48 May 2001/Vol. 44, No. 5 COMMUNICATIONS OF THE ACMPrinciplesFor Digital LibraryDevelopmentBuilding a digital library is expensive and resource-intensive. Before embarking on such a ven-ture, it is important to consider some basic principles underlying the design, implementation,and maintenance of any digital library. These principles apply not only to conversion projectsin which analog objects are converted to digital form, but to digital libraries in which the objectshave always been in digital form (“born digitally”) and to “mixed” digital libraries in which theobjects may be of both types. The principles are, in some sense, self evident, yet it is easy to losesight of them when under pressure to build a system, despite limited resources and time. COMMUNICATIONS OF THE ACM May 2001/Vol. 44, No. 5 49Adhering to the following set of 10 principles(see Figure 1), as well as to the practices that evolvefrom them, benefits those responsible for thedesign and continued development of any digitallibrary system, and, perhaps, more important, con-tinues to pay off over the long-term. The principles are derived from our experiencedeveloping digital library systems over the pastdecade [7]. We migrated one digital library (cre-ated in the early 1990s, before any thought of serv-ing its contents over the Internet) into a morerecently created system called Profiles in Science(profiles.nlm.nih.gov/). Even though the originalhardware and software in the two systems werecompletely different, the migration was successful,because each was designed with the same basicprinciples in mind. Expect change. It may not be apparent why thechanging technology landscape is such a thornyproblem for digital library projects. Consider, forexample, a conversion project in which docu-ments are converted to some digital format. If thechosen format is part of a proprietary system,viewable only through a proprietary interface,when the company that markets the interface nolonger supports the system and format, the digi-tized documents are all but lost. Consider, too, ascenario in which a document is created in a par-ticular word processing program and the docu-ment is attached to an email message sent to anotable person. Suppose the goal is to preserve allof that person’s email messages for future genera-tions. We are all too aware of our dependence onour email technology for reading such attach-ments. Imagine what today’s platform limitationswill mean to future generations, when the contentAlexa T. McCray and Marie E. Gallagher{{Principles, Tools, and Issues in Building DLsWant a library’s content to persist and be accessible no matterwhich computer, browser, or digital format is used? Follow theseprinciples and practices, as well as their implied promises.of the attachments is likely no longer accessible. Although the Internet, together with the Web, hasmade digital libraries possible, this fact may also con-tribute to unforeseen problems if library designersdepend too much on today’s paradigms and tools.They might be tempted to create a Web site withHypertext Markup Language (HTML) pages andWeb-accessible digital images of objects and docu-ments—that may all be obsolete when HTML changesor is superseded by something else. Changing tech-nologies can quickly outpace the ability of designers tomaintain a particular digital library. An approach thatanticipates and plans for change is needed to providelasting access to its information.Know your content. For users, content is the mostinteresting and valuable aspect of a digital library (seeFigure 2 for a sample exhibit page in the Profiles in Sci-ence system). Creators of digital libraries need to man-age and make decisions about their content, includingselecting the objects to be included, digitizing itemsthat exist only in analog form, possibly marking-upitems using standard languages like the Standard Gen-eralized Markup Language (SGML), and assigningmetadata describing the content and other attributesof each object.It is important for developers to decide on thenature and number of metadata elements early in aproject. Although some elements may be added overtime, significant costs might be associated with assign-ing metadata retroactively to already tagged and cata-loged items in a collection. Some metadata elementsdescribe the content of an item, including, say, its title,creator, date of publication, and subjects discussed.Other elements might be assigned for managing the col-lection; examples include scan status, quality-controlstatus, and internal notes, as well as the technical aspectsof the digital objects, such as file format and size. Important, too, is deciding on the basic conceptualunits, or objects, the system will include, such as indi-vidual documents, photographs, videos, or lab note-books. This decision affects the level at whichmetadata is assigned (for example, to an entire book orto each chapter in the book) and how the materials areorganized, accessed, and archived. Sometimes over-looked is the practice of assigning each conceptualobject its own unique identifier linking it to its meta-data record and to other objects in the collection.Involve the right people. Ideally, individuals from avariety of backgrounds and offering a variety of exper-tise contribute to building a digital library. In practice,this may not be the case, but even when it’s not, know-ing that building the system requires insight from anumber of fields yields a better digital library. The two fields involved most directly are computerscience and library science. Computer scientists appre-ciate the possibilities, as well as the limitations, of tech-nology and are generally the ones who actually buildthe system. Librarians, including catalogers, indexers,and archivists, have long been the custodians of infor-mation resources, understanding not only the infor-mation needs of diverse audiences but the issuesinvolved in preserving materials for continued accessand use. Digital library research and development havemeant that each group has had to understand theother groups’ perspectives. Illustrating the importance of multiple perspectivesis the development of the Dublin Core metadata stan-dard [8]. Computer scientists are concerned with thesemantic interoperability a digital library metadatastandard affords in the very large Internet informationspace; librarians already have deep experience indexingand cataloging and recognize the importance of