Storyspace, Eastgate, and the literary hypertext tradition
The academic hypermedia tradition described in the last chapter had a parallel that ran inside literature departments and writers’ rooms rather than computer science departments. It produced a body of work — novels, stories, poems, essays — that was hypertextual in form: nonlinear, multi-pathed, requiring choices by the reader to assemble what they were reading. The works are small in number compared to the volume of conventional literary output, but they exist, they were taken seriously by critics, they were taught in graduate programs, and they constitute the only sustained attempt in the history of literature to produce art in the hypertext medium. They are the answer to anyone who asks what hypertext could be as a literary form rather than a technical apparatus. The chapter is about how the works got made, the tools they were made with, the community that surrounded them, and why, when the web arrived, the form did not migrate to the new medium as cleanly as one might have predicted.
The form before the tool
It is worth establishing that nonlinear writing predates the hypertext systems that were eventually used to produce it. Jorge Luis Borges’s “The Garden of Forking Paths” (1941) is the classical reference: a story whose narrative is structured as branching paths, in which any decision the reader takes changes the events that follow, but in which the branches do not exclude each other — each branch is real, and the story is the totality of branches rather than any one of them. Julio Cortázar’s Hopscotch (Spanish Rayuela, 1963) is a novel with a “table of instructions” at the front that proposes two reading orders for the chapters: one starting at chapter one and proceeding sequentially, the other following a jump-list that produces a different sequence and includes additional chapters not in the linear reading. Italo Calvino’s If on a winter’s night a traveler (1979) interleaves a frame narrative with the opening chapters of ten other novels, structured so that each chapter is the beginning of a different book whose continuation is interrupted. B. S. Johnson’s The Unfortunates (1969) was published as a box of unbound chapters that the reader could shuffle and read in any order. Marc Saporta’s Composition No. 1 (1962) did the same, earlier.
These books exist. They are good books. They establish that the impulse toward nonlinear narrative is not a product of digital technology; the technology made certain kinds of nonlinear structure easier to produce and easier to navigate, but the impulse was already in twentieth-century literature. The hypertext systems that began to be built in the 1980s landed in a world that had already produced a small canon of nonlinear printed work, and the writers who took up the new tools had already-existing models for what they were trying to do.
Storyspace
The dominant authoring tool for literary hypertext, throughout the period the form was most active, was Storyspace. It was created in the mid-1980s by Jay David Bolter, Michael Joyce, and John B. Smith. Bolter was a classicist who had become interested in computing as a tool for thinking about written form; Joyce was a fiction writer who had been thinking about nonlinear narrative since the late 1970s; Smith was a computer scientist at the University of North Carolina who provided much of the technical work. The collaboration was unusual. Storyspace was not designed by computer scientists who hired a writer for advice; it was designed jointly by a writer, a humanist, and a computer scientist, each with a stake in the system’s substantive form.
Storyspace’s data model was a set of “writing spaces” arranged in a two-dimensional plane and nestable inside each other. A writing space could hold text. It could also hold other writing spaces. The result was a hierarchy of containers, like nested folders in a file system, but with positional information: the spaces had spatial relationships within their parent space, and the spatial arrangement was an authorial signal about how the spaces related. Spaces were connected by links. A link could be a simple jump from one space to another, or it could be guarded by a “guard field” — a condition, expressed in Storyspace’s small scripting language, that determined whether the link was available to the reader. Guard fields were typically conditioned on what the reader had previously read: a link might become available only after the reader had visited certain other spaces, or might be unavailable until certain spaces had been read.
The guard field was the system’s signature feature. It allowed the writer to control not just what links existed but when. A reader proceeding through a Storyspace work would find that the same link, clicked at different points in the reading, took them to different places, or did not appear at all. The author shaped the reader’s path by controlling the visibility of paths. This is structurally different from the academic hypermedia systems treated in the last chapter, which were largely about giving the reader more freedom to navigate; Storyspace’s design was about giving the writer more control over the reader’s experience even in a nonlinear medium. The two impulses — toward reader freedom and toward authorial control — coexist in the literary hypertext canon, and individual works strike the balance differently.
Storyspace had several other features that mattered for the works produced in it. It supported a “default link” — a link followed by default if the reader pressed Return without selecting a specific link. Many Storyspace works use the default link as the spine of the work: pressing Return repeatedly takes you through one possible reading; the other links open the alternatives. It supported a “path” concept, in which the writer could specify a sequence of spaces that constituted a reading; multiple paths could be defined for the same body of spaces, and the reader could choose which path to follow. It had a map view that let the writer see the spatial arrangement of spaces and their links, and a reader view that hid this from the reader and presented the text. The two-view structure made writing a Storyspace work feel different from writing a novel: the writer was both writing the text and shaping the underlying graph, and the graph view was where most of the structural decisions got made.
Eastgate
Eastgate Systems, the publisher and developer that brought Storyspace and most of the literary hypertext canon to market, was founded by Mark Bernstein in 1982. Bernstein was a chemist by training — he had a doctorate from Harvard in 1983 — who had become interested in hypermedia and turned a small software company toward publishing hypertext fiction. Eastgate acquired Storyspace from its original developers in 1990 and continued to develop and sell it for the next several decades. Eastgate published, on diskette and later on CD-ROM, the works that constitute most of what is canonical in literary hypertext: Michael Joyce’s afternoon, a story; Stuart Moulthrop’s Victory Garden; Shelley Jackson’s Patchwork Girl; Carolyn Guyer’s Quibbling; Joyce’s later Twelve Blue; Judy Malloy’s its name was Penelope; and others. The works were sold as physical media — a floppy disk or a CD in a small box, with a slim printed booklet — at prices in the same range as expensive trade hardcovers, fifty to one hundred dollars each. They had ISBN numbers. They were reviewed in literary journals, including some that did not normally cover digital work. They were taught in graduate programs.
Eastgate also developed Tinderbox, a personal hypertext tool by Bernstein himself, which has continued to be sold and updated through the present and is the closest thing the world has now to a maintained desktop hypertext authoring system. Tinderbox is used by writers, researchers, lawyers, and other professionals for personal note-taking and structured thinking; its install base is small but devoted, and it descends directly from Storyspace’s structural assumptions about how writing should be organized.
The canonical works
Michael Joyce’s afternoon, a story is the work usually cited as the first major literary hypertext. Joyce completed it in 1987 and Eastgate published it commercially in 1990. afternoon is a story about a man named Peter who, driving to work in the morning, sees an accident on the road and gradually comes to suspect that the people he saw may have been his ex-wife and son. The work consists of 539 writing spaces; depending on the path the reader takes, the story can be read as a meditation on memory, an account of marital betrayal, a psychological study of trauma, or simply a series of impressions of a morning’s ambiguous events. Joyce’s use of guard fields means that the same passage, read in different orders, takes on different meanings, and some passages are reachable only after the reader has visited specific others. The work is short by novel standards — readable in a couple of hours along any one path — but the totality of possible readings is much larger, and most readers have come back to afternoon multiple times to follow different paths.
Stuart Moulthrop’s Victory Garden (1991) is more political and structurally more ambitious. Set during the first Gulf War in 1991, it interleaves the lives of a group of academics and students at a Southern university with the news coverage of the war and a series of dreams and reflections about violence, place, and history. The work is much longer than afternoon — over 900 writing spaces — and uses Storyspace’s branching more aggressively. Moulthrop, an academic at the University of Baltimore for most of his subsequent career, wrote about Victory Garden in his subsequent critical work and is one of the major theoretical voices in literary hypertext alongside George Landow, Jay Bolter, and N. Katherine Hayles.
Shelley Jackson’s Patchwork Girl (1995) is the work most often taught and most often cited as evidence that hypertext can produce art at the highest level. The premise is that Mary Shelley, in the aftermath of writing Frankenstein, builds a female monster for the creature — the female counterpart that Victor Frankenstein refuses to make in the novel — and that the female monster (the patchwork girl of the title) speaks in the work, telling stories about the women whose body parts she was assembled from. The structure is a body: the writing spaces are arranged anatomically, and the reader navigates by selecting parts of the body. Jackson’s prose is dense, allusive, and beautiful; the work’s structural choices reinforce its thematic preoccupation with assemblage and fragmentation; the relationship between hypertext form and the text’s content is closer than in any other work in the canon. Patchwork Girl has been the focus of more academic critical writing than any other single hypertext, and is the work most often handed to a skeptic who doubts that hypertext can be literature.
The canon is not large. Eastgate’s catalog of original literary hypertext extends to perhaps twenty or thirty significant works over the period 1990 to roughly 2005. Other small presses and academic projects produced more; the Electronic Literature Organization, founded in 1999, has tried to maintain a directory. The total population of works that anyone has called important literary hypertext is in the low hundreds. By the standards of conventional literary publishing this is tiny; by the standards of an experimental form working in a niche medium it is substantial.
The critical apparatus
Literary hypertext had, almost from the beginning, a critical apparatus that took it seriously. Robert Coover, the novelist and Brown professor, published “The End of Books” in The New York Times Book Review on June 21, 1992, an essay that announced literary hypertext to the broad reading public and made the case that the form represented a serious next stage in literature. Coover taught a graduate course at Brown in hypertext fiction beginning around 1990, with Michael Joyce and other practitioners as visiting figures; the course was the first sustained academic instruction in writing hypertext as a literary form. Bolter’s Writing Space (1991) was the most influential theoretical book on hypertext as a writing technology. George Landow’s Hypertext (1992), already mentioned in the previous chapter, made the case that hypertext realized in technical form what poststructuralist theory had been describing about the nature of textuality. N. Katherine Hayles’s later work, especially Writing Machines (2002) and Electronic Literature (2008), took the field’s critical apparatus into the post-web era. Espen Aarseth’s Cybertext (1997), in a slightly different register, treated nonlinear texts of all kinds — hypertexts, interactive fiction, games — under a unified concept of “ergodic literature,” literature requiring nontrivial effort from the reader.
The combined effect of this critical work was to give literary hypertext a place in the academic conversation about what literature was. The place was contested — many critics outside the small community of practitioners and theorists remained skeptical — but it was real. By the mid-1990s, “hypertext fiction” was a thing the literary studies world knew existed, and the canonical works were in syllabi at a number of universities.
The community
The community that produced and read literary hypertext was small and densely connected. Most of the writers knew each other. They met at the ACM Hypertext conferences, at the smaller “Hypertext, Hypermedia, and Hyperliterature” gatherings that grew up alongside, at the Modern Language Association, and at events Eastgate sponsored. Mark Bernstein at Eastgate was the institutional center: he published the works, attended the conferences, wrote criticism of his own, and maintained the social network that kept the field cohesive. The writers also taught each other’s work: Moulthrop assigned Joyce, Joyce assigned Moulthrop, Jackson assigned both, and the canon was self-curated through the syllabi of the small number of programs teaching it.
The community had a magazine of sorts in the American Book Review, which published occasional essays on hypertext, and in the early online journal Postmodern Culture, founded in 1990 and one of the first peer-reviewed humanities journals to publish exclusively online. There was an annual Hypertext anthology edited at various points by Eastgate. Cathy Marshall, then at Texas A&M (and earlier at PARC), wrote frequently about the form. There was a critical mass.
The community also had the partial sponsorship of literary celebrities. Coover’s involvement was the most visible, but Robert Pinsky, John Barth, and other established writers expressed interest at various points. Donald Barthelme had been writing nonlinear pieces for The New Yorker and elsewhere in print form since the 1960s; his work was often cited as a precursor. The form did not lack for prestige in the years it was active.
Why the form did not migrate to the web
The web arrived in the middle of literary hypertext’s most active period and produced an obvious question: why not publish hypertext fiction on the web, where it could be read by anyone, instead of on disk through Eastgate? The question was raised at the time and answered in several ways, none of which fully account for why the migration did not happen at scale.
One answer is that the web’s link model was structurally inferior for literary purposes. Storyspace’s guard fields — links that became visible based on reading history — had no native counterpart on the web. To replicate them, you needed JavaScript, which in the mid-1990s was new, fragile, and not universally implemented. You could not, in 1996, reliably build a Storyspace-like work on the web. You can now. The lag of about five years was enough to break the momentum.
A second answer is that Eastgate was a publisher with an economic model and the web was a medium without one. Eastgate sold disks for thirty to a hundred dollars. The literary hypertext community paid the writers, modestly, through these sales. The free web disrupted this not through hostility but through default expectation: readers had been trained, by 1996 or so, to expect that what they read on the web was free. Moving the works to the web meant moving them out of the economic relationship that had supported their production. Some works did move. Joyce’s Twelve Blue (1996) was published on the web rather than on disk and is still available there. Other works followed. But the economic transition was not solved, and the rate of new production slowed.
A third answer is that the web’s authoring culture absorbed the kind of writer who might have written hypertext fiction. The blogging tools of the late 1990s and early 2000s gave writers a much larger audience than Eastgate could provide. The audience came with the cost that the form was different: a blog is sequential and dated, not branching and structural. Many of the writers who had been working in hypertext on disk shifted to blogs or other web-native forms. They did not stop writing; they stopped writing hypertext in the structural sense.
A fourth answer, less material but still relevant, is that the cultural moment for the form passed. The early 1990s had been a period of considerable enthusiasm about hypertext as the future of literature. By the early 2000s the conversation about literature on the internet had moved on, to blogs, to fan fiction, to online journals, to e-books, and the specific structural ambitions of the 1990s hypertext canon were no longer at the center of anyone’s attention. The form did not die; the conversation around it became smaller and more specialized.
What is being recovered
Twine, an open-source tool first released by Chris Klimas in 2009, has been the focus of literary hypertext’s recovery in the 2010s. Twine is a Storyspace-descendant: it lets writers create branching narratives with conditional logic, publish them as standalone web pages, and share them freely. Twine has been taken up enthusiastically by a generation of writers, particularly in interactive fiction and game communities, and has produced a substantial body of work — Anna Anthropy’s queers in love at the end of the world, Porpentine’s various works, Michael Lutz’s The Uncle Who Works for Nintendo, many others — that constitutes a second wave of literary hypertext in web-native form. The Twine canon is larger than the Storyspace canon, more diverse demographically, and operating outside the print-publishing economy through patronage, game-distribution platforms, and direct community support. It is a genuine recovery of the form, in a different shape than Eastgate’s.
Interactive fiction more broadly — the descendants of the Infocom text adventures of the 1980s, surviving through community tools like Inform 7 and through new platforms like Choice of Games — is a parallel tradition that overlaps with literary hypertext in important ways. The Interactive Fiction Database catalogs thousands of works. The IF community produces criticism, holds annual competitions, and continues to evolve. It is not the same as the literary hypertext canon, and most of its participants do not think of themselves in the Storyspace lineage, but the structural concerns — branching narrative, reader agency, choice as a meaning-making act — overlap substantially.
The dynamic document tradition treated in chapter twenty has, separately, recovered some of what hypertext was attempting. Bret Victor’s explorables, the Observable notebooks, and Glamorous Toolkit are all attempts to produce documents that are also computations, in which the reader’s interaction with the document changes what the document shows. This is not literary hypertext, exactly, but it is a continuation of the underlying claim that a document can be a thing the reader does, not just a thing the reader looks at.
What does not survive
The Eastgate canon itself is at risk. The works were published on diskette and on CD-ROM, in formats requiring specific operating systems and Storyspace versions. afternoon runs on classic Mac OS; Patchwork Girl runs on classic Mac OS or early Windows; both require Storyspace, which Eastgate continues to maintain but which is not part of any current operating system’s standard environment. Reading these works in 2026 requires either emulation or one of a small number of preserved authoring environments. The Electronic Literature Organization and various academic preservation projects have been working on the problem; the works have been re-issued in newer Storyspace versions periodically; but the long-term preservation question is open. A literature that depends on a specific software environment is, at the limit, a literature with an expiration date.
This is a loss the broader web has avoided through its own technical conservatism: a 1995 HTML page is still readable in 2026, because the web’s standards have evolved compatibly. The literary hypertext canon does not have this property. The works require their tools, and the tools are no longer running on the current generation of machines without help. The form has produced a body of work, and the body of work is in danger of being legible only to specialists who maintain the emulation environment.
The lesson
Literary hypertext is the answer to the question of what hypermedia could do as an art form rather than as a technology. The answer is that it could do real things — could produce works that critics took seriously, that programs of study were built around, that other writers took as models. The form did not become mainstream and probably could not have, given the constraints of the medium and the economic structure that supported it. But the form existed, the works exist, and the question of whether literary hypertext was a viable art form is not, as the dismissive accounts sometimes suggest, an open one. It was viable, briefly, in a particular institutional and technological window, and the closing of that window in the late 1990s was a loss not because the form was guaranteed to produce a thousand more masterworks but because the form had not been exhausted.
The next chapter goes to the third strand of hypermedia, less literary and less ambitious than either Brown’s or Eastgate’s, but in some ways the most directly recovered today. NoteCards at Xerox PARC, KMS at Carnegie Mellon, and the smaller knowledge-base tradition that grew up around them treated hypertext primarily as a tool for thought rather than as a tool for publishing or for literature. The systems they produced were used by their builders to do intellectual work — research, theorem-proving, software design, writing — and the use revealed properties of hypertext that the other traditions had touched less directly. The recovery, in Roam and Obsidian and Logseq, has been substantial.