They fell upon the last part of the baggage train and drove the men of the rear guard, who were protecting the troops in front, down into the valley below. In the skirmish that followed, they slaughtered every last one of those men. Once they had looted the baggage train, the Basques, under the cover of darkness, since night was then coming on, quickly dispersed in every direction. The Basques had the advantage in this skirmish because of the lightness of their weapons and the nature of the terrain, whereas the Franks were disadvantaged by the heaviness of their arms and the unevenness of the land. Eggihard, the overseer of the king's table, Anselm, the count of the palace, and Roland, the lord of the Breton March, along with many others died in that skirmish. But this deed could not be avenged at that time, because the enemy had so dispersed after the attack that there was no indication as to where they could be found. (Einhard 21-22)

Thus reads the only extant historical mention of a person named Roland in the service of King Charlemagne. Charlemagne's biographer Einhard writes that this "skirmish," which in reality was a massacre that badly shook Charlemagne's empire, took place in 778 A.D. in the Pyrenees.

Einhard does not mention the mountain village Roncesvalles in his Vita Karoli Magni. But that place is now firmly associated with Roland, thanks to an epic poem composed following the battle. The Song of Roland was not written down until three hundred years later, by which time the Basques had been transformed into Pagans and the ambush became a consequence of betrayal on the part of a presumed Charlemagne loyalist. Roland became Charlemagne's nephew, stepson to the traitor Ganelon, and the focal point of the story.

From then on, Roland has been constantly evolving. His character was appropriated – and re-molded in various ways – by every major European culture; and he has made his way into North America as well. Roland has become a cultural hypertext.1 It is with his unique fate as a multifaceted and often self-contradictory fictional character that the present dissertation is concerned.

Charlemagne has besieged Saragossa for seven years. The Pagan (Muslim) king Marsile, who rules over Saragossa, sends an ambassador to Charlemagne to beg for mercy. If Charlemagne goes home to Aix-la-Chapelle, Marsile promises to follow him and convert to Christianity. In truth, however, the Pagan has no intention of following through on his promise.

Charlemagne solicits advice from his Twelve Peers (the Carolingian legendary cycle equivalent of the Knights of the Round Table), among them his nephew Roland and Roland's stepfather Ganelon. The latter is sent as an ambassador to Marsile, with a tentatively affirmative answer. Ganelon is a coward, and resents being sent into a dangerous place at the suggestion of his stepson, whom he already despises for familial reasons.2 In Saragossa Ganelon convinces Marsile to ambush Charlemagne's rearguard (which, he is certain, will be captained by the proud and hot-headed Roland) while it is crossing the Pyrenees. Should the Twelve Peers be taken out, Ganelon says, Marsile will have nothing more to fear from the old and frail Charlemagne.

Marsile agrees, and Ganelon returns to camp. As predicted, Roland volunteers to command the rearguard, aided by his best companion Oliver. The journey begins; the rearguard is ambushed; the Franks are outnumbered twenty to one. They fight well but cannot withstand the onslaught. Oliver repeatedly urges Roland to sound his horn Oliphant and summon Charlemagne's army to their aid, but Roland is too proud to call for help until it is too late. Seeing that almost all 20,000 of France's best warriors have been killed, and himself mortally wounded, Roland finally sounds the horn and dies facing the enemy.

Charlemagne returns to the Pyrenees pass to find a gruesome scene of carnage. He grieves for his nephew and soldiers, and soon avenges them by pursuing and killing those of Marsile's army who escaped the battle alive. Marsile's wife surrenders and converts to Christianity. The Twelve Peers are brought back to France for a proper burial. Oliver's sister and Roland's betrothed Alde dies upon learning of Roland's death, and is buried with religious martyrs. Ganelon is judged a traitor and executed by being drawn and quartered. The archangel Gabriel calls Charlemagne to his next great deed in the name of Christianity.

Small heart had Carlon to journey and to fight;
"God!" says the King, "how weary is my life!"
He weeps, he plucks his flowing beard and white.

Here ends the geste Turoldus would recite. (Sayers 203)

Roland's faces are many. Twelfth-century Germans made him a martyr (Konrad); Renaissance Italians, a madly devoted, clinging lover (Boiardo and Ariosto); medieval Welsh Saxons, a single-minded warrior (Rejhon); modern Americans, a Civil War hero (Low), a roadie (Silverstein) and the ghost of a headless Thompson gunner (Zevon). He has been melodramatic comic relief in France (Fricker), a puppet in Sicily (De Felice), a comic book hero somewhere between Santa Barbara and Brazil (Amaya et al.). He is the subject of drawings, sculptures, musical works.

Roland is no longer a single character. He is not always "an epic hero," not always "mad," and certainly not preoccupied with battle at all times. Roland has long been a vehicle, a vessel carrying whatever it is filled with to its next socio-geographic destination. I will continue to refer to Roland as "Roland" and "he" – to mean this generic, all-encompassing, multi-contextualized entity.

I referred to him above as a cultural hypertext. The definition of the word hypertext, as well as its usefulness as a critical term, is still being hotly debated, four decades after the word was coined. Following are two definitions by Theodor Nelson on which I have based my approach to this ambiguous term. One is his 1963 definition of literature – "an ongoing system of interconnecting documents." (Literary Machines 2/9) The other is Nelson's definition of hypertext itself. It has mutated since 1965, but I would like to return to the original: "a body of written or pictorial material interconnected in such a complex way that it could not conveniently be presented or represented on paper" (Complex Information Processing 96).

The present study traces a character entity that is defined, in its multiplicity, by cultural transmission. Regardless of the medium of transmission, what matters are the connections and the specific cultural exigencies expressed in different treatments of the character. In his thousand-year travels, Roland is a notable example of Nelson's system.

At the same time, the Roland document system cannot be easily represented as a flat entity. A better visual representation of Rolandiana would be a jigsaw puzzle as MC Escher would conceive one. Individual cultural objects are the pieces of this puzzle, and the borders at which they connect are themes and imagery common to them. By themselves these pieces reveal a limited subset of their intercultural context. But put them together and a larger picture emerges – and yet we will be unable to explain all of the connections at once without hitting some walls. Because of this, it is easier to represent Roland in explicitly hypertextual form, using semantic encoding and hyperlinks, which are described in Appendix F.

In Roland^HT I combine Nelson's two definitions to study a group of texts that belong to a diverse set of cultural traditions and media.3 Rolandiana spans a millennium and includes literature (prose, poetry, epics), theatre (puppet and live), visual arts (sculpture, stained glass, drawings, comics) and music (opera, rock-n-roll).4 Yet this diverse lot is held together by a set of common characteristics, and in this it is the very definition of a corpus. In Roland^HT I propose that the works under consideration form a corpus. The principles according to which it is formed are as follows:

1. Each corpus object contains a character named Roland. Characters with similar characteristics but different names (such as, for example, Don Quixote) are excluded.

2. The Roland portrayed in each corpus object has personality traits traceable to one of three nexus points: medieval France, medieval Scandinavia, and Renaissance Italy (see Chapter 2, "An Ur-Roland?"). Such traits may include: exceptional battle skill; death on a battlefield; overconfidence/arrogance; engagement to his best friend's sister; possession of blessed or enchanted arms and other magical artifacts; madness following enchantment; knighthood; conception as a result of incest; moral virtue; impulsiveness. At least two connecting characteristics, plus recurrent themes/imagery, must be present in each corpus artifact.5

3. Roland's character must be used more substantively than as a passing mention. That is, he must be a primary or secondary character present at some time in the course of the text; or else he must be mentioned in a way that explicitly or implicitly highlights a socio-cultural issue recurrent in the corpus.

At the outset, this proposal was intuitive. It is easy to see certain thematic connections among certain Roland works: for example, the child conceived through incest in the Old Norse Karlamagnús Saga who goes on to become Charlemagne's right hand is clearly the same Roland as the French epic hero.6 Some ties, however, are not so easy to see. Following on the work I did for my Master's thesis, I have semantically encoded excerpts from corpus objects using XML (eXtensible Markup Language), to show recurrent themes and imagery as well as characters. I will discuss the encoding process in more detail in "Heuretics" below, as well as in Chapter 3.

Four problems with this formulation arise immediately. I will first summarize the problems, and then address them in order:

A. The argument seems circular: the corpus is defined by the encoding, but the encoding emerges from the intuitive statement that this set of objects is a corpus;
B. My use of the word corpus conforms to its theoretical definition but not to any of its prior use in practice;
C. In point 2 above, traceability is a notion of dubious objectivity; and
D. Point 3 above is ambiguous as well: when does the use of a character become "substantive"?

A. The main contribution of this thesis is a set of proposed boundaries to a Roland corpus, which to date has not been discussed as a corpus. Primarily this is achieved through a close comparative reading of the corpus objects. The semantic encoding serves two purposes: to systematically record small individual discoveries in a very large corpus; and to provide a structured dataset to make a bird's-eye view of the corpus possible. The same dataset can later be utilized for visualizations and analysis of new potential corpus objects, but the technical expertise and time necessary to accomplish these place them outside the scope of the present work.

XML is the tool that allows me to test a hypothesis. Roland^HT is an experiment, encouraged by the success of a similar experiment in 2001 (Zafrin) and the flexibility of approach provided by the institutional framework in which the experiment is performed: a Special Studies degree in Humanities Computing. For more on the role of experiment in literary-artistic research, see Chapter 3.

B. The notion of corpus is used in the study of languages (corpus linguistics), literature, and objects (architecture, archaeology). Definitions of corpus provided by several different dictionaries of English and of literary terms are variations of "a related 'body' of writings, usually sharing the same author or subject-matter" (Baldick 52). Curiously, it is almost never further defined in the context of specific studies. Instead corpus is taken for granted, and is found far more often in section headings than in discussion. When it is used in the text itself, the set of objects under consideration invariably has an author, time period and/or geographical area in common. None of these things can usefully delimit the Roland corpus; its unifying threads are semantic, more subjective than authorship or geographic origin attribution. However, some past uses of corpus would justify the use of the word to describe works about Roland.

Most often the contested word is used in corpus linguistics. One book in particular defines corpus as "a subset of an ETL [electronic text library], built according to explicit design criteria for a specific purpose" (Ghadessy et al. 179). This is the only non-dictionary definition I have found. Unlike dictionary definitions, which point to corpora as purely descriptive objects, the corpus linguistics definition implies the possibility of an idiosyncratic approach to corpus construction. Delimiting the boundaries of a Roland corpus is a subjective process; if we substitute "Western artistic production" for "ETL," Ghadessy et al.'s phrasing becomes relevant to the construction of the Roland corpus. The purpose of defining it is to shed light on intercultural transmission – specifically, on why its protagonist is compelling enough to have survived so unusually long while at the same time undergoing fundamental changes. However, the subjective nature of the definition process itself demands a willingness to alter design criteria as texts are encoded (see Appendix F and Chapter 3 for discussion of encoding and Document Type Definition construction).

In their study of how people understand the notion of region in Northern Greece, Lagopoulos and Boklund Lagopoulou describe corpus as "a flexible unit [that] can extend from a collection of sentences to a book by an individual author, to his complete works, to the variants of myth, or to the whole of a semiotic system of a particular period." Their own corpus being records of interviews and filled-out questionnaires, the authors turn to Greimas and Courtés' 1979 Sémiotique to distinguish between corpus and sample. It turns out that there isn't much difference unless one is working in corpus linguistics. "For [Greimas and Courtés]," they write, "the extension of the concept of corpus beyond linguistics leads to a corpus which is neither closed nor exhaustive but only representative, and to a corresponding model which is only hypothetical." Having accepted the limitations of their corpus, the authors look for a way to construct a reliable one. In Greimas and Courtés they find two such ways: "statistical sampling, and saturation of a model, i.e., beginning with an initial model constructed on the basis of an initial corpus, one arrives at a final model after a series of new applications" (75-6). This suggests a flexible corpus whose content evolves in the course of its analysis.

Likewise, in the beginning of her Infiltrating Culture: Power and Identity in Contemporary Women's Writing (1996), Mireille Rosello does not actually call "the juxtaposition of texts found in the following chapters" (18) a corpus in her discussion, only in the section heading ("A word on the corpus"). She highlights some of the similarities among her texts (common language, time period, writers' gender), but emphasizes the differences. "Even infiltration is not a unified category" (19), she writes, except implicitly and/or as a result of her analysis.

So it is with the Roland corpus. Its unifying subject may be a default connecting thread by virtue of his name, but since the actual connecting threads are recurrent themes and imagery, the way to ascertain a work's status as a corpus object (or not) is a close reading of it. In this case, micro-results of the close reading are recorded by semantic encoding. Thus, the corpus emerges in the course of analysis (see "Heuretics" below).

Why not use words that have been used more frequently to describe Roland, such as myth or legend? Both of these terms imply a coherency that the current collection of texts does not possess. It has branched out too much to be a single myth, and the legends that comprise it are also many. Similarly, the terms closest to corpus – oeuvre and canon – are limiting to the point of inaccuracy: the corpus has too many authors to be called oeuvre and too little institutionalized authority to be a canon.

C/D. In the analysis of art works, objectivity is an unreachable ideal that is nevertheless useful to strive for. To offset doubts regarding the approach, in addition to encoding recurrent thematic elements I have encoded comments into the text segments and multimedia files that draw attention to details that support the overall argument of corpus existence. These comments are viewable by clicking on the quill icons, which are visible wherever a comment is present. They are generally short – a sentence or two long – and are easiest to take in when presented in the context of the primary source(s) to which they pertain. None of them individually is sufficient to support my argument, but together they help weave the complex picture of the Roland corpus.

An examination of themes recurrent in the Roland corpus offers insight into the evolution of written literature itself, as well as the relationship between orality and literacy and between literature and other forms of artistic expression. This insight sheds light on our literary present a millennium after Roland's entry into the public consciousness, the dawn of an age of widely accessible electronic writing, when the processes of composition and transmission are changing again, and the borders between verbal and non-verbal arts are blurring more than ever before.

A related question is, why Roland in particular? He was quick to anger, and his refusal to admit that he cannot deal with a problem alone got him – and twenty thousand of Charlemagne's best men – killed. As presented in the epic, he cost his lord dearly. Yet he became France's national hero, and the very symbol of a Christian champion. Why, and how?

One possible explanation is that the Song of Roland was a medium through which the Franks dealt with the devastating consequences of the Basque ambush. Charlemagne's policy of conquest at the time required the repeated and sustained mobilization of forces. Historian Pierre Riché points out that the Basque ambush was only one of a series of crises that befell Charlemagne in 778. By the following year the king had begun "[taking] initiative" (Carolingians 93) to counteract the crises; the rapidity and (at least momentary) effectiveness of his action suggests the use of some sort of propaganda to mobilize his vast realm.

Another possibility, which may coexist with the above, is rooted in Roland's branding as Charlemagne's nephew. Einhard merely mentions him as captain of the Breton March; the Song of Roland proceeded to establish a blood tie between Roland and his sovereign. It did not stop there, either: as the Roland legend developed, more genealogical details were added. His mother, for example, was named Bertha. She had gotten pregnant out of wedlock by a duke named Milon, and the two were exiled by the king for their transgression. Bertha escaped to Italy, where Roland was born and they lived in poverty. Legend has it that, when he was seven years old, Roland attempted to steal food from Charlemagne's travel camp (Forrester). The king recognized the child, and then his mother; emotions ran high, Bertha was pardoned and returned to the family fold.

The above echoes multiple people in the real-life Carolingian dynasty. Three years after coming into power following their father's death, Charlemagne's younger brother Carloman died. The two had had a hostile relationship; Charlemagne annexed Carloman's lands and exiled his widow Gerberga and their two children. The latter fled to Italy, where they found refuge at the court of King Desiderius of Lombardy, who happened to be Charlemagne's father-in-law. By that time the Emperor-to-be was gaining confidence; he broke off his marriage to Desiderata and sent her back to her father. Three years later Charlemagne took Desiderius prisoner and became King of Franks and Lombards, bringing the new home of his sister-in-law into the fold of his growing empire (Riché, Carolingians 86-7).

In 780, Charles traveled with his [new] family to Rome, where his sons Pippin and Louis were anointed by the pope as kings of Italy and Aquitaine, respectively. "This step marked an important advance insofar as Pippin became not king of the Lombards but of Italy," writes Riché. "The boy... remained his father's delegate and acted accordingly" (Carolingians 99). The legend, starting with the Song, continually emphasizes that Roland was Charlemagne's administrative right hand, instrumental to expanding and retaining the empire. Here, he speaks to his mythical sword Durendal:

"On me he girt the, the noble Charlemayn.
With this I won him Anjou and all Bretayn,
With this I won him Poitou, and conquered Maine;
With this I won him Normandy's fair terrain,
And with it won Provence and Aquitaine,
And Lombardy and all the land Romayne,
Bavaria too, and the whole Flemish state,
And Burgundy and all Apulia gained;
Constantinople in the King's hand I laid;
In Saxony he speaks and is obeyed;
With this I won Scotland, [Ireland and Wales,]
And England, where he set up his domain;
What lands and countries I've conquered by its aid,
for Charles to keep whose beard is white as may!" (Sayers 140)

Roland's accumulated characteristics make him a composite reflection of a dynasty that was portrayed by both folk culture and official historians as an embodiment of Frankish successes, and of the Franks' unity as an ethnic group. Discussing the Royal Frankish Annals in her History and Memory in the Carolingian World, Rosamond McKitterick notes that the previous Merovingian dynasty is never mentioned. "Rather than thinking of the annal entries as year-by-year jottings," she writes, "they should be recognized as a skilfully constructed and highly selective portrayal of the careers of the Carolingian rulers whose fortune and success is identified with that of the Frankish people" (102). In his first century of fictional life during the period when the Annals were written, Roland became a catch-all Carolingian figure.

Troubadours were poets; jongleurs were performers. Troubadours' earliest extant songbooks (chansonniers) date from around 1100, exactly the same period during which the Song of Roland was first written down. This may be a coincidence, but it is more probable that vernacular writing flourished as a recording device for artistic production (as opposed to legal matters, for example) precisely around the turn of the twelfth century.

As F.R.P. Akehurst and Judith Davis say in A Handbook of the Troubadours, history and poetry were more or less the same thing in late-medieval France. Gestes were how history was passed on. Add to that the pre-print-culture understanding of what faithfulness to a text entailed – not a word-for-word reproduction but rather a rendition faithful to the spirit of the original – and we are left with the question: how to draw the line between a Song of Roland variant and a derivative work? For the purposes of this dissertation, what matters is not the precise location of this line but a work's status as a corpus object. Striking semantic differences that add to our understanding of Roland are included; features such as spelling variations are not.

We do not read linearly. Among several different types of movements of which our eyes are capable, one – saccade, "a brief, rapid movement of the eye from one position of rest to another," ("saccade") – physicaly prevents us from reading in a linear fashion. During this movement, the fastest the human body can produce using an external organ, (Judge) our eyes "jump" from one word to another and then frequently retrace backwards before jumping again. During the saccadic movement itself most visual information we perceive is not processed. This allows us to maintain an illusion of a smoothness to our surroundings, textual and otherwise, but physiologically remains a non-linear way of reading (Judge Fig. 2).

We do not think linearly, and do not have a linear memory. Specifically as regards reading, this issue has been addressed by Minnesota-based electronic writer and designer Rob Wittig, advisor to the Electronic Literature Organization and the driving force behind TANK20 literary studio. Wittig points out the obvious which we deny: we do not always, or even often, read deeply. His characterization of what is implied by a seminar question "Did you read the assignment?" is striking:

The work was read straight through, without interruptions from any other reading. The work was read in a state of existential detachment, not influenced by the events and emotions of everyday life. The work and the reader are constants; i.e. a given book will produce always produce the same experience when read by the same reader. (Wittig)

"Now we've wandered into the realm of pure fantasy," Wittig quips. Readers are distracted, they multitask. They eat, listen to music and even watch television while reading. As a result, memories of readings will necessarily be fragmented, and different from one reading session to the next. The associative and highly multilinear nature of memory (see Chapter 3, "Working Differently") enables us to generate new insights when we think about a subject more than once. This is a great boon to cognition, but does unequivocally emphasize that neither phenomenological perception nor the memory of an artifact is ever stable.

Finally, the body of works under consideration was not produced in a linear fashion. Roland's character developed in parallel in France, Italy, Scandinavia, the British Isles and elsewhere. The three nexus points (see "Roland^HT is a corpus" above) of which all other Roland works can be considered derivations, however distant, attest to this parallel development.

All of this suggests that, as a representation of a semantic hypertextual corpus that evolved in a multilinear fashion, it would be not only extremely difficult but incongruent with its nature to make an argument for Roland^HT in a linear medium such as a book. The electronic medium, on the other hand, allows me to record and express Roland's multilinearity without the spatial constraints of a book. In addition, the electronic Roland^HT enables variant visualizations of the same dataset, removing the necessity for a single sequential line of argument, and thus allowing a complex, multilinear argument to be made.

In his contribution to the 1995 Scholarly Editing: A Guide to Research, G. Thomas Tanselle defines texts as, "in a broad sense, ...the arrangements of elements in artifacts, and we can speak about the texts of vases and paintings as well as the texts of pieces of paper inscribed or imprinted with words or with musical or choreographic notation" (10). In Roland^HT I adopt this definition of text to apply to the content of each cultural artifact in the Roland corpus, regardless of original medium. Tanselle goes further to distinguish document (equivalent to what I refer to here as cultural artifact or cultural object) from work (the abstract conception of an artistic product, before it is instantiated in an object). Within the context of a scholarly editor's task, he writes:

In thinking about works in intangible media—works of literature, music, dance, cinema (the media of which are language, sound, movement, and light)—we must keep in mind the fundamental fact that the artifacts we work with cannot be the works themselves and thus that we must constantly distinguish the texts of documents from the texts of works. This distinction has been the basis of the discipline of textual criticism throughout its long history: the reason for questioning the texts of documents is that they are not the texts of works and that they may be faulty witnesses to those texts. (12)

In Roland^HT I have used Tanselle's classification, with the addition of superwork, a Library and Information Studies term encompassing all manifestations of a work in all media (Svenonius). Thus, the superwork of the Song of Roland includes works such as the French Chanson de Roland and the Middle English Cân Rolant. The document subset of the Chanson includes the Oxford Digby 23 manuscript housed at the Bodleian Library. The set of words that makes up the Chanson de Roland in Digby 23 is the text of that document.

What, then, would be the text of the work called Song of Roland? Tanselle admits that, "however much evidence survives, the production of the texts of works always involves critical judgment. It is in the nature of works in intangible media that the very constitution of those works is a matter of conjecture" (12). Roland^HT's current concern is with variant manifestations of the Roland character and what they may signal about the cultures that created them. The texts of works are thus less meaningful to this endeavor than the texts of documents, which (necessarily in this case, considering the oral origins of the corpus) contain more variations. Documents have an advantage over works here: they bear the potential for a composite Roland portrait that is more informative with regard to cultural transmission.

Because minor textual variations within the same document are extremely unlikely to have an effect on Roland's overall portrayal, I have chosen to use only one document to represent each work. In the encoding documents are thus designated with the more general <work> element.7 The text of each work (as opposed to the paratext – meta-information about the work's provenance, and informational comments about the work as a whole) is labeled as <excerpt>s, to stress that no work is reproduced in its entirety. This approach works with digitized images of tangible media, as well: even an image that encompasses the entirety of a sculpture contains only visual data recorded from a single vantage point, and no data at all pertaining to its non-visual aspects or its context.

As my main methodological approach to studying Roland, I am using heuretics. The term was coined by Gregory Ulmer in his 1994 Heuretics: The Logic of Invention. Ulmer writes that theory finds its way into the humanities by one of two paths: "critical interpretation" or "artistic experiment" (3). Heuretics is the latter. It is a heuristic approach to theory – a reading process that, instead of attempting to theorize "what might be the meaning of an existing work," guides, in Ulmer's words, "a generative experiment: based on a given theory, how might another text be composed?" (5)

My generative experiment began with the hypothesis that there exists a corpus of Roland works that is united by themes and imagery that recur in its constituent artifacts. The new text composed as a result of this experiment is manifest in two forms. One is the collection of semantically encoded excerpts from corpus objects, contained in the electronic file named rolandht.xml. The other text form is the website as a whole, which includes: the XML file; the Document Type Definition (DTD)8 that expresses the syntax rules according to which the XML file was constructed; transformation stylesheets and JavaScript functions that control how the XML data is displayed in the browser; images; and film and sound clips.

The contents of all of these files, including the DTD, were created iteratively. My approach to semantic encoding of cultural objects combines the relatively simple syntax of XML (see Appendix F) with allowing the encoding vocabulary – the English-language terms defining themes and imagery that constitute Roland^HT's semantic tagset – to develop in the process of encoding. These terms became the analytical descriptors for the corpus. Some were eventually discarded, usually because they occurred in insufficient quantity or were combined with other terms; others were added retroactively.

Roland^HT is the next object in the Roland corpus. By encoding idiosyncratically, creating new tags as necessary without waiting for a system to emerge before starting to encode, I propose another text. I use the largely non-hierarchical way in which we tend to think as a tool instead of the obstacle it tends to become when writing linearly. As salient features of the Roland corpus reveal themselves, the analytical framework arises as an emergent property instead of being imposed upon the primary sources. Looked at in this light I am using XML as a system for defining markup languages in which critical argument is made (for example, "here the author talks about death" as contrasted with a descriptor function: "this is a proper name"), and the encoded elements come as close to being empirical statements as is possible in literary-artistic criticism.

Associative thinking depends on associative memory, and in research this can be both a boon and a hindrance. The more associations we make and keep active on a given subject, the more likely we are to make new associations conform to what we already know. This is why, periodically, it is useful to share work with colleagues – to get a fresh perspective. Working on an informed autopilot, recording associations as my mind makes them, lessens this effect: having a record absolves me from the obligation to actively remember, which is a significant boon with a corpus as heterogeneous as Roland.

To achieve the near-empiricism described above, I may play, approaching primary source material with a hunch that there may be something valuable in it and rummaging around to see what I find. At the same time I come to the corpus armed with a critical framework that combines past criticism of Roland texts and, more generally, structuralism (more on which in Chapter 2). Digital humanities scholar Stephen Ramsay has advocated for this approach. During a panel on reconceiving text analysis that took place during the ALLC-ACH conference in Tübingen in 2002, he acknowledged: "It is, of course, possible to go to a literary text armed with a hypothesis, but we do better to go to it with a hunch borne of our collective musings – a sneaking suspicion that looking at it this way will turn up something interesting." Although jarring to a traditionalist ear, Ramsay's proposal merely brings to the fore what literary critics have always done: close reading of primary sources. Roland^HT's difference from prior approaches is the stark way in which it records and exposes thought processes, usually hidden in conference papers and articles that present the results of these thought processes. Thus Roland^HT further approaches the experimental paradigm: it can be repeated. Disagreement with what has been encoded, or how it has been encoded, may be expressed through a different – or, better yet, supplementary – encoding. So we will play, until Roland's manifold image is satisfactory.

Many excellent books have been written about Roland. Missing from them almost entirely is an awareness of the Roland corpus as defined in Roland^HT, although much of the work done in the field so far is intertextual. This dissertation aims to fill in the missing gap, and bring past scholarship to bear on a wider look at Roland's influence on the development of national and religious identities. The thematic vastness of the aims will lead far beyond the study of Roland's character and into problems of orality and literacy, genre and cultural transmission.

This part of the literature review includes only the critical work concerning the Roland legend and related history. Other topics, such as different literary-critical methods and new media theory, will be discussed in future chapters.

Because of its topical breadth, the literary overview cannot aim to be exhaustive. Below are the resources I have found useful for their critical perspectives and primary-source references; from them I drew the beginnings of the corpus. Excluded, although pertinent in the grand scheme of Roland's character, are close unitextual character studies and histories that build in more detail upon their preceding, more general counterparts where all that's needed for Roland^HT is an understanding of the sequence of historical events. Overall, I will emphasize that although the critical works described below have been helpful to this study, no collection of existing critical material adequately represents the heterogeneity of the corpus – or even, as stated above, acknowledges the existence of such.

Being the most famous French chanson de geste, the Song of Roland has received critical attention steadily throughout the last century and a half. Perhaps the most versatile and well-rounded is Eugene Vance's Reading the Song of Roland. Vance applies his signature multitude of approaches to the poem. He proposes that the Song is fundamentally about "human consistency and not change" (11, emphasis author's), that its formulaic language is a choice made to rally the troops, and that it is a tragedy no less moving or personal than the Iliad (68). These representative examples of Vance's critical approach support the hypothesis that Roland's legend survived for this long partly because it was composed as a way of dealing with the trauma of the real-life Roncesvalles battle.

Vance's reading is rounded out by his brief synthesis of past Roland criticism, which positions his own book as a mediating force between those of Joseph Bédier and Ramón Menéndez Pidal in the question of whether Roland originated from a single mind (Bédier) or a collection of creative forces (Menéndez Pidal). This question has largely been settled in favor of an oral tradition predating the earliest extant manuscript of the Song in Oxford.

Robert F. Cook performs a similar reading in The Sense of the Song of Roland. Alongside some interesting points made – for example, that its geographic situation in Saragossa does not matter, so much as it serves as a symbolic stand-in for the Other – Cook makes similar observations that seem to miss their mark. For instance, he claims that Charlemagne's name "is not a reference to a historical person but the invocation of a positive, civilizing force in the world." (3) A disputable claim: as I have referenced above, the Song of Roland is one of a cycle of Frankish-propaganda texts that specifically exhalt the Carolingian dynasty, of which Charlemagne is the brightest example.

Cook's work shines not in this first part but in the second, eponymously titled with the book. Cook explores characters' motivations by contraposing their words and deeds, and his elucidation of the poem's political ideology paints a nicely blended combination of feudalism and Christianity – a bold move in a subject where other critics have felt compelled to take one side or the other.

Other close readings and contextualizations of the Song were done by André Burger (Turold, poète de la fidélité), Cesare Segre (La tradizione della Chanson de Roland) and Ramón Menéndez Pidal (La Chanson de Roland et la tradition epique des Francs). Burger's work is mostly synthetic and repetitive of others', adding bits here and there – for example, he shows Turold (if such a poet did, in fact, exist) as a Norman. Segre's study is an impressively thorough comparison of the Chanson's different manuscripts, which unfortunately is not useful for Roland^HT's purposes, as its concern is limited to a single corpus object. Most valuable in Segre is the chapter on the transition from hagiographic to geste poetry, and on the differences between the two: hagiographic poems are presented as elargazioni ("monks offer to the public ignorant of Latin texts that allow for better participation in a celebration of a feast or a saint"), and chansons de geste as servizi ("the singer... impassions his audience in exchange for compensation") (84, translations mine).

Menéndez Pidal's greatest contribution is another synthesis of the criticism that went before him, illustrating its contexts and evolutionary paths. He arrives at a short history of Roland studies and proceeds to situate the Chanson in its cultural context using linguistic and literary analysis, and discussing at length the complexity introduced by variants.

The Song's literary and cultural circumstances are inseparable from its evolution, and to understand them I turned to André de Mandach's Naissance et dévéloppement de la chanson de geste en Europe. It is the most detailed history of three Roland corpus objects from medieval France: La geste de Charlemagne et de Roland, La Chronique de Turpin's Anglo-Norman text, and La chanson d'Aspremont. Much of the three volumes is devoted to manuscript versioning questions. Mandach pays particular attention to borrowings of the legend, particularly by English scribes, and points out an increased inattentiveness to transcription accuracy in 14th-century England. His maps detailing the diffusion of Turpinian texts and of the Roland legend, though not dated, do provide a good visualization of where the transmission hubs were, and which parts of the world saw which strains of the legend. (See particularly Mandach, map C1.) Mandach draws parallels between Turpin's Charlemagne and King Alphonse VI, hinting at a possible patron for the stories' diffusion.

Paul Zumthor's Toward a Medieval Poetics, a translation of his famous Essai de poétique médiévale, theorizes all of the major literary genres of the Franco-Italian Middle Ages – romance, novella, chanson de geste and others – analyzing their audiences, lexical styles and mutual influences. Zumthor remains fully grounded in the Romance world, so for Roland^HT's broad outlook his work is of limited usefulness. However, his description of the epic as a way to singer autonomy dovetails with Albert Lord's study of oral storytellers' social power in The Singer of Tales.

Intertextuality is at Roland^HT's heart, but some investigations of single corpus objects have provided information unavailable elsewhere. Barton Sholod's Charlemagne in Spain: The Cultural Legacy of Roncesvalles addresses the question of Roland's reception on the enemy side – in Spain. Sholod acquaints his reader with Charlemagne's influence on the Spanish religion and his significance as a symbol of some Crusades (in which he could not have participated). Sholod also highlights Roland's appearances in the Spanish epic production. Of particular note are the few pages that discuss a possible carnal sin committed by the histocial Charlemagne in his youth, which is not only (per Sholod) hinted at by histocial documents but (as I show elsewhere) also appears in the Scandinavian saga that describes Roland as a product of Charlemagne's incestuous relationship with his sister Gille. Such a detour from the Iberian peninsula is a striking departure from Cultural Legacy's main thrust; Sholod's offhand intertextual references in the text and notes have proven informative, and have made the work more useful.

The Italian epic cycle has been treated with particular interest by critics. Lewis Einstein's Luigi Pulci and the Morgante Maggiore is the most concise study of the hefty Morgante to date, which nevertheless touches upon all the major inter- and contextual points necessary to situate Morgante in its romantic-poetic milieu. Einstein dedicates a particularly attentive several pages to Pulci's linguistic register, initially far more serious than those of later Italian writers. He argues that Pulci was the first to elevate Carolingian poetry from folk to high-culture status. While this is debatable on a European scale, it certainly seems to be true in Italy.

Adriano Bozzoli's La chiave dell'Orlando Furioso attempts to answer the question of whether Orlando's madness was a given in Boiardo's epic poem Orlando Innamorato – of which the Furioso was a sequel – or invented entirely by Ariosto. Although studies such as this one are often interesting, and contribute much to our understanding of individual works, they are not (yet) essential to the present broad investigation of the Roland corpus.

Edoardo Sanguineti and Luca Ronconi's book documenting their theatrical production of Orlando Furioso is a unique and important record of the only Roland corpus object of its kind: a performance that took place on multiple stages simultaneously, with the audience free to wander among them (and inevitably miss some of the action). In an interview, Sanguineti discusses his adaptation of a clearly non-theatrical text to the stage, and the tensions – not least that of continuity – that emerged. The performance photographs that follow demonstrate the use of multiple stages and special-effects machines, as well as the placement of the (participating) audience with respect to the action. The script itself is followed by a series of short essays on Sanguineti's Furioso. Notable among them are Giuseppe Bartolucci's take on an emerging "new writing" which plays with simultaneity, and in this is hinted to be a precursor to hypertext. Giuliano Zingone's piece on polyphony in Sanguineti's language echoes the polyphony of the corpus itself.

Donald Hair analyzes Browning's "Childe Roland to the Dark Tower Came" in his Robert Browning's Language. Hair calls attention to the poem's most evocative points of connection with its tradition: Roland's horn as an instrument of survival (bodily and legendary); and his inner struggle, which may or may not apply to the original Song but is certainly a plausible reading. Most strikingly, Hair cites Browning's own qualification of the "Childe Roland" text as "one of Action in Character, rather than Character in Action" (103). This formulation is another way of saying that Browning thought of Roland as a meta-character, a container for processing larger philosophical questions than his direct actions. This meta-quality will be taken up by Stephen King in his Dark Tower series. Both Browning and King present to us versions – not complete, but in each case unique – of a Proppian ur-Roland.

Intertextual studies of Roland seem to gravitate toward two major areas: the reception of the French Roland elsewhere, and intertextuality in the Italian Roland. Of the latter, a good beginning is Maria Beer's Romanzi di cavalleria: Il Furioso e il romanzo del primo cinquecento. Beer's main thematic interest is the role of dreams, prophetic or symbolic, in the Furioso tradition and its descendants. She treats the Furioso in an intertextual context, but – refreshingly – not one with Orlando Innamorato. Instead, Beer draws attention to the links between the work and Ariosto's own Cinque Canti, as well as lesser-known works like La morte del Danese and La dragha de Orlando Furioso (the latter by F. Tromba da Gualdo). Beer's focus in the vast question of intertextual ties is on direct borrowing of plot and thematic elements, particularly on genealogical relationships. But her book's greatest gift is a thorough, clear account of all Italian printings of chilvalric romances between 1470 and 1600. She provides the most reliable data on the Roland corpus in the Renaissance Italy.

Supplementing Beer's study is Giuseppe Sangiardi's Boiardismo Aristotesco: Presenza e trattamento dell'Orlando Innamorato nel Furioso. Although limited in scope, the book does address the new chivalric romance genre as it developed through late 15th and early 16th centuries. Even that section fails to venture far from the Innamorato-Furioso-Morgante holy trinity of Italian Orlandiana. Sangiardi does, however, present a fairly thorough look at the "strategies of continuation" (27, translations mine) employed by Ariosto in order to ensure a cohesion with Boiardo. The author pays particular attention to "thematic-narrative connections" (37), including a dynastic theme, a multitude of characters (and the stability of usual-suspect protagonists), and "syntagmatic" references, "meaning the sutures through which is established a connection with the broken thematic strands of Boiardo's romance" (40). Sangiardi points out all of Ariosto's direct references to Boiardo's text.

One study of Roland's reception in Italy that stands out among others. F.T.A. Voigt's Roland-Orlando dans l'épopée française et italienne is a detailed discussion of Roland in Italy before Pulci. Voigt's work has unique bibliographic references, and is a well-paced story of Roland's transformation from tragic hero to a comic or even burlesque one.

The above intertextual treatments of Roland are crowned by a 1996 collection of essays edited by Karen Pratt and titled Roland and Charlemagne in Europe: Essays on the Reception and Transformation of a Legend. Each of the articles stands on its own and is again of a somewhat limited scope, but together they present a more well-rounded overview of Roland in the Romance and Germanic worlds than other critical works. Some articles supplement other books in this review, such as David Hook's "Roland in the Spanish Medieval Epic" complementing Sholod's Charlemagne in Spain. Others present new material: Pratt and Roger Middleton's "A Fragment of an Unknown 'Roland' Epic" introduces a new corpus object, and Mark Chinca discusses Pfaffe Konrad's German version of the Song in relation to its original and its patron.

Two excellent bibliographic sources are Susan Farrier's The Medieval Charlemagne Legend: An Annotated Bibliography and Joseph Duggan's A Guide to Studies on the Chanson de Roland. Farrier's hefty tome (over 600 pages) is truly a treasure cove, and at the same time reveals gaps present in the critical literature to date. This is most apparent in the section of the Roland chapter titled "Roland's birth," which cites origins in (and only in) the "Franco-Italian poetry, first half of the 14th century" (243). The list entirely neglects the most fascinating aspect of that theme, namely Roland's incestuous beginnings in the Norse Karlamagnús Saga. This motif is not mentioned anywhere else in the bibliography either, though the saga appears often. Farrier is not to blame: nothing has been written about this startling addition to the corpus by the Scandinavians either. Duggan's work is a much smaller one, with more focus and some references that do not appear in Farrier.

For a historical base I have leaned primarily on the work of Rosamond McKitterick and Pierre Riché. Riché's Carolingians and Daily Life in the World of Charlemagne provide insights into the formation of the great dynasty on one hand, and the life of most commoners on the other. McKitterick, on the other hand, is more interested in learning and the transmission of knowledge. Her Books, Scribes and Learning and History and Memory in the Carolingian World supplement Mary Carruthers' The Book of Memory with information specific to the Frankish kingdoms.

The critical works described in this overview have contributed invaluable guidance to the construction of Roland^HT. Their application in the present work is manifest mostly in the form of their influence upon my frame of reference during encoding. Few explicit references to these works occur in the electronic work (those that do are mostly clarifications of proper names, and occur in the in-text notes indicated by quill icons9). However, when put together, these materials have provided the foundation for the most important extemporaneous decisions of the encoding process: the choices of what is important to encode. This foundation consists of the following overarching topics:

• Historical framing: the impact of the Carolingian dynasty on the socio-political, religious, educational and artistic development of Europe, and the interplay between the monarchy and the feudal system in Europe;
• The role of emotion in artistic production and reception in the context of Rolandiana: rallying troops; art as an instrument of dealing with trauma; the epic hero's imperfections and their effect on audiences' ability to relate to the hero; and the function of performative arts as a unifying and empassioning influence upon their audiences;
• Specific paths of cultural transmission; and
• Interrelationships between the poetics of different artistic genres and media, and their influence on an intertextual approach to the study of cultural production.