Museums and the Web Conference, March 16-19, 1997-Los Angeles
Senior Systems Developer
Office of Technology Initiatives
United States Holocaust Memorial Museum
100 Raoul Wallenberg Pl, SW
Washington, DC 20024
Coordinator for Imaging Projects
Digital Library Research Program
University of Illinois
Urbana, Illinois 61801
Organizing Information: Choosing and Using Database Structures
The Concept of Integrated Information Systems
Visualizing Museum Information in Terms of Data Structures
Metadata as a Solution for Working with Mixed Media
Standards Development for Descriptive Metadata
Textual Metadata Formats
The Museum Educational Site Licensing Project
Ever wish you could put your fingers on all of the information about a specific topic in a museum, regardless of whether it was drawn from the objects collection, exhibit catalogues, the library's holdings, prints and slides collection? Or your interest might even extend beyond a single department. With computerization and public access projects, museums are increasingly called upon to provide information drawn from a great deal of heterogeneous material. The advent of the World Wide Web has placed even more pressure on information holders.
Museum information has been gathered and described in disparate forms -- either because of the nature of the material or the nature of the data that can be gathered about museum collections. For a variety of reasons -- availability of software, variations in standards, the needs of collections -- heterogeneity can be expected to persist and in many cases, it should be encouraged. In spite of these variations of form, style, emphasis, and content, in reality museum information takes only a few forms -- text, imagery, and occasionally sound. Web publishing enables museums to provide access to this information through innovative means, ranging from exhibits created in hypermedia to samplers of collections, to searchable bodies of text that can be queried directly by the user, with the potential for obtaining results that closely match the searcher's information needs. Search engines that enable querying multiple information sources drawn from diverse formats are changing the nature and the expectations of the role of museum information systems.
This paper investigates fundamental approaches to constructing integrated museum information systems. A key element in the process of building these systems is the development of a thorough understanding of the data structures and formats within your organization. Also critical is the need to determine how data ought to be stored and shaped, and how a museum would like the data to be displayed, once it is retrieved. Practical examples are drawn from projects in which the authors have participated, including the Oregon Historical Society's Collections Access Project, sponsored by the U. S. Dept. of Education, and the Museum Educational Site Licensing Project, sponsored by the Getty Information Institute, and The United States Holocaust Memorial Museum. An overview is provided of current web and database technology that supports integrated systems development, and consideration is given to the ways in which these technologies match existing information access systems.
As information managers, museums differ from other kinds of organizations in the way that they accumulate in disparate forms and in disparate ways contextual materials that give meaning to the objects that they maintain. A history museum might have gathered records about individual objects with the usual descriptions and commentaries, it might at a later date have acquired the business papers of the company that made those objects, and a historian might have done research on an industry of which that company was a part for an exhibit and filed those papers in a vertical file of topics about the collections. An art museum might have sketches and information from sketchbooks and diaries about a painting that was no longer in public view, a file of artist's biographies created by students for a university art history class, and a collection of books about the artist and her circle. All of these resources - the objects, the research, the documents, and the records themselves - come to be part of the fabric of information that gives an object life in a museum. All of these resources shed light on the museum's collections - some more than others, some more independently than others - and while not all are of equal significance, none are discarded by the prudent museum information manager. This diversity makes managing museum information a more difficult proposition than keeping track of the stock of a stationery store or tracking overnight express deliveries or even developing a profile of a consumer's spending habits. The sources of material that museums draw on need not be congruent or even conceived as being related.
In most museums, the written information that describes collections takes many forms, may rest in several departments, and may have been gathered at different times for purposes. It is likely that different pieces of information gathered in different ways over time may overlap, and it is equally likely that some parts of existing descriptive information is unique. For example, staff at the Oregon Historical Society discovered that the Library, the Archives, and the Collections departments all possessed some descriptive information about particular companies in Oregon. As the departments discussed the types of information that they had about these entities, they realized that it would be useful for each department to be able to use the other's information because in parts of the organization there were gathered facts about the history of companies and in others, information about the output of those companies. The photo archives had photographs taken of and by them, and oral history included conversations with people related to them. In fact, in some cases, one department might have a very thoroughly documented narrative histories that would benefit all of the departments. If the organization were creating a shared database, to which all departments had access, there was no need for each department to maintain duplicate narratives on particular topics.
The Concept of Integrated Information Systems
There are several ways to facilitate the search and retrieval of information that is common to each of the internal information systems that an organization maintains. The first and simplest approach involves creating a high-level menu that lets users choose their database, format their own searches, and requires that they move sequentially from one to the next database, with no integration across systems. Typically, users can adapt to different database command structures. However, the question is not whether they can adapt, but rather whether they want to spend the time doing so. The less time a user needs to spend negotiating search commands, the more time that user will have to spend searching your collections. The second approach involves keeping the information in each of the stand-alone systems that are operating (collections database, library online catalog), and using a search engine to execute sequential searches across those systems, each time formulating a query using the search language that matches the particular system. The drawbacks of this approach include the difficulty of programming a search engine to perform a variety of search queries and to return the results in different formats, and the concerns over exposing internal production systems to potentially high volumes of public use and unwanted hacking. A third approach might involve an organization choosing one database and search engine to serve all of its needs, which would necessitate the translation of several data formats into one (using the lowest common denominator principle), thereby producing a leveling effect on all of the data produced by different departments. While some organizations may be at a point where they are ready to negotiate new contracts for each of their internal database systems, the likelihood of this occurring is rare, as is the probability that all departments might agree to using the same system, due to the specialized format and descriptive needs involved in working with many types of information (e.g., objects, manuscripts, photos). This paper advocates yet a fourth approach-that of the integrated system-a system that can be built to collect metadata from the various legacy systems within an organization to create a new database for public use. This new database is not maintained-the legacy systems still serve as production systems within the museum. Rather, it is a file that can be refreshed on a periodic basis, which can be searched in a variety of ways, through which a museum can establish links among similar types of information across departments, without having to re-structure its internal production systems. The ideal integrated information system would be one which retrieves directly from each production database within and organization, with a minimum of overhead-only the formatting and linking information between similar information across systems. While this ideal is not commercially available today, the realization of this goal is close, given current work on interoperability and semantic retrieval systems.1 The premise behind advocating this approach is that there are distinct reasons why the museum community shall maintain a variety of data formats and classification schemes-arguments that are deeply rooted in the need to preserve contextual validity and meaning, and the inability to shoulder the significant economic investment in large-scale conversion or re-creation of existing metadata.
Visualizing Museum Information in terms of Data Structures
In planning to bring disparate pieces of information together through one integrated system, data should not be considered simply in the way it relates to one specific purpose, like a collections catalogue. Data needs to be visualized and prepared so that it is in a format that can be used for many purposes. Materials from such museum activities as exhibits need to be considered so as to be integrated with other information in an organization. Similarly, information about collections that span multiple institutions needs to be incorporated into the mix of resources. Data ought to be organized from the start and considered as a whole in order to accommodate all of these possibilities in developing an integrated system, the digital counterpart to an institution, a side the public may most easily access apart from the physical collections.
Of course, that is much easier said than done. Museums accumulate, and for all kinds of very good reasons, the materials that they gather are not "organized from the start" or considered as wholes. Excluding for the moment that items within the collection are apt to be information resources for the collection (think of vertical files full of company catalogues or collections of architect's papers or the notes of a 19th century horticulturist sketching mushrooms in the wild), the information resources of a museum have developed very distinct ways of being recorded. Several common types can be noted just to serve as reminders of what the task of integration involves:
Not surprisingly, the formal information systems that museums have developed reflect these different approaches. Museum organization traditionally reflects the form that materials take - the departments of the U.S. Holocaust Memorial Museum include the Archives, the Photo Archives, Oral History, the Library, Film and Video, and Collections (which means all types of objects, often including archival, photographic, and library materials that is perceived as being significant as "objects.") Though other museums may be more or less strictly divided (the splitting apart of all things thinner than a dime except paintings from all things thicker except books is the common structure for small historical societies), divisions along the lines of object form are common. The fact of that practical and political division has done much to encourage "special" ways of recording information about each form. Hybrids exist - library management tools have been adapted to recording data about objects and there are a variety of home-grown approaches to dealing with all of those materials that don't fall clearly under the purview of librarians on the one hand and object curators on the other - but the fact of these hybrids only helps to point towards the fact of the differences.
To an extent, hierarchical approaches to thinking about museum information ranging from adapting the Dewey Decimal System to the uniform application of the Art and Architecture Thesaurus in all information systems have helped to bridge the divides that prevent different forms of materials from being integrated into a single system. For a time, the Margaret Woodbury Strong Museum used the hierarchical categories described in Nomenclature for Museum Cataloguing to delineate its departments rather than the more traditional forms that materials took, and the Henry Ford Museum has a long history of creative organizational approaches to collections, but such extreme experiments in integration have been rare in historical museums. Natural history collections are often more interesting models in this regard, and art museums where there may be less difference in the form of collections are often organized by topic - usually geographical, chronological, or stylistic similarity. These institutional structures and reliance on standard topical approaches suggest some models for integration that are interesting. Of course, sometimes the form that they take is no more complex than the use throughout a museum of the Library of Congress subject headings as subjects for books, as a data element in recording information about objects, and as a standard vocabulary to describe the content of photographs.
Similarly interesting have been the electronic tools (primarily) that have developed in recent years to establish relationships within single parts of organizations as well across departments. The relational model now familiar in database tools begins to make itself felt in museums as a consideration as the example of the Oregon Historical Society's realization that when its information holdings were viewed as a whole - apart from departmental divisions - there was a great deal of commonality. Of course, recognizing that the Rose Lumber Company about which the archives has prepared a company history for the finding aid organizing the company papers, for which the photo collection has views, from which the objects collection has artifacts and recorded commentary about the provenance of those artifacts, and so could be related in an organization and actually realizing such integration are two very different things. Beginning to recognize that links could exist is an important step forward.
Any system that is going to integrate the data that is managed by museums must take into account that there may be as many good reasons for keeping data that has been gathered about collections by form separate as there are for combining it, that there will be hierarchical constructs that may be more or less available and of greater or lesser importance to data managers, and that there may be many more potential relationships within museum data than have been identified or developed. In addition, because administrative structure so often mirrors data structure and compatibility, the problems of integration may take on political overtones not warranted by the data. And finally, because data accumulates in museums, even if the one perfect system could be found to accommodate all users so that integration could be assured as all data entered the institutional systems, the likelihood of working backwards through layers of existing materials is small. On that account, an alternative approach that involves generating metadata from any existing data sources has been followed with considerable success by the authors.
Images and non-textual Media
One of the most appealing aspects of multimedia technology is that it enables museums to extend the way in which they represent their collections by sharing still images, video, and audio with the public. Images and non-textual media can be used in a very flexible manner, but it is important to accommodate their richest use, and render these data once from analog to digital format, using a database (preferably the same one that you are developing for other applications). So, it is important to identify the materials to keep track internally of what you have captured, in what format and condition you captured it, what technical specifications were followed to accomplish the capture, and what type (if any) of changes were made to the digital objects to render them easily usable in your intended applications. For instance, you may capture a digital image of a photograph once by scanning it at a fairly high resolution, storing a master archival file in a lossless format. In order to render the image on the web, you will need to downsize it considerably, and probably change its format to a lossy format. In doing so, you need to remember to do two things:
Filenaming, format, and server storage schemes are necessary, and must be planned in advance. With some formats like TIFF (tagged image file format) it is possible to embed text tags in the image file as searchable identifiers. For example, full or fielded text, images (still and moving), or audio are usually held in multiple departments within an organization-the objects collection, the library, the archives and manuscripts collection, the prints and photographs collection. While there are similarities in handling text documents and multimedia documents, there is one major distinction: whereas text documents are enhanced by images, video, and sound, media objects are usually dependent on text to identify or describe them within the context of a multimedia presentation to users.
Although storage and delivery of images and multimedia are not linked thematically to metadata, working with multimedia requires careful planning for the amount and types of storage, and the type of computing and networking that will support the effective delivery of media along with text. Images, video, and audio require significantly more storage than text, and ideally they need as high a network bandwidth as possible for reasonable delivery speeds. The experience of those who work with Web and other networked multimedia applications have had to make trade-offs between high image fidelity and the limits of network bandwidth to deliver large files to the user desktop. Typically, images and video may be produced at a high archival quality, then reduced or compressed in size in order to accommodate effective delivery across a network. The types of derivative files that are created out of archival master files need to be somehow linked with the master file. Somewhere in the museum's record-keeping systems a definitive record of these copies ought to be kept.
The approach we find most useful for developing the integrated version of a museum's data resources has been to develop a very simple format for accumulating metadata - a text representation of any data component that is perceived as discrete - an object record, a finding aid, a book catalog card, an entry out of a catalogue, a photo caption, or a collection of specimens. By adding a few administrative fields beyond the text block as well as a way to identify elements from each text block that might be candidates for interpretation with hierarchical tools/approaches, e.g. an indication of an object's classification, and elements likely to be parts of formally defined links, e.g. personal names, and place names, the challenge of giving access to integrated data is greatly simplified. The issues become those of conversion and parsing from native data formats into the metadata container, of providing normalization for those elements that are hierarchical or linked, and of course implementing access to the metadata with a search engine capable of providing satisfactory results, allowing recursive or subsequent queries, generating links, and providing hierarchical access to related materials where appropriate. These are not trivial tasks, but they are greatly simplified by the existence of the single container.
Of course, nothing is quite so simple. Reducing all data to a single container assumes that there is something like equality among each thing that is the occasion for a record, that there are not vertical relationships between parts, and that data elements reflect things that are kind of comparable. For most "things" a museum keeps track of, these may not be issues. A user seeing a list of items like that noted in the previous paragraph - object, finding aid, book, photo - might be well satisfied with the results of a query against the metadata. Limiting the scope of a search to a particular media, allowing access to related items, offering links to other items would be matters of user's choice. But additional measures seem appropriate to reflect variance in the scale of what a single element of metadata represents. In the case of the Oregon Historical Society, this has taken the form of a relatively simple distinction of data elements that bear parent/child relationships to one another being reflected as such. This allows searches on an item level or on a collection level (or both) to be supported with the ability to move from items to collections and collections to items. In addition, it is clear that some pieces of metadata-especially those involved in hierarchic relationships or which reflect from hierarchic standards-serve more as encyclopedic comments on other data elements than comments on items in the collection. Accordingly, some elements have been designated as "encyclopedic," i.e., the description of what was meant by the term "chest-on-frame" or what was meant by "Pre-Raphaelite" or what constituted the "ponticum series" of rhododendrons - all pieces of data apt to be found in museum information systems, but rarely very accessible to a questioning public.
Below are examples of metadata from the Archives and the Photographs departments of the U. S. Holocaust Memorial Museum:
Exhibit 1 A search of the combined Library and Archives collections, reprinted with permission of the U. S. Holocaust Memorial Museum.
Exhibit 2 Archives record display in the Inquery system, reprinted with permission of the U. S. Holocaust Memorial Museum.
Exhibit 3 Photograph record display in the Inquery system, reprinted with permission of the U. S. Holocaust Memorial Museum.
There are essentially two ways to bring together different data formats and present them to the public as a cohesive collection-either ask everyone to conform (on some level) to one fixed data format, regardless of the type of information, or manipulate the data using innovative database construction techniques that are built into some software programs. The first approach involves normalizing existing data so that it is a consistent format. For example, a large international agency that held metadata in different formats for approximately twenty countries recently converted all of these data into one SGML format. The second, a simple approach - creating a single source for metadata - along with designating parent-child relations and an encyclopedic quality for some instances of that metadata lacks the sophistication of many more ambitious approaches to unifying data. But, considering that it is very simple to implement, it generates surprisingly satisfying results. An example of this approach would involve each department (photographs, objects, archives, library) first generating an output file of data in their native structures, although the complete and identical dataset would not need to be generated. Therefore, information in the collections database about acquisition and physical condition might not be included in the file of metadata that is generated for public searching and viewing.
If you want to share data, between departments, or with other institutions, standards become important. The diversity of approaches across several disciplines poses challenges to the cohesive development of the categories of descriptive metadata for the combination of visual and textual resources in the museum and archival communities. Within the museum community considerable diversity of descriptive work exists due to the uniqueness of collections and the approaches to catalog, organize, describe, and present museum collections. Although this section focuses on standards for describing museums' digital information, the discussion cannot be held in isolation from the considerable prior work that has been done with traditional approaches to creating metadata (cataloging, indexing), as well as established methods for organizing, describing, and classifying collections. It appears that one of the biggest challenges facing museums that already have extensive manually indexed collection systems is using that data effectively without having to re-create or re-format significant portions of it. Bell, in writing about the descriptive work of art curators, comments that while technology has made it feasible to exchange information about art collections, not every curator chooses the same terminology to describe works of art. 2 This observation is readily extended to all of the disciplines which are encompassed in museum collections-not everyone speaks a common language, even in cases where two museums contain objects and information on the same subjects.
Several controlled vocabularies and classification schemes have been used or recommended as standards for categorizing the content of original works, such as the Art and Architecture Thesaurus (AAT), the Library of Congress Subject Headings (LCSH), the Thesaurus of Graphic Materials, ICONCLASS, Categories for the Description of Works of Art (CDWA), and Nomenclature for Museum Cataloguing. The list of approaches to organizing and describing objects, media, and text is extensive, and it represents the myriad of perspectives on manual classification and description. The schemes vary depending on the focus of the organization and may not be adequate to serve the needs of all museums. It is understandably difficult to define areas of these schemes that have common meaning, given the diversity of museum collections. Work toward developing descriptive standards in this area, particularly in the arts, has been under way for some time. The Visual Resources Association (VRA) has developed a list of "core categories" for describing visual resources which covers information about the original object, its creator(s) and the surrogate digital image. 3 The VRA core categories were developed so that they reference the corresponding MARC cataloging fields, and so that they correspond to the Categories for the Description of Works of Art, developed by the VRA and a number of museum professional groups. The VRA core also takes into account the use of controlled vocabulary subject and name terms. The work of this group addresses an overarching framework of three general categories of image file description, including two categories that are used to describe the original object and who created it, and a category of information used to describe the surrogate or digital file for the object:
Object Type/ Techniques/ Materials/ Dimensions/
Titles/ Larger Entity Names/ Dates/ Subjects/
Repository Name/ Repository Number/ Notes
Image Type/ Image Owner/ Image Owner Number/ Source
OCLC (Online Computer Library Center), in collaboration with the CNI (Coalition for Networked Information), has sponsored a series of workshops over the past two years that have focused on the development of a core of elements that can be used as metadata to describe digital information. In 1995 a group of thirteen elements, labeled the "Dublin Core" was developed by this group. 4
|Subject:||The topic addressed by the work.|
|Title||The name of the object.|
|Author||The person(s) primarily responsible for the intellectual content of the object.|
|Publisher||The agent or agency responsible for making the object available in its current form.|
|Other Agent||The person(s), such as editors, transcribers, and illustrators who have made other significant intellectual contributions to the work.|
|Date||The date of publication.|
|Object type||The genre of the object, such as novel, poem or dictionary.|
|Form||The physical manifestation of the object, such as PostScript file or Windows executable file.|
|Identifier||String or number used to uniquely identify the object.|
|Relation||Relationship to other objects.|
|Source||Objects, either print or electronic, from which this object is derived, if applicable.|
|Language||Language of the intellectual content.|
|Coverage||The spatial location and/or temporal duration characteristics of the object|
Exhibit 4: The Dublin Core Elements 5
The Dublin Core represented the first attempt at developing a common group of elements that could be used consistently to describe networked information resources. This core was expanded upon earlier this year at a second metadata workshop held in Warwick, England. The goal of the Warwick conference was to develop a framework for deploying electronic resource description. Dempsey and Weibel note that the utility of the Dublin Core lies in its simplicity and its flexibility:
The Dublin Core is intended to fill the niche between the terseness of the unstructured full-text web indexes and the structured description of more complex models such as MARC. It is intended to be sufficiently rich to support useful fielded retrieval but simple enough not to require specialist expertise or extensive manual effort to create. 6
A significant outcome of the Warwick conference on metadata deployment was a proposed convention for embedding metadata in HTML (Hypertext Markup Language) .7 On September 23 and 24, OCLC and CNI hosted a workshop on metadata for networked images. The goal of this conference was " to promote convergence among alternative approaches to describing images and image databases in networked environments." 8 The outcome of the conference was a shared realization that images and text documents could be described in similar ways, with a recommendation for slight revision of the existing Dublin Core. 9
Metadata can take on one of two formats-structured or unstructured. Examples of structured metadata can be found in many of the collections management systems that were built in the 1970's and 1980's. The MARC cataloging record consists of structured metadata, with specific fields, set field delimiters and other tags. The previous section reviewed standards for working with structured metadata. However, there is an increasing need to be able to work effectively with unstructured information. Examples of unstructured metadata include full-text documents like collection catalogs, journal articles, telephone directories, or training handbooks. HTML (Hypertext Markup Language) represents one set of standards that accommodate text formatting for delivery and display across the Web, such as whether the title is bold or italicized, where the line and paragraph breaks occur, or the beginning and end of a document. HTML is limited, however in its inability to convey the content and meaning of the elements within a full text document. SGML (Standard Generalized Markup Language) represents another set of standards that support the identification of content elements in full text, such as the author, title, or subject of a work. Museums, libraries, publishers, and archives have developed SGML DTD's that can accommodate the kinds of different content of the information that these institutions handle. The work of the AITF (Art Information Task Force) and other groups has led to the development of an SGML DTD for museum information. The results of project CHIO (Cultural Heritage Information Online) will prove interesting since they will prove the feasibility of the use of the CIMI DTD as well as the use of a Z39.50 client to enable a search across the participating institutions who have provided metadata about their collections using this particular DTD. 10 Similarly, work in the archival field has resulted in a DTD for creating SGML marked up finding aids. 11
Once the desired museum information is identified, the next step in creating an integrated system is to determine the structure of the database in which that information will be stored. The purpose of constructing a database with such great care is to ensure that all of the important relationships are firmly established within and between data elements and record types at the outset, so that these relationships can be fully exploited as the database grows. Worthington and Robinson emphasize the role of the database in their definition-- "The necessary feature which characterizes a database is its capacity not merely to store data items but also to record meaningful bindings or relationships among those items and to support multiple perspectives on both the data and their bindings." 12 Worthington and Robinson identify essentially two fundamental types of database models that exist-the hierarchical or network model, and the relational model. The hierarchical model imposes a tree-like data structure, similar to a classified catalog, where super-categories subsume lesser or more specific categories of information. A collection arranged according to the Dewey Decimal system which originally presumed to organize all known information, represents a hierarchical model. The strength of hierarchical models is that they support a clear delineation of the relationships between the information elements in a collection. Geographic areas present a good example of effective use of hierarchies. For example, it is easily agreed that North America is the continent where one can find the United States, wherein one can locate the state of Illinois, and more specifically the city of Chicago. Hierarchical data models were employed in virtually all of the library online systems of the 1970's and 1980's, and many of the museum collections databases of that same time period as well as the tools that were developed in support of museum information such as Nomenclature and The Art and Architecture Thesaurus. However, their chief weakness lies in the fact that they do not support multiple relationships between or among data elements, at different levels.
The relational data model uses tables in which attributes for a particular entity type are defined, with specific values that are assigned to that entity. Relationships between entities are indicated by either duplicating the attributes, or by establishing links between one table and the table with the shared attributes. Relational database software is the most commonly available commercial database product today (e.g., Microsoft Access, Foxpro). A theatre costume and set design database provides an interesting example of a relational database implementation at the University of Illinois. Several tables were set up within one Microsoft Access database, one for costumes and set descriptions, one for information from the theatrical productions such as reviews and cast lists, and one for technical notes on how images of the sets and costumes were produced. Relationships between entities that needed to share attributes were established in the database software. For instance, in an indexing record for a costume design for the character Portia from the Shakespeare play Merchant of Venice was linked to the entity in the production table that contained the information about that particular production.
Many existing software solutions for bringing together diverse formats of metadata operate on the relational database model, because more flexible relationships among data elements can be established once the data is moved from a hierarchical into a relational database structure, without corrupting the integrity of the original format of the data. Most of the search engines available, however, tend to favor hierarchical models because they treat a set of data as a whole without many ways to establish relations between database parts.
Creating a system that enables users to retrieve information across various existing systems and data formats has benefits and drawbacks. The benefit is that information across departments can be brought together in a meaningful way without the user having to move physically (or virtually) from one collection to another. The drawback stems from the fact that the merging of data in different formats inherently dilutes hierarchical control and poses the challenge of working with multiple formats for information. For example, a search in a hierarchically organized collections database for all objects created by a particular artist retrieves a result that will only change if the museum adds another work created by that artist to the database. However, the same search across a combined file of objects, manuscripts, photographs, and library materials for materials by or about a particular artist might produce a very different, less predictable result, depending on the search engine, the way in which the merged data is structured and indexed, and the way in which the query is executed across the existing data. While this collocation of information is a boon for the user, one could say that it might play tricks on the minds of conscientious indexers, curators, and catalogers. It is important to note here that a number of studies have indicated that professional indexers, expert searchers, and naïve users rarely choose the same words to describe the same information. 13 Some basic examples of the currently popular types of search engines may provide the reader with a deeper understanding of the ways in which search engines affect the ultimate results of a query. Numerous examples of virtually all of these systems can be found on the Web.
Turtle and Croft 14 point out that there are essentially three types of information retrieval models that are currently in use:
Exact match search systems are the most prevalent on the commercial market. They operate best with indexing or cataloging records (fielded and terse data), and can accommodate hierarchical control structures such as thesauri or other classification schemes. Search terms can be truncated and stemmed, boolean and proximity operators can be employed to increase the likelihood that a user's query will find matching information in the database. Since the vector space model is mainly in an experimental stage, it will not be discussed further in this context. Statistical methods are increasingly being employed to enhance the retrieval of information that is similar to that requested in a user's query, but may not otherwise be related to the search terms in a database. The probabilistic retrieval model is one statistical method that has been applied to the searching of full text documents and for databases with a combination of formats (which are essentially treated as full text).
There are a number of useful examples of the development of metadata for the description of images that are currently accessible on the Web. The Museum Educational Site Licensing Project (MESL) is an important effort that is now serving as a testbed to determine how this and other descriptive metadata is presented and used.
The MESL was initiated in February, 1995 with partial sponsorship from the Getty Information Institute and MUSE Educational Media. The goal of the project is to test the feasibility of developing site licensing arrangements between museums and educational institutions in the United States. Seven U. S. universities are collaborating with six museums and the Library of Congress to provide networked access to over 8,000 images and their corresponding text descriptions. Faculty in Art History, and other arts and humanities disciplines are using the images for classroom teaching. The images and text are mounted locally by each of the seven universities. The project has a two-year duration. MESL participants include:
|American University||Fowler Museum of Cultural History (UCLA)|
|Columbia University||George Eastman House|
|Cornell University||Harvard University Art Museums|
|University of Illinois at Urbana-Champaign||Library of Congress|
|University of Maryland||The Museum of Fine Arts, Houston|
|University of Michigan||National Gallery of Art|
|University of Virginia||National Museum of American Art|
The primary goal of the project is to "define the terms and conditions under which digitized museum images and information can be distributed over campus networks for educational use." 17 Related objectives include:
Critical among these goals has been the development and testing of standards for image information and its related text. A practical approach was adopted from the outset to generate data from existing museum sources. The text descriptive information for the images supplied by museums was drawn from a combination of their exhibition catalogs and their collections maintenance databases. In order to provide some means of searching across structured data fields from different systems and museums, the participating museums and universities developed a structured data dictionary and used that as a guide for the export of information from their collections databases. Museums had to extract data from their existing collections systems or other databases and re-format it, mapping discrete elements of their data to the structured data dictionary fields. The MESL data dictionary contains thirty-two fields, with some of these being repeatable. The elements of the Dublin Core, where applicable, can be found. The goal of the data dictionary was to represent the types of elements that museum curators and art historians felt were critical to the identification of images. These elements include data such as artist/creator, title, date, subject, description accession number, material, type of art work. However, in mapping existing data structures into a new, common denominator data structure, not every museum provided the same depth of descriptive information, and it was difficult to discern commonly employed standards in museum artifact classification, due to the uniqueness of the materials that museums collect. Therefore, the level of subject analysis, and the method of classification employed may differ radically from one museum to another. For example, the Fowler Museum of Cultural History has a rich collection of Peruvian artifacts. Their descriptive information for the Peruvian materials will undoubtedly include more specific terms than that of another museum that does not have an in depth collection in this area.
The retrieval possibilities, however, are impressive. At the University of Illinois, the images are accessible through a Web site that is restricted to the University community, in accordance with the terms of the site license agreement. Users can either browse creator and title lists within a museum, or they can submit a search using a Web form, which is submitted to an SQL (Structured Query Language) search engine. Perhaps the most significant aspect of mounting this database is that users can search within one museum, or across all seven of the museums simultaneously. The search in the SQL search engine retrieves words or phrases that match the user's search terms, across eight of the thirty-two possible fields. The following example search for the terms "bridge" AND "painting" AND "monet" retrieves images which have in their text descriptions these three words, occurring anywhere in the eight fields which are searched by the SQL engine. The search retrieves eight matches in a thumbnail and brief text record list:
Exhibit 5 Search Results, bridge AND painting AND monet; MESL Search, University of Illinois at Urbana-Champaign (thumbnail image and text description of Claude Monet's Waterloo Bridge, Gray Day reprinted with permission of the National Gallery of Art, Washington, DC)
The terms in the search are truncated, with an implicit boolean AND inserted between the terms, unless otherwise specified. The terms, adjusted for truncation and boolean operators, are then posted for keyword matches against seven of the thirty-two possible data dictionary fields within an SQL table. The matches that are returned are based on the occurrence of those terms within the fields searched. The user has several display options, including a view of a medium and a full-size version of the digital image, and a view of the complete indexing record for the item. While there are challenges to what the library world knows as consistent subject access, the example search suggests that the descriptive text that is supplied by the museums can be stored and manipulated to produce some very useful and interesting retrieval results for users.
The MESL group recognizes that using structured data, and the process of museums extracting specific information from their existing collections databases can be cumbersome. Museums are also using unstructured full text, and SGML (standard generalized markup language) with a DTD (document type definition) developed for either museum or archival information, such as those provided by the CIMI (Computer Information for Museum Interchange) group and by the library and archival community, which has developed the EAD (encoding for archival description) DTD.
The most impressive aspect of the array of possibilities before museums today - the search engines, database standards, reasoned approaches, digital images, and digital data - is that museums actually have developed feasible approaches to represent in meaningful ways the collections and the information that they manage. It may not be getting much easier to do, but there are possibilities. The success of the MESL project, the success of those few instances where search engines have been employed in either actual or test situations for museums, and the lively conversations that have developed in the museum community about providing meaningful public access to museum information indicate our progress. While the use of the web for these efforts has perhaps had only an indirect effect, the fact that the web has compelled museums to revisit the ways that they distribute information to the public and the tools that have been developed to support the kinds of things the web is used for has shaped the way that museums work in ways that we are only just beginning to see.
1 "Digital Libraries Computation Cracks Semantic Barriers Between Databases," Science Volume 272 (June 7, 1996).
2 Bell, Lesley Ann. "Gaining access to visual information: Theory, analysis, and practice of determining subjects-a review of the literature with descriptive abstracts." Art Documentation Vol. 13, no. 2 (Summer, 1994), p. 89.
3 Visual Resources Association. "VRA Core Categories", August, 1996.
4 Weibel, Stuart, Jean Godby, Eric Miller, and Ron Daniel. "OCLC/NCSA Metadata Workshop Report." March, 1995.
5 Dublin Core Metadata Element Set: Reference Description. August 29, 1996.
6 Dempsey, Lorcan and Stuart L. Weibel. "The Warwick Metadata Workshop: A Framework for the Deployment of Resource Description." D-Lib Magazine, July/August, 1996.
7 Weibel, Stuart (reported by). "A Proposed convention for embedding metadata in HTML." June 2, 1996.
8 "CNI/OCLC Metadata Workshop." September 24-25, 1996, OCLC, Dublin, Ohio, USA).
9 Weibel, Stuart and Eric Miller. "Image Description on the Internet: A Summary of the CNI/OCLC Image Metadata Workshop" D-Lib Magazine (January 1997).
10 Project CHIO demonstrates the use of two international standards cited in the CIMI Standards Framework: SGML and Z39.50. CHIO Structure. The project appears to be unfolding in two segments-the first which demonstrates the use of an SGML DTD among participating museums to mark up their descriptive metadata, and the second which involves the development of a Z39.50 search client that will enable the search and retrieval of similar information across the online databases of the participating institutions. Supported by the Dept. of Commerce (TIIAP program) and the NEH. CIMI (Consortium for the computer interchange of museum information).
11 Pitti, Daniel V. "Finding Aids for Archival Collections." 1996).
12 Worthington, Bill and Brian Robinson. "The medium is not the message: mixed mode document technology." In Mary Feeney and Shirley Day (eds.) Multimedia Information. New York: Bowker and Sauer, 1991, p. 56.
13 See, for example, the discussion in Dumais, Susan T. "Improving the retrieval of information from external sources." Behavior Research Methods, Instruments, & Computers. 23(2) (1991) p. 229.
14 Turtle, Howard R. and W. Bruce Croft. "A comparison of text retrieval models." The Computer Journal. Vol. 35(3) (1992), 279.
15 VIBE system home page.
16 Information about the Inquery full text retrieval system can be found at the Center for Intelligent Information Retrieval at the University of Massachusetts at Amherst.
17 Getty Art History Information Program. "Museum Educational Site Licensing Project. Goals and Objectives." February 22, 1995.
18 Getty Art History Information Program. "Museum Educational Site Licensing Project. Goals and Objectives." February 22, 1995.
Return to Digital Services and Development Unit Home Page