Part 4

Drawings are postscript files, not meeting international standards, but at least designed to go across platforms. Images, or rather the real archival forms, consist of the best available slides, which are being digitized. Much of the catalog material exists in database form--a form that the average user could use, manipulate, and display on a personal computer, but only at great cost. Thus, this is where the concession comes in: All of this rich, well-marked-up information is stripped of much of its content; the images are converted into bit-maps and the text into small formatted chunks. All this information can then be imported into HyperCard and run on a mid-range Macintosh, which is what Perseus users have. This fact has made it possible for Perseus to attain wide use fairly rapidly. Without those archival forms the HyperCard version being demonstrated could not be made easily, and the project could not have the potential to move to other forms and machines and software as they appear, none of which information is in Perseus on the CD.

Of the numerous multimedia aspects of Perseus, MYLONAS focused on the textual. Part of what makes Perseus such a pleasure to use, MYLONAS said, is this effort at seamless integration and the ability to move around both visual and textual material. Perseus also made the decision not to attempt to interpret its material any more than one interprets by selecting. But, MYLONAS emphasized, Perseus is not courseware: No syllabus exists. There is no effort to define how one teaches a topic using Perseus, although the project may eventually collect papers by people who have used it to teach. Rather, Perseus aims to provide primary material in a kind of electronic library, an electronic sandbox, so to say, in which students and scholars who are working on this material can explore by themselves. With that, MYLONAS demonstrated Perseus, beginning with the Perseus gateway, the first thing one sees upon opening Perseus--an effort in part to solve the contextualizing problem--which tells the user what the system contains.

MYLONAS demonstrated only a very small portion, beginning with primary texts and running off the CD-ROM. Having selected Aeschylus' Prometheus Bound, which was viewable in Greek and English pretty much in the same segments together, MYLONAS demonstrated tools to use with the Greek text, something not possible with a book: looking up the dictionary entry form of an unfamiliar word in Greek after subjecting it to Perseus' morphological analysis for all the texts. After finding out about a word, a user may then decide to see if it is used anywhere else in Greek. Because vast amounts of indexing support all of the primary material, one can find out where else all forms of a particular Greek word appear-- often not a trivial matter because Greek is highly inflected. Further, since the story of Prometheus has to do with the origins of sacrifice, a user may wish to study and explore sacrifice in Greek literature; by typing sacrifice into a small window, a user goes to the English-Greek word list--something one cannot do without the computer (Perseus has indexed the definitions of its dictionary)--the string sacrifice appears in the definitions of these sixty-five words. One may then find out where any of those words is used in the work(s) of a particular author. The English definitions are not lemmatized.

All of the indices driving this kind of usage were originally devised for speed, MYLONAS observed; in other words, all that kind of information-- all forms of all words, where they exist, the dictionary form they belong to--were collected into databases, which will expedite searching. Then it was discovered that one can do things searching in these databases that could not be done searching in the full texts. Thus, although there are full-text searches in Perseus, much of the work is done behind the scenes, using prepared indices. Re the indexing that is done behind the scenes, MYLONAS pointed out that without the SGML forms of the text, it could not be done effectively. Much of this indexing is based on the structures that are made explicit by the SGML tagging.

It was found that one of the things many of Perseus' non-Greek-reading users do is start from the dictionary and then move into the close study of words and concepts via this kind of English-Greek word search, by which means they might select a concept. This exercise has been assigned to students in core courses at Harvard--to study a concept by looking for the English word in the dictionary, finding the Greek words, and then finding the words in the Greek but, of course, reading across in the English. That tells them a great deal about what a translation means as well.

Should one also wish to see images that have to do with sacrifice, that person would go to the object key word search, which allows one to perform a similar kind of index retrieval on the database of archaeological objects. Without words, pictures are useless; Perseus has not reached the point where it can do much with images that are not cataloged. Thus, although it is possible in Perseus with text and images to navigate by knowing where one wants to end up--for example, a red-figure vase from the Boston Museum of Fine Arts--one can perform this kind of navigation very easily by tracing down indices. MYLONAS illustrated several generic scenes of sacrifice on vases. The features demonstrated derived from Perseus 1.0; version 2.0 will implement even better means of retrieval.

MYLONAS closed by looking at one of the pictures and noting again that one can do a great deal of research using the iconography as well as the texts. For instance, students in a core course at Harvard this year were highly interested in Greek concepts of foreigners and representations of non-Greeks. So they performed a great deal of research, both with texts (e.g., Herodotus) and with iconography on vases and coins, on how the Greeks portrayed non-Greeks. At the same time, art historians who study iconography were also interested, and were able to use this material.

******

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ DISCUSSION * Indexing and searchability of all English words in Perseus * Several features of Perseus 1.0 * Several levels of customization possible * Perseus used for general education * Perseus' effects on education * Contextual information in Perseus * Main challenge and emphasis of Perseus * +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

Several points emerged in the discussion that followed MYLONAS's presentation.

Although MYLONAS had not demonstrated Perseus' ability to cross-search documents, she confirmed that all English words in Perseus are indexed and can be searched. So, for example, sacrifice could have been searched in all texts, the historical essay, and all the catalogue entries with their descriptions--in short, in all of Perseus.

Boolean logic is not in Perseus 1.0 but will be added to the next version, although an effort is being made not to restrict Perseus to a database in which one just performs searching, Boolean or otherwise. It is possible to move laterally through the documents by selecting a word one is interested in and selecting an area of information one is interested in and trying to look that word up in that area.

Since Perseus was developed in HyperCard, several levels of customization are possible. Simple authoring tools exist that allow one to create annotated paths through the information, which are useful for note-taking and for guided tours for teaching purposes and for expository writing. With a little more ingenuity it is possible to begin to add or substitute material in Perseus.

Perseus has not been used so much for classics education as for general education, where it seemed to have an impact on the students in the core course at Harvard (a general required course that students must take in certain areas). Students were able to use primary material much more.

The Perseus Project has an evaluation team at the University of Maryland that has been documenting Perseus' effects on education. Perseus is very popular, and anecdotal evidence indicates that it is having an effect at places other than Harvard, for example, test sites at Ball State University, Drury College, and numerous small places where opportunities to use vast amounts of primary data may not exist. One documented effect is that archaeological, anthropological, and philological research is being done by the same person instead of by three different people.

The contextual information in Perseus includes an overview essay, a fairly linear historical essay on the fifth century B.C. that provides links into the primary material (e.g., Herodotus, Thucydides, and Plutarch), via small gray underscoring (on the screen) of linked passages. These are handmade links into other material.

To different extents, most of the production work was done at Harvard, where the people and the equipment are located. Much of the collaborative activity involved data collection and structuring, because the main challenge and the emphasis of Perseus is the gathering of primary material, that is, building a useful environment for studying classical Greece, collecting data, and making it useful. Systems-building is definitely not the main concern. Thus, much of the work has involved writing essays, collecting information, rewriting it, and tagging it. That can be done off site. The creative link for the overview essay as well as for both systems and data was collaborative, and was forged via E-mail and paper mail with professors at Pomona and Bowdoin.

******

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ CALALUCA * PLD's principal focus and contribution to scholarship * Various questions preparatory to beginning the project * Basis for project * Basic rule in converting PLD * Concerning the images in PLD * Running PLD under a variety of retrieval softwares * Encoding the database a hard-fought issue * Various features demonstrated * Importance of user documentation * Limitations of the CD-ROM version * +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

Eric CALALUCA, vice president, Chadwyck-Healey, Inc., demonstrated a software interpretation of the Patrologia Latina Database (PLD). PLD's principal focus from the beginning of the project about three-and-a-half years ago was on converting Migne's Latin series, and in the end, CALALUCA suggested, conversion of the text will be the major contribution to scholarship. CALALUCA stressed that, as possibly the only private publishing organization at the Workshop, Chadwyck-Healey had sought no federal funds or national foundation support before embarking upon the project, but instead had relied upon a great deal of homework and marketing to accomplish the task of conversion.

Ever since the possibilities of computer-searching have emerged, scholars in the field of late ancient and early medieval studies (philosophers, theologians, classicists, and those studying the history of natural law and the history of the legal development of Western civilization) have been longing for a fully searchable version of Western literature, for example, all the texts of Augustine and Bernard of Clairvaux and Boethius, not to mention all the secondary and tertiary authors.

Various questions arose, CALALUCA said. Should one convert Migne? Should the database be encoded? Is it necessary to do that? How should it be delivered? What about CD-ROM? Since this is a transitional medium, why even bother to create software to run on a CD-ROM? Since everybody knows people will be networking information, why go to the trouble--which is far greater with CD-ROM than with the production of magnetic data? Finally, how does one make the data available? Can many of the hurdles to using electronic information that some publishers have imposed upon databases be eliminated?

The PLD project was based on the principle that computer-searching of texts is most effective when it is done with a large database. Because PLD represented a collection that serves so many disciplines across so many periods, it was irresistible.

The basic rule in converting PLD was to do no harm, to avoid the sins of intrusion in such a database: no introduction of newer editions, no on-the-spot changes, no eradicating of all possible falsehoods from an edition. Thus, PLD is not the final act in electronic publishing for this discipline, but simply the beginning. The conversion of PLD has evoked numerous unanticipated questions: How will information be used? What about networking? Can the rights of a database be protected? Should one protect the rights of a database? How can it be made available?

Those converting PLD also tried to avoid the sins of omission, that is, excluding portions of the collections or whole sections. What about the images? PLD is full of images, some are extremely pious nineteenth-century representations of the Fathers, while others contain highly interesting elements. The goal was to cover all the text of Migne (including notes, in Greek and in Hebrew, the latter of which, in

## particular, causes problems in creating a search structure), all the

indices, and even the images, which are being scanned in separately searchable files.

Several North American institutions that have placed acquisition requests for the PLD database have requested it in magnetic form without software, which means they are already running it without software, without anything demonstrated at the Workshop.

What cannot practically be done is go back and reconvert and re-encode data, a time-consuming and extremely costly enterprise. CALALUCA sees PLD as a database that can, and should, be run under a variety of retrieval softwares. This will permit the widest possible searches. Consequently, the need to produce a CD-ROM of PLD, as well as to develop software that could handle some 1.3 gigabyte of heavily encoded text, developed out of conversations with collection development and reference librarians who wanted software both compassionate enough for the pedestrian but also capable of incorporating the most detailed lexicographical studies that a user desires to conduct. In the end, the encoding and conversion of the data will prove the most enduring testament to the value of the project.

The encoding of the database was also a hard-fought issue: Did the database need to be encoded? Were there normative structures for encoding humanist texts? Should it be SGML? What about the TEI--will it last, will it prove useful? CALALUCA expressed some minor doubts as to whether a data bank can be fully TEI-conformant. Every effort can be made, but in the end to be TEI-conformant means to accept the need to make some firm encoding decisions that can, indeed, be disputed. The TEI points the publisher in a proper direction but does not presume to make all the decisions for him or her. Essentially, the goal of encoding was to eliminate, as much as possible, the hindrances to information-networking, so that if an institution acquires a database, everybody associated with the institution can have access to it.

CALALUCA demonstrated a portion of Volume 160, because it had the most anomalies in it. The software was created by Electronic Book Technologies of Providence, RI, and is called Dynatext. The software works only with SGML-coded data.

Viewing a table of contents on the screen, the audience saw how Dynatext treats each element as a book and attempts to simplify movement through a volume. Familiarity with the Patrologia in print (i.e., the text, its source, and the editions) will make the machine-readable versions highly useful. (Software with a Windows application was sought for PLD, CALALUCA said, because this was the main trend for scholarly use.)

CALALUCA also demonstrated how a user can perform a variety of searches and quickly move to any part of a volume; the look-up screen provides some basic, simple word-searching.

CALALUCA argued that one of the major difficulties is not the software. Rather, in creating a product that will be used by scholars representing a broad spectrum of computer sophistication, user documentation proves to be the most important service one can provide.

CALALUCA next illustrated a truncated search under mysterium within ten words of virtus and how one would be able to find its contents throughout the entire database. He said that the exciting thing about PLD is that many of the applications in the retrieval software being written for it will exceed the capabilities of the software employed now for the CD-ROM version. The CD-ROM faces genuine limitations, in terms of speed and comprehensiveness, in the creation of a retrieval software to run it. CALALUCA said he hoped that individual scholars will download the data, if they wish, to their personal computers, and have ready access to important texts on a constant basis, which they will be able to use in their research and from which they might even be able to publish.

(CALALUCA explained that the blue numbers represented Migne's column numbers, which are the standard scholarly references. Pulling up a note, he stated that these texts were heavily edited and the image files would appear simply as a note as well, so that one could quickly access an image.)

******

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ FLEISCHHAUER/ERWAY * Several problems with which AM is still wrestling * Various search and retrieval capabilities * Illustration of automatic stemming and a truncated search * AM's attempt to find ways to connect cataloging to the texts * AM's gravitation towards SGML * Striking a balance between quantity and quality * How AM furnishes users recourse to images * Conducting a search in a full-text environment * Macintosh and IBM prototypes of AM * Multimedia aspects of AM * +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

A demonstration of American Memory by its coordinator, Carl FLEISCHHAUER, and Ricky ERWAY, associate coordinator, Library of Congress, concluded the morning session. Beginning with a collection of broadsides from the Continental Congress and the Constitutional Convention, the only text collection in a presentable form at the time of the Workshop, FLEISCHHAUER highlighted several of the problems with which AM is still wrestling. (In its final form, the disk will contain two collections, not only the broadsides but also the full text with illustrations of a set of approximately 300 African-American pamphlets from the period 1870 to 1910.)

As FREEMAN had explained earlier, AM has attempted to use a small amount of interpretation to introduce collections. In the present case, the contractor, a company named Quick Source, in Silver Spring, MD., used software called Toolbook and put together a modestly interactive introduction to the collection. Like the two preceding speakers, FLEISCHHAUER argued that the real asset was the underlying collection.

FLEISCHHAUER proceeded to describe various search and retrieval capabilities while ERWAY worked the computer. In this particular package the "go to" pull-down allowed the user in effect to jump out of Toolbook, where the interactive program was located, and enter the third-party software used by AM for this text collection, which is called Personal Librarian. This was the Windows version of Personal Librarian, a software application put together by a company in Rockville, Md.

Since the broadsides came from the Revolutionary War period, a search was conducted using the words British or war, with the default operator reset as or. FLEISCHHAUER demonstrated both automatic stemming (which finds other forms of the same root) and a truncated search. One of Personal Librarian's strongest features, the relevance ranking, was represented by a chart that indicated how often words being sought appeared in documents, with the one receiving the most "hits" obtaining the highest score. The "hit list" that is supplied takes the relevance ranking into account, making the first hit, in effect, the one the software has selected as the most relevant example.

While in the text of one of the broadside documents, FLEISCHHAUER remarked AM's attempt to find ways to connect cataloging to the texts, which it does in different ways in different manifestations. In the case shown, the cataloging was pasted on: AM took MARC records that were written as on-line records right into one of the Library's mainframe retrieval programs, pulled them out, and handed them off to the contractor, who massaged them somewhat to display them in the manner shown. One of AM's questions is, Does the cataloguing normally performed in the mainframe work in this context, or had AM ought to think through adjustments?

FLEISCHHAUER made the additional point that, as far as the text goes, AM has gravitated towards SGML (he pointed to the boldface in the upper part of the screen). Although extremely limited in its ability to translate or interpret SGML, Personal Librarian will furnish both bold and italics on screen; a fairly easy thing to do, but it is one of the ways in which SGML is useful.

Striking a balance between quantity and quality has been a major concern of AM, with accuracy being one of the places where project staff have felt that less than 100-percent accuracy was not unacceptable. FLEISCHHAUER cited the example of the standard of the rekeying industry, namely 99.95 percent; as one service bureau informed him, to go from 99.95 to 100 percent would double the cost.

FLEISCHHAUER next demonstrated how AM furnishes users recourse to images, and at the same time recalled LESK's pointed question concerning the number of people who would look at those images and the number who would work only with the text. If the implication of LESK's question was sound, FLEISCHHAUER said, it raised the stakes for text accuracy and reduced the value of the strategy for images.

Contending that preservation is always a bugaboo, FLEISCHHAUER demonstrated several images derived from a scan of a preservation microfilm that AM had made. He awarded a grade of C at best, perhaps a C minus or a C plus, for how well it worked out. Indeed, the matter of learning if other people had better ideas about scanning in general, and, in particular, scanning from microfilm, was one of the factors that drove AM to attempt to think through the agenda for the Workshop. Skew, for example, was one of the issues that AM in its ignorance had not reckoned would prove so difficult.

Further, the handling of images of the sort shown, in a desktop computer environment, involved a considerable amount of zooming and scrolling. Ultimately, AM staff feel that perhaps the paper copy that is printed out might be the most useful one, but they remain uncertain as to how much on-screen reading users will do.

Returning to the text, FLEISCHHAUER asked viewers to imagine a person who might be conducting a search in a full-text environment. With this scenario, he proceeded to illustrate other features of Personal Librarian that he considered helpful; for example, it provides the ability to notice words as one reads. Clicking the "include" button on the bottom of the search window pops the words that have been highlighted into the search. Thus, a user can refine the search as he or she reads, re-executing the search and continuing to find things in the quest for materials. This software not only contains relevance ranking, Boolean operators, and truncation, it also permits one to perform word algebra, so to say, where one puts two or three words in parentheses and links them with one Boolean operator and then a couple of words in another set of parentheses and asks for things within so many words of others.

Until they became acquainted recently with some of the work being done in classics, the AM staff had not realized that a large number of the projects that involve electronic texts were being done by people with a profound interest in language and linguistics. Their search strategies and thinking are oriented to those fields, as is shown in particular by the Perseus example. As amateur historians, the AM staff were thinking more of searching for concepts and ideas than for particular words. Obviously, FLEISCHHAUER conceded, searching for concepts and ideas and searching for words may be two rather closely related things.

While displaying several images, FLEISCHHAUER observed that the Macintosh prototype built by AM contains a greater diversity of formats. Echoing a previous speaker, he said that it was easier to stitch things together in the Macintosh, though it tended to be a little more anemic in search and retrieval. AM, therefore, increasingly has been investigating sophisticated retrieval engines in the IBM format.