# Curating Simulated Storyworlds — Chapter 8: Case Study: Diol/Diel/Dial

> 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 8 of 14 only: Case Study: Diol/Diel/Dial (~9,300 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

My interactions with *World* tended to work like this: fire up a new instance, start simulating the history of its storyworld, walk away for a while, terminate the simulation procedure, and then explore the generated history (stored in working memory) by constructing the appropriate Python queries. If I was showing the system to someone else, this worked as a kind of *livecoding* [collins2011live], but I felt that I could only scratch the surface of a storyworld's emergent material, and I found myself forming the same essential queries over and over again. [Note: As I explain in Chapter 10, the notion of simulation querying as performative livecoding is central in my capacity as *Bad News* wizard. Indeed, these informal sessions of querying the emergent material of *World*'s storyworlds were probably the seed for what would evolve into the *Bad News* wizard-of-Oz [dahlback1993wizard] configuration.] To facilitate this kind of material excavation, I coded up a simple procedure that would take a given storyworld entity as an argument—e.g., a character or ship—and print out a brief description of this entity. The result would look something like this:

> ‘Name: Kelsey Flinn

Sex: Male

Born: 676, Wilderness of Nthyiseunt, Nthyiseunt

Died: 749 (age 73; natural causes), Flinn Creek, Minneapolis

Spouse: Iolanthe Flinn

Parents: Ehud Flinn and Tasia Flinn

Siblings: Walt Flinn, Cy Flinn, Prunella Worth, Gibb Flinn, Zondra Flinn, Quinlan Worth, Newton Flinn, Carmella Worth

Children: Catarina Flinn, Franklin Flinn, Antonetta Flinn, Allyce Flinn, Legra Worth, Secunda Flinn, Dominique Flinn, Roger Flinn, Andriette Major, Ulrika Flinn, Judie Major, Carmon Worth, Lilli Major, Sherye Flinn, Silvanus Flinn, Ola Flinn, Isa Major‘

 Though these printouts were not particularly readable, they did make it easier to investigate the emergent phenomena of a *World* storyworld. However, they only made it easier for *me* to do such investigation, since using the procedure required knowledge of the underlying codebase.

At this time in my life, my labmate Peter Mawhorter was an important mentor, and moreover a confidant with whom I was comfortable sharing my work; this was only my first year at UC Santa Cruz, and I was still new to computer programming more broadly. I had made a habit of showing him updates on the *World* project, and upon seeing my new entity descriptions, he remarked that it would be more effective if they were *hypertext*—that is, if references to other entities (such as other characters) functioned as hyperlinks that could be clicked to jump to descriptions of those entities (and so forth). This seemed like a great idea, and I set out to implement it. Initially, I sought out a means of doing this in a Python interpreter, but Peter later suggested a more general approach: generate a web page for each entity, with filenames corresponding to a unique entity identifiers, and then link between these web pages by treating the unique filenames as URIs. This was very easy to do, and within a few hours I had generated a first storyworld encyclopedia; Figure 8.1 shows a screenshot from one of these early examples. [Note: Soon after, this technique would give me the idea for *GameNet* [ryan2015what], a tool for videogame discovery in the form of a network of games that are interlinked according to how related they are to one another. My original prototype for the tool actually reused this encyclopedia code.] Though rudimentary, these encyclopedias represented my first foray into curationist emergent narrative.

[Figure omitted]

While my first generated encyclopedias made it easier to understand a storyworld's entities and the relations among them, they did not include information about the *events* that had emerged in that world. To capture this material, I developed the functionality that is associated with what I would now call a *chronicler* module. As I explained in Section 5.3.1, in the curationist architecture a chronicler is tasked with maintaining a *chronicle* of a storyworld, which is Hayden White's term for a raw historical record that lacks narrative structure. [Note: Incidentally, in the *World* code I happened to refer to these records as ‘chronicles'.] By this functionality, which I will explain later in this chapter, the system generates a simple sentence recounting each emergent event, as it transpires, which it then appends to collections of such records that are associated with the respective entities that were involved in the event. As a result, at the end of world generation, the life of a character—or the history of another entity, such as an island or ship—may be investigated by reading a simple concatenation of all of the sentences recording all of the events of her life.

Once this system was in place, I added a field into the entity encyclopedia entries so that these prose accounts would also be included in them. When I showed this new addition to Peter Mawhorter, he remarked, “Ah, so you've got a story generator now”. Until that point, I had never thought of my work as being in the realm of story generation—rather, I was intending to make a videogame—but it was interesting that a simple recounting of events seemed to have a narrative quality. This is probably where I began to think of my simulation practice as being an effort in emergent narrative.

[Figure omitted]

My remaining development work on this project aimed at improving the visual presentation of the generated world encyclopedias. Here, I decided to employ a scroll skeuomorph, since that matched *World*'s vaguely historical aesthetic; Figure 8.2 shows an example screenshot. When I finished this work at the end of 2014, I had built a system that could generate a storyworld encyclopedia for any given *World* instance, including cases where a player lived out the life of a character through *Islanders* gameplay. In the final version of that game, an encyclopedia would be generated upon the player character's death and the player's web browser would be triggered to automatically open the encyclopedia, beginning with her character's entry. I submitted a paper on this work to the Procedural Content Generation workshop, but it was rejected. Later, in June 2015, I decided to generate an example encyclopedia and host it online as a standalone project—I called this *Diol/Diel/Dial*, since its subject storyworld was variously called ‘Diol', ‘Diel', and ‘Dial' across its several language families. In one last attempt to make this work known, I submitted it to the Art Exhibition at the 2015 International Conference on Interactive Digital Storytelling, but that submission was also rejected. Though I found it conceptually interesting, it was not surprising to me that the work was not especially well-received—I can now articulate my sense that this was due primarily to its lack of curation, as I will explain in Section 8.4.

In this chapter, I provide an overview of *Diol/Diel/Dial*—with particular attention to the procedure for encyclopedia generation by which it was created—and situate the project against related work. Primarily, however, this chapter is a case study, as its title communicates. Below, I will provide a critique of the work in terms of the intellectual framework that I developed in Part I of this dissertation (to appraise all of emergent narrative). Specifically, I will highlight its successes in terms of the *pleasures* of emergent narrative that were articulated in Chapter 3, and then I will emphasize its failures with regard to the *pains* of emergent narrative that I identified in Chapter 4. Lastly, I will discuss this project in terms of its instantiation, or lack thereof, of *curationist emergent narrative*. This is how my subsequent case studies will be structured, as well. As a final note, while it is briefly mentioned in the paper on language modeling in *World* [ryan2016diegetically, p. 8], in essence I am presenting *Diol/Diel/Dial* for the first time in this chapter. [Note: *Diol/Diel/Dial* is hosted online at https://users.soe.ucsc.edu/~jor/diol.]

## Overview

*Diol/Diel/Dial* is a hypertext encyclopedia for a *World* storyworld that was generated in June 2015. A gargantuan tome of nearly 100M words, it houses biographical accounts and other information for 52,754 entities—characters, islands, wilds, settlements, ships, houses, and languages—that at one time existed in the 408-year history of the simulated world, variously called ‘Diol', ‘Diel', and ‘Dial' across its several language families. Each character's entry contains a written record of the person's entire life, from birth to death—the longest of these is nearly 27,000 words, the length of a novella. [Note: I had considered submitting the project to the National Novel Generation Month (NaNoGenMo) [nanogenmo], but I never did. NaNoGenMo is an alternative to National Novel Writing Month [jones2013wonks] that calls for the submission of computer programs that can generate 50,000-word novels (or other text works of commensurate length).] From an architectural standpoint, *Diol/Diel/Dial* is actually a collection of 52,754 individual web pages. [Note: My later collaborators Matt and Chris Antoun were appalled to discover this and urged me to convert it into a more elegant web app that would dynamically generate each entry as needed (from a data store). I never found the time to do this, though.] The encyclopedia is a hypertext in that each entry includes hyperlinks to several other entries. As I noted above, it is presented using a scroll skeuomorph. A reader interacts with it by navigating through its entries using conventional browser interactions: scrolling, clicking hyperlinks, and using the browser's ‘Back‘ and ‘Forward‘ buttons as needed; it does not accept interaction beyond mouse clicks and conventional browser navigation.

## Encyclopedia Generation

While *Diol/Diel/Dial* recounts a specific storyworld, Diol, it was generated by a framework that is capable of producing encyclopedias for any *World* world. In each case, the data contained in a given encyclopedia is recorded as the simulation proceeds—this is the task associated with the *chronicler* module in the curationist architecture that I presented in Chapter 5. [Note: In the code for this project, there is no explicitly demarcated chronicler module, but rather that functionality is carried out as part of the simulation's operation. Nonetheless, I think it is helpful to conceive of this functionality as being bundled together conceptually, and I will refer to this conceptual grouping as the ‘chronicler'.] Each encyclopedia contains entries for all entities that have ever existed in its storyworld: the entire world, as well as its ocean, islands, settlements, and wildernesses; all characters that have ever lived; all ships and homes that have been constructed; and all past and present languages (along with a few prehistoric ancestors). Each individual entry, then, contains general information about its subject entity, as well as a prose account of that entity's existence in the simulation.

#### Entry Contents

The segment of an encyclopedia entry that lists general information about the entity includes things like a person's hometown and family members, a ship's builder, an island's discoverers, and so forth, with references to those other entities being stylized as hyperlinks to their own entries in the encyclopedia. The HTML for this portion of an encyclopedia entry is composed at the time of encyclopedia generation by querying data pertaining to these various types of information, which the chronicler tracks for each entity in the simulation.

Each encyclopedia entry also includes a prose component (under the title ‘Narrative') that recounts all the storyworld's emergent events that have pertained to that entity. More specifically, this section provides a detailed yearly account of an entity's entire existence in the simulation, from its origination (or the beginning of the simulation, if it already existed then) up to the time at which the encyclopedia was generated. These prose accounts are stored as lists of strings, where each list corresponds to a year of the storyworld's history and each string recounts a specific event in rudimentary natural language that is written in the third person.

[Table]
{l | l | X}
  **Subject** & **Birthplace** & **Template**

  Newborn & Settlement & ‘[newborn.name]‘ was born to ‘[mother.name]‘ and ‘[father.name]‘ in the town of ‘[mother.location.name]‘, on the island of ‘[mother.location.location.name]‘.

  Newborn & Wilderness & ‘[newborn.name]‘ was born to ‘[mother.name]‘ and ‘[father.name]‘ in the wilderness of the island of ‘[mother.location.location.name]‘.

  Newborn & Ship &  ‘[newborn.name]‘ was born to ‘[mother.name]‘ and ‘[father.name]‘ aboard the ‘[mother.location.name]‘.

  Parent & Any & ‘[mother.name]‘ gave birth to a ‘[newborn.nominal_ gender]‘ named ‘[newborn.name]‘, of whom ‘[newborn.father.name]‘ is the father.

  Sibling & Settlement & ‘[sibling.name]‘'s younger ‘[newborn.relation _to(sibling)]‘, ‘[newborn.name]‘, was born in the town of ‘[mother.location.name]‘, on the island of ‘[mother.location.location.name]‘.

[Example templates used to produce rudimentary natural language expressions of emergent *World* events.]{Example templates used to produce rudimentary natural language expressions of emergent *World* events, so that they may be recounted in generated encyclopedias. For each type of event that may emerge in the simulation, I authored several template variants corresponding to different ways of recounting that event (depending on variable events details and which entity is to be treated as the subject). This was a tedious authoring task that preceded (and precipitated) my work on natural language generation.}{}

#### Event-Description Templates

The text fragments recounting emergent events are generated from templates that are associated with specific event types, such as the birth of a character or a mutiny aboard a ship. For most event types, I authored several template variants that differ according to details like which character is treated as subject or where the event took place, among other things. Table 8.1 shows a few of the many template variants that are used to generate accounts of a character's birth. The templates that I authored pertain to events of various magnitudes, from very minor events (such as a character regretting leaving their home island or worrying about a loved one who left on a ship) to very major events (such as an attempted mutiny aboard a ship). In the case of some major events, I authored multiple component templates corresponding to specific aspects of the event, which are then filled in and concatenated together to recount an entire event.

In the end, I authored templates for nearly every simulated event type, which means the chronicler is very thorough in its record keeping. Still, however, an array of simulated phenomena are not recorded, such as the deliberation processes that characters carry out in making decisions. As I have noted several times above, it is generally infeasible to perfectly capture the transpiring phenomena of a computer simulation—this is due to memory constraints, but also to the authorial burden of defining the functionality by which a chronicler may dynamically store a record of each type of phenomenon that is to be captured.

Generally, this authoring process was extremely tedious. Due to the number of event types and the number of template variants that were needed to recount each one, I was forced to craft several thousand templates. As I wrote in my rejected workshop paper on the project, the experience left me wanting more robust methods for procedural text generation (citations inserted for clarity):

> But because there were very many event types for which I authored several templates each, authorial burden in this task was still considerable; I wrote several thousand templates. [...] Moving forward, I am interested in exploring new, smarter simulationist authoring practices by which one may hand-author even smaller content units—ones corresponding to particular *aspects* of in-game events—and recombine them more smartly so that the resulting composition is more readable. Here, I might consider utilizing Tracery [compton2014tracery], a grammar-based authoring tool that handles low-level issues of recombination in a way that promotes authorial leverage. I might also consider a policy that is used in [Aaron Reed's *Almost Goodbye*] [reed2012sharing], in which certain language usages are kept track of and may force regeneration of text segments that reuse them too quickly.

 At this time, I was just beginning to explore the prospect of natural language generation in games [ryan2014toward,ryan2014combinatorial], and within a few months I would design an authoring tool for text generation called *Expressionist* [ryan2015natural], which was developed soon after with the help of several undergraduates [ryan2016expressionist]. [Note: These were Taylor Owen-Milner, Max Fisher, Ethan Seither, and Tyler Brothers, the latter of whom is now leading development on the tool. We hope to release an initial version soon.] Indeed, *Expressionist*'s design was inspired by *Tracery*, and in my own projects that have utilized it, I have implemented procedures that limit repetition in the style of *Almost Goodbye*, though by a probabilistic mechanism instead. In Chapter 12, I will discuss this style of text generation as it is utilized in my project *Sheldon County*. Suffice it say that my negative experience with template-driven text generation in *Diol/Diel/Dial* precipitated my quest for more expressive, generative, and controllable methods for text generation in computational media.

#### Notorious People

Finally, I should note that the storyworld's entry in a generated encyclopedia (which acts as a portal to the other entries) will include a list of ‘notorious people', which lists entities who are outliers with regard to various dimensions of fame or infamy. This list is constructed by querying over every character who has ever lived with regard to the kind of data that is tracked by the system's chronicler. Though not full-fledged narrativization, this is the result of curation that does involve some basic story sifting—a more satisfying curation procedure, however, would result in generated stories recounting these remarkable biographies. Here is what this component looks like in *Diol/Diel/Dial* (the entity names are stylized as hyperlinks that point to the entries for those entities):

> Lived the longest: Winifred Gilles (114 years)

Had the most children: Reese Guillermo (24 children)

Has the most descendants: Neville Vaclav (38065 descendants)

Most experienced ship captain: Capt. Iain Prasun (14 years)

Discovered the most islands: Capt. Kevin Vaclav (2 islands)

Founded the most settlements: Ella Vaclav (3 settlements)

Died or is living the furthest from birthplace: Anny Bert (9766 miles)

Built the most ships: Enya Gilles (2 ships)

Built the most houses: Aguste Vaclav (10 houses)

Deadliest murderer: Capt. Talbert Shumeet (25 victims) [Note: This information is presented in the entry for the world of Diol itself, which is available here: https://users.soe.ucsc.edu/~jor/diol/diol.html.]

## The Pleasure of *Diol/Diel/Dial *
In what ways does *Diol/Diel/Dial* succeed? This is the primary consideration of this section and, as I have noted above, my discussion here will rely in large part on the ideas that were presented in Chapter 3.

#### Artist's Statement

For any artistic work, the most generous appraisal of the piece probably appears in its attached artist's statement, so perhaps I should start there. [Note: To be fair, I will use these words against me in the next section.] Here is the description of the work that I included in my rejected submission to the Art Exhibition of the International Conference on Interactive Digital Storytelling, whose curatorial theme was *fragmentation*: [Note: Here is the call for artworks: “Fragmentation can refer to the way the processes of remediation and transmediation are juxtaposing or contrasting stories from old or traditional media (oral storytelling, written stories, theatre, comics, radio, television, film), and how this fragmentation shapes how we experience and understand interactive stories in the new medium. At some point we need to re-fragment or de-fragment these stories so that we can make sense of new ways of telling and experiencing stories through interaction with respect to the affordances of the new medium. Perhaps construction and sense making of these micro and mini narrative fragments also depends on the signs of time, to the zeitgeist specifically?” [erkut2015call, n.p.].]

> *Diol/Diel/Dial* is an explorable hypertext encyclopedia of a virtual world generated by a simulationist text adventure called *Islanders*, also by the artist. A gargantuan tome of nearly 100M words, it houses biographical accounts and other information for 52,754 entities—charact- ers, islands, wilds, settlements, ships, houses, and languages—that at one time existed in the 408-year history of the simulated world, variously called “Diol”, “Diel”, and “Dial” across its several language families. It evokes the old medium of the encyclopedia, but also the older medium of biography: each character's entry holds a written narrative of the person's entire life, from birth to death. The longest of these is nearly 27,000 words, the length of a novella, and each is written from an omniscient point of view that dips at times into the characters' mental and emotional worlds. In this sense, the work diverges from these old media into the computational: it could not have been made by hand (it is too massive) and could not have been made about a real world (its author, a suite of algorithms, is all-seeing and all-knowing with regard to its subject matter). While the entire volume is a perfect, totalizing account of Diol, its magnitude is such that the reader can only come to understand the world through smaller narrative fragments. A fragment, in this sense, is a reader's traversal across the hypertext—she starts at some character, clicks to read about her mother, thence the mother's birthplace, thence the ship at port there, thence its builder and his father and his language and its last speaker, and so forth. In carrying out such a traversal, the reader combines and recombines narrative fragments to form her own conception of Diol and the simulation that generated it. As such, interaction with the piece is not like exploring so much as accumulating narrative material with which to sculpt a coherent (though fragmented) view of the world. It is influenced in this way by the sculptural storymaking of Aaron Reed's *18 Cadence*, and more generally by the sheer immensity of Tarn and Zach Adams' *Dwarf Fortress*. While engaging *Diol/Diel/Dial* already feels like interactive storytelling, the piece itself is generated from a work of interactive narrative—one of the characters who lived in the simulation was controlled by a human playing *Islanders*, but unlike in player-centric game worlds, his interactions and their byproducts only ripple across the massive sum total of simulated events recounted in the larger piece. In this way, *Diol/Diel/Dial* is also an expression of the simulationist approach to interactive storytelling, and of the artist's view that non-player characters should be foregrounded ahead of human players.

#### Connection to Sculptural Fiction

In Aaron Reed's *18 Cadence* [reed2013cadence], the player sifts through an accumulation of material that captures the affairs in and of a fictional house over the course of the twentieth century; it was first described in his master's thesis [reed2011sequence, pp. 32–37], then later in his dissertation [reed2017changeful, pp. 126–130]. While it is not emergent narrative, since Reed himself authored all its materials, it is certainly curationist: players assemble fragments of the material—represented using a *magnetic poetry* skeuomorph, as Figure 8.3 shows—into lyrical recounts of what has happened in the house. This instantiates a curatorial gameplay mode that Reed identifies as *sculptural fiction* (citation inserted for clarity):

> Sculptural fiction attempts to address these shortcomings by giving the player a different kind of role. Rather than centering the act of navigation through a fixed graph of story nodes, the primary focus becomes the act of exploring the space of possible connections. Through mechanics for linking and unlinking nodes, the player builds a story, rather than moving through an existing structure. [...] The term “sculptural” is meant to suggest the constructive and iterative process of the artistic practice of sculpting, particularly the tradition of “assemblage” where existing objects are arranged to form an artistic result [rawson1997sculpture, p. 15]. The term also suggests the sculpture itself, an artifact produced by this act and designed to be exhibited and shared [reed2017changeful, p. 104]

[Figure omitted]

This concept is clearly related to the notion of curation that I have propounded in this dissertation, particularly human-powered curation, but it differs from curationist emergent narrative in that sculptural fiction does not depend on its accumulated material being generated. As such, my framework is intended as an intervention in the area of emergent narrative, which I view as working analogously to nonfiction (see Section 3.1.1), while Reed's framework is meant to constitute a new mode of interactive fiction. In any event, at the time of creating *Diol/Diel/Dial*, I viewed the work as an effort in *generative* sculptural fiction, one built primarily on the pleasure of sifting. [Note: As the next section will show, however, such an appraisal is too generous, due to the its lack of curatorial affordances, the presence of which critically structures *18 Cadence*'s gameplay.]

#### Evoking the Aesthetics of Emergent Narrative

Let us consider this project in relation to the aesthetics of emergent narrative that I identified in Section 3.2. In a sense, *Diol/Diel/Dial* exhibits the *aesthetics of the coauthored*, since a player comes to understand the world of Diol through the narrative (fragments) that she constructs about it. Moreover, in that every player will likely make a unique traversal through the encyclopedia, constructing a unique narrative understanding thereby, the *aesthetics of the personal* are likely activated to some degree.

In evoking ‘the sheer immensity' of *Dwarf Fortress* in my artist's statement, I was identifying what I would now called the *aesthetics of the vast*, and there is no doubt that *Diol/Diel/Dial* is vast—it total nearly 100M words, making it an inexhaustible tome. Due to its size, each of its embedded narrative threads weaves into the massive backdrop of a record of an entire world, and in this way the *aesthetics of a larger context* are yielded. Moreover, because no human could write such a tome, the encyclopedia clearly exhibits the *aesthetics of the unauthored*. Generally, the pleasure of the work is constituted in its inclusion of the kind of emergent stories and images that I identified in Section 7.4. In that such phenomena are rare—signals in the noise—the actual emergent narrative of *Diol/Diel/Dial* has the *aesthetics of the improbable*, and the reward for sifting it out is the *aesthetics of the uncovered*. When improbability is at play, the *aesthetics of the actual* and the *aesthetics of the ephemeral* are often attendant, though in the case the latter are perhaps inhibited in that a written account persists.

Lastly, the *aesthetics of the uncanny* are clearly present in this work, if only for the ridiculously ubiquitous blood feuds that dominate the storyworld's history. This was a feature that was implemented just before generating the encyclopedia, and it is clear that the frequency of killing (and especially killing to avenge) should be tuned to be much lower. At the pinnacle of its uncanniness, *World*'s characters may even murder themselves to avenge their own killings of their own loved ones—this occurs when a character becomes obligated to avenge her victim's death. While this is a bug, its expression suggests a peculiar honor code in *World*'s procedural societies. When this is not viewed as a bug, but rather as a strange truth about these storyworlds, the aesthetics of the uncanny may obtain. [Note: Of course, there a limit to this, and in any event the merits of uncanniness are highly subjective, as I discussed at length in Section 3.1.4.] In Chapter 10, I will explain how a bug in the *Bad News* code that handles romantic attraction produced a peculiar aesthetic hallmark of the work.

#### Narrative Portals and Emergent Images

While full-fledged story sifting is not carried out in this project—as the next section argues, the task is basically impossible due to bad chronicling—an approximation appears in the form of the ‘Notorious People' section (introduced above) of the entry for the world of Diol itself. This is the result of a basic curation procedure, and the links to these (in)famous entities can work as portals to potential narrative intrigue. For example, the “deadliest murderer” Capt. Talbert Shumeet is listed as having taken the lives of twenty-five victims. In viewing his encyclopedia entry, an intriguing emergent story is encountered:

> 290. Talbert Shumeet is trying to gather up a group of people in Cummington for an expedition on the Microbarograph, which Talbert Shumeet has agreed to lead. [...] Talbert Shumeet's shipmates have decided that he would make the best captain, though Dimitri Gilles is hurt by the decision. Talbert Shumeet was designated captain of the Microbarograph, which will depart Cummington shortly; he has never captained a ship before. The Microbarograph, under command of Capt. Talbert Shumeet, has left the harbor at Cummington, Waterford. The Microbarograph is heading N at quarter-speed. Capt. Talbert Shumeet married Indira Gilles on the Microbarograph; they now share quarters aboard the Microbarograph. Indira Gilles is now named Indira Shumeet. As an act of mutiny, Dimitri Gilles is going to attempt to murder Capt. Talbert Shumeet, whose decision-making aboard the Microbarograph has been seen by some passengers as reckless and potentially life-endangering. Dimitri Gilles attempted to murder Capt. Talbert Shumeet as an act of mutiny, but failed. A punishment is now being decided for Dimitri Gilles and his conspirators in the failed mutiny attempt: Max Noland, Hezekiah Aldis, Cassey Jordan, Uriel Prasun, Ariel Vaclav, Max Zollie, Jefferson Vaclav, Shep Gilles, Yigal Shumeet, Casandra Noland, Philis Melvin, Misha Vaclav, Jae Zollie, Hodge Gilles, Anna-Diane Zollie, Gerladina Aldis, Catarina Aldis, Pip Melvin, Emylee Aldis, Nate Vaclav, Bert Aldis, Frederica Shumeet, Breena Melvin, and Eddie Aldis. Since there are no nearby uninhabited islands on which the conspirators could be marooned, they will be hanged aboard the ship. Capt. Talbert Shumeet ordered the hanging of 26-year-old Max Noland at sea aboard the Microbarograph as capital punishment for a failed mutiny attempt in which Max tried to murder Talbert. Capt. Talbert Shumeet ordered the hanging of 23-year-old Hezekiah Aldis [...] Capt. Talbert Shumeet ordered the hanging of 44-year-old Eddie Aldis at sea aboard the Microbarograph as capital punishment for a failed mutiny attempt in which Eddie tried to murder Talbert. This was his 25th victim. [...] Capt. Talbert Shumeet is worrying about his father Capt. Addie Shumeet, who 11 years ago left for a voyage on the Microbarograph and hasn't been heard from since.

291. Capt. Talbert Shumeet is now 25 years old. The Microbarograph is heading NE at full-speed. The Microbarograph sank 275 miles off the coast of Judgement; there were 30 people aboard. Capt. Talbert Shumeet's wife, Indira Shumeet, drowned in the open ocean after the Microbarograph was shipwrecked; she was 27 years old. Capt. Talbert Shumeet drowned in the open ocean after the Microbarograph was shipwrecked; he was 25 years old. [Note: This encyclopedia entry is available at: https://users.soe.ucsc.edu/~jor/diol/5845211088.html.]

As the generated prose expresses, Shumeet's victims were in fact failed mutineers whom he ordered to be hanged as capital punishment. This scene is mildly interesting, and along with his marriage aboard the ship just prior, it forms part of an interesting year in a generated character's life. Moreover, clicking on the entry for his father, whom he worried about, reveals that the former was actually murdered by mutineers on the voyage under question. [Note: The entry for Shumeet's father, Capt. Addie Shumeet, is hosted here: https://users.soe.ucsc.edu/~jor/diol/5505259280.html.] The real intrigue of this story, however, is in its emergent punchline: karma shipwrecks the storyworld's deadliest adjudicator off the coast of an island called Judgement. Fragments like these (and the other emergent images recounted in Section 7.4) are what provides the pleasure of *Diol/Diel/Dial*, and they are the tokens of *World*'s meager success.

#### It Is Mounted

Finally, I would like to note that I am satisfied with this project in the sense that I actually mounted the generated content in a full-fledged media work—there is no ‘failure to mount' here, which was the curation pain that I identified in Section 4.2.5. There is a problem, however, in that a raw chronicle, rather than a set of curated narrative artifacts, is what is mounted in the experience. I will discuss this further in Section 8.5.

## The Pain of *Diol/Diel/Dial *
In the style of Miguel Sicart's (2015) takedown of Miguel Sicart (2008)—[sicart2015loops] vis-à-vis [sicart2008defining]—in this section I will brutalize my own work in the third person (and in doing so I will traverse a strange loop).

#### The Aesthetic of Big Numbers Is Dead

James Ryan's *Diol/Diel/Dial* (2014) is ostensibly a work of emergent narrative, imbued with the aesthetics of emergent narrative that Ryan identifies in his dissertation [ryan2018curating, pp. –], but it suffers primarily for the reason that its actual emergent narrative is difficult to identify (and moreover, to isolate). In his rejected submission to the art exhibition of the 2015 International Conference on Interactive Digital Storytelling, Ryan describes the piece as “a gargantuan tome of nearly 100M words”—but as Mike Cook has declared, “the aesthetic of big numbers is dead” [heaven2016when, n.p.]. It is trivial to generate massive volumes of text, and as such, the size of an artifact should never be its selling point. Yes, magnitude in this sense pertains to what Ryan calls “the aesthetics of the vast” [ryan2018curating, p. ], but such aesthetics are constituted in contextual (and conceptual) intrigue that is meant to *amplify* the appeal of an actual narrative artifact or storied experience—vastness alone is a hollow entertainment, as Cook declares eloquently.

#### No Curatorial Affordances

In his failed exhibition submission, Ryan seems to imply that merely traversing the encyclopedia results in a vaguely storied understanding of the material that is encountered thereby:

> While the entire volume is a perfect, totalizing account of Diol, its magnitude is such that the reader can only come to understand the world through smaller narrative fragments. A fragment, in this sense, is a reader's traversal across the hypertext—she starts at some character, clicks to read about her mother, thence the mother's birthplace, thence the ship at port there, thence its builder and his father and his language and its last speaker, and so forth. In carrying out such a traversal, the reader combines and recombines narrative fragments to form her own conception of Diol and the simulation that generated it.

 Ryan then goes on to compare this experience to the sculptural fiction of Aaron Reed's *18 Cadence* [reed2013cadence,reed2011sequence,reed2017changeful], but in doing so he fails to recognize a critical component of the latter that is not present in *Diol/Diel/Dial*: Reed's interface provides *curatorial affordances* by which a player can actually assemble material into manifest narrative artifacts. In Ryan's piece, there is only jumping across links, a whirlwind tour from entry to entry to entry, the culmination of which—a mere termination, actually—leaves the player's understanding jumbled and fragmentary. To be clear, this is because she has no means with which to actually assemble a coherent understanding (let alone a narrative artifact).

Yes, there may be a kind of storied nature to the mere traversal of *Diol/Diel/Dial*, but as Ryan himself notes in his dissertation, this is trivially characteristic of all human experience:

> When narrative psychologists write of the “storied nature of human conduct” [sarbin1986narrative], they do not mean that human conduct itself constitutes narrative—that the raw unfolding of time is literally a story—but rather that *we can only understand it as a kind story*, since we process the world and our experiences in it through a mechanism of narrative sensemaking. [ryan2018curating, p. ]

 Of course, a player who earnestly seeks to behold Diol's emergent narrative could make use of inscription technologies, such as a pen and pad, to emulate the essential experience of *18 Cadence*—but who would be so inclined? As Ryan himself notes in his dissertation, such player inclinations are typically found only in popular videogames—where they depend on attachment to a personally experienced storyworld—and indeed the more probable *dis*inclination toward curation may account for the general failure of academic emergent narrative:

> this pattern of human curation does not seem to be prevalent outside the realm of popular videogames. In the case of non-interactive story generation, and indeed many other interactive systems, humans may not be so inclined to tame the welter. To again invoke my coadvisor Michael Mateas's critique: without curation, emergent narrative is *just one damn thing after another*. [ryan2018curating, p. ]

 ‘One damn thing after another', indeed (Mateas is my coadvisor as well). This is unfortunately an apt description of *Diol/Diel/Dial*'s ‘narrative' entry components, and with all due respect to the project's creator, it is not surprising that his exhibition submission was rejected. While *18 Cadence* invites human curation, this piece makes it unpleasant work.

#### It Mounts a Raw Chronicle

If a work's output does not incentivize human curation of narrative artifacts, or even cultivation of a storied understanding, then the success of the project will depend on the system doing its own curation. This is the subject of Ryan's dissertation—he calls the framework ‘curationist emergent narrative'—and from his description of the underlying simulation engine *World*, it is clear that such automatic curation would be infeasible in *Diol/Diel/Dial*. This is due primarily to the system's shoddy recordkeeping about past events—to use Ryan's term against him, *World*'s ‘chronicler' is deficient. Intriguingly, it fails in the same way that Arthur Danto's *ideal chronicler* [danto1962narrative] fails: while it maintains an essentially perfect record of the storyworld *as events transpire*, such online recordkeeping does not produce narrative because the latter depends on a postmortem vantage (one that affords reasoning about causality in the large). Here, I will let Danto explain (‘I.C.' stands for ‘ideal chronicler'):

> Is not the I.C. definitively complete? [...] Of course it is complete—but complete in the way in which a witness might describe it, even an Ideal Witness, capable of seeing all at once everything which happens, as it happens, the way it happens. *But this is not enough*. [...] The whole truth concerning an event can only be known after, and sometimes only *long* after the event itself has taken place. And this part of the story historians alone can tell. It is not something which even the best sort of witness can know. What we deliberately neglected to equip the Ideal Chronicler with was knowledge of the future. [danto1962narrative, p. 154]

To be clear, in Ryan's curationist architecture [ryan2018curating, pp. –], a chronicler is not supposed to know the future. Instead, the ‘story sifter' and ‘narrativizer' modules are to operate over the chronicle after it has been constructed, with such a postmortem vantage enabling the construction of what Danto calls ‘narrative sentences' [danto1962narrative] and what Hayden White calls ‘motifs' [white2014metahistory]. The problem in *World* is that the chronicler does not store records that enable the reasoning that is required to develop such narratively and rhetorically charged material. I have learned from Ryan himself that the simulation engine does not store representations of events beyond the prose strings that compose the narrative components of *Diol/Diel/Dial*'s encyclopedia entries. [Note: Intrapersonal communication, June 14, 2018.] As such, story sifting in this framework becomes a matter of natural language processing, and that is a difficult challenge.

More ideally, the system would store *structured data* about past events, which a story sifter could then reason about automatically without having to understand natural language. However, even if the system could perfectly understand the event expressions, reasoning about causal relations among them is an outstanding issue that would require commonsense knowledge about which kinds of events tend to cause others. As such, *World*'s chronicler is further deficient in its lack of ‘causal bookkeeping' [ryan2018curating, p. ]. So while the pain of *Diol/Diel/Dial* is rooted primarily in its lack of curation—it mounts a raw chronicle as a media experience—this is more specifically due to poor chronicling, inasmuch as that module fails to record critical information about past events.

#### Horrible Prose

Beyond its poor curation, {which Ryan himself acknowledges} [ryan2018curating, p. ], *Diol/Diel/Dial* clearly suffers from additional presentation issues in terms of its generated prose. Here is an example passage:

> Travers Zollie is now 18 years old. Travers visited his friend Nevil Gilles, who has fallen ill and is likely to die soon. Travers visited his friend Dory Noland, who has fallen ill and is likely to die soon; they first met 18 years ago in Doyle, Waterford, where they meet now for the last time. Travers visited his friend Murphy Zollie, who has fallen ill and is likely to die soon; they first met 17 years ago in Doyle, Waterford, where they meet now for the last time. Travers visited his friend Cesya Melvin, who has fallen ill and is likely to die soon. Travers visited his friend Boyce Zollie, who has fallen ill and is likely to die soon; they first met 11 years ago in Doyle, Waterford, where they meet now for the last time. Travers visited his cousin, Armando Melvin, who has fallen ill and is likely to die soon. Travers visited his friend Friedrich Guillermo, who has fallen ill and is likely to die soon; they first met 10 years ago in Doyle, Waterford, where they meet now for the last time. Travers Zollie moved to Cummington, 11 miles away from Doyle. [...]

 Ryan acknowledges this deficiency in his rejected submission to the Workshop on Procedural Content Generation:

> Even though I authored a considerable amount of content, the encyclopedias are generally not pleasing to read. For one, they are quite repetitive; this is in spite of me having written stylistic variants for most templates. More pointedly, though, they do not read well because I do not enact any discourse planning in their composition. When a new event happens, a text representation of it is composed and then appended to the existing encyclopedia representation.

 As this admission points out, the poor presentation is in part rooted in the issues that I have just outlined with regard to chronicling. Discourse planning is a difficult prospect when the content of the next sentence cannot be known because it will recount an event that has not yet transpired. [Note: As Stephanie Boluk and Patrick LeMieux have noted, the prose of *Dwarf Fortress*'s Legends mode [dflegendsmode] suffers from the same issue: “the historical sequence appears strange and arbitrary, with no transitional expressions or conjunctive adverbs to tie together the sentences” [boluk2013dwarven, p. 144]. I am not sure whether these descriptions are also generated in an online fashion; if not, poor discourse planning would instead be the cause.] To be fair, such *just-in-time discourse* does characterize certain genres—*sports commentary* [rhodes2010enhancing] comes to mind—but this is not how discourse works in the encyclopedia genre. So while the quality of the prose does suffer due to Ryan's simple approach of template-based text generation, the real issue is that there is no reasoning at the level of discourse, since events are recounted in an online manner, as they transpire.

#### An Aesthetic Hodgepodge

It is worth also noting some issues at the level of *World*'s character simulation. A primary issue in this regard is that the system's modeled domain is hodgepodge of mixed concerns whose peculiar gestalt yields a dubious aesthetics. The world is vaguely historical, yet it features the distinctly modern phenomenon of divorce; Ryan cites a Polynesian influence [ryan2018curating, p. ], but its character names are largely western-coded (and taken together constitute another strange admixture).

#### Simulation Pains

In terms of the simulation pains identified in Section 4.1 of Ryan's thesis, *World* does not lack. First, the system exhibits some granularity extremes, namely its low granularity with regard to its modeling of time, which proceeds by coarse year-sized chunks. While the engine is meant to generate stories that take place over the course of decades or centuries of story time [ryan2018curating, p. ], its simulation detail is low enough to significantly limit the system's possibility space. At the other extreme, the modeling of character language phenomena is conspicuously rich relative to the modeling of other phenomena—though it may represent a technical contribution to the computational modeling of language change, the subsystem is probably too detailed in that it entailed significant authoring effort while producing little narrative potential in return. Moreover, we find at least one critical modeling gap: while the simulation dips into the internal worlds of characters, the social simulation is extremely rudimentary. All of this modeling disparity violates Richard Evans's directive that “Different systems should be of similar levels of granularity” [evans2010modeling, n.p.].

Though the *World* does model personality, character behavior is hardly driven by such concerns. In terms of authoring, a lack of modularity has further hindered the project, as Ryan indicates in his rejected workshop submission:

> But because there were very many event types for which I authored several templates each, authorial burden in this task was still considerable; I wrote several thousand templates. [...] Moving forward, I am interested in exploring new, smarter simulationist authoring practices by which one may hand-author even smaller content units—ones corresponding to particular *aspects* of in-game events—and recombine them more smartly so that the resulting composition is more readable.

While one pitfall in emergent narrative is that all the events generated in a given storyworld may be boring—uninteresting or not *tellable* [labov1997narrative]—event sequences may be paradoxically boring if they are *too* interesting, as Ryan himself notes in his disseration:

> 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* [ryan2018curating, p. ]

 In the case of *World*, and thereby *Diol/Diel/Dial*, there are simply too many murders, mutinies, and blood feuds. Of the storyworld Diol's 32,233 character deaths, 3,107 are due to murder—that is too high a rate, and the resulting onslaught (in the encyclopedia) of murder after murder makes each one unremarkable. As such, even though murder is what Roger Schank calls an ‘absolute interest' [schank1979interestingness] and what Robert Wilensky terms a ‘human dramatic situation' [wilensky1982points], emergent cases of it in *Diol/Diel/Dial* are not ‘reportable events', to use William Labov's phrase [labov2006narrative]. This does not mean that all of the storyworld's murders are uninteresting, it just means that none are *inherently* interesting; instead, intrigue depends on additional circumstances that may contextualize a murder in a remarkable way.

#### No Contingent Unlocking

Finally, *World*'s most acute blunder is its lack of *simulation feedback*, which limits both its narrative possibility space and the prospect of curating that space. In his dissertation, Ryan identifies a feedback mechanism at work in Sheldon Klein's murder-mystery generator [klein1971program,klein1973automatic] whereby emergent events may contingently unlock subsequent emergent events, which over time generates storylines weaved with clear causal threads:

> in this way, emergent actions in Klein's system *unlock* future emergent actions that may build upon them to grow full-fledged emergent storylines—such unlocking is an expression of the contingency relationship, and thus such emergent storylines are precisely what I have termed ‘emergent contingency structures'. I call this method *contingent unlocking*. For me, the phenomenon suggests the image of a plant growing out of a spongy forest bed: meaning is constituted against the soft backdrop of entropy. [ryan2018curating, p. ]

As Ryan notes, this kind of feedback mechanism is useful not only for its potential to generate coherent emergent storylines, but to automatically *identify* them as well, by a process he calls ‘causal bookkeeping'. That is, whenever an emergent event is explicitly contingent on a past event, the system records a causal relation between the two, and an emergent storyline may then be identified by reasoning over causal chains. In *World*, there are only a few kinds of events that are defined according to a contingency relation. For example, revenge killings are contingent on the earlier killings for which they exact retribution, and from this simple mechanism chains may emerge in the form of blood feuds. These are recognizable by the system—though unfortunately *Diol/Diel/Dial* does not showcase them—but the resulting sequences are not especially interesting because they are composed by a single action type. However, it should be noted that *World* does model event contingency in its mutiny simulation, as Figure 8.4 illustrates.

#### No Causal Bookkeeping

Unfortunately, as I have already explained, the system fails to store data about simulated events, which means that mutiny sequences cannot be reasoned about (or even retrieved) after they transpire. If event representations *were* stored in the chronicle, and if the system did do causal bookkeeping, then *Diol/Diel/Dial* could operationalize William Labov's procedure of *narrative pre-construction*:

> Before a narrative can be constructed, it must be pre-constructed by a cognitive process that begins with a decision that a given event is reportable. Pre-construction begins with this most reportable event and proceeds backwards in time to locate events that are linked causally each to the following one, a recursive process that ends with the location of the unreportable event—one that is not reportable in itself and needs no explanation. [labov2006narrative, p. 37]

[Figure omitted]

 In the case of an emergent mutiny storyline, a story sifter might begin at the mutiny attempt itself—the most reportable event—and then chain backward across causal links through the series of events that precipitated the attempt, terminating in either the captain's affront or some earlier event (pending an operational notion of tellability). Likewise, the system could chain forward in this same way, thereby retrieving the remainder of the emergent storyline. This extracted material could then be handed off to a narrativizer module that would be tasked with generating a full-fledged narrative artifact recounting the mutiny. Finally, the curated story could be mounted in *Diol/Diel/Dial*, perhaps in an entry dedicated to the mutiny itself. Moreover, if the simulation engine modeled other pockets of its narrative possibility space according to explicit contingency relations—so that the kind of structure shown in Figure 8.4 would also characterize other possible storylines—then other kinds of stories could also be curated and mounted into the encyclopedia. In this way, the experience of *Diol/Diel/Dial* would not be that of an overwhelming encounter with endless dubious prose, but one of encountering a compilation of actual stories recounting the history of a storyworld. To recapitulate, this would require both contingent unlocking and causal bookkeeping.

As he explains in the final chapters of his dissertation, the various failings of *World* and *Diol/Diel/Dial* would ultimately drive Ryan's simulation and media practice toward a series of refined techniques that would eventually solidify in the form of his curationist architecture.

## Curationism and *Diol/Diel/Dial *
While a series of curation pains were just articulated in the last section, briefly I would like to specifically outline how the curationist architecture, introduced in Section 5.3, is instantiated in *Diol/Diel/Dial*. As my look-alike critic noted above, this project is peculiar because it mounts a raw chronicle as a media experience. This violates the curationist call to process chronicles so that actual narrative artifacts may be constructed (through the procedures of story sifting and narrativization), which are what should actually be mounted into experiences. Curiously, while it fails to meet this call, *Diol/Diel/Dial* does go halfway in that it avoids the ubiquitous pitfall (identified in Section 4.2.1) of avoiding curation altogether to instead mount the storyworld itself as a media experience (a move typically coupled with the conflation of the raw transpiring of simulation with something that could rightly be called ‘emergent narrative').

[Figure omitted]

For this case study and the following, I will answer the following question: *for each curationist architectural component, who or what in this project acts as that component?* Figure 8.5 illustrates these concerns with regard to *Diol/Diel/Dial*, which I will also outline in prose, component by component:

-  **Experiencer.** *Diol/Diel/Dial*'s *experiencer* is a human who visits the online encyclopedia to read through some of its entries. In the case of an encyclopedia being generated from an *Islanders* storyworld, there is another experiencer who interacts with the simulated storyworld that is itself recounted in the encyclopedia. In this case, the experiencer of the encyclopedia may be the same person as the *Islanders* player, or she could be someone else. Figure 8.5 does not capture the case of *Islanders* gameplay.

-  **Simulated storyworld.** A *World* simulation instance (the one in which Diol obtained) is the *simulated storyworld*.

-  **Chronicler.** As I noted above, there is no distinct chronicle module at work in *Diol/Diel/Dial*, but rather the associated functionality is built into the *World* code. To simplify, we can say that *World* is its own *chronicler*.

-  **Chronicle.** A *chronicle* in this project comprises all of the various data that is included in the generated encyclopedia; in particular, the narrative components of its entries (concatenated string representations of events) most resemble Hayden White's notion of a chronicle. As such, we can say that *Diol/Diel/Dial* is itself a chronicle, and curiously one that has been mounted as a media experience. *Dwarf Fortress*'s Legends mode [dflegendsmode], likely the most similar existing work to *Diol/Diel/Dial*, has itself been likened to a Whitean chronicle [boluk2013dwarven, pp. 143–149].


-  **Story sifter.** Here is where the curation pain begins—in *Diol/Diel/Dial*, the human encyclopedia reader is herself tasked with carrying out the duties of the *story sifter*. This is because the system does not do this automatically, which is in turn due to the simulation not being crafted in a way that would facilitate automatic sifting (see Section 8.4).

-  **Material.** In this project, the sifted *material* is simply a mental construct in the mind of the human encyclopedia reader. As she traverses through a series of entries, knowledge capturing the information she has encountered so far accumulates in her mind, and in total this is the sifted material. Alternatively, if the reader brings inscription technologies to bear—for instance, by taking notes using a pen and pad—then the material also obtains in the form of her inscriptions.

-  **Narrativizer.** The human encyclopedia reader is also forced to carry out the duties of the *narrativizer*. If she comes to discern emergent storylines in the materials that she has excavated—be they mental constructs, inscriptions, or a combination—that she has performed narrativization.

-  **Story.** If the human reader does indeed perform narrativization, she produces a narrative artifact that recounts Diol's emergent phenomena. Specifically, these phenomena will have transpired in Diol before being captured in the chronicle (*Diol/Diel/Dial* itself) and then sifted out of the chronicle in the form of material, which will have finally been narrativized into a *story*. This story is likely another mental construct that obtains in the mind of the reader, in which case *mental curation* (see Section 5.3.2) has occurred. Of course, she could also render the story in another form by mounting it in a media experience in the style of Tim Denee's *Oilfurnace*, in which case *feedforward curation* would be at work. The other architectural variant, *feedback curation*, is not possible here because it would require the reader to inject her constructed story back into the storyworld of Diol, but she has no means to do so. (In fact, Diol is gone forever and can never be retrieved—unlike *Talk of the Town* and *Hennepin*, storyworlds do not have associated seeds, which means they cannot be recovered once a *World* session ends.)

-  **Media experience.** *Diol/Diel/Dial* is of course the *media* experience in this project, but again it is curiously also the chronicle component. To reiterate the takeaway of this chapter, the mounting of a raw chronicle as a media experience is a major blunder. Also, as I have just noted, it would be possible for a reader to curate the emergent material captured in the chronicle (in *Diol/Diel/Dial* itself) to construct stories that could be mounted in a second-order media experience. To my knowledge, no one has done this.

#### Summary

To summarize, *Diol/Diel/Dial* employs the degenerate architectural variant of *mental curation* (see Section 5.3.2), since it leaves story sifting and narrativization as exercises for the reader. However, while in most examples of mental curation the storyworld is not curated at all, in this case a chronicle is automatically produced (and then mounted as the media experience, rather than the storyworld itself). While *feedforward curation* is possible here, it would require the reader to mount her curated stories about Diol into a second-order media experience, and to my knowledge this has not occurred. As I will discuss in the next case study, *Bad News* represents a novel experiment in having the experience itself do its own curation, but through the mechanism of a wizard-of-Oz hybrid human–computer scheme. Later, in Chapter 12, I outline my ongoing exploration of fully automatic curation in the generative podcast *Sheldon County*.