Overview of the database (VILD)

Welcome to the Verbal Images in Literature Database (VILD). At the time of writing [30/11/23] VILD contains 94 manually pre-annotated verbal images from a variety of authors and genres. Each verbal image occupies a row of the database; each column adds relevant contextual information (i.e. the poem, play or novel hosting the image, its genre, author, the critic who noticed it) as well as stylistic information (e.g. the size of the image counted in syllables and words, its location within the hosting text, the presence of metaphor and deixis, and more). Stylistic data is either quantitative (discrete, as in the number of syllables; or continuous as in imageability ratings) or qualitative (nominal: as in the critical comments provided or the semantic domains assigned; or nominal and binary as the presence or absence of a characteristic).  

VILD is an offshoot of the larger project “Verbal to visual: the image in poetic discourse”, financially supported by the Research Council of Lithuania under grant Nr 09.3.3-LMT-K-712-19-0204 and which run from September 2020 to August 2022.

Overview of the parameters (columns)

Verbal images: text and context

• Verbal image: an imageable stretch of text (see Sections ‘What are verbal images’ and ‘How are verbal images’ below)
• Verbal image + co-text: the verbal image and surrounding text
• Verbal image (orig. if not in EN): a verbal image in languages other than English
• Author: the author of the verbal image, that is, who crafted it, e.g., Shakespeare or Whitman; in a few cases the author is anonymous or not relevant (e.g., as when the cliché ‘a broken heart’ is called an image)
• Text: the text hosting the image, e.g., Othello or In a station of the metro.
• Genre: has to do with the form and function of the text: poetry, drama, prose (fiction, essay, news)
• Line(s): if the text is in verse (poetry or drama), then the line(s) where the image occurs is/are indicated (e.g., ‘petals on a wet, black bough’, l. 2 = second line)
• Noted by: reference to the critic or scholar who noted and commented the verbal image. It is provided in author-date style, e.g., Frye (1957: 123)
• Full comment: the commentary by the critic or scholar, including the crucial word ‘image’ which was one of the conditions to select the example in the first place (see ‘How was the database compiled?’ more below)

Verbal images: stylistic information (textual characteristics)

• Sense(s) elicited: the key five external sense(s) likely to be elicited by the image (visual, aural, olfactory, haptic, gustatory). Most often the key sense is the visual (e.g., ‘a pearl on forehead white’), but additional senses are possible (e.g., ‘the banner at daybreak is flapping’ suggests the auditory sense on top of the visual through the semantics of the verb ‘to flap’).
• Perception flag (PF): words that explicitly signal that an act of perception is involved or about to occur (e.g., apparition, invisible, the eye, saw, resemble, in front of us)
• Metaphor: the image has a metaphorical mapping, that is, a conceptual structure that brings together and compares features from different domains. For example, in ‘the aggressive light / strikes’, the light is tacitly compared to a weapon. Following standard practice, conceptual metaphors are indicated in capital letters (LIGHT IS WEAPON for the aforementioned example)
• Deixis: words that have indexical value, that is, that point to something and are intrinsically linked to a context. For example, you, today, here, this.
• Foregrounding (position): images at the beginning or end of a text, or at the beginning or end of a stanza, are likely to stand out more and be remembered better.
• Foregrounding (devices): images that are written in parallelistic structure (e.g., ‘we have been able to rise above the brutes / we have often sunk to the level of the demons’), or use negation, or list of three (e.g., ‘her scrubbed and sour humble hands’, with three adjectives) or rhyme, are likely to stand out more and be remembered better.
• Size (syllables, words, characters, lines): the length of the verbal image counted in syllables, orthographic words, characters (space excluded) or the number of lines the image runs across if found in drama or poetry.
• IMAG rate of Head: the imageability rating of the most vivid word in the verbal image. The ratings are taken from The Glasgow Norms, a psycholinguistic database of ratings for 5500 words. Basically, each word was rated by participants on various dimensions, including imageability (the easiness to produce a mental image). In VILD, ratings are given both in full and in approximate format, rounded off at the first decimal.
• Thematic core: this is the semantic domain, that is, the topic, of the main word in the verbal image. For instance, in ‘these faces in the crowd’ the key word is ‘face’ and its semantic domain is that of the body, marked as B1. All labels (B1, L3, S2 and so on) are taken from the USAS semantic system tagset, which the reader can consult to elucidate their meanings.

Finally, nearly all these parameters (except for ‘Sense(s) elicited’, ‘Size’ and ‘Thematic core’) are likely to act as image boosters, enhancing the perceived vividness and memorability of the image.

What are verbal images?

Verbal images are stretches of text that have a higher than chance potential to trigger mental images and vivid sensations in readers. Verbal images construct sensorially rich fictional worlds whilst acting as a magnet for critics’ interpretive efforts. In other words, they elicit immersive/presence effects, and also pave the way to symbolic significance.

How are verbal images made?

Verbal images often coincide with descriptive phrases, narrative clauses, and image-metaphors, but I argue that they are an even more fundamental unit of literary meaning and interpretation. Verbal images can be formalized into a set of prototypical characteristics (Castiglione, in preparation): the most prominent are imageable vocabulary and metaphor, as well as an average length of 18 syllables or 13 words, which makes them compact enough to be accommodated in working memory for online processing. There are also recurrent ancillary characteristics, such as specific stylistic strategies (lists, parallelism, presence of deixis, topic shifts, negation) which can be argued to function as imagery boosters, i.e., they bring the fictional scene or entity vividly to the fore of readers’ imagination, thus functioning as foregrounding devices. Typical semantic domains are those related to the body, people, animals, plants, and the environment: in short, those related to perceptual experience rather than to abstract thought or social institutions.

Although no individual characteristic accounts for 100% of verbal images (that is, no characteristic alone is both necessary and sufficient condition for the classification), the occurrence of either an imageable word and/or a metaphor and/or a set of imagery boosters is both a necessary and sufficient condition for a stretch of text to be classified as a verbal image. For a more detailed description of each characteristic, refer back to the ‘Verbal images: stylistic information (textual characteristics)’ section.

How was the database compiled?

The verbal images have been manually collected from a range of academic books and articles, mostly in the field of literary criticism (see the reference list here). For a verbal image to be inserted in the database, a few conditions had to be met:

1. The scholar used the word ‘image(s)’ in his commentary; near-synonyms such as ‘scene’, ‘figure’ or ‘picture’ have been excluded.
2. The word ‘image(s)’ in the commentary explicitly refers to a specific stretch of text which is either reported in the co-text (immediately before or after the commentary) or is easily and unequivocally recoverable (e.g., through wordings such as ‘the last two lines of poem X construe an image…’). Therefore, loose or metaphorical uses (e.g., as in ‘the image of the woman in 18th century Britain’, or ‘he despised his self-image’) have been excluded.

Whilst condition 1 could or could have been automated through a corpus search, manual inspection was necessary to ensure that condition 2 was also met. I owe around 1/3 of the images to the invaluable help of my former student Ugnė Spečiūtė, who painstakingly went through hundreds of articles from various literary journals and found new occurrences that met both the above conditions. Overall, this procedure gives the database a strong intersubjective basis and minimizes those biases tied to the preferences of individual scholars. Crucially, I am not in the list of critics (that is, I have not added any image on my own) to prevent my own theorizing on imagery to affect the authenticity of the data.

What languages are represented in the database?

Mostly English; there are a few examples in French, Italian and Spanish, all with accompanying English translations.

What can I do with the database?

Users can find unique or related verbal image(s) by keying in specific parameters and keywords in the search box. For instance, say that you want to see if an author you are fond of is in the database: you key in his or her name (e.g., Shakespeare, Huxley) and the verbal images tagged with that name will be displayed. Or maybe you want to select only the verbal images found in poetry (or in fiction, or in drama): all you have to do is to either key in the genre or select it from the top-down menu to filter the results. Or perhaps you are interested in specific stylistic characteristics and want to find all the images that have a metaphor, are less than 10 words or more than 20 words long, display parallelism, or are about a specific topic: you can do the same by selecting all the relevant parameters to get a set of verbal images that share one or more characteristics. As the dataset is still small, I recommend you limit your search to one or two characteristics at a time.

Why would I want to do that?

Well, say you are:

• a writer, poet, artist or content creator and are looking for inspiration and wants to know how specific entities (animals, objects and so on) have been described by canonical authors;
• a teacher and you want your students to learn specific techniques or practice imagination or interpretation through customized sets of examples rather than on the static pre-determined lists from textbooks;
• a literary scholar or a (cognitive) linguist: the former will probably be interested in the symbolic potential of verbal images, that is, in how verbal images have been interpreted (see Full comment column); the latter will probably be more interested in the linguistic resources used to express or re-enact sensory experience;
• a psychologist or psycholinguist and you need to select stimuli in a controlled manner for experimental purposes; for instance, a verbal image can be reconceptualized as a ROI (Region of Interest) in psycholinguistics, and eye-tracking studies can investigate if these textual loci are dwelt upon for longer under various reading conditions;
• an IT developer or IT tester in the field of AI art generation, and you want to test your software by keying in linguistic prompts (the verbal images) that are highly picturable and much more well-thought than random descriptions;

Can I contribute to the database in the future?

Sure you can! The database is an open-ended project and a growing repository of vivid/imageable literary language in English and beyond. The more data (and the more accurate and complete the information in it) the better! As we know, patterns emerge and stabilize only when a lot of data is gathered.

How can I contribute to the database?

There are a few ways to contribute:

• By finding new occurrences of verbal images that meet both conditions outlined before (see the ‘How was the database compiled?’ section) and email them to me at davide.castiglione@flf.vu.lt (make sure you send the original paper in pdf format or photographic evidence of the example); I will inspect, annotate and upload the new examples;
• By assisting me with a compilation of a corpus of academic articles in literary criticism to find further examples of verbal images that meet the conditions listed before;
• By flagging inaccuracies or omissions in the database, and by suggesting improvements or integrations; in particular, verbal images still lack annotation for meter and syntax, and the help of a specialist in these areas would be very much welcomed.

More generally, feel free to contact me if you want to be involved in my research on verbal images more generally. Any help with recruiting participants for empirical studies, designing surveys and questionnaires, performing statistical analysis, and applying my model (including on languages other than EN) would be much appreciated; and I would also welcome the opportunity to co-author an article in case our research interests converge.

Can I contribute to the database with verbal images from other languages?

In principle, yes: verbal images are likely to be a universal of (literary) language and it is important that my model of imagery reflects as much variety as possible. However, as I cannot understand most languages, I will have to train you to annotate the verbal images, and you will also have to provide a reliable and accurate English translation. Therefore, you should be a linguist yourself or be willing to learn some linguistics.

I am a student or a passionate reader, not an academic researcher; can I still contribute?

Sure! As mentioned before, one former student of mine helped me to find more occurrences of verbal image. Don’t underestimate your potential.