Chapter 16—
A New Consortial Model for Building Digital Libraries

Raymond K. Neff

Libraries in America's research universities are being systematically depopulated of current subscriptions to scholarly journals. Annual increases in subscription costs are consistently outpacing the growth in library budgets. This problem has become chronic for academic libraries that collect in the fields of science, engineering, and medicine, and by now the problem is well recognized (Cummings et al. 1992). At Case Western Reserve University, we have built a novel digital library distribution system and focused on our collections in the chemical sciences to investigate a new approach to solving a significant portion of this problem. By collaborating with another research library that has a strong chemical sciences collection, we have developed a methodology to control costs of scholarly journals and have planted the seeds of a new consortial model for building digital libraries. This paper summaries our progress to date and indicates areas in which we are continuing our research and development.

For research libraries in academia, providing sufficient scholarly information resources in the chemical sciences represents a large budgetary item. For our purposes, the task of providing high-quality library services to scholars in the chemical sciences is similar to providing library services in other sciences, engineering, and medicine; if we solve the problem in the limited domain of the chemical sciences, we can reasonably extrapolate our results to these other fields. Thus, research libraries whose mission it is to provide a high level of coverage for scholarly publications in the chemical sciences are the focus of this project, although we believe that the principles and practices employed in this project are extensible to the serial collections of other disciplines.

A consortium depends on having its members operating with common missions, visions, strategies, and implementations. We adopted the tactics of developing a consortial model by having two neighboring libraries collaborate in the initial project. The University of Akron (UA) and Case Western Reserve University

(CWRU) both have academic programs in the chemical sciences that are nationally ranked, and the two universities are fewer than 30 miles apart. It was no surprise to find that both universities have library collections in the chemical sciences that are of high quality and nearly exhaustive in their coverage of scholarly journals. To quantify the correlation between these two collections, we counted the number of journals that both collected and found the common set to be 76% in number and 92% in cost. The implications of the overlap in collecting patterns is plain; if both libraries collected only one copy of each journal, with the exception of the most used journals, approximately half of the cost of these subscriptions could be saved. For these two libraries, the cost savings is potentially $400,000 per year. This cost savings seemed like a goal worth pursuing, but to do so would require building a new type of information distribution system.

The reason scholarly libraries collect duplicative journals is that students and faculty want to be able to use these materials by going to the library and looking up a particular volume or by browsing the current issues of journals in their field. Eliminating a complete set of the journals at all but one of our consortial libraries would deprive local users of this walk-up-and-read service. We asked ourselves if it would be possible to construct a virtual version of the paper-based journal collection that would be simultaneously present at each consortium member institution, allowing any scholar to consult the collection at will even though only one copy of the paper journal was on the shelf. The approach we adopted was to build a digital delivery system that would provide to a scholar on the campus of a consortial member institution, on a demand basis, either a soft or hard copy of any article for which a subscription to the journal was held by a consortial member library. Thus, according to this vision, the use of information technology would make it possible to collect one set of journals among the consortium members and to have them simultaneously available at all institutions. Although the cost of building the new digital distribution system is substantial, it was considered as an experiment worth undertaking. The generous support of The Andrew W. Mellon Foundation is being used to cover approximately one-half of the costs for the construction and operation of the digital distribution system, with Case Western Reserve University covering the remainder. The University of Akron Library has contributed its expertise and use of its chemical sciences collections to the project.

It also seemed necessary to us to want to invite the cooperation of journal publishers in a project of this kind. To make a digital delivery system practical would require having the rights to store the intellectual property in a computer system, and when we started this project, no consortium member had such rights. Further, we needed both the ongoing publications and the backfiles so that complete runs of each serial could be constructed in digital form. The publishers could work out agreements with the consortium to provide their scholarly publications for inclusion in a digital storage system that would be connected to our network-based transmission system, and thus, their cooperation would become essential. The chemical sciences are disciplines in which previous work with electronic libraries

had been started. The TULIP Project of Elsevier Science (Borghuis et al. 1996) and the CORE Project undertaken by a consortium of Cornell University, the American Chemical Society, Bellcore, Chemical Abstracts, and OCLC were known to us, and we certainly wanted to benefit from their experiences. Publications of the American Chemical Society, Elsevier Science, Springer-Verlag, Academic Press, John Wiley & Sons, and many other publishers were central to our proposed project because of the importance of their journal titles to the chemical sciences disciplines.

We understood from the beginning of this effort that we would want to monitor the performance of the digital delivery system under realistic usage scenarios. The implementation of our delivery system has built into it extensive data collection facilities for monitoring what users actually do. The system is also sensitive to concerns of privacy in that it collects no items of performance information that may be used to identify unambiguously any particular user.

Given the existence of extensive campus networks at both CWRU and UA and substantial internetworking among the academic institutions in northeastern Ohio, there was sufficient infrastructure already in place to allow the construction and operation of an intra- and intercampus digital delivery system. Such a digital delivery system has now been built and made operational. The essential aspects of the digital delivery system will now be described.

A Digital Delivery System

The roots of the electronic library are found in landmark papers by Bush (1945) and Kemeny (1962). Most interestingly, Kemeny foreshadowed what prospective scholarly users of our digital library told us was their essential requirement, which was that they be able to see each page of a scholarly article preserved in its graphical integrity. That is, the electronic image of each page layout needed to look like it did when originally published on paper. The system we have developed uses the ACROBATR page description language to accomplish this objective.

Because finding aids and indices for specialized publications are too limiting, users also have the requirement that the article's text be searchable with limited or unlimited discipline-specific thesauri. Our system complements the page images with an optical character recognition (OCR) scanning of the complete text of each article. In this way, the user may enter words and phrases the presence of which in an article constitutes a "hit" for the scholar.

One of the most critical design goals for our project was the development of a scanning subsystem that would be easily reproducible and cost efficient to set up and operate in each consortium member library. Not only did the equipment need to be readily available, but it had to be adaptable to a variety of work flow and staff work patterns in many different libraries. Our initial design has been successfully tailored to the needs of both the CWRU libraries and the Library at the Univer-

sity of Akron. Our approach to the sharing of paper-based collections is to use a scanning device to copy the pages of the original into a digital image format that may be readily transmitted across our existing telecommunications infrastructure. In addition, the digital version of the paper is stored for subsequent retrieval. Thus, repeated viewing of the same work would necessitate only a one-time transformation of format. This procedure is an advantage in achieving faster response times for scholars, and it promotes the development and use of quality control methods. The scanning equipment we have used in this project and its operation are described in Appendix E. The principal advantage of this scanner is that bound serials may be scanned without damaging the volume and without compromising the resulting page images; in fact, the original journal collection remains intact and accessible to scholars throughout the project. This device is also sufficiently fast that a trained operator, including students, may scan over 800 pages per average workday. For a student worker making $7.00 per hour, the personcost of scanning is under $0.07 per page; the cost of conversion to searchable text adds $0.01 per page. Appendix E also gives more details regarding the scanning processes and work flow. Appendix F gives a technical justification for a digitization standard for the consortium. Thus, each consortium member is expected to make a reasonable investment in equipment, training, and personnel.

The target equipment for viewing an electronic journal was taken to be a common PC-compatible computer workstation, hereafter referred to as a client. This client is also the user platform for the on-line library catalog systems found on our campuses as well as for the growing collections of CD-ROM-based information products. Appendix B gives the specification of the workstation standards for the project. The implications for use of readily available equipment is that the client platform for our project would also work outside of the library-in fact, wherever a user wanted to work. Therefore, by selecting the platform we did, we extended the project to encompass a full campuswide delivery system. Because our consortium involves multiple campuses (two at the outset), the delivery system is general purpose in its availability as an access facility.

Just as we needed a place to store paper-based journals within the classical research library, we needed to specify a place to store the digital copies. In technical parlance, this storage facility is called a server. Appendixes B and C give some details regarding the server hardware and software configurations used in this project.

Appendix C also gives some information regarding the campuswide networks on both our campuses and the statewide network that connects them. It is important to note that any connected client workstation that follows our minimum standards will be able to use the digital delivery system being constructed.

Because the key to minimizing the operating costs within a consortium is interoperability and standardization, we have adopted a series of data and equipment standards for this project; they are given in Appendixes A and B.

Rights Management System

One of the most significant problems in placing intellectual property in a networked environment is that with a few clicks of a mouse thousands of copies of the original work can be distributed at virtually zero marginal cost, and as a result, the owner may be deprived of expected royalty revenue. Since we recognized this problem some years ago and realized that solutions outside of the network itself were unlikely to be either permanent or satisfactory to all parties (e.g., author, owner, publisher, distributor, user), we embarked on the creation of a software subsystem now known as Rights Manager(tm). With our Rights Manager system, we can now control the distribution of digitally formatted intellectual property in a networked environment subject to each stakeholder receiving proper due.

In this project, we use the Rights Manager system with our client server-based content delivery system to manage and control intellectual property distribution for digitally formatted content (e.g., text, images, audio, video, and animations).

Rights Manager is a working system that encodes license agreement information for intellectual property at a server and distributes the intellectual property to authorized users over the Internet or a campuswide intranet along with a Rights Manager-compliant browser. The Rights Manager handles a variety of license agreement types, including public domain, site licensing, controlled simultaneous accesses, and pay-per-use. Rights Manager also manages the functionality available to a client according to the terms of the license agreement; this is accomplished by use of a special browser that enforces the license's terms and that permits or denies client actions such as save, print, display, copy, excerpt, and so on. Access to a particular item of intellectual property, with or without additional functionality, may be made available to the client at no charge, with an overhead charge, or at a royalty plus overhead charge. Rights Manager has been designed to accommodate sufficient flexibility in capturing wide degrees of arbitrariness in charging rules and policies.

The Rights Manager is intended for use by individuals and organizations who function as purveyors of information (publishers, on-line service providers, campus libraries, etc.). The system is capable of managing a wide variety of agreements from an unlimited number of content providers. Rights Manager also permits customization of licensing terms so that individual users or user classes may be defined and given unique access privileges to restricted sets of materials. A relatively common example of this customization for CWRU would be an agreement to provide (1) view-only capabilities to an electronic journal accessed by an anonymous user located in the library, (2) display/print/copy access to all on-campus students enrolled in a course for which a digital textbook has been adopted, and (3) full access to faculty for both student and instructor versions of digital editions of supplementary textbook materials.

Fundamental to the implementation of Rights Manager are the creation and maintenance of distribution rights, permissions, and license agreement databases.

These databases express the terms and conditions under which the content purveyor distributes materials to its end users. Relevant features of Rights Manager include:

• a high degree of granularity, which may be below the level of a paragraph, for publisher-defined content

• central or distributed management of rights, permissions, and licensing databases

• multiple agreement types (e.g., site licensing, limited site licensing, and payper-use)

• content packaging where rights and permission data are combined with digital format content elements for managed presentation by Web browser plug-in modules or helper applications

Rights Manager maintains a comprehensive set of distribution rights, permissions, and charging information. The premise of Rights Manager is that each publication may be viewed as a compound document. A publication under this definition consists of one or more content elements and media types; each element may be individually managed, as may be required, for instance, in an anthology.

Individual content elements may be defined as broadly or narrowly as required (i.e., the granularity of the elements is defined by the publisher and may go below the level of a paragraph of content for text); however, for overall efficiency, each content element should represent a significant and measurable unit of material. Figures, tables, illustrations, and text sections may reasonably be defined as individual content elements and be treated uniquely according to each license agreement.

To manage the distribution of complete publications or individual content elements, two additional licensing metaphors are implemented. The first of these, a Collection Agreement, is used to specify an agreement between a purveyor and its supplier (e.g., a primary or secondary publisher); this agreement takes the form of a list of publications distributed by the purveyor and the terms and conditions under which these publications may be issued to end users (one or more Collection Agreements may be defined and simultaneously managed between the purveyor and a customer).

The second abstraction, a Master Agreement, is used to broadly define the rules and conditions that apply to all Collection Agreements between the purveyor and its content supplier. Only one Master Agreement may be defined between the supplier and the institutional customer. In practice, Rights Manager assumes that the purveyor will enter into licensing agreements with its suppliers for the delivery of digitally formatted content. At the time the first license agreement is executed between a supplier and a purveyor, one or more entries are made into the purveyor's Rights Manager databases to define the Master and Collection Agreements. Optionally, Publication and/or Content-Element usage rules may also be

defined. Licensed materials may be distributed from the purveyor's site (or perhaps by an authorized service provider); both the content and associated licensing rules are transferred by the supplier to the purveyor for distributed license and content management.

Depending on the selected delivery option, individual end users (e.g., faculty members, students, or library patrons) may access either a remote server or a local institutional repository to search and request delivery of licensed publications. Depending on the agreement(s) between the owner and the purveyor, individual users are assigned access rights and permissions that may be based on individual user IDs, network addresses, or both.

Network or Internet Protocol addresses are used to limit distribution by physical location (e.g., to users accessing the materials from a library, a computer lab, or a local workstation). User identification may be exploited to create limited sitelicensing models or individual user agreements (e.g., distributing publications only to students enrolled in Chemistry 432 or, perhaps, to a specific faculty member).

At each of the four levels (Master Agreement, Collection Agreement, Publication, and Content-Element), access rules and usage privileges may be defined. In general, the access and usage permissions rules are broadly defined at the Master and Collection Agreement level and are refined or restricted at the Publication and Content-Element levels. For example, a Master or Collection Agreement rule could be defined to specify that by default all licensed text elements may be printed at a some fixed cost, say 10¢ per page; however, high value or core text sections may be individually identified using Publication or Content-Element override rules and assessed higher charges, say 20¢ per page.

When a request for delivery of materials is received, the content rules are evaluated in a bottom-up manner (e.g., content element rules are evaluated before publication rules, which are, in turn, evaluated before license agreement rules, etc.). Access and usage privileges are resolved when the system first recognizes a match between the requester's user ID (or user category) and/or the network address and the permission rules governing the content. Access to the content is only granted when an applicable set of rules specifically granting access permission to the end user is found; in the case where two or more rules permit access, the rules most favorable to the end user are selected. Under this approach, site licenses, limited site licenses, individual licensing, and pay-per-use may be simultaneously specified and managed.

The following use of the Rights Manager rules databases is recommended as an initial guideline for Rights Manager implementation:

1. Use Master Agreement rules to define the publishing holding company or imprint, the agreement's term (beginning and ending dates), and the general "fair use" guidelines negotiated between a supplier and the purveyor. Because of the current controversy over the definition of "fair use," Rights

Manager does not rely on preprogrammed definitions; rather, the supplier and purveyor may negotiate this definition and create rules as needed. This approach permits fair use definitions to be redefined in response to new standards or regulatory definitions without requiring modifications to Rights Manager itself.

2. Use Collection Agreement rules to define the term (beginning and ending dates) for specific licensing agreements between the supplier and the purveyor. General access and permission rules by user ID, user category, network address, and media type would be assigned at this level.

3. Use Publication rules to impose any user ID or user category-specific rules (e.g., permissions for students enrolled in a course for which this publication has been selected as the adopted textbook) or to impose exceptions based on the publication's value.

4. Use Content-Element rules to grant specific end users or user categories access to materials (e.g., define content elements that are supplementary teaching aids for the instructor) or to impose exceptions based on media type or the value of content elements.

The Rights Manager system does not mandate that licensing agreements exploit user IDs; however, maximum content protection and flexibility in license agreement specification is achieved when this feature is used. Given that many institutions or consortium members may not have implemented a robust user authentication system, alternative approaches to uniquely identifying individual users must be considered. While there are a variety of ways to address this issue, it is suggested that personal identification numbers (PINs), assigned by the supplier and distributed by trusted institutional agents at the purveyor's site (e.g., instructors, librarians, bookstore employees, or departmental assistants) or embedded within the content, be used as the basis for establishing user IDs and passwords. Using this approach, valid users may enter into registration dialogues to automatically assign user IDs and passwords in response to a valid PIN "challenge."

While Rights Manager is designed to address all types of multimedia rights, permissions, and licensing issues, the current implementation has focused on distribution of traditional print publication media (text and images). Extensions to Rights Manager to address the distribution of full multimedia, including streaming audio and video, are being developed at CWRU.

The key to understanding our approach to intellectual property management is that we expect that each scholarly work will be disseminated according to a comprehensive contractual agreement. Publishers may use master agreements to cover a set of titles. Further, we do not expect that there will be only one interpretation of concepts such as fair use, and our Rights Manager system makes provision for arbitrarily different operational definitions of fair use, so that specific contractual agreements can be "enforced" within the delivery system.

A New Consortial Model

The library world has productively used various consortial models for over 30 years, but until now, there has not been a successful model for building a digital library. One of the missing pieces in the consortial jigsaw puzzle has been a technical model that is both comprehensive and reproducible in a variety of library contexts. To begin our approach to a new consortial model, we developed a complete technical system for building and operating a digital library. Building such a system is no small achievement. Similar efforts have been undertaken with the Elsevier Science TULIP Project and the JSTOR project.

The primary desiderata for a new consortial model are as follows:

• Any research library can participate using agreed upon and accepted standards.

• Many research libraries each contribute relatively small amounts of labor by scanning a small, controlled number of journal issues. Scanning is both systematic and based on a request for an individual article.

• Readily available off-the-shelf equipment is used.

• Intellectual property is made available through licensing and controlled by the Rights Manager software system.

• Publishers grant rights to libraries to scan and store intellectual property retrospectively (i.e., already purchased materials) in exchange for the right to license use of the digital formats to other users. Libraries provide publishers with digital copies of scholarly journals for their own use, thus enabling publishers to enrich their own electronic libraries.

A Payments System For The Consortium

It is unrealistic to assume that all use of a future digital library will be without any charging mechanisms even though the research library of today charges for little except for photocopying and user fines. This is not to assume that the library user is charged for each use, although that would be possible. A more likely scenario would be that the library pay on behalf of the members of the scholarly community (i.e., student, professor, researcher) that it supports. According to our proposed consortial model, libraries would be charged for use of the digital library according to the total pages "read" in any given user session. It could be easily worked out that users who consult the digital library on the premises of the campus library would not be charged themselves, but if they used the digital library from another campus location or from off-campus through a network, that they would pay a per-page charge analogous to the cost of photocopying. A system of charging could include categorization by type of user, and the Rights Manager system provides for a wide variety of charging models, including the making of distinctions of usage in soft-copy format, hard-copy format, and downloading of

a work in whole or in part. Protecting the rights of the owner is an especially interesting problem when the entire work is downloaded in a digital format. Both visible and invisible watermarking are techniques with which we have experience for protecting rights in the case of downloading an entire work.

We also have in mind that libraries that provide input via scanning to the decentralized, digital library would receive a credit for each page scanned. It is clear that the value of the digital library to the end user will increase as higher degrees of completeness in digitized holdings is achieved. Therefore, the credit system to originating libraries should recognize this value and reward these libraries according to a formula that charges and credits with a relative credit-to-charging ratio of perhaps ten to one; that is, an originating library might receive a credit for scanning one page equal to a charge for reading ten soft-copy pages.

The charge-and-credit system for our new consortial model is analogous to that used for the highly successful Online Computer Library Center's cataloging system. Member libraries within OCLC contribute original cataloging entries in the form of MARC records for the OCLC database as well as draw down a copy of a holding's data to fill in entries for their own catalog systems. The system of charging for "downloads" and crediting for "uploads" is repeated in our consortial model for retrospective scanning and processing of full-text journal articles. Just as original cataloging is at the heart of OCLC, original scanning is at the heart of our new consortial model for building the library of the future.

Data Collection

One of the most important aspects of this project is that the underlying software system has been instrumented with many data collection points. In this way we can find out through actual usage by faculty, students, and research staff what aspects of the system are good and which need more work and thought. Over the past decade many people have speculated about how the digital library might be made to work for the betterment of scholarly communications. Our system as described in this paper is one of the most comprehensive attempts yet to build up a base of usage experience data.

To appreciate the detailed data being collected by the project, we will describe the various types of data that the Rights Manager system captures. Many types of transactions occur between the Rights Manager client and the Rights Manager server software throughout a user session. The server software records these transactions, which will permit detailed analysis of usage patterns. Appendix D gives some details regarding the data collected during a user session.

Publishers And Digital Libraries

The effects of the new consortial model for building digital libraries are not confined to the domain of technology. During the period when the new digital distribution system was being constructed, Ohio LINK, an agency of the Ohio

Board of Regents, commenced an overlapping relationship with Academic Press to offer its collection of approximately 175 electronic journals, many of which were in our chemical sciences collections. Significantly, the Ohio LINK contract with Academic Press facilitated the development of our digital library because it included a provision covering the scanning and storage of retrospective collections (i.e., backfiles) of their journals that we had originally acquired by subscription. In 1997, Ohio LINK extended the model of the Academic Press contract to an offering from Elsevier Science. According to this later agreement, subscriptions to current volumes of Elsevier Science's 1,153 electronic journals would be available for access and use on all of the 57 campuses of Ohio LINK member institutions, including CWRU and the University of Akron. The cost of the entire collection of electronic journals for each university for 1998 was set by the Ohio LINK-Elsevier contract to be approximately 5.5% greater than the institution's Elsevier Science expenditure level for 1997 subscriptions regardless of the particular subset these subscriptions represented; there is a further 5.5% price increase set to take effect in 1999. Further, the agreement between Ohio LINK and Elsevier constrains the member institutions to pay for this comprehensive access even if they cancel a journal subscription. Notably, there is an optional payment discount of 10% when an existing journal subscription (in a paper format) is limited to electronic delivery only (eliminating the delivery of a paper version). Thus, electronic versions of the Elsevier journals that are part of our chemical sciences digital library will be available at both institutions regardless of the existence of our consortium; pooling collections according to our consortial model would be a useless exercise from a financial point of view.

Other publishers are also working with our consortium of institutions to offer digital products. During spring 1997, CWRU and the University of Akron entered into an agreement with Springer-Verlag to evaluate their offering of 50 or so electronic journals, some of which overlapped with our chemical sciences collection. A similar agreement covering backfiles of Elsevier journals was considered and rejected for budgetary reasons. During the development of this project, we had numerous contacts with the American Chemical Society with the objective of including their publications in our digital library. Indeed, the outline of an agreement with them was discussed. As the time came to render the agreement in writing, they withdrew and later disavowed any interest in a contract with the consortium. At the present time, discussions are being held with other significant chemical science publishers about being included in our consortial library. This is clearly a dynamic period in journal publishing, and each of the societal and commercial publishers sees much at stake. While we in universities try to make sense of both technology and information service to our scholarly communities, the publishers are each trying to chart their own course both competitively and strategically while improvements in information technology continually raise the ante for continuing to stay in the game.

The underlying goal of this project has been to see if information technology

could control the costs of chemical sciences serial publications. In the most extreme case, it could lower costs by half in our two libraries and even more if departmental copies were eliminated. As an aside, we estimated that departmentally paid chemical sciences journal subscriptions represented an institutional expenditure of about 40% of the libraries' own costs, so each institution paid in total 1.4 times each library's costs. For both institutions, the total was about 2.8 times the cost of one copy of each holding. Thus, if duplication were eliminated completely, the resulting expenditures for the consortium for subscriptions alone would be reduced by almost two-thirds from that which we have been spending. Clearly, the journal publishers understood the implications of our project. But the implications of the status quo were also clear: libraries and individuals were cutting subscriptions each year because budgets could not keep up with price increases. We believed that to let nature take its course was irresponsible when a well-designed experiment using state-of-the-art information technology could show a way to make progress. Thus, the spirit of our initial conversations with chemical sciences publishers was oriented to a positive scenario: libraries and the scholars they represented would be able to maintain or gain access to the full range of chemical sciences literature, and journals would be distributed in digital formats. We made a crucial assumption that information technology would permit the publishers to lower their internal production costs. This assumption is not unreasonable in that information technology has accomplished cost reductions in many other businesses.

In our preliminary discussions with the publishers, we expressed the long-term objective that we were seeking-controlling and even lowering our costs through the elimination of duplication as our approach to solving the "cancellation conundrum"-as well as our short-term objective-to receive the rights to scan, store, and display electronic versions of both current and back files of their publications, which we would create from materials we had already paid for (several times over, in fact). Current and future subscriptions would be purchased in only one copy, however, to create the desired financial savings. In exchange, we offered the publishers a complete copy of our PDF-formatted current issue and backfiles for their use, from which they could derive new revenue through licensing to others. Since these once-sold backfiles were being considered on the publishers' corporate balance sheets as a depleted resource, we thought that the prospect of deriving additional revenue from selling them again as a digital resource would seem to be an attractive idea. In the end, however, not one publisher was willing to take us up on this exchange. To them, the backfiles that we would create were not worth what we were asking. One chemical sciences journal publisher was willing to grant the rights to backfiles for additional revenue from our consortium. But this offer made no sense unless the exchange could be justified on the basis of savings in costs of library storage space and the additional convenience of electronic access (the digital library is never closed, network access from remote locations would likely increase marginal usage, etc.). When we saw the proposed charge, we

rejected this offer as being too expensive. Another publisher did grant us the rights we sought as part of the overall statewide Ohio LINK electronic and print subscription contract, but this arrangement locked in the current costs (and annual increments) for several years, so the libraries could not benefit directly in terms of cost savings. With that particular limited agreement, however, there still is the definite possibility for savings on departmentally paid, personal subscriptions.

When we began to plan this project, it was not obvious what stance the publishing community would take to it. Our contacts in some of the leading publishing houses and in the Association of American Publishers (AAP) led us to believe that we were on the right track. Clearly, our goal was to reduce our libraries' costs, and that goal meant that publishers would receive less revenue. However, we also believed that the publishers would value receipt of the scanned backfiles that we would accumulate. Thus, the question was whether the backfiles have significant economic value. Clearly, libraries paid for the original publications in paper formats and have been extremely reluctant to pay a second time for the convenience of having access to digital versions of the backfiles. In our discussions, the publishers and AAP also seemed interested in doing experiments in learning whether a screen-based digital format could be made useful to our chemical sciences scholars. Thus, there was a variety of positive incentives favoring experimentation, and a benign attitude toward the project was evinced by these initial contacts with publishers. Their substantial interest in the CWRU Rights Management system seemed genuine and sincere, and their willingness to help us with an experiment of this type was repeatedly averred. After many months of discussion with one publisher, it became clear that they were unwilling to participate at all. In the end, they revealed that they were developing their own commercial digital journal service and that they did not want to have another venue that might compete with this. A second publisher expressed repeated interest in the project and, in the end, proposed that our consortium purchase a license to use the backfiles at a cost of 15% more than the price of the paper-based subscription; this meant that we would have to pay more for the rights to store backfiles of these journals in our system. A third publisher provided the rights to scan, store, display, and use the backfiles as part of the overall statewide Ohio LINK contract; thus this publisher provided all the rights we needed without extra cost to the consortium. We are continuing to have discussions with other chemical sciences journal publishers regarding our consortium and Ohio LINK, and these conversations are not uncomplicated by the overlap in our dual memberships.

It is interesting to see that the idea that digital distribution could control publisher costs is being challenged with statements such as "the costs of preparing journals for World Wide Web access through the Internet are substantially greater than the costs of distributing print." Questions regarding such statements abound:For example, are the one-time developmental costs front-loaded in these calculations, or are they amortized over the product's life cycle? If these claims are true,

then they reflect on the way chemical sciences publishers are using information technology, because other societies and several commercial publishers are able to reflect cost savings in changing from print to nonprint distribution. Although we do not have the detailed data at this time (this study is presently under way in our libraries), we expect to show that there are significant cost savings in terms of library staff productivity improvement when we distribute journals in nonprint versions instead of print.

As a result of these experiences, some of these publishers are giving us the impression that their narrowly circumscribed economic interests are dominating the evolution of digital libraries, that they are not fundamentally interested in controlling their internal costs through digital distribution, and that they are still pursuing tactical advantages over our libraries at the expense of a different set of strategic relationships with our scholarly communities. As is true about many generalizations, these are not universally held within the publishing community, but the overwhelming message seems clear nonetheless.

Conclusions

A digital distribution system for storing and accessing scholarly communications has been constructed and installed on the campuses of Case Western Reserve University and the University of Akron. This low-cost system can be extended to other institutions with similar requirements because the system components, together with the way they have been integrated, were chosen to facilitate the diffusion of these technologies. This distribution system successfully separates ownership of library materials from access to them.

The most interesting aspect of the new digital distribution system is that libraries can form consortia that can share specialized materials rather than duplicate them in parallel, redundant collections. When a consortium can share a single subscription to a highly specialized journal, then we have the basis for controlling and possibly reducing the total cost of library materials, because we can eliminate duplicative subscriptions. We believe that the future of academic libraries points to the maintenance of a basic core collection, the selective acquisition of specialty materials, and the sharing across telecommunications networks of standard scholarly works. The consortial model that we have built and tested is one way to accomplish this goal. Our approach is contrasted with the common behavior of building up ever larger collections of standard works so that over time, academic libraries begin to look alike in their collecting habits, offer almost duplicative services, and require larger budgets. This project is attempting to find another path.

Over the past decade, several interesting experiments have been conducted to test different ideas for developing digital libraries, and more are under way. With many differing ideas and visions, an empirical approach is a sound way to make

progress from this point forward. Our consortium model with its many explicit standards and integrated technologies seems to us to be an experiment worth continuing. During the next few years it will surely develop a base of performance data that should provide insights for the future. In this way, experience will benefit visioning.

References

Borghuis, M., Brinckman, H., Fischer, A., Hunter, K., van der Loo, E., Mors, R., Mostert, P., and Zilstra, J.: TULIP Final Report: T he U niversity LI censing P rogram. New York: Elsevier Science, 1996.

Bush, V.: "As We May Think," The Atlantic Monthly, 176, 101-108, 1945.

Cummings, A. M., Witte, M. L., Bowen, W. G., Lazarus, L. O., Ekman, R. H.: University Libraries and Scholarly Communication: A Study Prepared for The Andrew W. Mellon Foundation. The Association of Research Libraries, 1992.

Fleischhauer, C., and Erway, R. L.: Reproduction-Quality Issues in a Digital-Library System: Observations on the Reproduction of Various Library and Archival Material Formats for Access and Preservation. An American Memory White Paper, Washington, D.C.: Library of Congress, 1992.

Kemeny, J. G.: "A Library for 2000 A.D." in Greenberger, M. (Ed.), Computers and the World of the Future. Cambridge, Mass.: The M.I.T. Press, 1962.

Appendix A Consortial Standards

Marc

• Enumeration and chronology standards from the serials holding standards of the 853 and 863 fields of MARC

- Specifies up to 6 levels of enumeration and 4 levels of chronology, for example

853¦aVolume¦bIssue¦i(year)¦j(month)

853¦aVolume¦bIssue¦cPart¦i(year)¦j(month)

• Linking from bibliographic records in library catalog via an 856 field

-URL information appears in subfield u, anchor text appears in subfield z, for example

856 7¦uhttp://beavis.cwru.edu/chemvl¦zRetrieve articles from the Chemical Sciences Digital Library

Would appear as

Retrieve articles from the Chemical Sciences Digital Library

TIFF

• The most widely used multipage graphic format

• Support for tagged information ("Copyright", etc.)

• Format is extensible by creating new tags (such as RM rule information, authentication hints, encryption parameters)

• Standard supports multiple kinds of compression

Adobe PDF

• Container for article images

• Page description language (PDF)

• PDF files are searchable by the Adobe Acrobat browser

• Encryption and security are defined in the standard

SICI (Serial Item and Contribution Identifier)

• SICI definition (standards progress, overview, etc.)

• Originally a key part of the indexing structure

• All the components of the SICI code are stored, so it could be used as a linking mechanism between an article database and the Chemical Sciences Digital Library

• Ohio LINK is also very interested in this standard and is urging database creators and search engine providers to add SICI number retrieval to citation database and journal article repository systems

• Future retrieval interfaces into the database: SICI number search form, SICI number search API, for example

0022-2364(199607)121:1<83:TROTCI>2.0.TX;2-I

Appendix B Equipment Standards for End Users

Minimum Equipment Required

Hardware: An IBM PC or compatible computer with the following components:

• 80386 processor

• 16 MB RAM

• 20 MB free disk space

• A video card and monitor with a resolution of 640 ³ 480 and the capability of displaying 16 colors or shades of gray

Software:

• Windows 3.1

• Win32s 1.25

• TCP/IP software suite including a version of Winsock

• Netscape Navigator 2.02

• Adobe Acrobat Exchange 2.1

Win32s is a software package for Windows 3.1 that is distributed without charge and is available from Microsoft.

The requirement for Adobe Acrobat Exchange, a commercial product that is not distributed without charge, is expected to be relaxed in favor of a requirement for Adobe Acrobat Reader, a commercial product that is distributed without charge.

The software will also run on newer versions of compatible hardware and/or software.

Recommended Configuration of Equipment

This configuration is recommended for users who will be using the system extensively. Hardware: A computer with the following components:

• Intel Pentium processor

• 32 MB RAM

• 50 MB free disk space

• A video card and monitor with a resolution of 1280 ³ 1024 and the capability of displaying 256 colors or shades of gray

Software:

• Windows NT 4.0 Workstation

• TCP/IP suite that has been configured for a network connection (included in Windows NT)

• Netscape Navigator 2.02

• Adobe Acrobat Exchange 2.1

The requirement for Adobe Acrobat Exchange, a commercial product that is not distributed without charge, is expected to be relaxed in favor of a requirement for Adobe Acrobat Reader, a commercial product that is distributed without charge.

Other software options that the system has been tested on include:

• IBM OS/2 3.0 Warp Connect with Win-OS/2

• IBM TCP/IP for Windows 3.1, version 2.1.1

• Windows NT 3.51

Appendix C
Additional Hardware Specifications

Storage for Digital Copies

To give us the greatest possible flexibility in developing the project, we decided to form the server out of two interlinked computer systems, a standard IBM System 390 with the OS/390 Open Edition version as the operating system and a standard IBM RS/6000 System with the AIX version of the UNIX operating system. Both these components may be incrementally grown as the project's server requirements increase. Both systems are relatively commonplace at academic sites. Although only one system pair is needed in this project, it is likely that eventually two pairs of systems would be needed for an effort on the national scale. Such redundancy is useful for providing both reliability and load leveling.

Campuswide Networks

Both campus's networks and the statewide network that connects them uses the standards-based TCP/IP protocols. Any networked client workstation that follows our minimum standards will be able to use the digital delivery system being constructed. The minimum transmission speed on the CWRU campus is ten million bits per second (M bps) to each client workstation and a minimum of 155 M bps on each backbone link. The principal document repository is on the IBM System 390, which uses a 155 M bps ATM (asynchronous transfer mode) connection to the campus backbone. The linkage to the University of Akron is by way of the statewide network in which the principal backbone connection from CWRU is also operating at 155 M bps, and the linkage from the UA to the statewide network is at 3 M bps. The on-campus linkage for UA is also a minimum of 10 M bps to each client workstation within the chemical sciences scholarly community and to client workstations in the UA university library.

Appendix D System Transactions as Initiated by an End User

A typical user session generates the following transactions between client and server.

1. User requests an article (usually from a Web browser). If the user is starting a new session, the RM system downloads and launches the appropriate viewer, which will process only encrypted transactions. In the case of Adobe Acrobat, the system downloads a plug-in. The following transactions take place with the server:

a. Authenticate the viewer (i.e., ensure we are using a secure viewer).

b. Get permissions (i.e., obtain a set of user permissions, if any. If it is a new session, the user is set by default to be the general purpose category of PUBLIC).

c. Get Article (download the requested article. If step b returns no permissions, this transaction does not occur. The user must sign on and request the article again).

2. User signs on. If the general user has no permissions, he or she must log on. Following a successful logon, transactions 1b and 1c must be repeated. Transactions during sign-on include:

a. Sign On

3. Article is displayed on-screen. Before an article is displayed on the screen, the viewer enters the RM protocol, a step-by-step process wherein a single Report command is sent to the server several times with different state flags and use types. RM events are processed similarly for all supported functions, including display, print, excerpt, and download. The transactions include:

a. Report Use BEGIN (just before the article is displayed).

b. Report Use ABORT (sent in the event that a technical problem, such as "out of memory," prevents display of the article).

c. Report Use DECLINE (sent if the user declines display of the article after seeing the cost).

d. Report Use COMMIT (just after the article is displayed).

e. Report Use END (sent when the user dismisses the article from the screen by closing the article window).

4. User closes viewer. When a user closes a viewer, an end-of-session process occurs, which sends transaction 3e for all open articles. Also sent is a close viewer transaction, which immediately expires the viewer so it may not be used again.

a. Close Viewer

The basic data being collected for every command (with the exception of 1a) and being sent to the server for later analysis includes the following:

• Date/time

• Viewer ID

• User ID (even if it is PUBLIC)

• IP address of request

These primary data may be used to derive additional data: Transaction 1b is effectively used to log unsuccessful access attempts, including failure reasons. The time

interval between transactions 3a and 3e is used to measure the duration that an article is on the screen. The basic data collection module in the RM system is quite general and may be used to collect other information and derive other measures of system usage.

Appendix E
Scanning and Work Flow

Article Scanning, PDF Conversion, and Image Quality Control

The goal of the scan-and-store portion of the project is to develop a complete and tested system of hardware, software, and procedures that can be adopted by other members of the consortium with a reasonable investment in equipment, training, and personnel. If a system is beyond a consortium member's financial means, it will not be adopted. If a system cannot perform as required, it is a waste of resources.

Our original proposal stressed that all existing scholarly resources, particularly research tools, would remain available to scholars throughout this project. To that end, the scan-and-store process is designed to leave the consortium's existing journal collection intact and accessible.

Scan-and-Store Process Resources

• Scanning workstation, including a computer with sufficient processing and storage capacity, a scanner, and a network connection. Optionally, a second workstation can be used by the scanning supervisor to process the scanned images. The workstation used in this phase of the project includes:

-Minolta PS-3000 Digital Planetary Scanner

-Two computers with Pentium 200 MHz CPU, 64 MB RAM, 4 GB HD, 21" monitor

-Windows 3.11 OS (required by other software)

-Minolta Epic 3000 scanner software

-Adobe Acrobat Capture, Exchange, and Distiller software

-Image Alchemy software

-Network interface cards and TCP/IP software for campus network access

• Scanner operator(s), typically student assistants, with training roughly equivalent to that required for interlibrary loan photocopying. Approximately 8 hours of operator labor will be required to process the average 800 pages per day capacity of a single scanning workstation.

• Scanning supervisor, typically a librarian or full-time staff, with training in image quality control, indexing, and cataloging, and in operation of image processing software. Approximately 3 hours of supervisor labor will be required to process 800 scanned pages per day.

Scan-and-Store Process: Scanner Operator

• Retrieve scan request from system

• Retrieve materials from shelves (enough for two hours of scanning)

• Scan materials and enter basic data into system

-Evaluate size of pages

-Evaluate grayscale/black and white scan mode

-Align material

-Test scan and adjust settings and alignment as necessary

-Scan article

-Log changes and additions to author, title, journal, issue, and item data on request form

-Repeat for remaining requested articles

• Transfer scanned image files to Acrobat conversion workstation

• Retrieve next batch of scan requests from system

• Reshelve scanned materials and retrieve next batch of materials

Scan-and-Store Process: Acrobat Conversion Workstation

• Run Adobe Acrobat Capture to automatically convert sequential scanned image files from single-page TIFF to multi-page Acrobat PDF documents, as they are received from scanner operator

• Retain original TIFF files

Scan-and-Store Process: Scanning Supervisor

• Retrieve request forms for scanned materials

• Open converted PDF files

• Evaluate image quality of converted PDF files

-Scanned article matches request form citation

-Completeness, no clipped margins

-Legibility, especially footnotes and references

-Minimal skewing

-Clarity of grayscale or halftone images

-Appropriate margins, no excessive white space

• Crop fingertips, margin lines, and so on, missed by Epic 3000 scanner software

-Retrieve TIFF image file

-Mask unwanted areas

-Resave TIFF image file

-Repeat PDF conversion

-Evaluate image quality of revised PDF file

• Return unacceptable scans to scanner operator for rescan or correction

• Evaluate, correct, and expand entries in request forms

• Forward corrected PDF files to the database

• Delete TIFF image files from conversion workstation

Notification to and Viewing by User of Availability of Scanned Article

Insertion of the article into the database

• The scanning technician types the scan request number into a Web form.

• The system returns a Web form with most of the fields filled in. The technician has an opportunity to correct information from the paging slip before inserting the article into the database.

• The Web form contains a "file upload" button that when selected allows the technician to browse the local hard drive for the article PDF file. This file is automatically uploaded to the server when the form is submitted.

• The system inserts the table of contents information into the database and the PDF file to the Rights Manager system.

Notification/delivery of article to requester

• E-mail to requester with URL of requested article (in first release)

• No notification (in first release)

• Fax to requester an announcement page with the article URL (proposed future enhancement)

• Fax to requester a copy of the article (proposed future enhancement)

Appendix F
Technical Justification for a Digitization Standard for the Consortium

A major premise in the technical underpinnings of the new consortial model is that a relatively inexpensive scanner can be located in the academic libraries of consortium members. After evaluating virtually every scanning device on the market, including some in laboratories under development, we concluded that the 400 dot-per-inch (dpi) scanner from Minolta was fully adequate for the purpose of scanning all the hundreds of chemical sciences journals in which we were interested. Thus, for our consortium, the Minolta 400 dpi scanner was taken to be the

digitization standard. The standard that was adopted preserves 100% of the informational content required by our end users.

More formally, the standard for digitization in the consortium is defined as follows:

The scanner captures 256 levels of gray in a single pass with a density of 400 dots per inch and converts the grayscale image to black and white using threshold and edge-detection algorithms.

We arrived at this standard by considering our fundamental requirements:

• Handle the smallest significant information presented in the source documents of the chemical sciences literature, which is the lowercase e in superscript or subscript as occurs in footnotes

• Satisfy both legibility and fidelity to the source document

• Minimize scanning artifacts or "noise" from background

• Operate in the range of preservation scanning

• Be affordable by academic and research libraries

The scanning standard adopted by this project was subjected to tests of footnoted information, and 100% of the occurrences of these characters were captured in both image and character modes and recognized for displaying and searching.

At 400 dpi, the Minolta scanner works in the range of preservation quality scanning as defined by researchers at the Library of Congress (Fleischhauer and Erway 1992).

We were also cautioned about the problems unique to very high resolution scanning in which the scanner produces artifacts or "noise" from imperfections in the paper used. We happily note that we did not encounter this problem in this project because the paper used by publishers of chemical sciences journals is coated.

When more is less: images scanned at 600 dpi require larger file sizes than those scanned at 400 dpi. Thus, 600 dpi is less efficient than 400 dpi. Further, in a series of tests that we conducted, a 600 dpi scanner actually produced an image of effectively lower resolution than 400 dpi. It appears that this loss of information occurs when the scanned image is viewed on a computer screen where there is relatively heavy use of anti-aliasing in the display. When viewed with software that permitted zooming in for looking at details of the scanned image (which is supported by both PDF and TIFF viewers), the 600 dpi anti-aliased image actually had lower resolution than did an image produced from the same source document by the 400 dpi Minolta scanner according to our consortium's digitization standard. With the 600 dpi scanner, the only way for the end user to see the full resolution was to download the image and then print it out. When a side-by-side comparison was made of the soft-copy displayed images, the presentation image quality of 600 dpi was deemed unacceptable by our end users; the 400 dpi image was just right. Thus, our delivery approach is more useful to the scholar who needs to examine

fine details on-screen. We conducted some tests on reconstructing the journal page from the scanned image by printing it out on a Xerox DocuTech 6135 (600 dpi). We found that the smallest fonts and fine details of the articles were uniformly excellent. Interestingly, in many of the tests we performed, our faculty colleagues judged the end result by their own "acid test": how the scanned image, when printed out, compared with the image produced by a photocopier. For the consortium standard, they were satisfied with the result and pleased with the improvement in quality that the 400 dpi scanner provided in comparison with conventional photocopying of the journal page.