1) Semiotic and cognitive models

This chapter will inevitably seem somewhat disjointed in places. Subsequent chapters rely on a range of models and theories drawn from a variety of disparate disciplines. Even if it were possible to assume the reader’s familiarity with most of these topics, the diversity of interpretations in some areas of theory is such that it is always advisable to clarify what one means by potentially ambiguous terms before deploying them in earnest. That is the purpose of this chapter. Whilst chapter 2 develops various elements of theory that are specific to narratology proper, this opening chapter is intended to form the (hopefully stable) foundations on which that speculative edifice will rest.

1.a) making sense of images

Before discussing how narrative images ‘work’ it is necessary to step back a little and consider how images in general work. That is to say how a particular pattern of light and shade comes to evoke a particular meaning in the mind of the observer. It has long been recognised that images can serve as cognitive short-cuts; communicative signals which can be recognised or decoded more quickly than a written text. Hence, for example, the ubiquitous ideographic warning signs (instead of their textual equivalents) that line the roadsides of any developed country; their warnings, prohibitions and recommendations can be communicated to drivers far more efficiently as simplified images than would ever be possible by spelling out their meanings in words. The same semiotic economy is also found in medieval images of saints. If, in a church, I see an image of a man with a halo, holding a knife, I immediately assume that it represents St Bartholomew. If, in Paris, I see an image of a decapitated man carrying his head, I assume it to be St Denis, and so on. These acts of recognition are almost instantaneous and normally dominate other clues which may be provided - even if there is a caption or titulus identifying the saint, anyone familiar with the relevant iconography will normally identify the figure depicted by means of the visual attributes long before they get around to reading the textual labels.

A possible explanation for this communicative efficiency of images might be found in the field of ‘schema theory’ - one of the many offshoots of gestalt psychology that emerged in the 1920’s. In essence, schema theory asserts that ‘… all new experiences are understood by means of comparison to a stereotypical model, based on similar experiences and held in memory’.1

After remaining for some time as a largely theoretical construct, of interest only to cognitive psychologists, schema theory was given a major boost in the 1970’s by Marvin Minsky’s development of the ‘cognitive frame model’ as a practical tool for research into artificial intelligence (AI), particularly in the problematic area of machine vision.2

Minsky proposed that the human mind makes sense of the world around it through a very large (but finite) set of templates, or ‘frames’. Each of these frames is composed of a set of testable criteria by means of which the details of any perceived phenomenon can be evaluated against expectations, their ‘goodness of fit’ calculated and the most appropriate frame selected. For instance, the frame for {coffee table} and the frame for {cow} might both feature the test ‘has four legs’ but the provision of other tests (e.g. ‘produces milk’, ‘found in the living room’, ‘covered in old magazines’) makes it simple to distinguish between the two frames and select the most appropriate one.3 Frames are comprised of ‘nodes’, each made up of a set of related ‘slots’ (individually testable criteria), the number of levels in this hierarchy depending on the required granularity of the analysis. Figure 1 illustrates this model in a rudimentary fashion with a hypothetical breakdown of the cognitive frame for {Adoration of the Magi}.

Figure 1 - Simplified Cognitive Frame for 'Adoration of the Magi'

Note that the cognitive frame shown in Figure 1 is not specific to any particular depiction of the Epiphany. Nor indeed it is specific to images, being equally applicable for understanding the event in a painting, a prose description of the setting, a school nativity play, or even an eye-witness experience of the original scene. Note also that each slot here could easily be treated as a node or a sub-frame and decomposed into further testable criteria. For example the node for ‘ox’ might have the slots ‘has horns’, ‘four legs’, etc. Fully resolved frames (in which every slot has been decomposed into the simplest testable criterion) are complex affairs with multiple hierarchical levels of sub-frame combining into what Minsky called a ‘frame network’. However, the basic architecture of the model remains reassuringly simple. One particularly helpful aspect of Minsky’s cognitive frame model is the principle of ‘defaults’. If the value for a particular slot is not supplied or is ambiguous, the brain automatically supplies default values associated with that slot until further evidence is obtained. For example, when viewing a table from a position where only three legs are visible, we assume the presence of the fourth leg. Similarly, confronted by what is ‘obviously’ an Epiphany scene but with only two Magi visible, or without the ox and ass, we assume the missing elements were present but for some reason not shown. Thus the Epiphany scene shown in Fig. 8.09A, does not include the ox and ass, yet even without them the scene is instantly recognisable - and the knowledgeable viewer can subconsciously add them to the scene. This aspect of the model ties in well with Heidegger’s notion of the hermeneutic circle. As the brain attempts to fit its stock of cognitive frames to a scene, it can test expected features (default values) against the available evidence in its effort to elucidate elements which may not have been immediately apparent. Frame matching is thus an iterative process in which our expectations and perceptions interact, progressively homing in on the most appropriate frame. Finally, the model incorporates what has become known in the informational sciences as ‘fuzzy logic’; what matters for effective cognition is not an exact match of some particular cognitive frame to the observed scene on all criteria but simply a better ‘goodness of fit’ than any of the other available frames.

The main limitation with Minsky’s cognitive frame model was that it was designed primarily to deal with static situations, rather than with processes of change, and is not therefore an obvious choice for discussing narratives. The missing temporal aspect was addressed by two other pioneers of artificial intelligence, Roger Schank and Robert Abelson, who worked together at Yale in the 1970’s. Building upon Minsky’s AI work and on the theories of Canadian sociologist Erving Goffman, they developed the concept of the script as part of their efforts to resolve one of the major challenges of natural language processing namely the problem of disambiguating polysemic utterances based on context.4

The simplest definition of a script in this sense is ‘...a set of expectations about what will happen next in a well-understood situation’.5 As ‘senders’ of signals, we use these scripts to guide our own actions and utterances in different situations and as ‘receivers’ we use them to understand the actions and utterances of others. 6 Thus (to borrow the example most frequently used by Robert Schank) if I go to a restaurant, my familiarity with the {restaurant} script means that I already know how to behave (more or less) and do not need to think too hard about the correct order in which to remove my coat, sit down, read the menu, order food, etc. Equally importantly, the waiter and the other diners are all acting in accordance with their own {restaurant} scripts which (assuming it is a European restaurant, rather than, for example, a traditional Japanese one) will normally be sufficiently similar to my own and to each other’s to ensure a smooth and harmonious level of social, culinary and commercial interaction for all.

In the 1990’s, models based on cognitive frames and scripts were adopted by literary theorists, particularly Manfred Jahn and David Herman, as a way of explaining how readers make sense of stories, both in terms of static situations (using frames) and processes of change (using scripts). The result has become known as ‘cognitive narratology’ and is the approach which underlies much of this thesis.7 The important thing about these schema models is that the receiver will normally understand a given scenario using the same basic frames or scripts regardless of whether they experience it personally in real time or second-hand as a received narrative. Thus if one is reading a novel in which the hero goes to a restaurant, it is not necessary for the author to describe in minute detail every aspect of the decor and every action of the hero, the other diners, the waiters, etc - instead the author can concentrate on describing only those things which are unusual or significant about this particular instance of ‘restaurant’ and the reader can supply all the missing information themselves from their own copy of the standard {restaurant} script.

The attractions of these models for considering how narrative images are read are obvious; the individual panels of a stained glass window, a sculpted frieze or a manuscript illumination typically present simplified scenes, represented through a small number of figures, usually with one or two obvious props or gestures to indicate the nature of the situation. Such images, with their limited number of elements, can easily be resolved into cognitive frame networks. Furthermore, because of the topos-rich character of Christian narratives (especially hagiographies), a relatively small number of schemata is sufficient to understand a high proportion of the images that feature in medieval narrative art. For example, any image of one figure driving another figure towards a building with an open door can instantly be recognised using the cognitive frame {imprisonment} (in the windows at Chartres, this frame recurs no less than eight times - see Fig. 2.10), while a {decapitation} almost invariably denotes a saintly martyrdom.

Just as cognitive frames are invoked in the (re-)cognition of individual scenes, ‘scripts’ deal with the temporal and causal processes that relate these scenes to each other within a narrative. As such they are an essential part in the viewer’s gap-filling activity when explaining the visible differences between a pair of consecutive images. Again, the relatively predictable course of our narratives make the process straightforward. For example when presented with a scene of a person about to be decapitated, followed by an image of the same person holding his own head, the viewer unconsciously applies the {cephalophore} script to fill in the intervening actions, correctly inferring that the head has been cut off and that the victim has then picked up their head and is now carrying it somewhere (see for example the St Valerie Châsses in Fig. 8.03). The cultural specificity of frames and scripts can often be observed in the bemused expressions of modern tourists confronted by scenes of St Denis’ martyrdom on various Parisian churches. A man holding his head in his hands is commonly recognised as an expression of grief or pain (most cultures have this cognitive frame) - but when the head in question is no longer attached to a neck, the image is one which does not fit any of the frames available to those viewers unfamiliar with medieval hagiography. The result is usually puzzlement or amusement. By contrast, a contemporary audience would have been well-equipped with suitable cognitive templates since such legends were very popular at the time - indeed, one folklorist recorded over 130 cephalophore saints documented in French hagiographic literature. 8

From the point of view of the story-teller, whether verbal or visual, schemata are something of a double edged sword. On the one hand, they obviate the need to clog-up narratives with endless minutiae. On the other hand, they create the risk that readers/viewers will simply flick mentally through the familiar episodes with as little attentiveness as a bored bedesman recounting his rosary. It is perhaps in order to counter this tendency that authors and artists alike have a range of meta-communicative devices to ‘re-engage the recipient’; a point to which I will return shortly.

1.b) not dead, just implied; the author-function in Christian narratives

The death of the author, so famously trumpeted by Roland Barthes in his 1968 essay of the same name, is of little consequence to the student of medieval art. Barthes himself made clear that his argument was only relevant to those post-medieval, post-Reformation cultures which had ‘...discovered the prestige of the individual’.9 In contrast to the individualist texts addressed by Barthes, the narrative corpus that is the focus of this thesis is marked by an abnegation of individual authorship. Modern scholarship holds that most of the Old Testament was written over the course of a millennium, from around the twelfth to the second centuries before Christ, in a mixture of Hebrew and Aramaic, by a number of very different authors and redactors. The versions of those texts familiar to medieval Christian authors were based mainly on the Septuagint (a koine Greek redaction compiled in the third and second centuries BC, supposedly by a divinely-inspired committee of seventy-two elders assembled by King Ptolemy II), although from Jerome onwards, elements of other Hebrew textual versions (some closer to the Masoretic Text) also found their way into the Latin biblical cannon. 10 Similarly, the attribution of the four Gospels to the Evangelists must be tempered by philological and other evidence that dates the texts themselves to between 70 and 115 AD. Thus the biblical ‘authors’ are composites, unknown and unknowable. Indeed, since much of the Old Testament was considered to have been divinely inspired, and the Gospels purported to be historical eye-witness accounts, one could say that the ‘authors’ of these texts, as individual creative entities, were actively ‘denied’ - the absence of authorial invention being axiomatic to the value of the texts as holy scripture.

Whilst it is true that the authorship of many hagiographic texts can be attributed to specific historical figures, such as Sulpicius (re. St Martin) or Walter Daniel (re. St Aelred), this is also a genre which frowned on invention and which, for very good theological reasons (see chapter 2.h), treasured conformity to traditional forms. Furthermore, our ignorance of these authors as entities outside their surviving texts protects us from the preconditioning that comes from being, as Barthes expressed it, ‘tyrannically centred on the author’.11

Even though our ‘real authors’ are inaccessible to us, that is not to say that we have to abandon any interest in what Foucault called the ‘author-function’.12 Instead we can turn to the ‘implied author’. This curious figure emerged from the theories of the Chicago School in the early 1960’s, where the ‘Neo-Aristotelians’ were driven by a growing sense of unease over the positivist belief in authorial intention. ‘Intentionalists’ (from Goethe onwards) argued that there is one true meaning of a text - the one which the author intended, and that a literary work should be interpreted or judged in terms of that intention and how successfully it was actualised. They also believed that one can understand this authorial intention by means of a careful examination of the rest of the author’s oeuvre, aided by supporting biographical material. These two assumptions; that one could feasibly access the author’s true intention, and moreover that what an author produced was what they had intended to produce, were challenged by a number of scholars, particularly Wayne Booth.13 Yet the author-function was too useful a figure to dismiss entirely. What Booth proposed instead was a proxy figure - the ‘implied author’ - who can be imagined as the set of views/positions which can be reconstituted or construed by the reader, solely on the basis of what is presented within the text itself. Thus if the real, physical author creates the text, the implied author is born out of it.

If the birth of the implied author is clear and uncontroversial, his subsequent life has been anything but. Booth himself explained the concept in a rather enigmatic fashion, which left the way open for each subsequent theorist to impose their own interpretation. Many writers have argued strongly against the validity or usefulness of the implied author model, seeing it as either ideologically unsound (on the grounds that it might be used to excuse the ‘real’ author for any sexist or imperialist biases apparent in the text), or else too vague to be of practical value. In response to such attacks, Seymour Chatman took a typically pragmatic view. As he put it ‘...the question is not whether the implied author exists but what we get from positing such a question’. 14 As I hope to demonstrate, when it comes to medieval narrative art, what we get from positing the implied author may be sufficiently useful to justify its existence.

1.c) activating the recipient; reader-response models

The inevitable corollary to Barthes’ obituary for the author was the announcement of the birth of the reader (or more generally the recipient) as an active agent in the production of ‘the text’. From the late 1960’s onwards, various theorists, both in Europe (Hans Robert Jauss and Wolfgang Iser of the ‘Constanz School’) and in America (notably Stanley Fish and Jonathon Culler), were developing reader-centric models of literature and literary history. Although their approaches and objectives differed, all shared an emphasis on the role of the reader/viewer in (re)constructing meaning from the incomplete data supplied in the text, interpreted in conjunction with the recipient’s own experience, their foreknowledge of the genre and the cultural context in which the act of reading takes place. Although the details and associated terminology vary according to the individual author, such models are usually known collectively as ‘reception theory’ or ‘reader-response models’.15

Hans Georg Gadamer set the stage for reception theory with his concept of the ‘horizon’ of the observer - a viewpoint comprising the particular set of ‘prejudices’ (devoid of any negative connotations) and preconceptions that determine every act of interpretation and without which no understanding could ever be possible. Crucial to Gadamer’s theory was the recognition that every such horizon, whether that of the present observer or of the past writer, is ‘historically situated’. In the act of reading a text, the reader is made conscious of the gap between the historically-situated horizon of the text and their own. The process of understanding is marked by a ‘fusing’ of these horizons (Horizontverschmelzig). ¹⁶ Meaning therefore is not fixed but changes during (and in its turn changes) the process of reception. For example, when one looks at a medieval tapestry, one’s reception of it as an object evolves both in the short term (as shapes and colours resolve themselves into figures and scenes) and in the longer term (as one recognises those scenes and their relationships and adds them to one’s mental store of cognitive frames) - and that process of reception will itself change one’s horizon in relation to further objects of that type or period (one acquires the cognitive machinery to read medieval tapestries more efficiently) and more generally (one’s world-view is forever subtly nuanced by this newfound understanding of medieval tapestries).

A key feature of the ‘horizon of expectations’ (as Hans Robert Jauss, a former student of Gadamer’s, called it) is that any act of reading is unique to the reader and the moment but is also situated within a specific cultural context. Recipients who share a common social, educational or religious background will also share overlapping horizons, forming what Stanley Fish called ‘interpretive communities’.17 These overlapping horizons will almost certainly be very different from those of the modern observer - but they can, however imperfectly - be reconstructed by careful study. This does of course bring us into another hermeneutic circle; we can only reconstruct the original reader by means of the evidence available within the surviving text - which in turn we seek to interpret according to our understanding of that hypothesised reader. With this in mind, as a counterpart to the implied author within the reader-response model, Wolfgang Iser proposed an ‘implied reader’, equally embedded within and defined by the text. 18 Although modern theory often stresses the individuality of reader response, within a religious institution one may postulate various different communities of audience (local clergy, visiting clergy, pilgrims, townspeople, etc) who would each operate under their own set of shared horizons of expectation and familiarity with the stories depicted. This opens the possibility of considering reader-response models within the cathedral according to the conditions pertaining to each of these different audience groups, or interpretive communities (this prospect becomes even more attractive when one considers that these different communities would have had different patterns of physical access to the visual narratives presented in, on and around the cathedral).

Although it has not been as widely accepted as the implied author, the concept of the implied reader remains useful, particularly when one considers the issues of indeterminacy and gap-filling. It is a widely acknowledged paradox that what makes narratives interesting are the missing bits. Information the writer omits, the reader supplies (via an appropriate cognitive schema). This ‘gap-filling’ activity lies at the heart of all reader-oriented theories of narrative. As Roman Ingarden noted in the 1930’s, gaps and gap-filling are omnipresent elements of cognition. 19 Robert Holub explained Ingarden’s phenomenological view thus:

All objects [...] have an infinite number of determinants, and no act of cognition can take into account every determinant of any particular object. But while a real object must have a particular determinant - a real object cannot be merely coloured; it must have a particular colour - the objects in a literary work [...] must retain some degree of indeterminacy. [...] In theory then, each literary work, indeed, each represented object or aspect, contains an infinite number of indeterminate places.20

The job of the reader, albeit one which is normally performed subconsciously, is to deploy an array of socio-cultural knowledge and linguistic skills to fill in these gaps as required to construct a satisfying and meaningful narrative.

Gaps and gap-filling are just as relevant to images as they are to verbal narratives, even if the nature of the gaps is different. To return to Holub’s example, the particular colour or tone of an object will always be concretised in a picture. The indeterminacy is instead shifted to a range of other criteria, such as depth, scale, texture, or durativity, as was discussed at length by Götthold Lessing in his famous Laöcoon essay. 21 Such indeterminacy should not be seen (as it was by Lessing) as a weakness of visual narratives, since it is key to making the viewer more involved in the communicative process; ‘… the pictorial medium has problems with narrativity and requires a ‘reader’ who is much more active in (re-)constructing a narrative than would be necessary in verbal texts’. 22 From the Church’s viewpoint, far from being a weakness this is a real asset - images draw the recipients in to the story more effectively than verbal story-telling, as they strive to reconstruct the narrative. Narrative image cycles always expose a particular type of gap, which is the temporal gap between successive events/frames, and it is in the awareness of, and infilling of these gaps that the viewer experience of visual narratives is most obviously created. Because of the viewer’s greater role in the construction of meaning, images may be better suited for promoting exegetical thought than texts. We interrogate texts, looking for further information to fill in the gaps. Images however interrogate us, demanding a deeper search of our stock of cognitive frames and scripts as we try to make sense of them - and in doing so, flagging up the intertextual connections that are the basis of exegesis. Their wilful lack of determinacy requires our more active participation in the interpretative dialogue. This is the ‘agency’ of the narrative image.23

For any text or image to evoke a desired effect, the level and location of its gaps must be such that the intended recipients would be able to fill them in a satisfactory fashion. This then is where the implied reader becomes useful (even if this is not exactly the role Wolfgang Iser intended for him). We can posit the implied reader as the ideal audience for a particular work - a member of an interpretive community equipped with all the necessary cultural machinery to make good sense of it. Putting these elements together and adding an optional pairing of narrator and audience within the text, the transmission (or arguably, creation) of meaning has been summarised by Seymour Chatman in what he called the 'narrative communication situation': 24

Figure 2 - Chatman's schema for the 'narrative communication situation'

Whilst this relatively simple interrelationship may be sufficient for the kinds of situations that concern most text-oriented narratologists, the visual narratives and their sources that are the focus of this thesis have an altogether more complex lineage. As suggested earlier, any study of Christian narratives and their medieval audiences must address the problems of transmission. Even for the literati, most biblical texts were only available through a lengthy sequence of translations, transcriptions and recensions that led to the Vulgate editions - not just Jerome’s Vulgate, but also Alcuin’s Vulgate (c.800), Stephen Harding’s Vulgate (1130’s), the Parisian Vulgate (1230’s), etc. But of course this was just the beginning - the real complexity comes from what Gérard Genette would call the ‘epitexts’ - the vast agglomeration of paraphrases, commentaries, glosses and exegeses which engulfed the biblical texts from the days of the Early Fathers onwards.25 In a manner that has few modern parallels, the core Biblical texts and the epitexts supplied by authors like Bede, Strabo, Hrabanus Maurus, Gilbert la Porée and others, became quite literally bound together; by the twelfth century, clerics typically studied the individual books of the Bible from manuscripts with the standardised commentary of the Glossa ordinaria arranged around the margins of the page. These epitexts told them how to read the text and thereby made the book a potent tool for constructing interpretive communities.26 Just to complicate matters, verbal and visual epitexts interacted, feeding symbiotically off each other. One common example mentioned earlier is the motif of the ox and the ass, who appear in just about every depiction of the Nativity, but who are not mentioned in any of the Gospels. Instead, the exegetical link to Isaiah 1:3 (‘The ox knows his owner and the ass his master’s crib; but Israel does not know’) created a visual tradition which then fed back into the general perception of the Nativity scene, becoming, in the process, canonical.27

Most important of all however was the special nature of the medieval reader. To an extent that is often lost on modern audiences, the very nature of the medieval theological mindset was governed by what one could call a ‘will to intertextuality’. Scholars brought up in the tradition of four-fold exegesis were trained to see a text not as a closed entity but as a participant in a network of typological, allegorical, tropological and anagogical relationships, not only with other texts but also with the physical and transcendent worlds around them.28

Though only of peripheral interest to narratologists who study the Bible as a text in its own right, these epitextual filters presents a major challenge to the study of medieval visual narratives. How can one analyse the transition from textual narrative to image cycle when the source narrative may actually have been a cocktail of several different texts, written by different authors for different purposes over a millennium or more - and assembled into a virtual, or ‘implied text’ which also incorporated a range of artistic traditions - and which existed only in the mind of the artist or ecclesiastic (or more likely a combination of the two) who designed a particular narrative image programme? While an answer to that question will have to wait (perhaps sine die), it does offer a reprieve of sorts for the implied author. For he can now stand-in for this complex, composite figure whose stories appear in windows, carvings, manuscripts, etc. This biblical implied author is part divinely-inspired Hebraic scribe, part Bede, part Isidore of Seville, part Peter Comestor, etc - and part custodian of the visual traditions of the various interpretive communities to which he belongs. And since our medieval ‘artist’ is, generally, a composite made up of anonymous theologians deciding on the subject matter and its general arrangement, plus the anonymous masters and artisans who implement that scheme, an additional abstraction of agency in the shape of an ‘implied artist’ will do no harm at all. Putting all this together, the ‘source’ of the visual narratives we see in stained glass, manuscript illuminations or wherever, is therefore the implied author of an implied text that serves as the raw material for an implied artist, who shapes and implements it for the consumption of an implied viewer. Our biblical implied author may be just an abstraction but he is an abstraction which, I suspect, may yet prove useful.

The situation for hagiographic texts is no less complicated. If we posit a virtual vita - the underlying implied text or fabula (see chapter 2.a) of a particular saint’s life-story that is to be visualised - then the sources that will have fed into that text turn out to be remarkably diverse. There will of course be the various specific stories of that saint’s life, which may include (now lost) local oral traditions, individual written vitae, gestae and libelli and compilations like the ‘Golden Legend’ - but there are also the various antetypes, ultimately Christ Himself, of whom all saints were considered to be ‘copies’ (see chapter 2.h). Episodes from other saints’ lives were often paraphrased or borrowed almost verbatim in order to ‘flesh-out’ a new vita and often we find that historical and ahistoric figures were conflated, confused and invented to feed the demand for saintly narratives. In all cases, if there was a demand for a saint to have a written biography, then the chances are that there would be demand for that biography to be visualised in some way. When it came to turning a vita into a new narrative image cycle, our implied artist might simply adapt or copy similar scenes from the story of some unrelated saint. They might select which episodes in the story to depict according to which episodes were most familiar from earlier cycles, or even from popular drama. Thus the ‘implied text’ that is actually being illustrated in a hagiographic window is a complex composite of general and specific biographical data nuanced by local and more general visual traditions.

Yet another mode of authorship appears when we turn to secular narratives. The medieval romances were far better known to contemporary audiences through their appearance in oral and visual media than as fixed texts. As James Rushing pointed out in his analysis of Ywain imagery:

We read Chrétien de Troye’s Yvain or Hartmann von Aue’s Iwein [...] and develop the conviction, whether we think about it or not, that Ywain existed only in those now canonical texts. In the Middle Ages, on the other hand, Ywain was not a prisoner of the texts, in which relatively few people encountered him. He experienced many more adventures in the lands beyond them, in the realm of orality and in the realm of images. Many times the number of people who heard or read Chrétien or Hartmann must have known the story through all the varieties of second hand narration, hearsay, and conversation that we may call secondary orality.29

Even within a relatively coherent corpus such as Arthurian romance, characters, events and stories were eminently reusable and interchangeable, drawn as required from a rich stock of standard models. One story-teller’s Ywain may well be another’s Gawain or Galahad, while an image of a knight fighting a lion or a wild-man might be any of these three or none (hence the confusion and uncertainty one often finds in exhibit captions and catalogue entries for medieval artworks depicting Arthurian scenes).

If this section has seemed at times to be unnecessarily obscure, it will, I hope, have served to make the point that the art-historian’s oft-asked question, ‘on which text are these images based?’ is usually evidence of an oversimplified view of visual narrative. This is not to denigrate in any way the value, nor the phenomenal knowledge of biblical and exegetical scholarship underlying the iconographical studies of great art historians like Emile Mâle, Erwin Panofsky and Adolf Katzenellenbogen. There is always however a temptation to seize upon the first text one finds that mentions whatever feature one is looking for and then assume the artist must have known/used that text. This can then lead to a tendency to interpret visual features according to what one is expecting to see, based on that particular text, rather than what is actually there. The implied texts we see visualised in medieval narrative art are best thought of as exercises in mythopoeic bricolage. Whilst it may often be possible to postulate which texts were the main sources (and I won’t hesitate to do so myself when appropriate), it would be entirely wrong to think in terms of a simple mapping from a single text fragment to a single image. The past excesses of over-interpretation which, for example, led M. D. Anderson to reconstruct a ‘lost’ Whitsun play based on the roof bosses of Norwich Cathedral (on the grounds that artists couldn’t have imagined such scenes without first having seen them performed by actors) may now raise a smile from more worldly-wise art historians - yet our propensity to fall into such traps remains strong. 30 Some typical examples of this kind of over-interpretation will be explored further in chapter 4 when looking at the Charlemagne window at Chartres.

1.d) defamiliarisation - ‘readerly’ and ‘writerly’ texts

In a pioneering essay of 1917, Victor Shklovsky declared that ‘...art removes objects from the automatism of perception’. 31 His demand that the artist’s first responsibility is to defamiliarise their message (to make the act of reading more demanding and thereby to engage the reader/viewer by forcing them into a more active mode of reception), was strangely prescient, not just of the obfuscatory twists and turns that western art has taken since he wrote it, but also regarding the conclusions of various reader-response theorists in the latter part of the twentieth century. Any text, to engage and maintain its reader’s attention, must steer the narrow path between banality and incomprehensibility, between the obvious and the obscure, or (in Roman Ingarden’s terminology) between excessive gappiness and over-determination. Roland Barthes, in the opening chapter of S/Z, drew a distinction between ‘readerly’ and ‘writerly’ texts (textes lisibles / textes scriptibles). 32 The former, usually characterised by a simple, linear narrative with a single, transparent meaning, is essentially a closed text, optimised for the ‘automatism of perception’ so reviled by Shklovsky. It is the textual equivalent of the road-signs with which I opened this chapter. The writerly text, by contrast, is dense and polysemic, requiring an active reader to explore and ‘co-produce’ its plurality of meanings.

These distinctions (terminology notwithstanding) are familiar ones to art historians. For example, Abbot Suger’s imagery on the western portal at Saint Denis (as Lindy Grant pointed out) uses a language of signs which are generally scriptural, literal & easy to understand - the visual embodiment of the readerly text. 33 The visitor, having first encountered these bold, easy-to-understand narratives would then be led through the ancient nave to the darker, more contemplative, bejewelled spaces of the ambulatory. Here, in the semi-circular chevet, the imagery is altogether more ‘writerly’; offering allegorical and typological readings on several levels, and demanding the more active participation of an educated viewer. The contrast between readerly and writerly imagery at St Denis does not necessarily indicate a targeting of two distinct audiences or interpretive communities (lay pilgrims versus clergy, for example) but may indicate a deliberate construction of the church as a psychological ‘performance’ in which the viewer’s experience has a temporal as well as a spatial dimension; their intellectual and emotional involvement climaxing at the most sacred part of the church, by the patron saint’s shrine.

Defamiliarisation of the visual text was not only achieved through reference to obscure theological topics. The elaborate medallion patterns of early thirteenth century stained glass or the irregular groupings of narrative voussoirs depicting the Incarnation and Public Ministry of Christ on the south portal of the west façade of Le Mans Cathedral are deliberately writerly texts which challenge the viewer to find the correct path through them - but they also encourage multiple readings, particularly in the famous typological windows of Bourges, Sens, Chartres, etc. Such arrangements are not unlike the structure of a computer hypertext document, with its multiple pathways that require the reader’s involvement to create one out of the many possible textual ‘threadings’. Such texts, where the narrative sequence may differ with each reading, are classed by literary theorists as ‘ergodic’ and are usually limited to literary experiments.34 Arguably though, the experience of any viewer exploring the myriad narratives encrusting a major cathedral without a suitable cicerone is always ergodic since they have no choice but to create their own path through the labyrinth of images.

Even more dramatic effects of re-engagement could be achieved through the use of deliberate paradoxes within what was otherwise a straightforward narrative image cycle. Such strategies can, at times, seem like precursors to the methods used by Laurence Sterne to deconstruct the fledgling novel form in his famous ‘Tristram Shandy’; a book much loved by Shklovsky himself. 35 For medieval artists, there was more than one way to ‘...remove [narratives] from the automatism of perception’. This is however a major subject in its own right and one that will be explored at length below, in chapter 6.

1.e) resembling or representing; the semiosis of medieval images

Art is not a copy of the real world. One of the damn things is enough.36

In the Historia naturalis, Pliny tells how the two greatest painters of their day, Zeuxis of Heraclea and Parrhasius of Ephesus, competed to see who could paint the most convincing image. The story goes that when Zeuxis pulled back the curtains concealing his work, the painting thus revealed was so convincing that the birds flew down and tried to peck at the fictive fruit. Yet when the time came for Parrhasius to unveil his own work, it transpired that the curtains concealing his painting were themselves just a painted fiction. Zeuxis conceded defeat, confessing that while he had succeeded in deceiving the birds, Parrhasius had deceived even him. The significance of this story is not of course what it tells us about the relative abilities of semi-mythical Greek painters but the fact that for Pliny (and for the Renaissance authors who mimicked him), the sine qua non of the artist was to reproduce the visual appearance of the experienced world as convincingly as possible.

Since Pliny, there have been a great many attempts to codify a semiotics of art - to explain how art means. Unfortunately nearly all of these discourses have been founded either on this same model of ‘art as illusion’ that is exemplified by Pliny’s story - or else on one or other of the myriad modernist ideologies of aesthetics. This assumption that art merely seeks to imitate nature (regardless of whether that ‘nature’ is the observable world or some higher, or more abstracted form of it) has informed many lengthy debates over whether it is correct to ever consider art as a system of signs, akin to the linguistic model propounded by Ferdinand de Saussure.37

The argument against a semiotics of art has traditionally been that the image of a thing does not ‘signify’ that thing; it only resembles it (or aspires to). It is therefore in a different class of relationships than we find in the Saussurean model of linguistic systems, where (with the occasional onomatopoeic exception) the relation between signifier and signified is based on convention, not on resemblance. This is not the place to review those protracted arguments which, for all their intellectual interest, pay no heed to the peculiarities of medieval image systems. 38 For the moment let it suffice that I mean to treat medieval art as a semiotic system of sorts, partly because (with apologies to Hans Belting) I see the medieval image as being not so much ‘before art’ as ‘beyond art’ - but mainly because it is useful to do so. 39 In seeking a semiotic system that treats ‘art’ in a manner more appropriate to the needs of this thesis, one could do far worse than turn to the Polish semiotician Mieczysław Wallis, one of the few theorists to have specifically considered the sign characteristics of medieval art.40 Rather than starting from scratch, Wallis built upon the tripartite model of Charles Saunders Peirce, the ‘conventionalism’ of Ernst Gombrich and above all on the semiotic theories of the Tartu-Moscow School (particularly Yuri Lotman and Boris Uspensky). 41 In particular Wallis set out to clarify the distinction between ‘iconic signs’, ‘conventional signs’ and ‘symbols’ - an important task since as he himself observed, in this field ‘...almost every writer has a vocabulary of his own’.42 He also added to the general vocabulary the highly useful (but generally ignored) notion of the ‘semantic enclave’, of which more anon. Wallis’ definitions of the key terms, which I will use throughout this thesis, can be explained simply with a few simple maxims: 43

• A sign is ‘...a sensibly perceptible object, made or used by a certain sender in order that, owing to its peculiar properties, it might evoke in a certain receiver a though about an object other than itself, [i.e. its referent].’44
• Iconic signs evoke thoughts of objects other than themselves primarily on the basis of resemblance.
• Conventional signs evoke thoughts of objects other than themselves primarily on the basis of a custom or convention (this is the Saussurean linguistic concept of the sign).
• A drawn, painted, sculptured or photographed likeness of a person is an iconic sign of that person (and also an index of the artist since it has a causal relationship with their agency), while their name or their heraldic device is a conventional sign.
• An iconic sign can also function as a conventional sign, having a different referent. For example, a tiny image thus; <, is an iconic sign for a 3.5” floppy disk (which it resembles) but in the right context, such as on the tool-bar of a word-processor, is also a conventional sign for the ‘save’ function (even in a world where such disks have become largely obsolete). Similarly, in medieval images the likeness of a golden disk behind the heard of a person, functions as a conventional sign for ‘sanctity’ whilst a woman thus nimbed who holds a miniature tower in her hand is a combination of iconic signs functioning as a conventional sign representing St Barbara (this is of course the sign-relationship exercised in medieval art by all such ‘attributes’).
• A symbol is ‘... a sensually perceptible object ... which is able to evoke in a recipient a thought neither on the basis of resemblance ... nor on the basis of a custom or convention ... but on the basis of some analogy between it and the object. A lion for instance can evoke in us the thought of power or courage ... Therefore a lion can be for us the symbol of courage’.
• A symbol can be represented by means of either an iconic or a conventional sign. Pictures and sculptures do not present the symbols but the iconic signs of those symbols. A depiction of the Judgement of Solomon may be an iconic sign of that episode but also a symbol evoking ‘good judgement’, or more generally, ‘wisdom’.
• The relationship between symbols and their referents is polyvalent in both directions; many symbols could evoke the Virgin and most of those symbols (the pearl, the lily, etc) were also capable of evoking various other referents. Generally, other contextual information (what I will describe later as the ‘{interpretive} frame’ - see chapter 3) is required to disambiguate the symbolic relationship. Such relationships may be ‘natural’, based on metonymy, or purely conventional (and in the middle ages were often codified in detail in works like the Rationale Divinorum Officiorum of Durandus).
• A semantic enclave is a part of a work of art consisting of signs of a different kind or from another semiotic system than the signs forming the bulk of the work. The most important examples in medieval art are textual inscriptions and coats of arms within images.

This then is the semiotic model to which I will attempt to adhere through the rest of this thesis