Textual: <Sorrow>

Textual: <Sorrow>

Example of a search:

User input:

Textual: <sorrow>



Object A is an ancient Chinese poem written in 100 B.C.E. and Object B is Expulsion from the Garden of Eden on the walls of the Brancacci Chapel in Florence. Object A was retrieved because of the association between the word “sad” and the word “sorrow” + the association between “lord” and “lightning” with the word “god.” If the cataloger were familiar with the poem, he or she might have associated it with “marriage” as well since the poem is actually about an unhappy marriage but even without that expertise, Object A would be retrieved. Object B was retrieved via both contextual and artistic tags. Contextual tags would have associated “Adam & Eve” with “marriage” and “God” and the story of expulsion with “sorrow”, and artistic tags would have associated Eve’s facial expression and Adam’s body language with “sorrow.”

There are many different perspectives on artistic creativity which span epistemological, artistic and social science venues. For my purposes here, let us adapt a theory about creativity that combines two generally acceptable theories: That artistic creativity is the amusing, serendipitous, and profound combination of ideas rooted in knowledge about extant art, artistic principles and their human contexts. That being established, an information system designed to inspire artistic creativity using extant art would need to retrieve relevant art objects using two layers of “tags,” one grounded in artistic principle, the other grounded in context. Descriptors and an Index language from which the catalogers may extract them would be needed to catalog the art objects within each layer.

The danger here is that the system could turn simply into a database of art objects, albeit an exhaustively cataloged one, but still just a database. This would particularly be a problem if the system was too literal, tagging a poem about betrayal with the word “betrayal” so that when a user types in betrayal, his search returns all instances of betrayal in art. This is nothing more than an index. Furthermore, it renders the artistic facet useless. One could theoretically derive an instance of “betrayal” from the techniques used in the art object but, this could be too subjective, and thus, inaccurate.

I believe a solution to this pitfall is an appropriate degree of abstraction via the use of what I will call idea modules. Borrowed from Arthur O. Lovejoy’s concept of unit-ideas, these idea modules would take common ideas portrayed in or by art (e.g. betrayal) and associate it with all kinds of descriptors (the same descriptors that will be used when cataloging the art objects), be it people, emotions, actions, etc. Take the example of “betrayal”; encapsulated within its idea module, one might find the emotions “jealousy”, or “sorrow.” One might also fine actions like “cheating” or “abandonment.” One may even find an association to “Benedict Arnold.” Therefore, in response to user input “infidelity,” the system would be required to associate it with the appropriate idea module(s), one of which would be “betrayal.” Then, all of the art associated (not with betrayal) but with betrayal’s descriptors, would be retrieved. Art objects dealing directly with betrayal would certainly be included in the result pool but so would the art objects with more nuanced instances of betrayal, or art objects that somehow convey the idea of betrayal within its composition.

How would this system work?

Textual input from the user would prompt the system to analyze the input, associate an idea module with it and then retrieve artistic object (text or graphics) from its bank which is determines are complementary to the user’s input. The process is similar to the creation of a mind map but the system’s job is to provide new ideas and associations to the user (both textual and graphical) in an attempt to elicit new and novel combinations of ideas for his or her use in artistic creation. The most challenging part of creating this system is the vast amount of cataloging that would be necessary to program the system to associate certain objects with certain words. Another challenge is finding a way to limit the pool of search results (because free associations could go on forever) without diminishing their quality. This will need to involve complex search parameters (the challenge of which is to limit the user’s difficulty in executing an accurate search.)

How would the information be stored?

There will need to be three separate sets of entity types. The first set is the entity types needed to describe the artistic principles and elements encompassed in the art object. The second set is the entity types needed to describe the context of the art object. The third is a set of entity types that the other two share, a common pool of descriptors with which the first two sets of entity types will be associated. Storing entity types separately will allow the system to easily exclude one or the other layer if they so choose. If a user wishes the system to only retrieve art objects that are contextually related to his input, then he can indicate such and the system can automatically devalue art objects with greater artistic than contextual associations.

The main relationship type is <isAssociatedwith>. This is the only relationship that catalogers will use but they will be able to indicate the degree of associated.