I can't tell you what I want, but I'll know it when I see it: terminological disconnects in digital image reference.

In mediating between image collections and image information needs, visual resource professionals--stock image and footage experts, slide librarians, video and film archivists, museum librarians and curators, and so forth--have traditionally assisted users in defining and expressing their image needs and helped to match those needs to appropriate images from a variety of sources. These requests are often among the most challenging to satisfy due to the disconnect between the words that users employ in image requests, and the words (or visual attributes) that are used by systems for image retrieval. This article presents a description of the challenges in providing digital image reference and presents the results of a study exploring these challenges in digital image reference practice. Data for the project include terms extracted from image requests, terms extracted from image answers provided by digital reference librarians, and the associated terms used to index these images on the Web.

**********

Image retrieval is essentially an act of translation. Users' cognitive image needs must be translated into external descriptions or depictions to communicate these needs to other humans and to information systems. The primary difficulty in this stems from translating a visual information need into a verbal expression. An additional challenge lies in the diversity with which images may be described by both users and by organizers of collections. To add to the dilemma, many digitized images available via the Web are not accompanied by any textual descriptions of their content. The practice in some institutions has been to describe images at the collection level rather than the individual image level.

Among the many challenges encountered by digital image archives are the problems associated with helping users translate their visual information needs into appropriate terminology for input into information systems designed to retrieve textual information. The difficulty users often have in translating their image needs into verbal or written expressions is exemplified by the patron who states, "I can't tell you what I want, but I'll know it when I see it!" The following are actual requests for images received by the Virtual Reference Desk Project at Syracuse University:

* I have a project in identifying the attached picture. The only clue I have is that it's "astronomical."

* I need to locate pictures on the Web depicting religious persecution, for example, the inquisition, etc.

* I am searching for a graphic depicting the rates of development in adolescent--physical, cognitive, and social development graphed versus time.

* I need pictures like this one (attached).

* Do you know of any database that compiles images from front pages of major papers, like the NYT?

It is one thing to get users to describe the content of images sought, but it is quite another thing to match that expression to an existing image. Surrogates such as keywords, titles, captions, or cataloging records function not only as attributes against which a query may be matched, but also provide support for browsing, navigation, relevance judgment, and query reformulation.

Although most retrieval systems are text-based, images frequently have little or no accompanying textual information. Historically, the solution has been to develop taxonomies and thesauri that reflect the unique characteristics of a particular collection or clientele and assign terms from a controlled vocabulary to describe the subject matter of an image, the objects depicted within an image, or both. Collections and the …