Bridging the divide between image collections across campus (Tuesday p.m.) The 25 Cal State image collections have built a collective database, cf http://worldart.sjsu.edu and use Excel template for data input and then put the data in Embark. They use approximately 3500 keywords and a kiosk with low-resolution images. Online portfolios are created by faculty for courses but this has been used mostly by humanities non-art faculty. They are talking to the libraries about shared searching. They’ve had about 15.6 million hits in a couple years. * LUCI is the library of UC images. There is a Digital VR Task Force, with Laine Farley as chair. UC collectively has 11.3 million analog images and perhaps 35 million digital images, about 56% of which are licensed. SPIRO and UCAI are being prepared for addition to LUCI. The UC campuses split subscription and license fees. * Smith College is building a centralized image database for all disciplines, with collaboration giving more clout and a common look. They are discovering that a diverse audience needs additional cataloging. * The University of Michigan library is sharing a Saskia subscription with the art history department, but two campus VR collections are using different cataloging systems.
Preparing for shared cataloging (Wednesday a.m.) Overview by Mary Elings: metadata standards as tool for sharing; Jewitt needed rules to build universal catalog in 1850s so he wrote rules; LC/NUC led to OCLC and then RLIN; California Digital Library built on principle of common cataloging approach, accept shared cataloging with minimal modification, review local practices in light of collaborative guidelines. Needs growing out of VR projects to date: tools, manual, rules, devotion to sharing. CCO is more a guide than rulebook. Some of current tools need rules (or guidelines) for application. Issues: collection development has been sporadic, in support of curriculum; attributions change and buildings move; “authoritativeness” (community as central agency for authoritativeness rather than absolute authority); equality of rigor in cataloging; tool development with as much machine support as possible (both software and hardware); ownership of central agency. * Harvard Fine Arts did a study of 900 slides: only about 100 found clear matches in UCSD’s Roger; about 116 matched in ARTstor (compared 9 titles that indicated similarity and found 3 were different and 6 were same work with different titles). Issues: no central agency; need central data structure; image is important as metadata; central database principally for search and discovery. * UCAI issues: coherent view; master vs clustered records; distribution; Google images is de facto union catalog with easy availability, false positives, and inadequate images. UCAI services being discussed and developed: data conversion; clustering; merging; enhanced searching through vocabularies; impediments: data format incompatibilities, idiosyncratic cataloging, work/surrogate, absence of “mature” descriptive standards, inconsistency between collections. * Clustering by image would be wonderful. * Classification and labels not as important in digital world. * CDL moving to union catalog rather than federated searching (control! both content and data). * XML schema separate descriptive and administrative data; slides have smushed.
Project CLiMB: using computational linguistic techniques to harvest image descriptors (Thursday a.m.) Columbia project to build tool to harvest metadata from scholarly texts: significant words or concepts determined, text brought in raw or TEI, paragraphs broken into sections, terms highlighted and can be used in cataloging; not magic yet. Target Object ID or TOI is the hook into a collection: building project for Greene and Greene; deity for Chinese paper gods; artist/work for North Carolina Museum of Art. NCMA project worked best because text was a collection catalog with regularized text and no transliteration problem. Looking at using controlled vocabularies in combination with words found in scholarly texts. Using computational linguistic tools such as occurrence, relation to vocabulary, existence in index. cf. http://www.columbia.edu/cu/cria/climb
Issues in cataloging contemporary art images (Thursday a.m.) Are terms such as “installation” and “multimedia” so ubiquitous as to be meaningless? Terms are meaningful if they can be differentiated from other terms. * Many VR collections use “New forms” or similar bucket(s) for otherwise unclassifiable works. * Geography not as meaningful. Users tend to know artist by name rather than thinking of genre or nationality. Issues enumerated by Amy Lucker: creator is usually known and may be findable; date and title usually known; full media description helpful.
Countdown to ARTstor (Thursday p.m.) 14 institutions in fall beta, with 16 more in spring 2004. * Hunter College desires: want it ready; office access; offsite access; more training; want to integrate with PowerPoint and other presentation tools; concern about image quality. Hunter’s Embark records have been merged as test with ARTstor records. * Princeton: searching of ARTstor and Almagest (Princeton product) not federated; need to continue to build toward curricular support; de-duping (clustering); one-on-one training of faculty has been necessary and worked well; general awareness of images use and images available; building new relationships across campus; Carnegie set is great to have again!; data cleanup is needed. * UCSD professor is using GroveArt as art history 101 text, in combination with ARTstor images. A theater professor used images without any zooming or other fancy ARTstor tools. A Chinese culture class used images for PowerPoint lectures even though loss of resolution, with much use of tools. * Hopkins using MDID for presentation in 5 art history, 3 studio classes. Art librarian trained faculty and grad assistants on reference desk. Issues: how to integrate personal images; how to keep personal images within classroom setting when publishing is priority. * ARTstor is to be marketed through JSTOR which has more than 1900 subscribing libraries.
An embarrassment of riches: developing robust schemas and interfaces to provide meaningful access to online cultural heritage materials (Friday a.m.) University of Minnesota shared database for architecture and landscape architecture * Digital Emblemata at UIUC - http://images.library.uiuc.edu/projects/emblems - using CONTENTdm, Dublin Core, ICONCLASS, topos (actual theme of picture), theme (abstract or sentiment), page turning and other display tools, working with Herzog August Bibliothek in Wolffenbüttel - Open Emblem Portal and metadata harvesting http://media.library.uiuc.edu/projects/oebp * Shatford attributes of subject: biography, subject, exemplified, relational - David Summers and Real space extends art history, subject as objective rather than layers of cultural judgment. * Who supplies the words? Who has authority? What are the words we are capable of supplying? (VR catalogers deal with non-verbal works and therefore are compelled to work on verbalizing.) * “New forms” may be described by delivery system rather than format. * Corbis catalogers are instructed to catalog the of-ness and not the about-ness. Corbis has about 100K terms in its thesaurus, though has bought some collections with only keyword access. They do transaction analysis of all searches to find potential candidate terms.
Complexities of the built environment: cataloging images of architecture (Friday a.m.) What is the work? Plan may be work in its own right, especially in museum context. First step in CCO is determining work or image. Advice: put image in smallest bucket you can; accessionable object should get a bucket. * What kind of relations can you define? attribute, component, qualifier, and does nature of relationship need to be defined? Examples: Latrobe letter to Jefferson with proposal for UVa quad; pictures of Rotunda.
Yours, mine and ours: the future of the personal collection in the digital age (Saturday a.m.) Susan Jane Williams working on Excel tool for faculty to use in data input. * Lion Share at Penn State Tool to integrate personal images with institutional, with most personal images having much metadata. Underlying concept is “learning objects” such as image file, video clip, text. cf. http://www.libraries.psu.edu/vius * MDID now includes capability to integrate personal collections, with a DC editor that only needs a title. Searching across VR collection and personal collections will use what it can and ignore irrelevant fields. cf. http://cit.jmu.edu/mdidinfo/
Notes compiled by Sherman Clarke, NYU Libraries
5 April 2004