At a King's College, London, symposium in 2000 John Unsworth introduced a tentative list of scholarly primitives as "a list of functions (recursive functions) that could be the basis for a manageable but also useful tool-building enterprise in humanities computing" (Scholarly Primitives). A slightly re-ordered application of these seven functions to Roland^HT is a good starting point from which to contextualize it as a digital humanities project, from both the builder's and the user's perspective.

Discovering. Departing from the knowledge that Roland appears in several cultural contexts over a number of centuries, I have deliberately sought out cultural artifacts featuring Roland in an attempt to discover the reasons for his popularity, especially when combined with the character's remarkable mutability in almost all respects. Because potential corpus objects exist in such abundance, I have taken search opportunities as they occurred, utilizing library and interlibrary resources at Brown University, the Oxford and Cambridge Libraries and the British Library, and taking advantage of being in France to travel to Angoulême, Reims and Chartres, photographing Roland art. Criteria for discovery processes emerged in this process, and were continually tested against newly discovered artifacts.

The corpus as introduced in the present work is not comprehensive, in part because there are numerous works in which Roland is mentioned once or twice in passing.10 Such works are not included in the current version of Roland^HT, but may prove interesting to study in the future. Sampling came into play when choosing which parts of which works to encode, and was largely a subjective process. It is discussed further in "Objectivity and Originality" below. Decisions about sampling criteria were based on previously discovered and annotated objects. For example, if the plot of a poem was similar to that of an already encoded poem, and the former did not present a different point of view or other unique material, then the new arrival was set aside for later consideration.

Annotating ultimately took two forms: semantic encoding of the excerpts chosen during the sampling process, and later addition of notes to entire texts and to specific words and phrases. The latter two features appear in the XML document as content within <textnote> and <note> elements, respectively, and have provided the project as a whole much more flexibility of expression than theme/imagery elements and attributes can provide on their own.11

The web interface design aimed to represent the corpus and illustrate semantic features that make it an interconnected whole. Hyperlinks that were created by applying styles to the raw encoded text, and that allow excerpts to be grouped according to user-selected criteria, encourage the user to examine art works that appear in the same groups and compare their other features (such as date and place of creation, medium, treatment of specific themes).

In the process of analyzing the corpus I also compared texts, themes and imagery through statistical analysis of the encoded excerpts (see Appendix D, "Theme Statistics").

Finally, referring is also manifest in multiple ways. Tagging different excerpts with the same semantic elements refers them to each other, and also to an emergent network of meaning overlaid on the corpus but abstracted from its individual components. The above-described hyperlinks serve as referents as well; Unsworth went so far as to equate the two ("linking, or referring, [is] another scholarly primtive"). Unsworth's emphasis on "the importance of stability in reference" – still a painful topic on the web, where links often point to a target that has been destroyed or moved – has been taken to heart in Roland^HT. Here, almost all of the hyperlinks the user sees are dynamically generated, limited to within the confines of the project, and relative as opposed to absolute. Changes in the styling of these links will make them unlikely candidates for obsolescence, since the semantic connections are in the encoding itself, not in a static web page that happens to be rendering them visible.

For all its wide-ranging span of production media, Roland^HT resonates particularly closely with the literary theories of Gérard Genette. Laying out the notion of archigenres as they relate to the romantic triad ("lyrical/epical/dramatic" (63)) in his The Architext: An Introduction (1992), Genette writes:

[They are] archi- , because each of them is supposed to overarch and include, ranked by degree of importance, a certain number of empirical genres that—whatever their amplitude, longevity, or potential for recurrence—are apparently phenomena of culture and history; but still (or already) -genres, because... their defining criteria always involve a thematic element that eludes purely formal or linguistic description. (64-5)

Roland^HT is an architext (see below). Its archigenre is a combination of the digital environment in which it is composed – the hypertext browser (on or off the web) – and specific decisions as to how to use that environment, such as the lack of a search engine (see Chapter 3). Roland^HT absorbs information from the primary sources' paper, stone, music, theater, film and other media; "flattens" them by expressing all texts in electronic form; and unifies them through their thematic elements. Eluding linguistic description goes only so far (Genette himself gets this point across using language). Markup is a meta-language that acts upon – and builds meaning in concert with – natural language; therefore marked up text is particularly well adapted for concepts that take poorly to being described only in natural language.

But Genette's archigenres also elude purely formal description: if there is an archigenre to Roland^HT, it is closely intertwined with its electronic medium. Genette's "lyrical, epical and dramatic" genres refer to the rherotical mechanisms by which works affect their audiences emotionally and intellectually. Medium choice is a rhetorical statement – in this case, the choice of making the project readable using freely available software is a statement of the desired accessibility of the project (see also Chapter 3, "Tools").

This implicitly makes a good case for the place that idiosyncratic semantic encoding may someday occupy in the academic toolset. Aside from the basic syntax of the XML metalanguage12 a Document Type Definition, which defines the syntax of an actual XML markup language, is the most purely formal aspect of semantic encoding. In the codification process I used, the tagset arose from the primary materials being encoded, and was only later regularized and tied together into a system (see also Chapter 3, "Tools"). Thus the formalism necessary for academic consideration of an argument is achieved – by informal means.

Roland^HT may be an inherently multimedia corpus, but most of its texts are word-based. This work resonates both with structuralism and poststructuralism, particularly (as briefly outlined above) with Genette's notions of architext and archigenre. The back cover summary of The Architext has Genette "assert[ing] that the object of poetics is not the text, but the architext—the transcendent categories (literary gentres, modes of enunciation, and types of discourse, among others) to which each individual text belongs." Architext's transcendence, and by implication flexibility, is one of its more attractive qualities. Paradoxically, it seems to be misinterpreted as a rigid structure for classifying texts. In Intertextuality (2000), British scholar Graham Allen writes: "Genette... employ(s) intertextual theory to argue for critical certainty, or at least the possibility of saying definite, stable and incontrovertible things about literary texts" (4). I suggest that the notion of architext is useful precisely because it is flexible enough to not require definitive statements about texts. Genette himself writes (discussing archigenre, whose differences from architext are ambiguous) that "any genre can always contain other genres, and in that respect the archigenres of the romantic triad [lyrical, epical and dramatic] are distinguished by no natural privilege" (65). If archigenres can be recombined, then architexts may well fit into more than one genre or combination of genres. Its precise placement in the multidimentional [archi]genre spectrum would be the result of its interplay with the reader/critic's context, always subjective and never, at least as regards its literary qualities, an absolute truth or complete picture.

Allen does point out that "the term intertextuality was initially employed by poststructuralist theorists and critics in their attempt to disrupt the notions of stable meaning and objective interpretation" (3) proposed by structuralists. Perhaps a more accurate contextual view of Genette's theory of architext is not as a shining example of structuralism but as a transition to poststructuralism.

A counter-argument may point out that in Architext (which first came out in France in 1979) Genette repeatedly references Northrop Frye, who had famously called for a more precise direction for literary criticism. "Criticism," Frye wrote in his 1957 Anatomy of Criticism, "seems to be badly in need of a coordinating principle, a central hypothesis which, like the theory of evolution in biology, will see the phenomena it deals with as part of a whole" (16). This sounds misleadingly simplistic, as though Frye were advocating for an all-encompassing "coordinating principle." But Genette does not take it that way. In his consideration of Julius Petersen's rosette-like system of genres, Genette claims that,

because its defining criteria are mainly formal, it cannot make thematic distinctions, such as those contrasting tragedy with comedy, or romance... with the novel... . For that, perhaps another compass would be needed, or even a third dimension, and undoubtedly relating the two to each other would be as difficult as fitting together the several concurrent—and not always compatible—grids composing Northrop Frye's "system." Here, too, suggestive power far transcends explanatory—or even simply descriptive—capacity. (55, emphasis mine)

Roland^HT's huge span, both in terms of genres and formats of the constituent works, benefits from a perspective similar to Genette's. The construction of the electronic resource was inspired in part by Stephen Ramsay's principle of "going in on a hunch" (2002), without a pre-existing hypothesis as to what the corpus may tell us about intercultural transmission. Readers without much prior knowledge of the Roland stories are at least as much a part of Roland^HT's intended audience as are researchers; the corpus is meant to introduce them to the multifaceted whole and also to suggest thematic directions they may want to follow, but it makes no assumptions about its reader beyond their ability to read in English and use a mouse.

Roland^HT spans not just genres and texts but also academic disciplines. Genette scoffs at the notion of 'disciplines,' proclaiming them to be useful only until a goal is reached:

[A] 'discipline' (let's use quotation marks of protest) isn't, or at least shouldn't be, an institution, but only an instrument, a transitional means that's abolished at once in achieving its goal—which may very well be only another means (another 'discipline'), which in turn... and so on: the whole point is to move ahead. (84)

Genette sees work on texts to be constantly evolving and always transdisciplinary, "endlessly forming and re-forming poetics, whose object, let us firmly state, is not the text, but the architext" (84, emphasis author's). The object of the present study is an architext that contributes to the knowledge of cultural transmission, that adds to an understanding of art's function in the writing of history without making a definitive pronouncement on it.

A critical interpretation addresses not an isolated text, but "the encounter between text and reader" (Leitch et al. 2). The primary purpose of the online Roland^HT is to facilitate such an encounter – with a composite of ideas, a collection of texts, a corpus. In the process of encoding excerpts I engaged in a generalized form of hermeneutics – "the art of understanding and interpreting discourse through systematic procedures" (Leitch et al. 610). All cultural artifacts, particularly those that survive at length, form part of an implicit cultural discourse, the story we tell to ourselves about ourselves. It is this implicit discourse on Western cultures that Roland^HT aims to address.

Perceptions of art's authorship – the ownership of beauty – have changed in a fundamental way relatively recently. Not too long ago, Oscar Wilde illustrated a single-author attitude, typical of late-nineteenth-century critics, in a dialogue.

Ernest. [S]urely you would admit that the great poesm of the early world, the primitive, anonymous collective poems, were the result of the imagination of races, rather than of the imagination of individuals

Gilbert. Not when they became poetry. Not when they received a beautiful form. For there is no art where there is no style, and no style where there is no unity, and unity is of the individual. (Wilde 900-1)

Wilde is right – unity is of the individual. Roland's status as a border-defying, polyphonic folk hero prevents the corpus from having a unity to it. This, combined with the diversity of genres and media employed in the corpus, makes Roland a poor candidate for a Proppian analysis. But, as I will show in the section below titled "An Ur-Roland?", the interconnected polyphony of the corpus is at the core of its artistic value.

The process of Roland^HT's composition is ultimately aimed at understanding texts' interaction with their cultural contexts. Kenneth Burke claims that textual analysis "move[s]... towards a clutter of tiny insights whose worth is impaired by the law of diminishing returns" (Burke 1276). Roland^HT's impairment by this law is in part mitigated by its electronic – and text-only, as regards the encoding itself – format, which makes it easy for another critic to play with the data using a different perspective, which may highlight different tiny insights.13

Considering Roland within the theoretical framework of Vladimir Propp's Morphology of the Folktale (1968), it is compelling to ask the question: does an ur-Roland exist, as an overarching character or else a pervasive storyline?

In a sense, and only to a point, yes – a conception of an ur-Roland is what holds the corpus together. In reality several of them exist, overlapping perceptions of the Roland archetype rooted in three nexus points: medieval France, medieval Scandinavia, and Renaissance Italy. In having multiple anchors as regards both geography and time period, the corpus (which is imperfectly interconnected – no corpus object has direct relationships with every other object) is a distributed network.

Recent discovery and widespread use of distributed networks in biological and computational contexts has confirmed them as optimal for ensuring efficient preservation and propagation of data. Their main benefit is redundancy: the more frequently a piece of data is used, the more likely it is to be stored in multiple storage points. This redundancy in turn makes it less likely that the knowledge in question will be lost. The Roland corpus' cultural "currency" are pieces of knowledge that help a society thrive. So it is reasonable to conclude that the corpus' distributed-network form is a contributor to Roland's remarkable, transformative dissemination over the past thousand years.

France's warrior and feudal lord informed many European retellings of the Song of Roland as well as some other stories (such as, for example, Firumbras and Otuel and Roland (O'Sullivan), as well as Browning's "Childe Roland to the Dark Tower Came"). In Norway he is a child contained entirely within Charlemagne's story, and later an impetuous, indiscreet youth who – on a lark – fails to keep a king alive at his uncle's request. This Roland's headstrong attitude is a large part of what ties him back to the French "original."14 Renaissance Italians and Sicilians began a comic tradition about Roland the star-crossed lover, the French adventurer knights, and the frail, almost senile Charlemagne – a tradition that found its way into Cervantes' Don Quixote and later back into French operatic melodrama.

Today's Roland has come full-circle back to the French epic hero. Zevon's hard-edged CIA agent, Amaya et al.'s protagonist, King's gunslinger, even the supporting character in Baker's Beowulf are all reflections of Charlemagne's knight. The one exception to this is Silverstein's Roland the Roadie, who can be traced back to Italy's comic tradition: he prefers [the] rose-colored glasses [of madness] while Gertrude the Groupie – much like the Pagan sorceress Angelica – charms him, displays disregard for his feelings, and takes another lover.

There may not be a single ur-Roland, but persistent elements of the character do exist. In a meta-sense (not completely portrayed in every story), a combination of fortitude, pride and fierce loyalty may make up an ur-Roland. However, considering him in only this light strips the character of nuance.

The remarkable thing about the Roland corpus is that it contains no ur-story. Themes and imagery recur, but their contexts – and, by extention, they – change radically. The corpus is a true Bakhtinian polyphony – its authors are active participants in dialogue with their characters, who seem to take on a life of their own with the help of a certain contingent of the corpus' audience that not only assimilates these stories but changes them and passes them on (Bakhtin 3-5). It is precisely the multitude of variations on the same themes that renders Roland^HT worth a careful examination.

Objectivity and originality of argument occupy a place of honor in humanities scholarship. But, as Robert Scholes, Nancy Comley and Gregory Ulmer point out in Text Book, originality isn't: "...the desirable quality we call 'originality' does not mean creating something out of nothing but simply making an interesting change in what has been done before you. One develops as a writer by playing with material already in existence" (150). They address an audience of young writers, but the statement is also effective if "writer" is replaced with "critic." Development of the critical mind is aided by playing with, imitating and otherwise responding to pre-existing critical texts. That, however, works only to a point, and is no substitute for playing with primary sources.

Scholes et al. also take on the issue of objectivity. Its platonic ideal is not achievable, they argue: "The eye of an adult human being is never innocent. [...] As observers of life, we go from ignorance to prejudice without ever passing through the mythical land of objectivity" (151). When encoding, I used little more than the skill of observation and prior knowledge of the subject to notice and tag interesting themes and imagery. As my familiarity with the corpus increased, I began to notice more relevant elements – which can become its own trap by predisposing me towards a certain emergent way of thinking. The goal became not objectivity but originality through consistency: for example, discovering new thematic threads midway through the encoding process prompted a return to the beginning and several re-encodings. The connections among corpus objects that are reflected in the encoding are presented as the project's main contribution to the study of Roland. Since I am unable to get away from my own vantage point, the only way for this corpus to become more objective is future consensus by interested parties.

Taken too far, subjective interpretation can become a problem. Susan Sontag takes critics to task in her essay "Against Interpretation": "Real art has the capacity to make us nervous. By reducing the work of art to its content and then interpreting that, one tames the work of art. Interpretation makes art manageable, conformable. This philistinism of interpretation is more rife in literature than in any other art" (8).

Does Roland^HT's form tame the corpus, flatten and homogenize it? In a trivial sense, yes: most obviously, the corpus's medium span is "flattened" into the digital. This is unfortunate, but will have to suffice until our holodeck technology improves; in this the project is yet another instance of responding to art using words on paper.

On a more significant level, Roland^HT counteracts this medium flattening by foregrounding the primary sources, not the critical text(s). Analysis is still there, but as far as the website user is concerned, it is mostly hidden in the code (except when the code is a tag for annotation, such as <textnote> and <note>, in which case commentary is visible after the appropriate icon is clicked). By default users do not see the raw XML, just as they would not see raw footage of a film: fundamental to the visible product, the code is nevertheless filtered through styling and presentation choices. At the same time, web-savvy users can deduce the location of – and thus access – the XML document by looking at the source code for the index page, finding the address of the JavaScript document, and looking there in turn.

This begs the question of when a document should be considered published on the web. There is no obvious way for a novice user to access the XML document, and yet it is not difficult for someone who knows the basics of web publishing to find out its address. Viewing the source code of the main page (a feature of all modern browsers) reveals the name and location of the JavaScript file being used to construct the site. The JavaScript file has purposely been written as clearly as possible, such that anyone with a modest familiarity with programming syntax will be able to make sense of it. Looking closely, the reader will notice that a function called startRoland calls another function called loadXMLDocument, which – in this case – loads rolandht.xml from a specific location (in this case, the same directory as the JavaScript file). Once this information is obtained, the user can load the XML file directly and see its raw code, including the commented-out (and thus invisible to the browser) sections.

Perhaps a for-profit publisher would hesitate to allow even this level of visibility. After all, it is not unlike looking at an author's personal notes; and visible raw code may be copied into another document altogether and thus "stolen." With a project such as Roland^HT, however, this is a twofold advantage – assuming a user comfortable enough with the technologies involved to find the raw data in the first place. If such a user sees a bug in the website, not only will she be more likely to report it at all, but she may quickly look at the code and pinpoint the problem. This is common practice in software development communities, and is more common in settings that remove as many obstacles to seeing the code as possible.

In addition, digital humanities scholars wishing to improve on the project will be more likely to act on that desire if they can paste relevant bits of it into their own text editor and experiment without having to spend time requesting the code and waiting for it to be shared (or not). Roland^HT is published on the web under the Creative Commons15 Attribution-Share Alike 3.0 License, which specifically allows the work to be modified and re-published, as long as Roland^HT's original creators (myself and Ethan Fremen16) are explicitly credited, and the new publication is issued under an identical license. This licensing choice removes as many barriers as possible to the production of new knowledge, while still asserting authorship of the original.

Notwithstanding the above, the ability to filter information by styling, making some of it invisible to the general user, has proven useful in an unexpected way. Weeks after I had encoded three stanzas of Shel Silverstein's "Roland the Roadie and Gertrude the Groupie" and sent an inquiry to the copyright holders asking permission to include it on the website,17 I received word that permission was, in fact, being withheld. No reason was given. As a compromise that both honors the copyright and preserves the work performed on the piece, I have decided to comment out the Silverstein excerpts but leave them in the XML document for the purpose of thesis archival.

The primary learning aid Roland^HT offers is a reading process different from nearly every other text, hyper- or otherwise. The reader is not just encouraged to actively participate in her learning process; it is impossible to read any one of the primary-source texts in its entirety. The same applies to film, comic, and electronic game excerpts. The only exceptions to this are digital images of Roland artworks – and even these are often presented without their immediate physical context. Emphasis is thus shifted from studying individual art works to "listening in" on a polyphonic record of European and American cultural histories.

In addition to clicking on hyperlinks, active reading of the electronic text involves a more active search for information. It is not immediately obvious where and when the excerpts originate: the reader must click on an excerpt title to see that. There is currently no easy way to pursue research structured by objective criteria (for example, by place and time period of the art work's origin). An obvious means of making such research available is a simple stylesheet-based search engine with "canned" (pre-defined) search options. But such an engine would be contrary to the electronic project's purpose; it would make Roland^HT behave like an archive. Because of its inherent subjectivity and non-exhaustive approach to inclusion, Roland^HT would make a bad archive. (See also next section.) In addition, providing a feature so familiar to students and researchers as a search engine would implicitly discourage exploration on the new level of thematic threads, and would shift the emphasis from the network of art back to individual art works.

Three special features of Roland^HT are, however, aimed specifically at students: text notes (shown as square "information" icons by the titles of some works), inline notes (indicated by quill icons within the excerpt texts themselves) and translations of difficult-to-understand Middle English words (available by mousing over any underlined word). Sample teaching modules utilizing these features may be found in Appendix G.

Many digital humanities projects dealing with primary sources are advertised as archives. As stated above, Roland^HT is not an archive in the sense that similar projects seem to use the word, and its unique qualities merit further consideration.

Archives, digital and otherwise, are usually repositories of primary texts and/or other artifacts having in common author(s), place(s), time period(s) or topic(s), plus critical materials (if any). If archival holdings are fragmentary, it is generally not by choice: the fragments are all that has survived, and if more of the artifact were available then it would be archived as well. Critical materials are kept in separate documents. Roland^HT does not strive for completeness, either within single works (only excerpts are presented) or within the corpus (which is representative but incomplete) and its criticism inseparable from primary texts.

Searching for "archive" in the titles of over 18,000 records accessible through the Intute: Arts and Humanities18 internet portal returns 392 records (Intute, 3 April 2007). Of these, a hundred were selected semi-randomly: records 1-25, 51-75, 100-125 and 151-175. Among the selected records, most contained either primary materials, usually accompanied by descriptions but not analysis, or only critical/analytical documents such as journal issues. Only around twenty percent of the selected records contained both primary and analytical materials,19 which were uniformly presented as separate documents. Semantic encoding, if any, generally reflects selection, description and/or editorial choices, as opposed to critique or analysis.

In Roland^HT, on the other hand, transcriptions of text-based primary sources and critical metadata are located in the same document, and the metadata extends beyond editorial choices. This mix of analytical and primary content renders Roland^HT an electronic entity different in both format and content from currently existing archives. The distinction is made more apparent by Roland^HT's intention as the opposite of a static collection of unchanging documents. Its primary-source contents, encoding and annotations are all meant to change as research progresses. (see Chapter 3, "Collaborative Possibilities" ff.)

The Roland legend began, and for a long time was transmitted, orally – through minstrel song, and in the south of Italy through puppet theater (De Felice).20 Although the mechanics of his legend's transmission changed over time to become heavily dependent on writing, the corpus as a whole has retained important aspects of orally transmitted cultural history.

Albert Lord's The Singer of Tales (2000), fruit of research on oral storytelling conducted in Yugoslavia with his mentor Milman Parry, bears evidence to this. Perhaps the most obvious similarity between the corpus and the most recent thoroughly studied oral society is the lack of a canonical version of a story. For modern scholars Roland may have started with the French Song of Roland, but even that story has been told in many different versions around Europe. Relatively recent agreement on referring to the Oxford Bodleian Library's Digby 23 manuscript as the definitive version is convenient for the purposes of scholarship but largely meaningless as regards cultural transmission of the legend. And of course, even after the Oxford MS became scholarly-canonical, the many translations of the Song within different temporal and cultural contexts have introduced variability to researchers' perceptions of the work.

One particularly telling observation Lord makes in his book is the utter certainty of Parry's research subjects that they were repeating the same story "word for word and line for line" regardless of how many times they told it, or how much time passed between retellings. Relying as they did on formulaic, on-the-spot composition, of course the singers were not singing exactly the same words. The reason for that discrepancy, Lord writes, is that the precision claimed in the quote above was "simply an emphatic way of saying 'like'... What is of importance here is not the fact of exactness or lack of exactness, but the constant emphasis by the singer on his role in the tradition" (27-8). Evidently, what these people valued in a [hi]story was not a precise sequence of events but conveying a moral lesson, or inspiring their listeners, or making them laugh – all the while enabling them to feel connected with their ancestors. Good storytelling was (and remains) nourishment and glue of society, story details – the spices, the personal touches that keep us coming back.

Some examples of such storytelling – archetypical plots and situation-specific details – can be found in Roland^HT. One such example appears in the Welsh Cân Rolant, excerpt titled "Battle Imminent." Ganelon's betrayal – a mainstay of Song of Roland retellings – is posited by the knight Oliver as a possibility. Contrary to his usual hostile attitude towards Ganelon in Cân Rolant and elsewhere, here Roland exclaims, "God forbid that I should suspect Gwenlwyd of faithlessness, and he a stepfather to me!" As the in-text note states, on this very rare occasion, Roland gives Ganelon the benefit of the doubt. It is no coincidence that this unusual turn of thought occurs in the Welsh work: of all the stories about the battle of Roncesvalles it emphasizes feudal and familial loyalty the most.

In the 1997 Hamlet on the Holodeck Janet Murray frames the recombinant, detail-oriented but largely detail-independent nature of oral storytelling in related terms: "At the highest level of organization, Lord's bardic singers assembled their thematic units into plots..., which exhibited both constant and variable elements"21 (193). The constant elements of Roland^HT's thematic units are the themes and imagery that bind it together; the variables are individual storylines and embellishments introduced by different cultures.22

Murray talks about "the computer as storyteller" (197). There is no storyteller in Roland^HT, or rather, there is not intended to be one. The user does construct her own story, but it is not "told" to her by a third party. The story is the interaction of the user with the corpus, and there is no external agent acting upon it after the encoding is done and interface built. Of course, the encoding and interface (which includes paratext written by me, as publisher of Roland^HT) is quite a lot of external agency and cannot be discounted, but it results in a potential for a story, not in a story as such. It lacks a plot, and even more so than famous hypertext fictions such as Jackson's 1995 Patchwork Girl (composed in StorySpace) and Ryman's 1996 253 (a web novel that also appeared in print) the corpus lacks an ending. In addition, there is no cohesive "writer's voice" (204) in Roland^HT, and neither the web version nor the raw XML is "procedural"23 (274-6) – both being central features of stories as Murray defines them, referring to both digital and analog works.

It is possible to imagine a future in which all art works that feature Roland are included in this corpus. But that future is distant enough that, for our purposes here, Roland^HT is an ever-unfinished polyphony.

In informal testing sessions users have expressed a desire to be able to "record" paths through the site that they find interesting. This feature is intended as part of the future development of the web interface. When it is constructed, the recorded paths will be complete stories within an amorphous whole lacking defined edges; and some of the users who compose these stories will go on to become our singers of tales.

What will Roland look like on the holodeck? Taking holodeck to mean any immersive virtual-reality environment used as a storytelling medium, it is easy to imagine environments for both procedural (story-based) and less structured ("jumping around") interaction with the character. Thus it is difficult to imagine the Roland corpus ever becoming a single, cohesive story. But it will remain the "mutable, kaleidoscopic world" (275) that Murray's stories create.