—weird tiny abstract worlds that work. Again, this does not mean celebrating what is merely bad, but rather what is surreal and striking. I return to this idea in Section 3.2, and in Section 4.2.6, I criticize Meehan's approach to *Tale-Spin* for its posturing toward a boring human aesthetic. Of course, these are highly subjective considerations, and in this section and the others I am certainly proceeding from my own personal inclinations. Nonetheless, I think it is worth at least *considering* what an unabashed *computer art brut* might look like, and for our purposes here, an alien emergent narrative.

## The Aesthetics of Emergent Narrative

Through its resemblances to the various forms outlined above, emergent narrative unlocks some of the pleasures that are associated with them. To conclude this chapter, I will briefly identify a set of such appeals, which are also constituted in additional peculiar features of the form—I view these pleasures as constituting an *aesthetics of emergent narrative*. Note that my aim here is to provide a brief first articulation, which makes a deeper engagement of these ideas in terms of existing aesthetic theories the object of future work.

### Aesthetics of the Actual

Due to its resemblance to nonfiction, articulated in Section 3.1.1, emergent narrative unlocks the hallmark pleasure of that form, which may be referred to as the *aesthetics of the actual*: when a story recounts (ostensibly) actual events, it will likely be more interesting than a fictional, but otherwise identical, counterpart. This is the basis for the marketing trope characterized by the phrase “based on true events”, and it is a pleasure that undergirds enjoyment of emergent narrative.

### Aesthetics of the Personal

Through its connection to stories of lived experience, identified in Section 3.1.2, (interactive) emergent narrative yields an *aesthetics of the personal*: when a story recounts events to which a recipient is personally connected, it will likely be more interesting to her (with all other things being equal). As I noted in my discussion above, this is primarily a feature of interactive works of emergent narrative.

### Aesthetics of a Larger Context

By an analogy to worldbuilding, as discussed in Section 3.1.3, emergent narrative is bolstered by the *aesthetics of a larger context*: a special intrigue is attached to stories that transpire against the backdrop of a larger storyworld. While all narrative works imply (and at times concretely model) a storyworld, world simulation is the organizing principal in emergent narrative.

### Aesthetics of the Uncanny

As a kind of simulationist art brut, following my argument in Section 3.1.4, emergent narrative produces a unique *aesthetics of the uncanny*. As I discussed above, there is a (subjective) gray area between the uncanny and the mere bad, but the tiny abstract storyworlds that are simulated in emergent narrative seem to buttress an uncanniness that does not manifest in other forms of (procedural) narrative. Stephanie Boluk and Patrick LeMieux have celebrated this feature in one of the form's major works: “*Dwarf Fortress* embraces the granularity of computation over the specularity of graphical realism, delighting in the strange, formal logics of the computer” [boluk2013dwarven, p. 126]. In a related argument, Ian Horswill has called for deeper consideration of genre and tone in computational media, since some orientations do not seem to work procedurally given the state of the art in available technologies [horswill2012punch]. As a specific example, he argues for the targeting of *farce*. Likewise, Tanya X. Short has recently advised game designers to “lean into the comedy” and embrace the peculiar dynamics of procedural characters [short2017designing, p. 114], a stance echoed in Kenneth Chen and Stefan Rank's recent study of humor in interactive emergent narrative [chen2018towards]. This kind of coming to terms with the broad aesthetic potentials of procedure is what I am getting at with my evangelism of the uncanny.

### Aesthetics of the Unauthored

As a form of procedural generation, and moreover the least authorial approach to story generation, emergent narrative is characterized by an *aesthetics of the unauthored*: there is a peculiar delight in beholding an artifact that was not created by a human. This is a phrase that I borrow from Jason Grinblat, who in a Twitter thread on generated world maps remarked, “I love procedural generation and the aesthetics of the unauthored” [grinblat2018maps, n.p.].

### Aesthetics of the Coauthored

In the case of interactive emergent narrative, a player may have the sense of sharing in the authorship of an ongoing experience—this supports an *aesthetics of the coauthored*. Ben Samuel's recent dissertation explores this notion at depth through his framework for *shared authorship*, which concerns patterns of interaction whereby a human and machine collaborate to compose a story that neither entity could have composed on its own [samuel2016crafting]. Related to this framework is Aaron Reed's notion of *sculptural fiction* [reed2011sequence,reed2017changeful], in which a player composes a story graph, instead of traversing one.

Due to its bottom-up workings, player interactions are more easily incorporated into emergent narrative than other forms of procedural narrative: in emergent narrative, the storyworld evolves in response to player actions according to the rules that govern its world simulation, whereas by interventionist approaches an experience manager must take action to *repair* the storyworld in response to player actions (e.g., [riedl2006believable,li2010offline,paul2011adaptive,young2013plans,robertson2016model]). As I explained in Chapter 2, this was the impetus for Ruth Aylett's influential exploration of emergent narrative. Because emergent narrative (in interactive works) depends more on bottom-up player actions than top-down story policies, shared authorship is more potent in the form, and the aesthetics of the coauthored are in turn amplified.

### Aesthetics of the Uncovered

In his feedback on an earlier draft of this thesis, reading-committee member Jonathan Lessard identified an additional feature of emergent narrative that is related to, but distinct from, the aesthetics of the coauthored:

> I really like the parallel you make between the work of the historian and the relationship to simulated storyworlds. To me, this should be pushed even further. In your list of aesthetics, I would add: “the aesthetic of the uncovered” [which] is perhaps the same or close to your aesthetic of the coauthored. It's about the pleasure of discovering/unearthing/excavating/witnessing an interesting story *yourself*. This is the historian's pleasure when finding something surprising when going through their archive. This is *your* pleasure when you find these superb nuggets from your simulations, and, as you become narrator, it becomes our pleasure as you know your audience and know what will be interesting to us. [Note: Personal communication, July 27, 2018.]

This is indeed a core aspect of the aesthetics of emergent narrative, and one that ties in nicely with the other patterns that I identify in this section. In this thesis, I argue at length that, with regard to computer simulation, interesting stories can only obtain through an act of curation. When the curator (or her audience) encounters such a story, her expended curatorial effort is exchanged for added aesthetic value, because the story has been *uncovered*—what would have otherwise gone unnoticed is now beheld, and this produces a distinct pleasure.

### Aesthetics of the Improbable

Part of the intrigue of true stories is that they are inherently *improbable*: coincidental real-world event sequences generally lack narrative intrigue, or even narrative structure, and as such it is a notable occurrence when something interesting actually does happen in the real world. This is the *aesthetics of the improbable*. In contrast to this, the very project of fiction writing is to come up with interesting events, and as a result all works of fiction tend to be ‘interesting' with regard to basic notions of narrative intrigue. {{But when every story is interesting, interestingness itself is highly probable, and so the bar for intrigue is raised—this process may culminate in what Guy Bergstrom has called the *Michael Bay school of storytelling*:}

> People want a thrilling ride? The Michael Bay School of Storytelling says OK, let's blow their minds with the most intense story ever. Except when everything is dialed up to 11, the audience goes numb. [bergstrom2011evil, n.p.]

This pitfall can also characterize works of procedural narrative in which top-down methods, driven by experience managers or other mechanisms of artifice, are employed in an effort to always produce interesting outputs. Again, when every story is interesting, that quality becomes suspect, and paradoxically the stories become boring. In emergent narrative, however, interesting event sequences are improbable, as in the real world, and so stories that do actually emerge tend to be inherently more interesting. As such, the aesthetics of the improbable is present in emergent narrative. [Note: Regarding the Bergstrom quotation above, Ian Horswill imparted the following in his feedback on an earlier draft of this thesis: “I don't think this is arguing what you want it to. You want to argue about improbability but I think this is arguing more about temporal dynamics—you have to have variations in intensity, within a given story, because humans are essentially sensitive to changes more than to absolute values. You see that in all time-based media—authors are always scheduling intensity and variation, putting in times to let the audience rest, foreshadowing and building up to an event to let the audience anticipate it, for example” (personal communication, July 28, 2018). This is a great point, and indeed reasoning about such temporal dynamics feature heavily in the computational systems that drive relatively top-down approaches to procedural narrative (for instance, in *Facade* the fundamental unit of narrative is the *dramatic beat* [mateas2003integrating]). What I am attempting to write about here is dynamics at the level of a system's *possibility space*: what are the intensity dynamics across the instances in this space? In this case, the instances are event sequences, and in this dissertation I am promulgating an approach whereby few instances in a system's possibility space are compelling. This may seem paradoxical, from the standpoint of media creation, but this is what makes true stories interesting in the real world—they are improbable with regard to the larger space of possible actual event sequences—and that is the kind of pleasure that characterizes emergent narrative.] In a nice example of this phenomenon, Stephanie Boluk and Patrick LeMieux describe how *Dwarf Fortress* players have celebrated the narrative quality of the improbable:

> Another example of translation can be seen in the painstaking research and attempts to narrativize the remarkable history of the dwarf Tholtig Cryptbrain, a statistical anomaly discovered in *Dwarf Fortress*'s Legends mode and the sole survivor of a century-and-a-half-long war in which she claimed 2,341 unique kills. Tholtig's story is a death-defying feat not of skill but of chance and probability. [boluk2013dwarven, p. 135]

As I explain in Chapter 5, curationist emergent narrative (the subject of this dissertation) is about automatically identifying interesting stories that emerge from simulations. A system that can curate a simulated storyworld in this way may produce interesting story after interesting story, but critically each one has improbably emerged from a simulation. When this improbability is expressed in the media experience that houses the curated stories, the aesthetics of the improbable are maintained. Thus, by relying on curation instead of artifice or intervention, interestingness and improbability are no longer incompatible, but are instead complementary.

### Aesthetics of the Vast

Emergent narrative is vast: through the mechanism of massive rich simulation, its storyworlds have considerable depth and breadth, and a higher-order vastness obtains with regard to the possibility space constituted by all a system's generable storyworlds. By centering on the world, and not the plot, emergent narrative overwhelms with its sheer accumulations of material. This can be overwhelming in a good way (*Dwarf Fortress* players love its overwhelming complexity and detail), but also in a bad way (*Tale-Spin*'s extraneous prose may obfuscate narrative intrigue). When vastness contributes to the success of a work, the *aesthetics of the vast* is at play. As Noah Wardrip-Fruin noted in his feedback on an earlier draft of this thesis, my articulation here evokes the concept of the *technological sublime* [costa1994sublime,nye1996american]. Indeed, others have made the connection between this notion and computational media (and sometimes procedural generation more specifically) [bolognini2004smsms,amaral2011festive,shinkle2012videogames,betts2014investigation]. As I have already noted, a deeper engagement of these ideas in terms of existing aesthetic theories remains for future work.

### Aesthetics of the Ephemeral

Because its stories are constituted in improbable event sequences occurring in simulated storyworlds, there is an appealing ephemeral quality to emergent narrative—this might be referred to as the *aesthetics of the ephemeral*. In beholding an emergent story, a reader (or player) can be assured that no one else will encounter this artifact: it has never existed before and will never exist again; in a sense, it is *just for her*. This sensation is especially present in projects that feature procedurally generated storyworlds, since the actual worlds, and all their characters, are likewise ephemeral.