Iconographic Description in DMVI

One of the principal aims of the Database of Mid-Victorian wood-engraved Illustration has been to develop a method of allowing images to be searched by content, as well as by artist, author, and title. Such a scheme involves 'tagging' or 'marking-up' the content of images, i.e. attaching particular search terms to a particular illustration, according to what is depicted. The user of the database can then search according to these tags, either by typing a word into a search box, or by browsing the set of available terms. The iconographic mark-up of images is not intended to describe a picture in full, or to somehow capture its essence. It is designed to facilitate searching, and is intended to allow users to find what they are looking for quickly, as well as creating links between images which users can follow.

For this project, it was felt that pre-existing tools for iconographic mark-up would be too blunt for our purposes, and that a tailor-made system, which was designed for Victorian visual art, and for illustrations in particular, would give the best results. This meant building up our own set of keywords for describing illustrated scenes. This process can initially appear straightforward. If a dog appears in the picture, the word 'dog' can be added to the search terms, and users looking for dogs will arrive at that image. So far, so good, but there are issues which make iconographic mark-up a more problematic endeavour, both in theory and in practice, than it can first appear. The two issues that cause the most problems are the need for a hierarchical set of categories and the tagging of abstract concepts.

Faceted Categories

When working with a large corpus of material it is important that tagging by pictorial content is consistent and logical as far as possible. The best way to ensure consistency is to use an organised system of iconographic terms (or keywords). The cataloguer who looks at an image and decides on suitable keywords therefore needs a list from which to draw. This list can be added to, but any additions must follow the logic of the previous entries. A single term therefore needs to exist in a hierarchy of standardised keywords, which allow users to consistently locate illustrations matching their search terms. DMVI uses faceted metadata to organise the categories. Facets are mutually exclusive top-level categories by which things can be classified, each of which can contain a hierarchy of sub-categories.

To go back to the example of 'dog', one possible structure for the hierarchy could be: birds and animals > animals > dogs. The hierarchy could, however, be organised in a different way: animals > mammals > dogs, or perhaps animals > domestic animals > dogs. The choice of structure would depend on the requirements of users and on the material in the collection. 'Dog' itself could have subsets: 'collies', 'alsatians', 'pugs'; or 'sheepdogs', 'guard dogs', 'lapdogs'. The important point is to choose a system which allows users to navigate in a logical way, facilitates the adding of new terms, and makes sense for the images in a particular collection. If a database had dozens of images of dogs, or was particularly concerned with the visual representation of animals, then subsets would be necessary to break up a long list, but if only a few dogs appeared then the single category 'dog' would probably be enough.

Abstract Concepts

'Dog' is not a particularly abstract term, but it can still cause problems. An image of a pack of fox terriers running across a field has a very different meaning from an image of a spaniel sitting under a table. Such differences raise questions when designing a system of iconographic description. Is it necessary to make such a distinction when marking-up images, or is the interpretation of cultural difference best left to the viewer once they have found all the images of dogs? Is it necessary to attempt to record the position of the dog: whether it is standing, sitting, lying down, running, eating, etc.? What about its mood or emotions: is it angry, frightened, hungry, happy? These kinds of questions arise when dealing with an ostensibly simple noun like 'dog', and it is easy to imagine how much more difficult concepts of emotion and characteristics can be when dealing with human figures.

Other kinds of abstract terms arise. What if users wish to search using words like 'love' or 'death'? Fitting such terms into a hierarchical system is problematic in itself, but even more difficult is trying to ensure that such subjective ideas are deployed consistently across the whole collection of images. 'Death' could be used in several ways: when a dead body is present, when a coffin or grave appears, when a death-bed scene is depicted, for a courtroom scene in which a criminal is sentenced to hang, when characters are shown in mourning dress, or when a memento mori (such as an hourglass or a skull) is present. The response of DMVI to these kinds of questions has been that it is better to display too many images than too few, and it should be up to the user to make the final decision of relevance. We would use the tag 'death' in all of the above cases, and let the user decide which versions were appropriate for their interests.

Other Problems

Even ostensibly simple tags, such as the period or era depicted in an image, can involve acts of judgement. One way in which to 'date' the content of an illustration is by the clothes which are worn or by the objects that appear in the image. Some images may relate to an eighteenth-century text, but contain figures in Victorian costume. Some images will contain figures wearing costumes which do not precisely match any particular era; sometimes costumes from several different eras appear in the same image; sometimes an anachronistic object will appear. How does the electronic mark-up of illustrations deal with these inconsistencies? Do we give a date according to the costume worn and the furniture depicted? Or do we rely rather on information derived from the written text? In the case of DMVI, we have attempted to alleviate some of these problems by designing tags which are broad and fluid, to work alongside others which are more precise. For example, date keywords include Georgian and Victorian, as well as eighteenth or nineteenth century. We have also allowed images to be tagged with mutually exclusive categories: an image can be both Georgian and Victorian if the date is unclear.

To help keyword searching, the main search tags are backed up using associated search terms. These are words which relate to the main keywords, either as synonyms or as more generally associated terms. Associated search terms allow the user to enter spelling variants, plurals, and other connected words into the search engine and still achieve appropriate results. For example, the term 'buying and selling' is used in the iconographic hierarchy. Users entering either 'buying' or 'selling' in the keyword box will bring up these results. Each keyword also has a number of associated search terms. In the case of 'buying and selling', the associated terms are: transactions, transacting, conducting business, dealing, trade, trading, purchase, purchasing, vending, hawking, peddle, peddling, shopping, haggle, haggling, and renting.

Throughout the course of DMVI we have attempted to be as consistent and objective as possible in designing and applying the iconographic tags. There are no straightforward answers to the questions raised by such a process, and this project has evolved a variety of strategies depending on the circumstances. In spite of the desire for objectivity, creating search terms for visual art is an inevitably subjective process. In some cases it is necessary to form judgements or to make decisions which exclude other options and interpretations. Nonetheless, the results can be immensely useful for people studying Victorian culture, enabling them to find material which they would not have come across by other means. Furthermore, the connections and categorisations which are constructed in the process of iconographic description can themselves give rise to new ways of looking at Victorian art and literature.

See also the guide to Iconographic Browsing.