Intermedia, Microcosm, and the academic hypermedia tradition

While Nelson worked on Xanadu and Engelbart on NLS, a third hypermedia tradition was forming inside research universities. It was less ambitious than Xanadu, less integrated than NLS, and less famous than either, but it shipped working systems, taught classes with them, published peer-reviewed research about them, and produced a set of formalisms — the Dexter Reference Model, the notion of open hypermedia, the standard vocabulary of anchors and links and webs — that still shapes serious thinking about hypertext when serious thinking about hypertext happens. The two centers of this tradition were Brown University in Providence and the University of Southampton in England, and the systems most associated with each are Intermedia (Brown, mid-1980s) and Microcosm (Southampton, late 1980s through the 1990s). They were not the only academic hypermedia systems; a small constellation of related projects existed at Aalborg, at Sun Microsystems, at Carnegie Mellon, and elsewhere. But Brown and Southampton are the two reference points, and the things they did that the web did not do are the substance of this chapter.

The Brown tradition

The hypermedia program at Brown was old. Andries van Dam, the Belgian-American computer scientist who arrived at Brown in 1965 and has been there ever since, began work on what became the Hypertext Editing System, or HES, in 1967, in direct collaboration with Ted Nelson. HES ran on an IBM/360 mainframe with a 2250 vector display, and it was, by Nelson’s own account, the first running hypertext system in the world. It was a significant project: NASA ended up using HES for the documentation of the Apollo program, in part because the branching structure of hypertext was a better fit for the cross-referenced technical materials than the linear structure of printed manuals. HES was an instrument, not a product; it ran on one machine in one room and required IBM-class hardware. Its significance was that it demonstrated, in 1968, that the hypertext idea could be implemented and used for actual work.

HES was followed at Brown by FRESS, the File Retrieval and Editing System, in 1969. FRESS removed several of HES’s limitations and added bidirectional linking — every link in FRESS could be traversed in either direction, and the system maintained the link database in such a way that the destination of a link knew which links pointed to it. This is one of the properties Xanadu had specified and the web later rejected, and FRESS had it working at Brown in 1969. FRESS was used to teach undergraduate English courses at Brown through the 1970s, including a poetry class taught by Robert Scholes, who used the system to let students annotate, cross-reference, and discuss texts in a way that traditional course mechanics could not match. The pedagogical experiments produced data: students who used FRESS were more engaged with the material, more willing to make connections across texts, more able to follow associative threads through a curriculum. The data was published; the use continued for over a decade.

Intermedia, the next system in the lineage, was built between 1985 and 1989 by the Institute for Research in Information and Scholarship (IRIS) at Brown, with funding from IBM and Apple. The project leader was Norman Meyrowitz, with Nicole Yankelovich, Bernie Haan, Karen Catlin, and others on the team. Intermedia was a serious piece of engineering. It ran initially on Apollo workstations, later on Macs under A/UX. It supported text editing, graphics editing, a timeline editor, and a viewer for digital video — a multimedia authoring environment in the genuine multi-medium sense, six years before the same idea reached the consumer market in compact disc-ROM format. It had bidirectional links, like FRESS, and it had something newer: linkbases.

A linkbase, in Intermedia’s vocabulary, was a separate database of links that could be applied to read-only documents. The links did not live inside the documents they connected; they lived in their own store. A reader could open a document and choose which linkbases to overlay on it. Different linkbases supported different uses: a student studying Victorian poetry might overlay the linkbase produced by their professor for one course; a researcher working on the same texts might overlay a different linkbase produced by a colleague at another institution. The texts were stable; the links were the contribution of each interpreter. This is a much more sophisticated linking model than the web’s. It separates the act of authorship of the document from the act of authorship of the citation network, and it treats the citation network as a first-class object that other people can produce, share, and combine.

Intermedia was used at Brown to teach two courses prominently: a course on English literature taught by George Landow, who became one of the major theoretical voices in early hypertext studies, and a course on cell biology taught by Peter Heywood. Landow’s course assembled the “In Memoriam Web,” an Intermedia document collection built around Tennyson’s elegiac poem with links to biographical, historical, literary-critical, and contextual sources. The web was developed by Landow with his students; it grew over multiple offerings of the course; and it became, for a period in the late 1980s and early 1990s, one of the most-cited examples of educational hypertext in the literature. Landow’s book Hypertext: The Convergence of Contemporary Critical Theory and Technology (1992) drew on the Intermedia experience and was, for a time, the standard humanities introduction to the hypertext concept.

Intermedia was not a commercial product and was not designed to be one. The system required workstation-class hardware that was beyond what most institutions could deploy. It was demanding to learn. It needed a server architecture that fit Brown’s setup and not many other places. The Intermedia team had hoped, in the late 1980s, to find a commercial partner who would turn the system into a product; the hope did not materialize, and when the funding from IBM and Apple wound down in 1990, the system stopped being supported. Norman Meyrowitz left Brown for the commercial software world; Nicole Yankelovich continued at Sun Microsystems, where she worked on a related system; the team scattered.

The Southampton tradition

Microcosm was developed at the University of Southampton in England, beginning in the mid-1980s, under the direction of Wendy Hall, with Hugh Davis, Les Carr, and others on the team. Microcosm took the Intermedia idea of separate linkbases and pushed it further into what Southampton called “open hypermedia.” The Southampton position, articulated in a series of papers through the early 1990s, was that the link should be fully decoupled from the document. A document on a Microcosm system was, in itself, just text — or just an image, or just a video. The links that connected the document to other documents were stored in a separate link service, accessed through the operating system, and applied to documents at viewing time. This had several consequences that the field considered important.

First, links could be applied to documents the linker had no permission to modify. A researcher could link from a passage in a published paper to another passage in another published paper without altering either paper. This is a property the web does not have. To link from a web page to another web page, you have to be the author of the source page. You can write commentary about another page, but you cannot decorate that other page with a link the way a reader marks up a book with marginalia. Microcosm let you do exactly that. Second, links could be generic. A Microcosm link could specify “the word ‘photosynthesis’ anywhere in any document,” and the link would be applied wherever that word appeared, in any document the user opened, until the link was deactivated. This kind of “generic link” is something the web has approximated only with browser extensions and overlays, in narrow domains, decades later. Third, links could be dynamic. A Microcosm link could be a query that produced its target at link-traversal time, by searching the document corpus. Fourth, multiple link services could be active at once, with their own scopes and policies, so a single document could be richly annotated by different communities of users with no conflict among them.

Microcosm was deployed in industrial settings, not just academic ones. Ford in the UK used it for technical documentation. The BBC used it for archival access. The Mountbatten Archive at Southampton itself was an extensively linked Microcosm corpus that historians used for research. The system was a commercial product through a Southampton spinoff, Multicosm Ltd., and continued in some form into the late 1990s before being absorbed and re-formed under various corporate identities. The research continued under different names — DLS (Distributed Link Service), Auld Linky, and others — and the Southampton group’s work eventually fed into the early Web Science Trust and into Tim Berners-Lee’s own institutional home, since Berners-Lee took up a chair at Southampton in 2004.

The Microcosm group made the argument, in a series of papers in the early-to-mid 1990s, that the web’s link model was structurally inferior to the open hypermedia model and that the web should be evolved toward something more like Microcosm. The argument was not heard in time. By the time it could have been heard, the web had a sufficiently large installed base that the cost of changing its link model was prohibitive. The Microcosm group, like most of the academic hypermedia community, redirected its work toward the web in the late 1990s and 2000s, contributing to the early development of the Semantic Web, RDF, and Linked Data — all of which can be read as attempts to introduce, on top of the web’s reduced foundation, some of the structural properties open hypermedia had provided natively.

The Dexter Reference Model

The academic hypermedia field formalized its understanding of itself in a series of workshops in the late 1980s. The most influential was held at the Dexter Inn in Sunapee, New Hampshire, in October 1988 and again in March 1989, organized by Frank Halasz and Mayer Schwartz, with a group of about thirty researchers from the major hypermedia systems of the time. The output was the Dexter Hypertext Reference Model, published in 1990 and refined over subsequent years. The model was an attempt to specify, abstractly, what a hypertext system was, in a way that could be used to describe all the existing systems and to evaluate proposed new ones.

The Dexter model has three layers. The within-component layer is the layer at which a document’s content is structured — text formatting, image layout, the internals of the thing being linked. The storage layer is where the model says the hypermedia structure lives: nodes, links, components, the graph that connects them. The run-time layer is where the model says the user interacts with the system — the application that displays nodes, navigates links, and presents the graph to the user. The model is deliberately ecumenical about implementation; it covers HES, FRESS, NoteCards, NLS, HyperCard, Intermedia, KMS, and several other systems, and it lets the systems be compared on common terms.

The most useful thing the Dexter model did, for the purposes of this book, was distinguish the components of hypermedia from each other. A link, in Dexter terms, is not a piece of a document. It is a relationship between anchors, which are positions inside or attached to components, with its own identity in the storage layer. This let the field name what the web later refused to name: that a link is a thing, that it has properties beyond its endpoints, that it can be inspected and modified and stored separately from the documents it connects. The web’s <a href> is a within-component element by Dexter’s classification, not a storage-layer link. This is a precise way of saying what the previous chapters have said less formally — that the web flattened the hypermedia model into the document, and that the flattening was a loss.

Frank Halasz also wrote, in 1988, a paper called “Reflections on NoteCards: Seven Issues for the Next Generation of Hypermedia Systems.” The paper named seven problems the field had not yet solved: search, composite nodes, virtual structures, computation in/over hypermedia, versioning, collaboration, and extensibility. The paper became the most-cited document in the field for the next decade; the ACM Hypertext conferences in the 1990s structured much of their program around these seven issues. The fact that the web, when it arrived, addressed none of them as a research question is one of the reasons the hypermedia community received the web with a mixture of relief and resignation. The web was popular. It was also, by the field’s standards, a step backward on every one of Halasz’s seven issues.

The ACM Hypertext conferences

The community held its first conference at the University of North Carolina at Chapel Hill in November 1987. The Hypertext ’87 conference was small — fewer than two hundred attendees — but it was where the field met itself for the first time, and where the relationships among the various systems and traditions became visible. Subsequent conferences followed in 1989 (Pittsburgh), 1991 (San Antonio), 1993 (Seattle), and then biennially in the 1990s and annually since. The proceedings are a complete record of the field’s research output through the period: Brown people, Southampton people, Aalborg people, PARC people, Carnegie Mellon people, the various commercial system vendors, the literary hypertext writers, and the theoreticians of hypertext as a cultural form all came to the conference and presented papers. The conference series continues — the ACM Hypertext and Social Media conference is the current name — though much of its scope has shifted to web research since the late 1990s.

What is striking about the conference proceedings from the late 1980s and early 1990s is how completely the field anticipated, named, and proposed solutions to problems the web would later face. The Hypertext ’91 proceedings contain papers on link integrity (link rot), on collaborative authoring, on personalization, on adaptive hypermedia, on the relationship between hypertext and database systems. Each is a contribution to a research program. The web, when it began in 1991, was at a level of sophistication that the conference proceedings would have classified as belonging to the early 1970s. The web’s authors — Berners-Lee in particular — were aware of the academic hypermedia tradition but chose, deliberately, to build something simpler than the academic systems specified. The reasons for the choice are treated in chapter fifteen; here it is enough to say that the simplicity was a feature for adoption and a loss for the field’s research agenda, which was rendered, almost overnight, mostly irrelevant to mainstream computing.

What the field had been doing

A summary, with no claim to be exhaustive, of what academic hypermedia in 1991 offered that the web did not:

Bidirectional linking, in HES, FRESS, Intermedia, NoteCards, KMS, Microcosm, and other systems. The destination of a link knew which sources pointed to it, and you could enumerate the backlinks of any document. The web has webmentions and external link analysis; it does not have native backlinks.

Linkbases and overlay annotation, in Intermedia, Microcosm, and Sun’s Link Service. Links could be applied to documents from outside the documents, by readers, in separate databases. The web has third-party annotation systems (Hypothes.is, since 2011; the W3C Web Annotation standard since 2017) that approximate this for some uses but are not native and not universally adopted.

Generic links, in Microcosm. A link could be specified to apply to “this term wherever it appears,” and the system would activate the link at every occurrence in any document. The web does not have this; it has search-on-the-page and browser extensions.

Versioned documents, in NoteCards, KMS, and several others. The history of a document was accessible; links to specific versions were stable. The web has the Internet Archive, layered on as third-party retrofit, and no native versioning.

Composite documents, in NoteCards in particular. A node could be a composite of other nodes, presented as a unit; the composition was a first-class object that could itself be linked to. The web has frames and iframes, which are degenerate composites; it does not have first-class composition.

Computation in/over hypermedia, the fourth of Halasz’s seven issues. Several systems treated this seriously, with mechanisms for embedded scripts, queries, and procedural attachments to nodes. The web answered this question, eventually, with JavaScript, but JavaScript took a decade to arrive and was not the hypermedia community’s preferred answer.

Multi-headed linking, in Intermedia among others. A link could have many sources and many destinations; the system would let the user pick which to follow. The web has, structurally, only one-to-one links and has to fake the others through interface conventions.

Each of these properties has been recovered, partially, in some context, by some system, for some application. None is recovered for the web as a whole.

What the academic field got wrong

The academic field had its own pathologies, and any honest treatment has to name them. The systems were institutional in the sense that they ran inside research universities, required institutional infrastructure, and were not deployable by individuals. The Microcosm-Ford partnership is a fair representative: the system was used in industrial settings, but the cost of deployment was a multi-year integration project, not a download. The academic hypermedia community did not, with the partial exception of HyperCard (which most of them did not really consider one of theirs), produce mass-market software.

The field also had a tendency, common in research communities, to treat its problems as more solved than they were. The Dexter Reference Model was a useful formalization but it papered over real disagreements between systems about what the basic objects of hypermedia even were. The field had structural sympathies for what we would now call “heavyweight” approaches: rich type systems, complex authoring tools, formal models of every entity in the system. The web’s “lightweight” approach — small specification, permissive parser, no formal model of anything — was the kind of thing the field had argued against, at length, in many papers, for sound reasons that did not turn out to be correct at internet scale. The field’s reasons were not foolish; the web’s success was not predictable on the field’s terms. But the field had its analytical apparatus too thoroughly invested in the heavyweight tradition, and when the web arrived the field was slow to recognize what was happening.

When recognition came, in the mid-1990s, the academic hypermedia community largely moved to the web. The annual conferences began incorporating web-research papers. The Semantic Web, articulated by Berners-Lee in a 2001 Scientific American article and developed through the W3C in the 2000s, was in significant part the academic hypermedia community’s attempt to retrofit some of what they had wanted onto the web’s substrate. The Semantic Web has produced significant infrastructure — RDF, OWL, SPARQL, the Linked Open Data cloud — but it has not, despite twenty-five years of effort, become the way most of the web works. It is, instead, a layer used by specific communities for specific purposes (libraries, government open data, biomedical knowledge bases) while the general web remains structurally as Berners-Lee specified it in 1991.

What survives

The academic hypermedia tradition is alive in a smaller form than its peak. The Hypertext and Social Media conferences continue. The field still has practitioners. The vocabulary the field developed — anchors, links, components, webs, linkbases, the Dexter model’s three layers — remains in use among researchers thinking about hypertext as a serious topic. The lineage runs through Web Science research at Southampton, through the digital humanities communities at Brown and elsewhere, through the various open-annotation projects, through the Open Annotation Collaboration and the W3C Annotation Working Group. The work continues. Most of it is invisible to the working web developer, who has been taught that the link is <a href> and that the rest of the hypermedia model is an academic curiosity.

It is not an academic curiosity. It is the description of what a serious hypermedia system would look like, produced by a generation of researchers who built and used such systems for two decades before the web subsumed the question. The recoveries in Part V are, in part, recoveries of academic hypermedia ideas. Webmentions are the IndieWeb’s attempt to recover Intermedia’s bidirectional links. Hypothes.is is an attempt to recover Microcosm’s overlay annotation. Roam Research’s block references are an attempt to recover composite nodes. The Semantic Web, in its more focused forms, is an attempt to recover the structured links the academic tradition specified.

The next chapter goes to the literary side of the same tradition. The hypermedia community that gathered around Storyspace, Eastgate Systems, and the Hypertext ’87 conference contained, alongside the engineers, a contingent of writers who took hypertext seriously as a form. They produced a body of work — afternoon, Patchwork Girl, Victory Garden, Twelve Blue, others — that is the closest thing the field has to a canon. The literary hypertext canon is the answer to anyone who asks what hypertext can do that ordinary writing cannot, and it is the part of the field that the web has done the least to absorb.