“It's always more work than anticipated, right? And the question is whether you keep ploughing. A lot of game concepts start out as something very small, right? They start as a joke. They start as a shitpost. They start out as a vibe. And it never seems to be as big an idea. Sometimes you have some clue, but you never know all the challenges when you first have the idea.” - P2

Most of the participants directly spoke about the Writing Burden of creating content for an interactive system. Although as P11 expressed, ‘the word count or the branch number or the play time or all these standard metrics that people use to measure the size and scope of games’ could be deceptive and not ‘map on really with the workload for the writing’. Dynamic Authoring, writing lots of conditional or variable content that can require different game-states to instantiate, could be lengthy or difficult due to its ‘combinatorial complexity’ (P4). It could also involve the ‘hard work’ of ‘creating grammars: taking sentences that you would like the work to be able to output and generalizing them... while still keeping underlying structure’ (P2).

Programming Labor — the creation of new game-behaviour, or developing new tools — was identified by some as a ‘frustration’ (P5), for others it was just a different kind of work that allows authors to use ‘different parts of [their] brain’ (P6).

Scopability, the possibility of judging the amount of work required for a project, was usually low at the start of making games. As P2 explained, ‘the idea in your head at the start of a project is never what it comes out at the end’, because ‘in order to make a project work, you have to nail down certain structural things which may change everything, and that includes what is the proper size of the thing’. This means that ‘it is not practical to scope a project once and then be done with it’. Scoping, then, was more often an ongoing process throughout the creation of the game, especially as Feature Creep would increase workload as new ideas emerged during creation or testing. What could seem at the start like ‘a quick, easy project that wouldn't take that long’, could soon ‘balloon out of control’ (P14). Feature creep was often seen as inevitable (‘we ran over, obviously, like you do in every game’ —P10), but this was not always a bad thing. Sometimes the author finds a better way to implement their work, often from early play testing, as P9 experienced: ‘eventually I saw how I could do things better... I started seeing these possibilities really as I started playing the game and started getting transcripts’.

“Just because I know what good writing looks like doesn't mean I can produce it.” - P4

Authors of interactive narrative have different abilities and interests. Authors can and do increase their technical and creative abilities, but at any given moment in a project there is a limit to what they can achieve on their own which varies substantially from author to author. They would come to each project with their own prior Writing Ability, or ‘design knowledge, something that the author builds up themselves in relation to what they're working on and the environment in which they're creating’ (P10).

Digital narrative is created with programming tools such as domain specific languages and game engines, and authors varied in their Tool Mastery. The tools themselves each have their own Technical Limitations to what can be achieved with them and sometimes ‘even just getting things to work’ could be ‘slow’ (P14). But moreover, once familiar with one tool, authors experience a considerable ‘cost in becoming familiar with something else’ (P1). This opportunity cost of time and effort in switching to more appropriate tools meant authors could often make do with what they were already familiar with.

“My usual approach whenever there's a lot of content, is if anybody goes through and reads all of it, I want them to feel rewarded rather than to be cheated.” - P1

An author of interactive narrative may have any number of goals for their work which will have concomitant impacts on the burden of work. Implementing high player Agency where a player ‘could actually make a real decision that had a major effect on the story’ (P4) or Replayability, were large factors driving the requirement to write lots of additional content. This extra content could vary in Legibility to the Player. A lot of work creating textual variation or deeper games mechanics could be done without it necessarily being ‘visible’ (P4, P13), ‘transparent’ (P8), ‘exposed’ (P7), or ‘surfaced’ (P11) to the player.

Content could be more complex to write if there was Puzzle Design needed. Authoring could be slower when attempting to maintain Narrative Quality, which some author expressed as paramount. As P7 expressed, ‘graphical fidelity and gameplay mechanics have gotten polished to a point that really narrative is the way you differentiate your IP’. When there were parallel paths through a narrative, maintaining Evenness of authoring effort could be an issue: ‘what felt the hardest was to have was to maintain roughly equal discoverability of each path’, P4 maintained, not wanting ‘to spend a significant amount of time on a path that only 0.1% of players would do’.

The Genre itself often directly shaped the strategies authors could employ to keep content production manageable: a comedy game with less consistency of tone might be easier to write for when the author ‘couldn't really come up with a viable choice of two actual serious things’ a player might do (P1). A horror game could afford to ‘brutally’ end stray plot branches so ‘only a handful of paths move forward’ (P3). Conversely, implementing Player Affordances, such as hint systems and additional responses or making an ‘effort with accessibility’ (P3) sometimes added to Feature Creep.

“Events and competitions are useful in that they bring a sense of community around, they bring deadlines... a lot of the projects I've actually finished have been for some kind of some kind of contest or some kind of event.” - P2

Interactive narratives aren't created in a vacuum. A work that isn't just a personal creation is made for an intended audience, which may or may not be commercial. The authorial goals are impacted by the feedback from non-authors related to the project: testers, publishers, and other collaborators. This Audience Feedback can impact design throughout the creation process, but also between works. Authors would reflect on ‘good feedback’ (P8) of previous games. The Publishing Context— whether the game was made for a game jam, competition, or a commercial market (and what that market was) — impacted the author's drive to finish a work, as well as shaping the content and size of the project. One author was motivated to finish a work for a ‘book deal’ (P1) while another ‘enjoy[ed] having the deadlines’ (P9) that competitions would bring.

Our largest theme, detailed in Table 3 was a list of strategies used by authors to manage the authorial burden. The design elements of interactive narrative (their underlying structure, and the ways the player can interact with them) aren't chosen solely due to the impact on writing, but rather there is an interplay with the author's goals for the game, and what they have to do to make it happen. Some strategies are implemented on a project-wide scale, others on a scene-by-scene basis. Some appear in the initial design, others, such as cutting-content, are implemented in an ad hoc way during creation as required. We found that the categories can and do overlap with one another.

4.5.1 Decoupling. “The content model that we had... allowed scenes to dynamically reshuffle themselves in ways that made sense, and yet did not require us to make as much new content in order for it to work.” — P7

Decoupling strategies are those in which content is written that doesn't depend on most of the earlier content happening before it, and so can be written much the same regardless of existing state. These strategies were often implemented on a project-wide scale. Content could be broken up in to distinct chapters with Episodification, ‘trying to keep [the chapters] separate mostly’ (P8). Works could be written with Modularisation where the same story segments could ‘happen at different points... flexibly’ (P13). Modular content could be ‘dynamically reshuffle[d]’ (P7) or delivered as ‘conversational buckets’ in the form of semi-independent story segments, sometimes described as Storylets or Plot Clouds.

4.5.2 Embracing. “Every idea you have, write it down, write it somewhere. It'll be useful somehow. Just get everything in. Don't editorialize yourself too much… write through it.” —P14

Rather than seek to change the work itself, the authorial burden could be embraced (much as Stern encouraged designers to do with combinatorial explosion, one element of authorial burden) [20]. This can be done on the technical side, through Automation of rote processes, or Tool Creation, making more appropriate languages, engines, generators and the like for the particular project. This ranged from ‘script validation’ (P7) to a ‘custom domain specific language’ (P10). These approaches are the sort often advocated in the academic literature.

Existing tools for Visualisation or Knowledge Management could be employed to better conceptualise the work being done. Often less technical means were used for this, such as ‘hand drawn flow charts’ (P3). The writing process itself could be adjusted, though authors differed considerably on what they found useful. Some authors reported success with a period of extensive Planning and Prototyping beforehand, including ‘engineering of the system... from a paper prototype’ (P10). During creation, Writing Out Of Sequence could be used, producing, for example, ‘chunks of the dialogue trees’ (P8) in whatever order inspiration struck them. A bold strategy was omitting self-editing and focusing on getting as much done with No Rewrites.

The capacity of the authors themselves could be extended. This could mean bringing in more team members with Collaboration. This might split the workload, but also help with the complexity overall, with P2 maintaining that ‘it is much easier to work in a group than it is to work… for sustained periods on complicated projects alone’. Authors could attempt to induce a greater experience of Flow to write for longer with fewer distractions. P3 wrote one project ‘in probably four or five very long sittings: five to ten hours of just sitting down and doing the writing’. Where commercial pressures such as wages and release windows weren't a factor, simply allowing More Time for the project was common: ‘the difference between a personal project and a studio project,’ P10 described, was that with a ‘passion project... if you're blowing your scope, you can say this is going to be a five year project instead of a two year project’.

4.5.3 Generating. “ProcGen... makes for very good snippets of narrative, but not for very good overarching narrative” - P2

There is a common belief among non-authors that content could be generated very quickly and cheaply from Artificial Intelligence or Procedural Content generation more broadly (e.g. in Dobrovsky et al [7]). However, the participants didn't typically conceptualise generation as a means to save authoring time: ‘sometimes this is phrased as [procedural generation] gives you twice the content for only five times the work’ (P2). Authors often had a more nuanced view of the power and expressiveness of procedural approaches. This was particularly explored by P14, who said, ‘I don't believe in procedural generation of content. I believe in procedural assembly of authored content. So you have you have a pool of interesting things which have been written by human because humans are interesting and computers aren't. And the computer is just gluing them, so we can maximize the usage of all the content that we write.’

A different generative approach was mentioned by P10: using Story Sifting techniques to create narratives out of generated content or emergent gameplay. ‘The idea is that you have a series of events that are presented to the system... And what the system does is say from here, which sequences am I interested in? Which one of those are good stories?’ (P10)

4.5.4 Reducing. “Can I actually justify to myself that I'm subjecting the player to this much text?... I can probably say the same funny thing in in one sentence and just delete everything before... And then of course, in doing that, you delete a bunch of nouns and there's a bunch of other stuff that you don't need to implement and just by saying less you make less to implement.” - P5

Perhaps the most straightforward way of limiting the authorial burden in a project after it has commenced is just to limit ambitions through Cutting content. In the initial design and beyond, the amount of varying content can be limited by Reducing Objects, the ‘extraneous things’ (P13) such as items, places, or characters. It can be limited by Reducing Actions, limiting the choices, verbs or other interaction methods available to the players. Implementing actions, especially where they create combinatorial complexity, ‘can cause a lot of headaches’ (P11) which can be ameliorated by limited the range of ways the player can interact with the work. Complexity can be reduced through greater decoupling of content segments by Reducing Tracking of game-state, limiting the number of ‘eternal flags’ (P8) being checked. Overall workload can be reduced by limiting through Reducing Play Time. This latter approach was explored by P2: ‘most of the work was about packing in an awful lot of experience into a very small time and that takes a lot of work and you can't have that kind of density of experience over a multi-hour project’.

The medium of the work itself can allow reduction of both static and dynamic authoring. Abstraction, such as cutting elements extraneous to the narrative that might add to the realism of the situation but would be more work to implement such as checking ‘who's in the room when you have a conversation’ (P7), or adding game-like elements that don't necessarily make sense in-fiction, including arbitrary limitations on action. Interactive narratives that are primarily based in text can leverage the benefits of using Words vs Assets, in that the author can declare things about the world much quicker and cheaply in text: ‘you can write a huge game and you're not having to worry about making the 3D models or finding all the voice actors... because you're just writing sentences’ (P5).

4.5.5 Reusing. “I had been keeping a collection of different story ideas and it just so happened that I figured out a way to to put several together.” —P3

Within a work, content can be reused. This can be through using Shared Assets, repurposing the same creation or ‘programming structure’ (P8) in different contexts. Another common way is using Structural Patterns such as ‘branch and bottleneck’ (P1), or using game loops, so that the same content can appear again (this can overlap with Moduluarisation and other Decoupling strategies, though content can be Decoupled without Reusing elements, and content can be Reused without necessarily Decoupling sections)

The author can use a World Model— ‘a sensible default with interesting overrides’ (P14). Having a set of rules for default behaviour allows locations, objects, and characters that can reappear (this is a core feature of text adventures). States can be tracked as variables that accumulate, such as relationship ‘affinity scores’ (P3), or in-game resources like money, reusing the same Cumulative Variables that can be changed in multiple places in the work. This can allow players to make choices that ‘would add to stats’ allowing agential play without immediately ‘branching the plot’ (P12).

Ideas and content can be reused between works, Recycling Work between projects. Institutional knowledge can be built up in a studio or publisher in sets of guidance on House Style, thereby ‘institutionalizing design knowledge’ (P10).

A model illustrating different forms of authoring labour: content creation, dynamic authoring, and programming and tool creation. Generative, Reusing and Decoupling strategies shift the burden of work between different forms of authoring, while Embracing strategies may increase authorial capacity to create. Authorial goals for a work are formed in the context of authors’ Capabilities and the Audience and Publishing context. Reducing strategies impact the Authorial Goals, changing the overall shape of the work itself.

Figure 1: A conceptual Model of the Authorial Burden

Figure 1 illustrates how we have brought the insights above together into a conceptual model of the authorial burden. Authors have a vision for the games they want to create which determines their Authorial Goals. These goals are shaped by their Capabilities (their skills in writing, the tools they know, their imagination etc.), and the Audience and Publishing Context (including competitions, game jams, commercial launches, publishers). Often the vision is incomplete, and they write to find out more!

The design emerging from these goals sets the initial Authorial Burden (which may or may not be known before they start writing), the relative size of which is also influenced by the author's capability (a given project would be easier for an experienced author with relevant skills than an inexperienced one without those skills). There are three different kinds of work involved:

Static Authoring: creating more unchanging content. (e.g. writing lots of descriptions)

Dynamic Authoring: creating content which might appear based on some conditions, or appear in different contexts. (e.g. writing lists of objects and descriptions that can be recombined to create many different objects)

Programming/Tool Creation: expanding system capacities (e.g. creating a new game engine or scripting language to do something that couldn't easily be done in existing tools)

As part of an authors’ initial design, or later when they realise there's too much work, they implement strategies for managing the burden. These strategies can:

REDUCE the overall workload (e.g. cutting content) effectively adjusting the goals

EMBRACE the project as it is, and try to increase capacity instead (more writers, more time taken, crunch, improving skills)

REUSE elements (writing, code, assets)

DECOUPLE content, separating it into self-contained sections to manage complexity

GENERATE content (e.g. through procedural generation)

The first two strategies impact the relative size of the burden by either adjusting the goals (REDUCE) or increasing the capability of the author (EMBRACE).

The last three strategies (GENERATE, REUSE, DECOUPLE) do not necessarily change the size of the burden but instead primarily change the nature of the work. For example, a description generator might reduce the amount of unique descriptions you need to write (Content Creation is reduced), but it is still work to make the generator to begin with (Tool Creation is increased), and to populate it with enough content so that it feels varied and interesting enough (Dynamic Authoring is increased).

Rarely can creative work be sped up without changing the process of creation and the form that the output takes, and often alleviating one aspect left more time for other work rather than reducing labour overall. As one panelist put it, after interrogating the design and removing content, ‘what I found was I didn't write less as a result... I was picking what I wanted to write about’. Our model of the authorial burden emphasises that every proposed strategy to alleviate authorial burden changes the kind of creative labour that is performed (and thereby not always reducing work absolutely), and that these changes may not align with author visions, interests or skill-sets, or be appropriate for the intended audience of the work. This was repeatedly born out in the interviews and confirmed in the expert panels: often authors will use a strategy to move work into a domain they find more interesting or are more comfortable creating within. For instance, game designers who are also tool-makers are often seen spending time programming a specific engine to meet their requirements, rather than trying to amend their vision to fit with an existing tool, even though that would usually be quicker. Conversely, less technically-minded authors will often stay with a tool long after it has ceased meeting all their requirements, due to the high opportunity cost of switching to something better suited or attempting to make a tool themselves.

Authorial ‘burden’ is often not experienced as burdensome, and one common feedback from the panelists was about the negative connotations of the term itself. Creative labour is the part that's enjoyable for the author, and there is little desire to abstract it away. As one panelist said, ‘sometimes the strategy doesn't reduce the workload, it frames it in a way that you actually enjoy doing’. While commercial games are made under context of time and financial constraints, hobbyists are more likely to approach the craft in a similar way to many novelists: as an enjoyable but not particularly profitable past-time.

One panelist who had worked with machine-generated prose pointed out that it was a lot of work to get it to output desired content, and the content created needed a lot of editing to ‘force disparate content to make sense’, but in taking a ‘bottom up’ approach, it could be used as a source of creative inspiration. Others were even less enthusiastic: ‘It's never “how I'm using it in my own work” and always “here's how it's going to disrupt that other industry that I know nothing about” ’, one said while another reflected that ‘there's no part of [my writing] where I'm like, “I wish an AI would do this for me”. If I didn't find it interesting I just wouldn't do it.’

The aesthetic texture of hand-crafted content is very different to the ‘procedural oatmeal’ that can arise through machine generation. So even when the designers do want to avoid authoring work (e.g., to meet commercial constraints), the techniques touted to alleviate authorial burden often typically would not meet the desired quality and specificity of output. Where generative approaches are sought by authors, they are picked for their specific affordances. One participant stressed the distinctiveness of generation as its own artistic medium: ‘you are not going to save work overall if you want the same kind of product’, but, ‘you can save some kinds of work and introduce some kinds of variety: the question is whether those kinds of variety are valuable to you.’

This same dynamic is present in non-generative strategies as well. For example, the use of illusory agency has been proposed as a way to avoid authorial burden by offering choices but always redirecting to the same branch regardless [9]. While merging branches has long been recognised as a fundamental technique of hypertext authoring [1], committing to only or primarily using illusory agency in a game that is positioned to players as being high agency is in itself a specific vision for a work, and one which would become obvious on replay or review of that work: the illusion cannot hold indefinitely. Authors make games that according to a vision, and that vision is rarely ‘pretend to be some other game-type entirely’. If having high player agency is the design goal that is creating the authorial burden, removing that player agency doesn't solve the burden, rather it creates some other game entirely.

During the panel some explanatory challenges arose for the model. The 5-part schema of strategies [13] was a clarifying lens to view authorial design choices but it elides the temporal aspects of design. For instance, one expert in the first panel discussed how she would ‘collapse continuities’ in her hypertext fiction as part of an iterative process of shrinking the initial workplan by merging different split branches. While this is both a reducing and reusing strategy (specifically, cutting some proposed branches by way of reusing others), it is also a strategy for a specific time in the design process: when an initial hypertextual pattern has been planned but has yet to be fully implemented.

Another way of splitting the strategies would be to consider which take place at which stage of authoring, and the scope of each strategy, from the level of a single scene or room, all the way to game-wide changes. So many of the strategies to managing the authorial burden happen at the outset of the project, with expectations of how much work it will be shaping the initial vision for a game. In Pinchbeck's examination of the design of Dear Esther, he remarks that the whole setting and structure of the game was made with ‘the economics of building’ in mind: setting the game on a remote island was explicitly a strategy to diegetically limit objects to implement, especially other people [17]. As per Kitromili [14], authoring interactive narrative is made up of different activities from ideation and planning to writing and editing, and strategies for managing work are appropriate to different stages.

One panelist offered the challenge: instead of the ‘authorial burden’, what about the ‘authorial responsibility’? ‘Burden’ has the connotation of something that must be shouldered or shirked. As a term, it leaves something to be desired. Still, the term does leave the correct impression that authorial choices each leave a weight of work to be done, and that the structure of interactive narratives, especially those with plot branches or variable narrative combinations, itself creates more of this work, whether or not it is felt as a burden.

Another panelist felt that author work in presentation, in user interface, and in marketing, is its own kind of labour, which has concomitant strategies that may or may not work (he described snake-oil salesmen promising ‘500 clicks with this one tool’). Indeed, care should be taken not to present ‘authorial burden’ as a synonym for all authorial labour, including the creation of paratextual elements, but rather as a description of the work that arises specifically out of the interactive nature of the story. The author might take care over how their hypertext is presented to a competition, but that care is typically independent of the labour that was needed to flesh out the different plot branches they decided to implement.

Finally, a panelist who was foremost a game engine designer and secondarily a game author, was keen to highlight the specific burdens in tool creation. Even before someone makes a game with a tool, the designer programmer must wrestle with the work that emerges out of their programming design decisions. While this programming labour can be an activity undertaken by some authors, and many of the work management strategies are similar (such as modularisation or loose-coupling), an author of a tool for making games has a different kind of aim than the creator of a game. A hypertext system or text adventure engine might embody some assumptions about possible stories (especially if they implement a world-model) but it isn't itself a story. The combination of variable state management and the management of narrative continuity is what makes interactive narrative authoring so thorny. They have all the problems that face the programmer and the novelist as well as the unique problems from arise from the intersection of the different crafts.