Library cataloging has historically been all about getting control over the bibliographic universe, knowing exactly what works and editions a library holds, and making sure that all items in the library catalog are uniformly described. One of the discussion points of the Task Group on the Future of Bibliographic Control was the question the use of the term bibliographic control. The group defined the term as “the organization of library materials to facilitate discovery, management, identification, and access.”⁶ The group also said, however:

The phrase “bibliographic control” is often interpreted to have the same meaning as the word “cataloging.” The library catalog, however, is just one access route to materials that a library manages for its users. The benefits of bibliographic control can be expanded to a wide range of information resources both through cooperation and through design. The Working Group urges adoption of a broad definition of bibliographic control that embraces all library materials, a diverse community of users, and a multiplicity of venues where information is sought.⁷

In this statement, the Working Group opened up the possibility that in the future bibliographic control may be more than what we think of today as cataloging and may take place beyond the confines of the library catalog. They confirm this with another statement:

The future of bibliographic control will be collaborative, decentralized, international in scope, and Web-based. Its realization will occur in cooperation with the private sector, and with the active collaboration of library users. Data will be gathered from multiple sources; change will happen quickly; and bibliographic control will be dynamic, not static. The underlying technology that makes this future possible and necessary—the World Wide Web—is now almost two decades old. Libraries must continue the transition to this future without delay in order to retain their significance as information providers.⁸

There is much here to frighten anyone who hopes that the library catalog will need just some minor tweaking to keep up with modern times. It's pretty clear, though, that the group was defining bibliographic control to mean something quite different from the creation of library catalogs as we know them today. Their expanded definition and the emphasis on the World Wide Web as the appropriate platform for reaching today's users greatly broaden the role that library metadata will have to fulfill.

There is an increasing use of bibliographic data on the Web in general, through services like Google Scholar and Google Book Search, Wikipedia and Wikimedia, LibraryThing, Open Library, and others. Some of these directly import library metadata, others create their own. The content of these sites is not limited to bibliographic data; they use bibliographic data within an information context. In some cases, the object of the bibliographic metadata is the focus of the document or page; in other cases, it serves as a pointer to other resources. In either case, though, it is clear that resources cataloged by libraries are part of the online information landscape. That landscape, however, does not make use of the MARC record, and there is no unifying standard for bibliographic data. There also is no concrete way to link data on the Web with the many instances of that data in library catalogs. Where specific item or record identifiers, such as ISBN or OCLC number, are available, it is often possible to link through WorldCat to library holdings, but that is a viable option only for OCLC member libraries and also doesn't provide links from other data elements to the bibliographic data. It is a partial implementation of integration of library bibliographic data to the Web, but only partial.

The idea that library metadata will be used widely on the open Web changes the meaning of cataloging: cataloging will no longer be limited to the creation of records for the library catalog, but will serve other functions as well, and users who may never directly make use of the library catalog. This is a true expansion of the role of library data, to the point where it can be used for any bibliographic function. However, the effort of cataloging need not increase: instead, the sharing of data can increase, and with some forethought the act of cataloging can draw on cooperative data sources. To be sure, redesign of cataloging systems will be needed.

Library metadata is, and has always been, a complex concept with many different points of information. Both technically and in terms of information content, the library record must be used as a whole. The record provides the context for each data element, and holds together all of the fields that describe a particular manifestation. A field taken out of this context would not be meaningful. A field like the following is not useful because it is only within the record that we know to what book it refers:

• 260 $a New York : $b Viking Penguin, $c 1994

The exact same information can be designed to have meaning both within a record and independently. This is done by providing explicit relationships and identifiers for the subject of the description:

• New York → is place of publication of → Raintree County
• Viking Penguin → is publisher of → Raintree County
• 1994 → is date of publication of → Raintree County

This form of data allows processing on individual data elements within their meaningful context. It creates more possibilities for machines to act on the data. For example, you may wish to know the earliest date of publication of a book, at least the earliest that the library has. With this type of data organization, it is possible to ask for all of the dates of publication for the book in question, and to receive the following as an answer:

• 1948
• 1994

Note that where the above examples have meaningful words (“Raintree County,” “is publisher of”), the actual data would have identifiers for those terms and concepts. This is because the words themselves could be ambiguous (“Raintree County” is both a book and a film, and each would need its own identifier), and in any case they are not globally unique. Someone else could develop a term “is publisher of” that has a meaning slightly different from the one that I am using. The unique identification of things and relationships assures that data can mix with other data without losing its specific meaning.

All of this facilitates machine processing, of course, but it also potentially provides some new capabilities for user interfaces as well. It should be much easier to create a function that will find other editions of the same book without requiring that the user perform a search. For example, if the user has entered the catalog from a link on a professor's reading list and all copies of the book are checked out, it should be possible to expand the search to the most recent other edition of the book. In a library catalog today, the user would need to perform a search and either read through a list of retrieved items or limit by date. With the catalog record reorganized in a linked data format, this becomes an automated offering, not a user task. The main reason to organize our data as separate and unambiguous data elements, however, is to allow that data to be used outside of the context of the library catalog and to be combined with other data.

Fortunately, the data in library records is coded in some detail, so a transformation from the record-based model to a data-based model is feasible. The whole of the information about a single item may still be wrapped together as a record, but the data within that record will be usable in many different contexts.

The closed system model used by libraries is related to the dependence on a record-based model. This has been the dominant model for all forms of data, not just in libraries. In a database management system, data is stored in a highly controlled environment with specific functions allowed to different categories of users: those who can modify the database management structure (the system administrators), those who can modify records (the catalogers), and those who can enter only through the user interface (the users). Regardless of how rich the data, users of this system can perform only actions that are offered to them through the interface.

In addition, the data in the database cannot easily interact with data outside of the database. Libraries have added some links to Web services and are able to import book covers and reviews from other sources, but dynamic interaction is difficult to achieve. It can even be impossible to link into the library catalog from outside, thus limiting users' ability to make reference to items held in the library. This means that library data cannot participate in the highly linked and linkable information environment on the Web, and this limits the visibility of libraries to Web users.

Users go to the library catalog to search for items in the library collection. To conduct a search, they must have something in mind: an author, a title, or a topic. Searching is a familiar first step in information seeking, but it is not the only way, and perhaps not even the predominant way, that users find the information they need. In real life—that is, “offline”—friends, colleagues, and mass media are common leads to information sources. Online, social sites have become powerful meeting grounds where users ask questions, find recommendations, and pursue a wide variety of interests. While it may be difficult to think of these actions as “information retrieval,” they do provide users with a great deal of information. Just because a person stumbles upon an interesting site or reads a book that is recommended by a friend does not mean that no information has been exchanged. These informal sources of information are not at all new: many experts cite other members of their profession as their main source of information and commonly begin an investigation in a new area by contacting a colleague who is already expert in that area.

Offline, we rely on a web of human connections to help us find information. Online, that web consists of links between resources and rich social interaction that help us select and evaluate resources. The search itself is only a beginning. The library catalog, however, offers little beyond search and retrieve. Navigation is generally limited to clicking on headings, which then performs another search. The catalog, therefore, serves only limited information-seeking behavior. It is no wonder that few users report that they begin their information searches in the library catalog and that the use of library catalogs is minor compared to the use of the Internet.

It is unlikely that searching will be eliminated from our information discovery toolkit, but we can expect that navigation capabilities will become increasingly available as more and more information moves to the Web. Today we follow links found within documents, but the Semantic Web promises even richer navigation possibilities, as well as the ability to actually pose questions to the Web, treating it much like a database of information.

We talked above about the FRBR model and the user tasks that have guided its development. A great value of the family of functional requirements is that they begin to define the entities that our metadata addresses: bibliographic resources, agents, topics. At this level, we can see some similarities already with linked data standards being developed elsewhere. For example, an early Semantic Web project is “Friend of a Friend” (FOAF), a metadata format for persons that can be used in social network situations.⁹ It is not identical to the way that libraries define persons, but it points to an area where data could be exchanged among different communities.

This step is similar to the initial creation of a database structure, where you define all of your data elements. Each data element will need to be defined according to certain requirements posed by the Semantic Web concepts. The Semantic Web view of data differs from that of a database, so developers will need some level of learning and adjustment.

One of the great advantages that we have in transforming our catalog data to the Semantic Web is that we have already made much use of controlled vocabularies. These help greatly in communicating with other communities because we can clearly delineate the possible meanings of certain elements through the finite vocabulary list that the element can carry.

Vocabularies can be simple lists of terms, but it is also possible to define each term in a vocabulary with a unique identifier. Identifiers are less ambiguous than language and also often create links back to the identifying agency and documentation about the term. For example, the term green can mean different things: in the context of politics, it may indicate an approach toward environmental issues or even the name of a political party. In a shoe catalog, it may be the color of the product. The use of an identifier provided by the entity that has developed the vocabulary means that each of these meanings will have a different identifier.

There is another level of definition that will usually be undertaken, although it is not strictly required by the Semantic Web, and that is the creation of application rules or an application profile. Application rules generally add constraints to your metadata, such as whether your element will be mandatory or optional, and rules for repeatability. For example, although you may define the element for title in your list of elements, yet in your actual application you may wish to limit the use of the title to one per description. Or you may wish to say that in your application the title is mandatory. The definition of these rules in a machine-readable form will allow others to understand the output of your applications.

The remainder of this report will illustrate these steps in greater detail.