After the document model for scholarly communication: some considerations for authoring with rich media

Content in digital spaces is inevitably duplicated into new contexts, often automatically. While many people may associate duplication with manually copying files (e.g., from a hard drive to a USB key), digital content is also duplicated any time a web page is loaded on the Internet. Web browsers (e.g., Mozilla’s Firefox and Apple’s Safari) interpret HTML documents and create a visual display based on the document’s structure. But barring a few exceptions (e.g., scalable vector graphics), these text-based documents do not actually contain video, audio, or images. Instead, they reference the unique resource identifiers (or URIs) of such media, and a browser then asynchronously assembles them. During these assembly processes, Internet files are downloaded to a user’s computer as they are displayed in the browser. These asset files are then cached and kept in offline storage for a period of time (or until they are manually removed). This way, they can be rapidly recalled or regenerated—indeed, rendered undead—when a person returns to a previously visited site (Chun 2008, 171).

Although seemingly trivial, such default duplication is crucial to contemporary culture, especially when we consider the materiality of digital content. As Matthew Kirschenbaum argues, "computers . . . present a premediated material environment built and engineered to propagate an illusion of immateriality" (2008, 135). In this particular case, an "illusion of immateriality" manifests because users do not need to know how a browser works in order for it to leave a breadcrumb trail of their browsing habits, and (depending on the reliability and speed of an Internet connection) assembly generally occurs at a rate fast enough to escape human perception. What’s more, digital content allographically persists throughout the reproduction process (Kirschenbaum 2008, 133; Goodman 1976). For example, a JPG image will appear formally identical to all sixteen copies of it, even if each copy is inscribed, or autographed, on a different server. In such climates—where routine access to and acquisition of digital content presumably demands little to no labour on the user’s part—illusions of immateriality enable a lack of attention to the procedures through which people gather and interpret information. Now central to humanities research and composition, these procedures include searching an array of domains, clustering content found across them, copying that content, scraping its metadata, annotating it, granulating it, analysing it (closely or distantly), editing it, and citing it (Turkel 2011). And, in the history of print and manuscripts, we have insufficient precedent for how to document these procedures or approach them critically. Even if the duplication of digital content is a constantly regenerating trajectory anchored in a long legacy of writing and inscription, it begs for new scholarly practices (however quotidian they may at first appear).

These new scholarly practices demand new platforms, two of which are the focus of this essay. By foregrounding the duplication of digital content into new spaces and exploring the implications of that duplication, in the following paragraphs we underscore the current need for platforms that help humanities practitioners: (1) facilitate the aggregation of web-based content, (2) responsibly embed and share that content, (3) interpret the content in a flexible and open way, (4) identify and create relationships across the content, (5) express those relationships through a variety of formats, genres, views, and arrangements (e.g., juxtaposition of media, text, metadata, and dynamic visualisations), and (6) ultimately think beyond the "document" model for scholarly communication. These gestures by necessity raise the following questions: what should digital scholarly communications look, sound, and feel like? How should people navigate them? How are writing and attribution reimagined through new platforms? How and where should digital content be stored? And finally, to what degree should audiences be able to personalise or tailor content based on their own contexts and preferences? To unpack these questions as well as our argument, we briefly look at the online publication platform, ThoughtMesh, followed by a more thorough analysis of the multimodal authoring platform, Scalar.

ThoughtMesh: an online publication platform

Produced in collaboration between the University of Southern California’s Vectors Journal and the University of Maine’s Still Water Lab, ThoughtMesh (Figure 1) allows authors to ease their print material into a distributed online environment. Released in 2005, it asks authors to granulate their uploaded articles by tagging each section with descriptive metadata, or keywords.

Figure 1: A screenshot of ThoughtMesh (at http://thoughtmesh.net/). Credit: Jon Ippolito, Craig Dietrich, John Bell, Chirag Mehta, and Vectors Journal.

Authors can create these keywords on the fly or choose from an existing list. Much like other web-based genres (e.g., blogs), readers of these articles can then click keywords (presented in a tag cloud) in order to reveal other chunks of text related to the same theme or concept as the section being read. However, a distinguishing feature of ThoughtMesh is that—when a keyword is clicked—related materials from both the current document and across the Internet are displayed. This subtle enhancement of the reading and discovery process provides non-sequential paths for audiences through domain-specific material. The otherwise closed, complete, or frozen scholarly document becomes one stress point among many in a longer line of inquiry. With ThoughtMesh, readers might focus on a keyword—such as "digital"—that they are researching and then follow it through related instances of text and associated media, perhaps looking specifically for what is duplicated and in what contexts. Along the way, they will survey a number of scholarly articles and sample the domain-specific design of each. In other words, ThoughtMesh helps demonstrate how duplication and repetition across the web do not imply homogeneity or sameness. One effect of ThoughtMesh’s metadata-oriented distribution of knowledge is to wonder how scholarly communications might be modelled if composed within this partly user-generated, partly automated network of contiguous-yet-connected spaces, each with different styles, investments, and areas of emphasis.

Scalar: a multimodal authoring platform

The multimodal authoring platform Scalar was designed precisely with this question of modelling scholarly communications in mind. Intended for publishing long-form communications—or books—online, Scalar’s mandate is to create a space for composing that incorporates responsible duplication and relational affordances directly into the authoring environment. A project of the Alliance for Networking Visual Culture (ANVC) in association with Vectors Journal as well as the Institute for Multimedia Literacy and the Center for Transformative Scholarship at the University of Southern California (USC), the Scalar platform particularly facilitates work that engages visual materials, and it enables users to assemble media assets from multiple sources and juxtapose them with their own writing in a variety of ways. Scalar’s team of developers recently released the platform’s source code, and a few scholars have already used it to compose and publish material. One such scholar is Virginia Kuhn, whose Scalar book, Filmic Texts and the Rise of the Fifth Estate (recently published in the International Journal of Media and Learning), is our tutor text for this essay.

In many ways, Scalar’s authoring environment is similar to other web-based content management systems (CMS), including WordPress. As indicated by Figure 2, authors are presented with a what-you-see-is-what-you-get (WYSIWYG) visual editor where text can be edited and formatted in HTML (with clicks and drags rather than actually writing HTML). Authors can opt to author in HTML as well.

Figure 2: A screenshot of Scalar’s WYSIWYG visual editor.

However, unlike most other systems, Scalar integrates Dublin Core (DC) metadata entry into a scholar’s workflow. Fields are available to enter any (or all) of the DC elements (e.g., title, creator, subject, description, date, type, and format). Scalar also includes a set of editor hyperlink buttons (teal in Figure 2) that tell the system to re-write hyperlink behaviour depending on the button chosen. For instance, some editor buttons facilitate linkages to pages within a Scalar project. When clicked, these linkages can either direct audiences to another page or layer one Scalar resource on top of another. A screenshot of Kuhn’s article (Figure 3) provides an example of such layering.

Figure 3: A screenshot showing layering in Scalar of Filmic Texts and the Rise of the Fifth Estate (at http://scalar.usc.edu/anvc/kuhn/). Credit: Virginia Kuhn.

Meanwhile, other buttons allow people to: (1) create pages that contain other pages (thereby creating a Scalar path), (2) assign a tag relationship between existing pages (where the tag itself becomes a node or resource in the project), or (3) link to media at certain time intervals within the media (thereby establishing a media annotation).

With these relationships and links in mind, one of the central premises of Scalar is that lines of scholarly inquiry often overlap or intersect, even if they do not physically appear near each other in a book. For example, in print, a claim in Chapter 1 might resonate with a point made in Chapter 3. Or, a primary source might be examined in Chapter 2 as well as in Chapter 4. Comparable to ThoughtMesh in its approach to keywords, Scalar can instantly visualise such relations and explorations within a book, offering both authors and audiences the opportunity to see (either up close or from a distance) how content is articulated. Fittingly enough, in Scalar these articulations are referred to as "paths." For a view of paths in their abstract form, see Figure 4, which is a screenshot of Kuhn’s Filmic Texts.

Figure 4: A screenshot of Filmic Texts and the Rise of the Fifth Estate (at http://scalar.usc.edu/anvc/kuhn/). Credit: Virginia Kuhn.

In this screenshot, the structure of Kuhn’s book is visualised, and—importantly—the content of any given path includes pages (e.g., "After Iraqi Doctors," associated with an orange dot), tags (e.g., "Models as Scaffolding," associated with a red dot), and media (e.g., "IML340 Student Responses," associated with a green dot). This combination of content types is key to Scalar’s design, and it is further discussed in the following paragraphs. For now, it is important to highlight that (much in the fashion of any book chapter) an author actively constructs these paths, often duplicating content that already exists elsewhere on the web. However, each resource in Scalar—a path, page, annotation, tag, comment, version, or media asset—is treated discretely. Which is to say: every resource has a URI. As in ThoughtMesh, these resources can be connected by descriptive metadata. For instance, Figure 5 is a snippet of a Filmic Texts page dedicated to the tag, "emergent technologies," which has its own location and description in Kuhn’s Scalar book.

Figure 5: A screenshot of Filmic Texts and the Rise of the Fifth Estate (at http://scalar.usc.edu/anvc/kuhn/). Credit: Virginia Kuhn.

This page simultaneously describes the tag and links to all resources it references (e.g., "Decoding Filmic Texts"). Figure 6 shows how a single media asset in Scalar (e.g., "IML340 Student Responses") can be referenced and interpreted multiple times in a single book, and audiences can see where, exactly, those references occur within the larger framework of the argument.

Figure 6: A screenshot of Filmic Texts and the Rise of the Fifth Estate (at http://scalar.usc.edu/anvc/kuhn/). Credit: Virginia Kuhn.

The same is the case for pages: they may be repeatedly used, and doing so underscores the relevance of context and the function of amplification (e.g., the repetition of a page for the purpose of emphasis) in Scalar. Yet more generally speaking, paths in Scalar demonstrate how all resources exist in many-to-many relationships enabled in part by the Resource Description Framework (RDF).

Scalar and RDF

As a technology, RDF is applicable to scholarly communications for both practical and philosophical reasons. In the case of Scalar, it helps the platform facilitate traces, citations, and views that have already been developed by the semantic web community. Practically, it offers a flexible way to model data from multiple sources and to contextualise material based on ontologies. Philosophically, it provides a solution distinct from the relational systems prevalent on the web today, and it is a device for expressing information derived from a variety of locations or structures. Data kept in RDF are expressed as "triples," which reduce knowledge to a subject-predicate-object set. Multiple triples come together to define a "graph." A large graph can describe a person, place, or thing, and when two or more triples describe the same subject, variability or even contradiction can emerge (especially when objects are defined differently). For instance, one statement may say that a plate is red, while the other says it is orange. An object can be an image or a video available on the Internet, or it can be a web page or even a word. These discrepancies are of extreme concern to semantic web specialists because any triple could potentially disrupt what is known about a subject. The content of triples can therefore be easily called into question, and validation may require an additional layer of trust. Yet, in a concentrated scholarly authoring environment, this concern can be sidestepped by assuming that—much like the process of writing a shared document—the environment involves a relatively small group of knowledge workers. So, while trust and validation are understandably important concerns for the broader semantic web community, they are of lesser concern to scholars who typically work in small-scale collaborations. And once these trust and validation issues are acknowledged and addressed, many other benefits of the semantic web can be applied to the workflows of digital scholarly communication.

For example, scholars working with content from archives (e.g., Internet Archive and Vimeo) across the Internet require URIs to point to the content they interpret. In RDF triples, the subject, predicate, and object can all be expressed as URI references (although, as we will see in the example below, neither subjects nor objects require them). This aspect of RDF is one of its more confusing features: it assumes all resources—whether they are the subjects being described or the predicates linking subjects with objects—have URIs that can indeed be attributed to each resource. Admittedly, for subject and object instances, the confusion should not be considerable. We are well familiar with the idea that there is stuff on the Internet and that each bit of stuff has an address. However, what is quite perplexing is that predicates, too, have URIs. Consider the following example, which is a single statement from Kuhn’s Scalar book:

The resource http://scalar.usc.edu/anvc/kuhn/index has the Dublin Core Title "Introduction to Filmic Texts and the Rise of the Fifth Estate" (in English).

This human-friendly version of the triple is rather clear. The subject (the index page for Kuhn’s book) has an object (the title of that page) written in English, and the predicate is "has title." While in this example the subject has a URI, the object does not and is thus referred to as a "literal." Now let us look at the same statement formatted in RDF/XML:

<rdf:RDF xmlns:rdf=“http://www.w3.org/1999/02/22-rdf-syntax-ns#” xmlns:dcterms=“http://purl.org/dc/terms/”><rdf:Description rdf:about=“http://scalar.usc.edu/anvc/kuhn/index”><dcterms:title xml:lang=“en”>Introduction to Film Texts and the Rise of the Fifth Estate</dcterms:title></rdf:Description></rdf:RDF>

Again, the predicate here points to the "Dublin Core Title" (dcterms:title), assuring all systems are aware that Kuhn’s publication is using the DC interpretation of "title" and its particular construction of meaning for the subject, http://scalar.usc.edu/anvc/kuhn/index.

Since all information placed into Scalar is expressed in RDF (and, when appropriate, using the DC ontology), the platform’s interface widgets can depend on consistent metadata, such as DC titles for every page, media asset URIs, tag references, and path relationships. Importantly, there are no hierarchies across the URIs for RDF subjects, objects, or predicates. In Scalar, a link to a resource does not judge the prominence of what it references; that resource could be the longest paragraph in Kuhn’s argument or her annotation of Berger’s Ways of Seeing. Consequently, RDF’s connection of URIs through a collection (or "graph") of triples allows systems mindful of the framework to promote a flattened hierarchy between content. In RDF-backed environments, an image holds the same "weight" as a web page, a video can contain audio, and that audio can be isolated and referenced accordingly. Such a flattening by necessity influences how authors compose in Scalar.

Scalar’s relationships with online archives

For Filmic Texts, Kuhn composed all of the writing in addition to some of the videos and images. Other media in the book include work by her students (a primary focus of the article) and related supporting material gathered from around the web. But with few exceptions, all of the media assets in Filmic Texts are housed at the Internet Archive (IA). When the article is assembled in a browser, content on IA’s servers is integrated with Kuhn’s primary content on the Scalar server. And in order to search for and connect media to her project, she used Scalar’s import feature, through which authors can discover and grab media from YouTube, Vimeo, and the platform’s affiliated archives (i.e., Critical Commons, the Hemispheric Institute Digital Video Library, HyperCities, and IA). This entire importing process can be conducted within Scalar. Once imported, media are available for linking to individual segments of text anywhere in a project, or they can be inserted directly into a page using Scalar’s inline feature. Since media assets imported into the platform remain on the servers of their source archives, the only content that is actually copied is an asset’s DC metadata (if the originating metadata is not formatted in DC or another ontology supported by Scalar, then it is converted using XSLT). Distributing resources in such a way may seem risky. After all networks like YouTube are largely user-driven, and content is constantly subject to removal, either by contributors or algorithms that detect copyright infringement. Yet, in order to anticipate the inevitable ephemerality of web-based content, Scalar partners with a range of archives serving distinct purposes. Among them is Critical Commons, which is particularly relevant to the kinds of arguments that Scalar facilitates. It is housed at USC and was specifically created to harbour otherwise protected content frequently used for teaching and writing in media studies.

It should be noted that, when following the workflow described above, authors are not creating small, isolated archives on the Scalar server. Although they can upload their own videos, audio, and images to that space, they are instead encouraged to house them with a partner archive. Or, in the case where assets are already online, they are encouraged to embed those assets in their Scalar projects. That way, systems point to existing URIs rather than duplicating resources and producing redundancies. Here, the advantage is that media playback is overseen by groups that not only have institutional support but also specialise in metadata, asset categorisation, provenance, and interoperability.

Scalar, duplication, and attribution practices

Another advantage of how Scalar partners with online archives is that it encourages share-with-attribution practices, facilitated by the transfer of DC metadata. Such distributed attribution practices are already routinised through the design of popular networks like Facebook, Twitter, Google+, and Delicious. Twitter’s "retweet" layer is a key component of its success. While Facebook has no clear way to create citations upon upload, it recently launched a "via" mechanism for sharing other people’s postings in a fashion comparable to sharing on Google+. Meanwhile, Delicious is premised entirely on bookmarking other people’s work. Whether for personal promotion or merely due to an interest in transmitting knowledge, participants in these networks demonstrate that there is at least a baseline effort to give attribution for duplicated, embedded, or referenced content. True, most people may not go out of their way to attend to attribution issues; nevertheless, they will probably give credit if provided simple and consistent mechanisms to do so. That said, scholars can adopt or re-configure these popular models to meet the needs of their own platforms and processes.

In the particular case of Scalar, when a media asset is imported or embedded into a book, its trail is foregrounded through a front-end interface clearly displaying its title together with easy access to its source URI and (ideally) metadata for most (if not all) of the DC fields taken from the object’s original metadata. By making such information readily visible to audiences, Scalar encourages authors to leave breadcrumbs of their scholarly production and to give necessary credit to fellow knowledge workers. Put differently: Scalar works to make embedding a form of citation in itself—a citation that is, quite importantly, machine-readable and traceable. Such citation practices seem especially important to digital humanities, where—as Mark Sample argues—"The ‘builders’ will build and the ‘thinkers’ will think, but all of us, no matter where we fall on this false divide, we all need to share. Because we can" (2011). Of course, resources are rarely—arguably never—shared in raw form, devoid of context or design. As platforms continue to facilitate the integration of web-based content into scholarly communications, developers will need to determine how things like media assets and their metadata are presented and expressed. They will also need to determine to what extent audiences can tailor, personalise, or re-arrange content.

Tailoring content in Scalar: a plurality of views

Designers such as Bret Victor approach the issues of tailoring, personalisation, and re-arrangement through the framework of context-sensitive design. Put simply, context-sensitive design is design that accounts for the perspectives and habits of users: their location, preferences, histories, and needs. It works against fixed or static documents. For instance, in "Explorable Explanations," Victor unpacks what he thinks "active reading" means today, especially as it concerns reading web-based content (2011). He writes: "People currently think of text as information to be consumed. I want text to be used as an environment to think in" (2011). He then provides three ideas for imagining text as an environment. One of them is the "reactive document," which "allows the reader to play with the author’s assumptions and analyses, and see the consquences [sic]" (2011). While Victor’s example of a reactive document focuses more or less on how to manipulate (e.g., through a click or a drag) the quantitative data supplied in an author’s claims (e.g., switching $10 to $25 in order to immediately see how the results differ), Scalar indirectly applies Victor’s principle of play by inviting audiences to change the "view" of content in front of them. On the authoring side, the platform provides a number of view templates that can be applied to individual resources. Each template creates a different arrangement between text, media, annotations, and tags. And some go further by visualising the relationship between otherwise discrete instances. As Figure 7 (a screenshot of Kuhn’s book) demonstrates, after an author publishes content, audiences can select which view or arrangement they prefer.

Figure 7: A screenshot of Filmic Texts and the Rise of the Fifth Estate (at http://scalar.usc.edu/anvc/kuhn/). Credit: Virginia Kuhn.

In this case, the default view in Figure 7 (selected by Kuhn and marked with an asterisk) is "Split Emphasis." Through design, equal attention is given to media (on the left) and text (on the right). However, many other views (thirteen, to be exact) are available to audiences. Which is to say: the default view may not be the view through which audiences ultimately interpret Kuhn’s argument. In Scalar, contexts, and thus the content, are subject to change; and presumably authors should compose accordingly. Consider a few examples.

Figure 8: A screenshot of Filmic Texts and the Rise of the Fifth Estate (at http://scalar.usc.edu/anvc/kuhn/) in "Visualization: Radial" view. Credit: Virginia Kuhn.

Exhibited in Figure 8, the "Visualization: Radial" view is one of five HTML5 visualisation styles available in Scalar. The radial view allows audiences to browse all content in a book from a distance, and it displays the relationships between multiple content types: pages, paths, comments, media, tags, annotations, and authors (when multiple people contribute to a single Scalar project). Through this kind of contained, circular expression, audiences can see, for instance, what media are embedded or referenced in a page (e.g., "IML340: Praxis…"), how that page is tagged with descriptive metadata, and in what path(s) the page may be found. Meanwhile, the "Visualization: Index" view in Figure 9 serves many of the same functions. It graphically visualises content (organised by type) in a Scalar book and allows audiences to view relations between them.

Figure 9: A screenshot of Filmic Texts and the Rise of the Fifth Estate (at http://scalar.usc.edu/anvc/kuhn/) in "Visualization: Index" view. Credit: Virginia Kuhn.

Whereas Scalar’s five visualisation types represent a project abstractly (or from a distance), the "History Editor" view in Figure 10 lists however many versions of a given resource are available and allows audiences to compare aspects (e.g., word choice, date published, and number of characters) of those versions (which are numbered and time-stamped).

Figure 10: A screenshot of Filmic Texts and the Rise of the Fifth Estate (at http://scalar.usc.edu/anvc/kuhn/) in "History Editor" view. Credit: Virginia Kuhn.

Using this view, authors can also revert to previous versions or delete ones they no longer wish to make publicly available. Importantly, audiences can also comment on any version in the list, and—following Scalar’s use of RDF—a distinct URI exists for each version.

Similar to the "History Editor" option, the "History Browser" view in Figure 11 surveys the history of a resource’s versions. Audiences can scan each version and compare versions at a glance.

Figure 11: A screenshot of Filmic Texts and the Rise of the Fifth Estate (at http://scalar.usc.edu/anvc/kuhn/) in "History Browser" view. Credit: Virginia Kuhn.

In so doing, they can get a sense of how the content and design of a project has changed over time, and they can also provide commentary on any version or on the entire view itself. Perhaps more persuasively than any other Scalar view, the History Browser exhibits the processes through which digital scholars compose, breadcrumbs and duplication included. And if these scholars make deliberate decisions to frequently version their work, then they can effectively document the iterative development of their arguments and lines of inquiry—steps that are too frequently deemed unfit for publication because they are ostensibly rough or ephemeral.

Figure 12: A screenshot of Filmic Texts and the Rise of the Fifth Estate (at http://scalar.usc.edu/anvc/kuhn/) in "Metadata" view. Credit: Virginia Kuhn.

Finally, as Figure 12 effectively demonstrates, the "Metadata" view lists the metadata for any given Scalar resource, and elements are encoded as links where necessary. Of note, the default view of the resource is documented in the metadata (using "scalar:defaultView"), and anything directly referenced by the resource is listed as a DC reference. This view is particularly useful for citing, duplicating, and archiving Scalar materials.

Aside from the variety of interpretations sparked by these views, not to mention the aesthetic and conceptual functions they foster, the plurality of arrangements and contexts in Scalar is important for at least four reasons. First, allowing audiences to change a page’s presentation can disrupt models for "captured" or "frozen" content. The flexibility of relations central to Scalar unlocks scholarly communication from a fixed representation, particularly when flexible architectures are in short supply and most content management systems implement rigid templates through both their data and presentation layers (Arola 2010). Second, through a form of context-sensitive design, this flexibility calls attention to the assumptions that subtend authorial intent, composition, and data expression. Ultimately, no one view is a direct representation of data. Rather, each shapes audience perception and constructs both a subject and an argument (which are steeped in disciplinary histories of interpretation) (Drucker 2011). Third, and quite practically, Scalar poses curious opportunities for scholars who want to reference, cite, or present a Scalar book as they interpreted it (e.g., how screenshots are used in this essay). While each resource comes with metadata facilitating attribution, Scalar’s "Metadata" view is but one of thirteen views. It is also one of the least likely to be read by a human user. As such, Scalar practically begs researchers to screencast a Scalar book as they are engaging it. Through the platform, perception and the content perceived are essentially collapsed. A Scalar "view" becomes a design feature joining book and critic. Yet at the same time, it is literalised. A view can be referenced (e.g., "Visualization: Radial" or "scalar:defaultView"), and it—not the file format (e.g., .html)—is included at the end of a Scalar URI (e.g., "iraqidocs340.vis"). Not only can content and form be expressed independently (Liu 2004), the action that links them can be isolated or alienated, too. Still, a Scalar view should not be reduced to a template mapped neatly onto some text, and text cannot be described simply as that which is shaped by its view. Instead, a view is also the behaviour that puts resources into conversation. Framed this way, resources in Scalar are best understood as composites, which are enacted at the intersection of computer and human vision—the automated and the volitional, the discrete and the continuous.

Fourth and finally, the plurality of arrangements and contexts in Scalar matters most because it corresponds with a cultural moment when people’s orientations are often personalised and information is almost always tailored. One possible consequence of such context sensitivity is complicity and consent, a filtering out of complexity in favour of the default and familiar. For example, consider how people increasingly rely on personalised search engines (e.g., Google) to discover information and produce knowledge. But another possibility is the formation of rich identities afforded by these ephemeral, contradictory, multimodal, and ubiquitous expressions of digital content. In other words, to create a context-plural platform with various views is to do more than embrace a multiplicity of perspectives; it is to simultaneously see through them, complicate what enables them, and compose with and against them. For scholars working with dynamic content, we think this option warrants the challenges it presents.

After the document model

While platforms like ThoughtMesh and Scalar are being developed, the "document" model of scholarly publication (i.e., where the parameters of digital publication are neatly contained within publication standards and formats) is alive and well in the humanities. As but one example, consider JSTOR, which allows scholars to download portable document format (PDF) versions of journal articles. These versions are fantastically convenient. They can be easily organised, printed, highlighted, searched, emailed, referenced, and annotated; they can also be quickly detached, or unbound, from the journal issues in which they are published. In short, they mimic a print mode of scholarly communication while also accommodating affordances (e.g., ease of circulation and duplication) unique to digital content.

Yet, as humanities composition continues to incorporate multimedia (e.g., video, audio, maps, images, and graphs) as well as multiple modes of attention (e.g., close listening, distant reading, scanning, bouncing, and repeated watching) into interpretative practices—and as research continues to produce analytical tools (e.g., Voyant, http://voyeurtools.org), experimental interfaces (e.g., Vectors, http://vectors.usc.edu), and distributed forms of iterative, large-scale expression (e.g., HyperCities, http://hypercities.com)—the proto-print document model will increasingly face resistances. This speculation does not imply that PDFs and other such formats will soon be obsolete. They will likely adapt through remediation, or they will continue to be one option among many for disseminating and accessing scholarly content. After all, as we argue throughout this essay, one exciting challenge facing scholars today is the plurality of contexts and views in digital environments. Since the web is becoming increasingly customisable alongside the neoliberalisation of networks, it is by extension increasingly slippery to assume that even two people are viewing the same content in the same fashion. Of course (at least for now), PDFs brush against this tendency in some ways. They freeze content; they make saving synonymous with printing in digital culture.

But we can certainly imagine a not-too-far-off scenario where—for instance—software libraries are created to manage scholarly content, which is displayed in a context-sensitive fashion (e.g., tailored to a reader’s profile, search history, geolocation, or interpretive preferences) (Victor 2006). Publications could thus be built to accommodate an array of audiences through a variety of media, stylesheets, and reactive behaviours (Victor 2011). Accordingly, scholars will need to determine how to best document the traces of their work (duplication included) across the personalised iWeb: to cite or reference not only the source of an idea but also how it was viewed. Whatever the ultimate avenue, the potential for scholarly communications to leverage the histories, content, contexts, and metadata of humanities research is embedded in the trajectories of its platforms.