Chapter 3: Image
Table of Contents
This dissertation chapter is a full-color, interactive document, originally formatted in HTML. It contains interactive plots, hyperlinks, and other dynamic elements. If you are reading this in another document format (PDF, DOCX, or paper), you are missing these dynamic elements. Please visit dissertation.jonreeve.com for the canonical edition of this text, and github.com/JonathanReeve/dissertation for its source code. Annotations are welcome in the hypothes.is sidebar to the right.
Color and shape converge into the image. Having modeled color vision in Chapter 1, and object vision in Chapter 2, we now turn to their aggregate, the image, in order to further interrogate the literary historical trends posited by these previous chapters. The analyses in Chapter 1 find an overall increase in color terms, and hued visuality more generally, across the nineteenth and twentieth centuries. Those of Chapter 2 find a general increase in the presence of objects and shapes, as well. What could account for these trends? I have already discussed some generic considerations, in the genre analysis of Chapter 1, and in some of the lists provided in Chapter 2. But visuality is a complex phenomenon, which takes place across genres, so more investigation is necessary.
In this chapter, I explore three interrelated modes which I argue account for the increase in visuality in twentieth century British literature: a descriptive mode, an impressionist mode, and an imagist mode. Although the term mode is inadequate to describe what others in computational literary analysis have called, speaking of their empirical findings, a thread or a signal, I prefer it to genre and type, which imply that these categories are even more mutually exclusive than they are. These modes operate on progressively smaller scales: description tends to work on the level of the paragraph or sentence, impression in the sentence, and image in the phrase.
Literary criticism has largely been ignoring these modes, as I will show, or treating them loosely. Outside of a few continental monographs from the 1980s, critics have been quiet about description, a grossly under-theorized literary phenomenon. There is more written about literary impressionism, although its connections to description and image are under-discussed. Finally, imagery and imagisms have fallen out of fashion among critics since the ’80s. While it is not the purpose of this chapter to revive interest in the literary merits of Imagism, or to argue for a resurrection in the thought surrounding literary impressionism, I will argue that we can’t explain the rise in visuality in twentieth century literature without them. All three of these modes, I will argue, along with their relations, are crucial to understanding literary visuality around the turn of the century, and in particular, the surge in color and object that we see at the beginning of the twentieth century.
For the more familiar phenomena of impressionism and imagism, I will
complicate the discussions surrounding them, contextualizing them in
literary history, and pluralizing them to impressionisms and
imagisms. I am also using the -ism and -ist
suffixes loosely. Although there is a well-defined movement in early
twentieth Century poetry called Imagism which I will discuss at length,
I won’t be restricting my discussion to those eight or so
poets,Richard Aldington, H. D., John Gould Fletcher, F. S.
Flint, D. H. Lawrence, and Amy Lowell, are the writers that appear in
the anthologies the most, although the first, Des Imagistes,
featured Ezra Pound, the movement’s founder, along with James Joyce,
William Carlos Williams, and several others (Pound, Des
Imagistes; Some Imagist Poets, Some
Imagist Poets; Some Imagist Poets, Some
Imagist Poets; Some Imagist Poets, Some
but will consider a wider plurality of imagisms. Similarly, although there is a movement in nineteenth Century French painting called Impressionism, with its own membership rosters, I’ll deal with implicit, as well as explicit, impressionisms as they appear in literature. And while description doesn’t have an associated school, its effects on the historical literary visuality appear just as strongly.
Still, these -isms of the turn of the century, a period which Mary Ann Caws terms A Century of Isms, carry with them rules, manifestos, prescriptions, and proscriptions, and a rich body of autoexegetic critical writing, perhaps more so than any other artistic movements (Caws). While these critical works should not be considered the sole means of interpreting their movements, they nonetheless provide a useful means of understanding their philosophical underpinnings.
Since imagisms and impressionisms are much harder to identify computationally, the quantitative exploration of this chapter will be one of description detection, and analysis of diachronic trends in literary description. While the boundaries and the shapes of descriptions change over time, confounding any efforts to measurement, as I will explain later, I find that literary description generally decreases through the nineteenth and early twentieth centuries. This expands on the findings of the past two chapters, since it suggests, for one, that descriptive modes on their own account for less of the literary visuality than one might expect. Rather, I argue that this surge in visuality can be attributed to an increased interest in subjective impressions and image, as conveyed in text.
Literary description, the first mode of visuality I’ll treat in this chapter, has received relatively little critical attention, much of which is dismissive of it and its role in fiction. The remaining criticism came into fashion briefly, in the 1980s, before passing away, and is primarily relegated to French and German literary theory, bearing the mark of the continental philological tradition.
José Manuel Lopes, in Foregrounded Description in Prose Fiction, contends that critics have largely considered it “purely ornamental, redundant, or even irrelevant,” and that since the nineteenth century, they’ve considered descriptions an “‘extra’ or as a dispensable ornament of narration, if not as a cumbersome set of textual segments that, standing in the way of storytelling, function as mere ‘catalysts,’ that is, as a series of devices to interrupt or delay the sequential unfolding of plots” (Lopes 6).
Lopes refers to Roland Barthes’ famous essay of 1968, “The Reality Effect,” in which he attempts to explain the presence of a detail in a short story by Flaubert. In this story, we are told that “an old piano supported, under a barometer, a pyramidal heap of boxes and cartons” (quoted in Barthes and Howard 141). About details, Barthes claims, that “they seem to correspond to a kind of narrative luxury, lavish to the point of offering many ‘futile’ details and thereby increasing the cost of narrative information” (quoted in Barthes and Howard 141). This economy of words seems almost to defend a brutalist architecture against a bourgeois Gothic revival, as if they were mutually exclusive, and existed in a zero-sum relationship.
This view is shared by Georg Lukács, who in his much-discussed essay “Narrate or Describe?” of 1936, pits narration against description. Comparing Zola’s Nana and Tolstoy’s Anna Karenina, he prefers Tolstoy’s style in that, “in Zola the race is described from the standpoint of an observer; in Tolstoy it is narrated from the standpoint of a participant” (Lukács 111). Of Flaubert’s descriptions in Madame Bovary, he complains that “to the reader they seem undifferentiated, additional elements of the environment Flaubert is describing. They become dabs of colour in a painting which rises above a lifeless level only insofar as it is elevated to an ironic symbol of philistinism. The painting assumes an importance which does not arise out of the subjective importance of the events, to which it is scarcely related, but from the artifice in the formal stylization” (Lukács 115).
Gotthold Lessing, as well, in his Laocoon: or, The limits of Poetry and Painting, rejects description as a valid mode for poetry, and for all other literary writing, claiming that dynamic actions are the appropriate subjects for poetry, while static bodies are the appropriate subjects of painting: “those objects which are co-existent, or whose parts are co-existent, are called bodies; consequently bodies, with their visible properties, are the legitimate subjects of painting. Those things, on the contrary, which are consecutive, or whose parts are consecutive, are termed, generally speaking, actions. Actions are therefore the legitimate subjects of poetry” (Lessing 150–51).
Part of why description is so often misunderstood is that it’s so hard to extricate from narrative. Guido Isekenmeier argues that description is largely considered “doubly ontologically anchored,” meaning that it is “definable in terms of both the kind of phenomena in the fictional world that are descriptively constituted and the set of linguistic devices in the literary text that constitute description” (Isekenmeier 80). To demonstrate this doubly self-constituted nature of description, consider a paragraph from Madame Bovary, the novel Lukács is criticizing.
Towards four in the morning Charles, well wrapped up in his cloak, set off for Les Bertaux. Still drowsy with the warmth of slumber, he let himself be lulled by the peaceful trotting of his horse. Whenever it stopped of its own accord at one of those holes bordered with thorns that farmers dig along the edge of their ploughed land, Charles, waking with a start, would quickly remember the broken leg, and try to recall all the fractures that he knew. It was no longer raining; day was breaking, and, on the leafless branches of the apple trees, birds sat motionless, fluffing out their tiny feathers in the cold morning wind. The flat landscape extended as far as the eye could see, the clumps of trees round the farms making widely spaced splashes of dark purple on that vast grey surface which, at the horizon, merged with the dreary tones of the sky. Charles, from time to time, would open his eyes; then, his mind growing weary and sleep returning unbidden, he would fall into a kind of doze in which, his recent sensations mingling with his memories, he saw himself as double, at once a student and a married man, lying, as he had just been doing, in his bed, and walking across a surgical ward as in the past.Vers quatre heures du matin, Charles, bien enveloppé dans son manteau, se mit en route pour les Bertaux. Encore endormi par la chaleur du sommeil, il se laissait bercer au trot pacifique de sa bête. Quand elle s’arrêtait d’elle-même devant ces trous entourés d’épines que l’on creuse au bord des sillons, Charles se réveillant en sursaut, se rappelait vite la jambe cassée, et il tâchait de se remettre en mémoire toutes les fractures qu’il savait. La pluie ne tombait plus; le jour commençait à venir, et, sur les branches des pommiers sans feuilles, des oiseaux se tenaient immobiles, hérissant leurs petites plumes au vent froid du matin. La plate campagne s’étalait à perte de vue, et les bouquets d’arbres autour des fermes faisaient, à intervalles éloignés, des taches d’un violet noir sur cette grande surface grise, qui se perdait à l’horizon dans le ton morne du ciel. Charles, de temps à autre, ouvrait les yeux; puis, son esprit se fatiguant et le sommeil revenant de soi-même, bientôt il entrait dans une sorte d’assoupissement où, ses sensations récentes se confondant avec des souvenirs, lui-même se percevait double, à la fois étudiant et marié, couché dans son lit comme tout à l’heure, traversant une salle d’opérés comme autrefois.
(Flaubert et al. 13)
These six sentences begin and end with action: Charles setting off for Les Bertaux in the first sentence, and Charles imagining himself and his past in the last. The long descriptive passage then, in the middle sentences, is bookended by this action. The description is subordinated to, and supported by, its narrative. The details are not, as Lukács suggests, “undifferentiated, additional elements of the environment,” but inextricably narrative in their own right. They are essential to understanding Charles’s emotional state. The “gray,” “dreary” landscape mirrors Charles’s mood, and is caused by it. While there are “dabs of colour”—the “splashes of dark purple on that vast grey surface” is remarkably paintery—they hardly amount only to “an ironic symbol of philistinism,” as Lukács argues. Rather, these details express liminalities which support, and constitute, Charles’s feeling of doubleness. His internal state, between sleeping and waking, between the experience of the present and memories of the past, are mirrored in a rainy day without rain; the time between night and day; a winter scene with birds on the trees; and the color gray itself: neither black nor white.
To choose a more recent example, consider this strikingly similar passage from Mrs Dalloway (one I will use to train my neural model, below), which is even harder to parse with a narration/description binary:
She had the oddest sense of being herself invisible, unseen; unknown; there being no more marrying, no more having of children now, but only this astonishing and rather solemn progress with the rest of them, up Bond Street, this being Mrs. Dalloway; not even Clarissa any more; this being Mrs. Richard Dalloway.
Bond Street fascinated her; Bond street early in the morning in the season; its flags flying; its shows; no splash; no glitter; one roll of tweed in the shop where her father had bought his suits for twenty years; a few pearls; salmon on an iceblock.
“That is all,” she said, looking at the fishmonger’s. “That is all,” she repeated, pausing for a moment at the window of a glove shop where, before the War, you could buy almost perfect gloves. And her old Uncle William used to say a lady is known by her shoes and her gloves. (Woolf, Mrs. Dalloway )
Here, Clarissa, also out walking early in the morning, has been out walking, running errands, and daydreaming. Like Charles, she imagines herself doubly, both Clarissa and Mrs. Dalloway. There are details, but they are not unnecessary or unimportant: they are threads which lead to Clarissa’s memories. They not only tell us a bit about Clarissa’s history, but show how her mind works, how she daydreams, how details from the street make her remember her father.
“That is all,” her deictic comment to herself, is at once a continuation of “this being Mrs. Dalloway,” but is connected to the dead salmon, at which she is staring when she says it. It denotes a lack which is repeated in Bond Street’s “no splash; no glitter.” In contrast, the other things in this scene are overloaded: they are more than what they appear, since they contain also family memories: the roll of tweed is “where her father had bought his suits” and the glove shop reminds her of something her Uncle William used to say. The way that, according to her uncle, “a woman is known by” her gloves also reprises the motif of “being Mrs. Dalloway,” in which a woman is known by her husband’s name.
So, in most cases, description is never separate from narrative, and never its antagonist, but does its work, using image. One would never say of a film—any film that is narrative—that certain of its images or scenes are descriptive, and perform no narrative functions. And one would never say of a sufficiently narrative painting—a painting that is said to tell a story, such as Rembrandt’s The Night Watch—that its details are unnecessary, that a stray orange on the ground, or a dog curled up in the corner, had no purpose and shouldn’t exist. Yet fiction often comes under scrutiny for its use of images. Part of this may have to do with fiction’s inherent intermediality.
In Lessing, writing is always defined in opposition to painting, and this intermedial definition abounds in the thought surrounding description. In Description in Literature and Other Media, Werner Wolf argues that “description appears to be not only a transgeneric but also a transmedial phenomenon” (Wolf and Bernhart 1). Among his given evidence are a scene from Alice in Wonderland, in which the narrator gestures towards an illustration on a facing page, saying, “if you don’t know what a Gryphon is, look at the picture” (Wolf and Bernhart 3).
And there are plenty of other, less explicit, intermedialities, as well, not the least of which is the ekphrastic tradition, which claims as its progenitor the Homeric shield of Achilles, and which renders a visible art object. Another genre is the eighteenth century French monograph genre of the Description. Philippe Hamon’s monumental study of the description, Introduction à L’Analyse du Descriptif, begins with a history of this genre, which includes books with such titles as Description historique de la ville de Paris, of 1742, or Description des curiosités des églises de Paris, of 1759 (Hamon 11). At a time when mechanized reproductions images was considerably more expensive, a picture may have been worth much more, as the cliché goes, than a thousand words. But in another sense these books were readings, in the literary critical sense, of their objects, not unlike those which visual artists create of their models or subjects.
Hamon also cites the encyclopedic tradition of Diderot and others which, born of a rationalist impulse to interpret nature, finds description to be one of the primary tools of science, or “natural philosophy.” The power of this kind of description, in science and literary prose, is therefore much more than decorative: it’s evocative—it does what art is meant to do, which is to evoke, call forth, in its etymological sense, a reaction, an emotion, a thought, or a mental image.
The importance of description is also evident in pedagogical texts: writing and composition textbooks of the early twentieth century. Although we cannot assume that these instructional manuals were read and respected by the writers I’ve discussed here, is still likely that they, or others like them, would have been. At the very least, they form part of the zeitgeist surrounding description. And in these texts, we see many of these same types of assumptions about description: that it is separate from narration, that it is opposed to narration, that it is a methodical, sometimes scientific mode which is therefore tedious, and so on. In a 1902 American textbook of composition by Mary Rose and Arthur Beatty, Composition and Rhetoric Based on Literary Models, we see a section on “the relation between narration and description” which begins:
If we study the English novel historically we shall find that the early novelists massed their description, giving us sometimes two or three pages of it at once. These extended descriptions interrupt the story, which is our main interest, and become very tiresome. Later writers, realizing how prone we are to skip the descriptive passages when massed in this way, have broken up this element into shorter paragraphs, or even into sentences, and have scattered it throughout the book, so that it no longer retards the action. (Kavana and Beatty 145)
Is description really “tiresome,” as Rose and Beatty suggest, something that merely “retards the action”? Do readers simply skip descriptions? Or are they the very elements which lend vivacity to the literary work? The hard narration/description dichotomy of Lukács is still taken for granted here, but Rose and Beatty acknowledge that description may be interwoven into passages, or paragraphs, that are otherwise narrative. A 1903 textbook, Composition-rhetoric from literature, presents a similar dichotomy, but is less dismissive of description. The chapter on literary description begins, “we have found that description enters largely into literary narration,” suggesting that they are more intertwined than separate (Mooney 81).
Richard Green Moulton’s 1915 textbook, The Modern Study of Literature, gives a different typology, one in which description and “presentation,” along with poetry and prose, are four modes or “cardinal points” of literary form:
These four things, description and presentation, poetry and prose, are the four cardinal points of literary form. They are not to be conceived as four kinds of literature; but, like the cardinal points of the compass, they represent four necessary directions in which literary activity can move. Literature, developing from its starting-point in the ballad dance, finds its movement bounded in these four directions. The result of the movement so bounded gives us the six elements of literary form. (Moulton 17)
Fig. 1 shows Moulton’s cardinal points in diagram. This view enables description and narration (“presentation”) to coexist, as orthogonal vectors, as with poetry and prose: poetic prose and prosaic poetry mirror narrative description and descriptive narration.
These theorists, if they don’t reject description entirely, note that it is subordinated to narration. In some cases, they argue that it is increasingly subordinated. These are claims that deserve to be empirically tested.
Experiment: Quantifying Description
Does visuality, which I model in the previous chapters with its
expressions in colors and objects, correlate with description? To
measure this, I train a convolutional neural network to recognize
paragraphs containing literary descriptions. I begin with a corpus of
manually identified paragraphs of literary descriptions, from a range of
novels from the nineteenth and twentieth centuries, such as Charlotte
Brontë’s Jane Eyre, Thomas Hardy’s Tess of the
d’Urbervilles, and Virginia Woolf’s Mrs Dalloway,
annotated in the machine-readable markup format TEI XML. I then create a
corpus of non-descriptions, by randomly sampling paragraphs from these
novels not labeled as
descriptions.Although description certainly occurs on a
sub-paragraph level, the paragraph is a useful semantic unit. For an
in-depth look at the paragraph as a literary unit, see (Algee-Hewitt et al.)
Here is an example of a training passage I’ve manually labeled as description, from Tess of the d’Urbervilles:
This fertile and sheltered tract of country, in which the fields are never brown and the springs never dry, is bounded on the south by the bold chalk ridge that embraces the prominences of Hambledon Hill, Bulbarrow, Nettlecombe-Tout, Dogbury, High Stoy, and Bubb Down. The traveller from the coast, who, after plodding northward for a score of miles over calcareous downs and corn-lands, suddenly reaches the verge of one of these escarpments, is surprised and delighted to behold, extended like a map beneath him, a country differing absolutely from that which he has passed through. Behind him the hills are open, the sun blazes down upon fields so large as to give an unenclosed character to the landscape, the lanes are white, the hedges low and plashed, the atmosphere colourless. Here, in the valley, the world seems to be constructed upon a smaller and more delicate scale; the fields are mere paddocks, so reduced that from this height their hedgerows appear a network of dark green threads overspreading the paler green of the grass. The atmosphere beneath is languorous, and is so tinged with azure that what artists call the middle distance partakes also of that hue, while the horizon beyond is of the deepest ultramarine. Arable lands are few and limited; with but slight exceptions the prospect is a broad rich mass of grass and trees, mantling minor hills and dales within the major. Such is the Vale of Blackmoor. (Hardy 47)
This is a very traditional scene description, which seems almost like a landscape painting. It abounds in specificity, giving us a half-dozen place names. The imaginary “traveler from the coast” provides an excuse for a vista which is even more explicitly like a landscape painting, in which the countryside is “extended like a map beneath him.” Most noticeably, there are many color descriptors, here. The bright sun makes the lanes “white,” and the atmosphere “colourless.” The hedgerows are “dark green” over the “paler green” grass, and the valley is “azure.” The model will infer, based on descriptions like this, that a description may involve colors, light and darkness, and natural features such as grass, trees, hills, and “corn-lands.”
Using the Python libraries SpaCy and Prodigy, from Explosion AI, I train a multi-label text categorizer, with the labels “description” and “nondescription.” The model uses position-sensitive embeddings, and uses a convolutional neural network, based on an existing SpaCy language model of English. The training process holds back 20% of the training data, in order to evaluate its guesses, and iterates through about 5,000 epochs, until it achieves an 85% accuracy. The result is a model which can guess whether a paragraph of text is a description or not. Since it’s a multi-label model, it gives two probabilties for each paragraph it sees: the likelihood of it being a description paragraph, and the likelihood of it being a non-description paragraph.
The proportions of paragraphs labeled as descriptions, when viewed diachronically, are shown in fig. 2.