Links and Structures from Michael Joyce to Twine

A look at some articles from the scholarship of literary hypertext, and thoughts about how their concerns and terms relate to current work in Twine and other procedural literature.

Recently I’ve been including some coverage here of academic materials that might be of interest to industry or hobbyist readers. Some of that’s been focused on recent work in story generation or interactive narrative in some way.

Today’s example, by contrast, begins with a reading from a rather older paper: A Nonce Upon Some Times, by Michael Joyce in 1997. (If you want a proper citation: Modern Fiction Studies 43.3 (1997) 579-597.)

This article came to my attention because when I was responding to Mark Bernstein’s piece The Fellow Who Caught Fire, he called out that I was unfamiliar with something he considers to be essential reading in the field of literary hypertext.

“A Nonce Upon Some Times” raises questions about how we categorize hypertext structures, and what we understand from the meaning of these structures. Occasionally this is reminiscent of Sam Kabo Ashwell’s work or perhaps my own on small-scale structures in CYOA: what does it communicate when the author chooses a particular relationship between the available lexia?

Joyce, however, is overtly contemptuous of branching narrative:

The workshop exercise with which I began this essay seeks to isolate a set of primitive choices that both prompt the visual kinetic of rereading in hypertext and, at the same time, isolate the elements of what Douglas calls “a narrative of possibilities.”… It engages working writers with aesthetic and readerly questions about linking rather than encouraging a choose-your-own-adventure sort of drearily branching fiction.

…and what interests him is not the question of how one might project oneself into the role of a protagonist; not how one might experience agency, constraint, or non-agency through this pattern of links.

Joyce’s taxonomy is easier to diagram than to explain in words. He proposes that one start with a four part structure, linked linearly, which ends by going back to one of the earlier texts. I would diagram it like this:

Now: what happens next, after we have re-entered the second of the four texts? Where do we go from there, and is it different from where we might have gone the first time around, and what do these different arrangements mean?

If we go around a second time just the same as the first time, this would be a recursus or cycle:

…since we could repeat the same sequence endlessly, a loop-and-grow with the “grow” portion left off.

(An observation here, not from Joyce: How we interpret this depends very much on how we’re pairing up our fabula and our syuzhet. if we are rereading, many times, the experience of the lovers falling in love and quarreling once, then that suggests a narrator or else an implied reader that is for some reason obsessively interested in this sequence and cannot stop herself from playing it over again like a rosary; if however there are as many quarrels as there are readings of the quarrel, then what we see is the story of a dystopic relationship, in which the lovers never learn to better their communication, never improve towards a more skillful and peaceful partnership.)

If on the other hand we exit the story after revisiting this section, like so:

Then we may regard the structure as a flashback, reviewing section 2 once before proceeding to the end of the story. Joyce calls this a timeshift. A renewal, meanwhile, is in Joyce’s terminology a structure in which, from section 2, new and different possibilities arise:

To me, these are useful structures to compare, and at the same time it seems very strange to me to distinguish these different structures by giving a different label to the link from “they reconcile”.

The way I would analyze things, the difference between these structures does not reside in that return link, either authorially or perceptually. In most hypertext tools I’m familiar with — and granted I’ve never written anything in Storyspace — the code controlling what happens on a revisit of “They fall in love” is most likely located in the “They fall in love” node. And from a readerly perspective, I also tend not to think, “that link from ‘They reconcile’ is a cycling link” but something more like “this whole sequence is cyclical”.

After suggesting the typology of links, Joyce goes on to contemplate what it means to reread and even unread, in the hypertext context, and that some texts reward study that explicitly unpacks their structure:

No one reads this way, of course, except the hacker or the literary critic. Or perhaps the writer. Although no one writes this way, to read this way while writing is to reread as a prospective reader and, in the process, unread the text in favor of what is not normally read within it.

But of course many of us do read interactive literature with attention to its structure, not just because we’re hackers or critics, but because some literature encourages or even requires it — think of the visual novels that must be played over and over to unlock many endings before a final ending emerges; think of the narrative maps explicitly provided in some interactive works, even some book-form CYOA. Which of course is not to say that Joyce, writing in 1997, should have accounted for genres and reading habits that have developed more since.

Joyce does now talk about the ambiguity of UNDO in a hypertext reading, quoting Jim Rosenberg:

“One may revisit a lexia simply to read it again,” says Rosenberg, simply throwing the baby of this current essay out with the golden carp’s dark bathwater, “or it may be a genuine ‘undo,’ perhaps the reader didn’t mean to follow that link at all” (22). His immediate, low-level, interest here is in how to represent the meaningfulness of an action for the reader. “These [backtrackings] are arguably different actemes,” he says, “though typically not distinguished by the hypertext user interface” (22).

Rosenberg’s article, “The Structure of Hypertext Activity,” is also available online. A key thing to take from Rosenberg is this passage of the abstract:

Acteme is a low-level unit such as link-following; episode is a collection of actemes that cohere in the reader’s mind, session is the entirety of contiguous activity.

Coming to this from game design and parser IF, I think: perhaps an acteme is related to what we might conventionally refer to as a verb, and it is defined not by the specific form of engagement with the interface, since clicking might mean almost anything, but by what action the reader performs or intends to perform by clicking.

For hypertext in which there is only one kind of connection between lexia, or in which the author doesn’t communicate about the kinds of connection, follow link might be the only available verb, just as if we were playing a parser game that understood only USE.

But there are hypertexts that do offer other actemes to the reader-player: “expand this text,” “replace this text,” “clarify a meaning about which you are presently being coy.” Sometimes clicking on a link signifies that the reader commits to and endorses what is written there, and sometimes it’s just the opposite, an act of challenging what is written and demanding that the text be supported or withdrawn by the author. (I think, here, of the point in Their Angelical Understanding where to click on a word replaces that word with “LOL”, allowing the reader to poke through the narrator’s false front to reveal her true feelings.)

In Rosenberg’s understanding, one of the main things a reader does with hypertext is search through it in an attempt to find or assemble coherent “episodes”, which have some clear sense and connectedness for the reader. He suggests that hypertext reading tools should actually allow the reader to name the episode they’re currently reading/assembling, which suggests that these groupings don’t somehow pre-exist in the work.

The display at the bottom of the screen in SPY INTRIGUE charts adjacent lexia and gives some clues about the type of content to which they will lead.

It is interesting to read Joyce’s paper having just recently played SPY INTRIGUE, where the map of the text is explicitly presented to the reader, and even given a fictional explanation to justify its presence. As a reader one frequently departs from a particular node with the knowledge one will be back soon, after a short or long digression.

A less strong but similar sort of knowledge is available in With Those We Love Alive, where link coloration distinguishes those links that are just going to briefly explain or expand upon an idea, and those that will march us forward through the narrative. Indeed many Twine games are structured this way, with hub texts and links that lead away to short explanations and glosses, which we enter knowing that we will return in a moment.

In this case, one of the links is (and must be) last, since it advances the plot, but the others may be visited in any order, with any degree of thoroughness.

Do we read the entire hub text again each time we return to it? I don’t, but neither would I want to get rid of it: those paragraphs operate as context, a reminder of how the other texts relate to each other.

Other work — Anya Johanna DeNiro’s Solarium comes to mind — use a hub that is revisited more rarely, with longer journeys away and back; in Solarium new passages explicitly unlock as we make progress through the story. Used well this type of approach achieves the cadence of chorus and verse (an observation I owe to Katherine Morayati’s NarraScope presentation about musical structures).

Occasionally the hub and verse text are even different in kind. Get Your Gun, Dragonfly uses poetry in some of the explication sections; Colorado Red turns short explanations into hovertext.

There are, of course, any number of other techniques — hubs that allow you to visit only a certain number of their subsidiary texts, but not all; hubs in which you must read some subsidiary texts before a final, more important link becomes available, as near the end of SPY INTRIGUE; lexia from which a dozen or two dozen links lead on to a bad ending, and only one, hidden amongst the rest, leads to a positive resolution, as in howling dogs or Impostor Syndrome, so that it is discernment (or rigorous elimination) that unblocks progress for the reader.

And that is without considering all the ways that texts stretch or shrink, mutate in place on the screen; how they deploy images; how they call out to external websites and force other pieces into an intertextual relationship, like Cayotaje; how they may in some cases accept a word or two of typed input; how they can allow the player to remix portions of the text to possibly arrive at a final version that the author never saw or anticipated. (This level of remixing is possible, I think, in First Draft of the Revolution; definitely in Endure.)

From the perspective of more than twenty years later, many of Joyce’s observations feel like first pen-and-paper cartographical attempts on a territory that has now been explored very extensively on foot. Perhaps there are few rigorous catalogs of link and structure effects in Twine and other hypertext IF storytelling, but those effects exist, diversely and numerously, beyond anything described in “Nonce.”

Loose Strands offers the player an explicit map of the territory to traverse, and foreknowledge of the futures, blacked out boxes, that have become inaccessible.

But for me a question also arises from this essay that to the best of my knowledge has not been answered thoroughly elsewhere. Which is:

Suppose that, like Joyce, we despise “drearily branching” fiction. Suppose we want to put aside protagonism as a way of reading hypertexts, perhaps exclude the reader entirely from any form of diegetic agency. Suppose the reader is not allowed to change what happens in the story, via any process of choice or selection or ordering. Our only engagement is to be with the syuzhet and none with the fabula, with the ordering of the experience.

If there is no agency at the narrative level, it possible to write in such a way that the reader is anything other than an aleatory device? What kind of intention is possible? And does this intentionality go beyond exploration — since “what are you interested in here?” is a question that a page of links always asks of its readers, even a page of Wikipedia or TV Tropes — into something more organized?

Or, to come at it in Rosenberg’s way, how can we help the player discover and shape episodes?

Content mapping in Iain Pears’ Arcadia suggests various productive ways of clustering content into meaningful episodes.

In his long Bestiary of Player Agency, Sam Kabo Ashwell identifies at most one form of agency that would qualify here, Focus; in my Alternative Taxonomy for Interactive Stories, I list just a few works that I feel invite readerly interaction. The Reprover is one, Strange Rain another. These are stories that allow the reader to intentionally pursue a thematic direction, or request clarifications and deepening of a particular idea. Arcadia is related, though the lines it presents for following are geographical and character-based.

Current IF community terminology for structures in choice-based narrative typically makes some assumptions — that the structure was created by the author, that it’s designed to curate an experience that definitively terminates, and that any interface that allows the player/reader to execute searches on the fictional context probably qualifies as something other than standard hypertext.

But if we include in our consideration works where the reader may do something other than click links, we could also see Her Story as a kind of hypervideo rigorously indexed for content that permitted associative traversals. Even a card-deck narrative, if printed on cards with suit and number, can facilitate intentionally thematic readings; see Family Arcana.

We do also at times open up the general question, “how do we select what the reader will see next?” This is the subject of Michael Mateas’ research on content selection architectures, which he recently presented to the London IF Meetup.

I think the idea of a content selection architecture, with allowances for thematic access that went outside simple link-clicking, would not seem out of sympathy with Joyce’s project(s). Some of Joyce’s own writing about hypertext feels reminiscent of a procedural storylet system rather than of Twine, hypothesizing systems that would let the user search for types of content in the lexia. (See for instance “Siren Shapes: Exploratory and Constructive Hypertexts,” reprinted in Of Two Minds and in the New Media Reader; this itself calls back to “Reflections on Notecards: Seven Issues for the Next Generation of Hypermedia Systems,” by Frank Halasz in 1987.)

At times, as in “Sirens,” Joyce also considers how the process of assembling hypertext becomes an exercise for the creator, in understanding how the different possible facts and anecdotes might join together, a use case closer to an author creating something with Scrivener than one deliberately developing a hypertext structure.

Both Joyce and Halasz are writing about nonfictional hypertext, but they imagine scenarios in which the user could search for tagged content and perhaps retain the index or link structure that resulted as a permanent artifact that would then always allow that reader to re-access the content in that same way. Halasz in fact suggests two kinds of query. The first, more familiar, is the content query:

In content search, all cards and links in the network are considered as independent entities and are examined individually for those entities whose content or properties match a given query. For example, all the cards containing the string ‘hyper* system’ would be a content query. Content search is more or less standard information retrieval applied to a hypermedia information base.

If applied to fiction, this sounds a bit like a version of Her Story in which, say, your search history is retained over the course of play. The more interesting to me is Halasz’s idea of the structure query:

In contrast, structure search examines the hypermedia network for sub-networks that match a given pattern. For example, a simple structure query might be: all sub-networks containing two cards connected by a ‘supports.’ links where the target card contains the word ‘hypertext’. This query contains a description of node content (i.e., contains the word ‘hypertext’). It also contains a structural description of a sub-network (i.e, two cards connected by a ‘supports’ link).

And here we’re assuming that the author has created an argument, and that links between lexia carry additional semantic meaning, the way edges in a graph database might carry meaning about the relationship between the nodes; so that it later becomes possible for the user to look explicitly for all the evidence supporting a particular contention.

In developing the knowledge model for Character Engine, we think and talk about allowing the player, in certain situations, a kind of query very close to this. “My mother was a horrible woman,” says a character, and the player replies, “why do you say that?”

Supposing the knowledge base contains the character’s personal history, including all the incidents that reflect badly on the late Mrs. Snaith, the character can choose an example and report it back to us. This interests us because creating characters who seem realistic and humane suggests that they ought to have some knowledge of, and ability to account for, their own traits and motives.

Outside the boundaries of a conversational interface, one could imagine hypertexts in which an acteme of challenging the narrator’s viewpoint could be handled procedurally rather than through hand-written responses; so that the player could accept or reject their understanding of events and motives; perhaps reject high-level theses and then reject or accept, prioritize or de-prioritize the supporting evidence.

All those approaches imply that the author has at least considered some organizing principles for the work, and offered a UI that makes that navigable to the user.

More procedural possibilities suggest themselves as we deal with larger and larger bodies of content. 52 lexia fit on a card deck. This is a number a human can write in a reasonable amount of time, and a deck another human can productively understand and sort in several configurations. But what about thousands, tens, hundreds of thousands of lexia?

In these days of doc2vec, we could also imagine an automated embedding of lexia in a virtual space, and a reading interface that would allow us to move along the vector axes, perhaps towards more romantic or less philosophical.

Or, at the extremes: I am listening to Spotify as I write this, in playlist radio mode: I created a playlist of about ten songs, but Spotify has gone off on its own, riffing on that playlist for much longer than the original playtime, coming up with other tunes, many of them previously unknown to me, that recapitulate the styles and genres found in the original playlist. It is uncanny to me how well this works, how rarely playlist radio comes up with anything that jars me out of the mood I was attempting to create, and sometimes discovering for me new artists I like but hadn’t known before.

Would something similar work as well with literature? What would it mean if it did? Who would be writing all these lexia? More than one person, I presume, or it would never be done. Hard to believe such a process would produce a good hypertext novel, but might it not produce an interesting poetic anthology, for instance, taking the public domain of world literature as its source, and offering as its output a custom curated traversal?

This feels like it is again something very different than either Joyce or Rosenberg had in mind.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s