Figure 2.1 TLG Search, May 17, 2006.
Writing, Phaedrus, has this strange quality, and is very like painting; for the creatures of painting stand like living beings, but if one asks them a question, they preserve a solemn silence. And so it is with written words; you might think they spoke as if they had intelligence, but if you question them, wishing to know about their sayings, they always say only one and the same thing. (Plato, Phaedrus 275d)
This chapter suggests directions in which an ePhilology may evolve. Philology here implies that language and literature are the objects of study but assumes that language and literature must draw upon the full cultural context and thus sees in philological analysis a starting point for the scientia totius antiquitatis – the systematic study of all ancient culture. The term ePhilology implicitly states that, while our strategic goal may remain the scientia totius antiquitatis, the practices whereby we pursue this strategic goal must evolve into something qualitatively different from the practices of the past.
Digital technology is hardly new in classics: there are full professors today who have always searched large bodies of Greek and Latin, composed their ideas in an electronic form, found secondary sources online and opportunistically exploited whatever digital tools served their purposes.2 Nevertheless, the inertia of prior practice has preserved intact the forms that evolved to exploit the strengths and minimize the weaknesses of print culture: we create documents that slavishly mimic their print predecessors; we send these documents to the same kinds of journals and publishers;3 our reference works and editions have already begun to drift out of date before they are published and stagnate thereafter; even when new, our publications are static and cannot adapt themselves to the needs of their varying users; while a growing, global audience could now find the results of our work, we embed our ideas in specialized language and behind subscription barriers which perpetuate into the twenty-first century the miniscule audiences of the twentieth.4
This chapter makes two fundamental arguments. First, it assumes that the first generation of digital technology has only laid the groundwork for substantive change in classics and the humanities. Second, it advances arguments about what form an optimal digital future should assume. While Greek and Latin provide the focus for this chapter, the arguments apply in various ways to many areas within the humanities.
At least six features distinguish emerging digital resources: (1) they can be delivered to any point on the earth and at any time; (2) they can be fundamentally hypertextual, supporting comprehensive links between assertions and their evidence; (3) they dynamically recombine small, well-defined units of information to serve particular people at particular times; (4) they learn on their own and apply as many automated processes as possible, not only automatic indexing but morphological and syntactic analysis, named entity recognition, knowledge extraction, machine translation, etc., with changes in automatically generated results tracked over time; (5) they learn from their human readers and can make effective use of contributions, explicit and implicit, from a range of users in real time; (6) they automatically adapt themselves to the general background and current purposes of their users.
Print culture gave us expensive distribution by which we could send static documents to a few thousand restricted locations. If we can deliver information to any point on the earth and we can tailor that information to the varying backgrounds and immediate purposes of many people, we can thus address audiences far beyond the physical and, indeed, cultural limitations which communication – oral and print – has imposed.
In the Phaedrus, Plato’s Socrates, a fictive rendering of a historical character scratched into life by pen and preserved as a pattern of ink, critiques writing – and thus the very medium in which he exerts a living presence to this day. A generation ago, Derrida famously expanded upon the observation that writing is not so much a cure as a poison for memory5 – in a look-up culture not only do our memories decay but we lose in some measure that instant and deep recognition which integrated knowledge alone can spark. The critique in the Phaedrus is profound and addresses all technologies which represent information abstracted not only from the brain but also from the personal context in which much learning occurs. Plato’s arguments have been echoed ever since, consciously or not – many of us in the first generation of a television society heard similar criticisms from our parents and, in turn, directed these to our net-oriented children.6
The quote that begins this essay, however, directs a criticism which is just as trenchant but has attracted less attention. All products of information technology – paintings and poems, novels and newspapers, movies and music – have been static since our ancestors first scratched diagrams in the dirt or pressed visions of their world on the walls of caves. Other human hands could add or destroy, but the products of our hands could do nothing but decay, prey to the scorching sun, the worm, or the slow fires of acid within. We can direct our questions to the written word or to the most lifelike painting, but we can expect only silence.
Now, however, we have created cultural products that can respond, systems that can change and adapt themselves to our needs. Millions of people around the world will, on the day that I compose these words, seek directions from a mapping service. Natural language, mathematical formulae, and visual representations of space will interact to generate tailored itineraries, with estimates of time and customizable maps illustrating the journey from one point to another, in some cases speaking their directions in an expanding suite of languages. We should not confuse the humble and well-defined goals of such tools with their significance in the evolution of humanity and indeed life.
The great question that we face is not what we can do but what we want to accomplish. The tools at our disposal today, primitive as they may appear in the future, are already adequate to create a dynamic space for intellectual life as different from what precedes it as oral culture differs from a world of writing. At one level, little will change – the Homeric epics, products of an oral culture ironically preserved in writing, are arguably as successful as cultural products as anything that followed: the ceiling of human creativity has not changed in three thousand years of increasingly sophisticated information technology – an observation that we should consider as we fret over the codex and print.7
Nevertheless, we can now plan for a world where ideas cross from language to language and from culture to culture with a speed and authenticity far beyond what we have ever experienced. Consider curious minds in Beijing or Damascus a generation from now who encounter something that sparks their interest in the Greco-Roman world. It could be a film or a popular novel translated into Chinese or Arabic or a game that carries them through a virtual space. It could even be something in their formal education which, as occasionally happens, fires their imagination. The internet as we have it has already increased the chances for such encounters and provided unprecedented opportunities for Beijing and Damascus to learn about ancient Greece and Rome or other cultures.
We can, however, do more. The intellectually alive mind asks about a Greek author, perhaps a widely translated one (such as Homer) or perhaps not. Background information and the text itself are translated into the Chinese or Arabic. The inquirer has developed a profile, not unlike her medical history, which can record the classes she has taken, the books she has read, the movies she has seen, the games she has played, and the questions that she has posed.8 The personal reading agent can compare this profile, eagerly developed and shared only in part and under strict conditions, against the cultural referents implicit in the author or text of interest, then produce not only translations but personalized briefing materials – maps, timelines, diagrams, simulations, glossary entries – to help that reader contextualize what she has encountered. As the reader begins to ask questions, the system refines its initial hypotheses, quickly adapting itself to her needs.9 As the system changes, it inspires new kinds of inquiry in the reader, creating a feedback loop that encourages their conversation to evolve. Far from the static and one-sided interaction of Plato’s complaint, this is the definition of dialectic.
As this chapter will suggest, we already possess the technology to build a system of this type that will be effective in many cases: the professional classicist moving into early modern Latin or even tracking developments in his or her own field, with text mining identifying trends in the secondary literature or phenomena in the source texts.10
The question that we face is much deeper than the challenge of producing more or, preferably, better articles and monographs. We must more generally ask what kind of space we wish to produce in which to explore the linguistic record of humanity – whether we are contemplating the Odyssey, administrative records from Sumer, or tracing mathematical thought through Greek and Arabic sources. More important perhaps than the question of what we can do may be the opportunity to redefine who can do what – to open up intellectual life more broadly than ever before and to create a fertile soil in which humanity can cultivate the life of the mind with greater vigor and joy.
The systematic application of computing technology to classical languages began in 1968, when David Packard had toiled with primitive computing in the basement of Harvard’s Science Center to produce a full concordance to Livy. The resulting massive print volumes were both a fundamental new tool and a staple at Harvard University Press remainder sales of the 1970s, illustrating both the potential of even simple electronic tools and the limitations of the codex. Three fundamental developments quickly followed.
First, the Thesaurus Linguae Graecae (TLG),11 founded in 1972, began developing what would be called a digital library of classical Greek literature. A third of a century later, the TLG has completed its initial goal of digitizing all published Greek literature up through ad 600 and has extended its coverage through the Byzantine period and beyond.12 The TLG thus provided the first digital well-curated collection of digital resources in classics.
Second, David Packard began in the 1970s to develop a system not only to work with collections such as the TLG but to provide the first computerized typesetting and word processing for Greek.13 At the Boston APA convention of 1979, for example, Packard could show a working Ibycus computer system. Based on a Hewlett-Packard minicomputer, the Ibycus included a unique operating system designed for classics. The Ibycus was, by the standards of the early twenty-first century, astonishingly expensive – it cost tens of thousands of dollars – but it provided scholars with services they needed not only to exploit the TLG but to write and publish. Its contributions were so important that more than a dozen departments raised the necessary capital.
Third, the TLG and Ibycus system were the products of two distinct organizations, thus promoting a separation of data from service providers and opening the way for a range of entrepreneurs to create additional services and solutions.14 The TLG website lists more than a dozen packages that were developed to work with the CD ROM texts.
A generation later, classicists still depend upon texts and services designed in the 1970s. Figure 2.1 illustrates the results from a sample search of the TLG in May 2006 as suggested on the TLG website. The system reflects decades of investment, both from subscriptions and from grants (e.g., a $235,000 grant in 2000 from the National Endowment for the Humanities that provided partial support for “restructuring of data and development of an online search and retrieval system for the Thesaurus Linguae Graecae.”15) The resulting in-house system provides a fast, reliable service on which Hellenists depend, especially since the TLG no longer updates its CD ROM and thus does not generally distribute source texts published after the February 2000 TLG E Disk.16
It would be hard to overstate the importance of searchable text corpora. Classicists are also fortunate to have access to the Packard Humanities Institute CD ROM for Latin literature, as well as proprietary commercial databases such as the Biblioteca Teubneriana Latina.17 Classicists have become accustomed to scanning wide swathes of Greek and Latin literature, with full professors today who have never known a world without searchable texts. Many take for granted this core infrastructure and, when asked, admit that these tools have had far more impact upon the questions that they ask and the research that they conduct than they readily articulate. An analysis of primary source citations in the classics journals of JSTOR would give us a better appreciation of the impact which these collections have had upon published scholarship.
Figure 2.1 TLG Search, May 17, 2006.
In the past thirty years more texts have been added but the essential services and underlying data model visible to the classical community have not changed. The TLG, as it appeared in May 2006, is selected for analysis because it has successfully served, and continues to serve, the field and provides a standard of excellence, in terms of continuity and quality of service, but the analysis offered below applies to many efforts in classics and the humanities. The goal is not to diminish the importance of what TLG and projects like it have contributed but, by describing the state of the art as it existed when this chapter was written, to suggest future movements for classics and the humanities.
The limitations described above have been acceptable because they support the practices of print culture. Textual corpora such as the TLG, whether on the web or on CD ROM, are immense, dynamic, flexible concordances. They thus support traditional work but also provide no incentive for innovative forms of publication. The monolithic website isolates classicists from the electronic infrastructure which supports them. If our goal is to produce more and better researched articles and monographs – if we think that the answer to the crisis in academic monographs is to produce more content – then the status quo will serve us well.34
At this point, we return to the six features that, at least in part, distinguish digital from print publication. While work remains at an early stage of development, progress is being made in all six areas. The following section illustrates these points primarily with work done associated with Perseus for classics, but Perseus and the field of classics are only components of a much larger process.35
Library subscription budgets shield many scholars – especially those at the most prestigious institutions – from the economic realities with which libraries struggle. Many – probably most – do not realize that the scholarly resources – much of it in the public domain – on which they daily rely are available only through expensive subscriptions. Various open access movements have attacked this problem – rarely with support, not infrequently with scorn, from academics: Project Gutenberg began in 1971 (one year before the TLG), hosts a library of 18,000 public domain books and downloads two million of these each month.36 More recently, Google Library and the Open Content Alliance (OCA) have set out to digitize the entire published record of humanity. Each pursues contrasting rights regimes: Google retains its collection for its proprietary use, while the OCA is building an open source collection: Yahoo and Microsoft are both backing OCA, with each planning to provide its own set of unique services to the shared content. Both Google Library and the OCA are, however, open access – the business models of Google, Yahoo, and Microsoft all depend upon maximizing their audiences.37 Open access seems to them to be a better engine for revenue generation than subscription models.38
Within classics, the Latin Library dramatized the hunger for open-source primary materials. Frustrated by proprietary text corpora, members of the community, most from outside of academia, have spontaneously digitized almost all classical Latin, and a growing body of post-classical Latin literature and made it freely accessible at a single site.39 It is easy to criticize this work: original scholarship resides along with texts bearing the unnerving label “from an unidentified edition,” while other texts combine multiple editions without substantive documentation.40 The site reflects a widespread and heart-felt desire to assemble a critical mass of freely accessible Latin texts. While professional scholars can criticize some of the texts, we should also ask ourselves why the community felt it necessary to do so much work to establish such a basic service. Were the publications that we composed with proprietary databases a greater contribution to intellectual life than a universally accessible library of primary texts?
From the beginning of its web presence in 1995, Perseus provided open access to all of its holdings not otherwise restricted by third party rights.41 More recently, members of the community – especially the rising generation of classicists – have argued forcefully that all core materials should be available under an open-source license, allowing third parties to repurpose what we have begun. We have thus moved beyond open access and to open source for all materials to which we have rights. We chose a Creative Commons attribution/share-alike/non-commercial license.42 Third parties may thus freely create new resources based on what we provide but they must make their additions available under the same terms and they cannot restrict access to these resources behind a subscription barrier. The non-commercial license does not exclude advertising-based revenue and we hope that internet services such as Google, Yahoo, and Microsoft will load everything that we produce into their collections.
Since spring 2005, we have provided a web service that exposes well-formed chunks of our data to third parties. In March 2006, we have made available under a Creative Commons license the TEI-compliant XML files for the Greek and Latin source texts that we have created that were based upon public domain editions: c. 13,000,000 words of text. While this collection is much smaller than the 76,000,000 words on the 2000 TLG E Disk or the 91,000,000 words on the spring 2006 TLG website, it does already contain most of classical Greek and many classical Latin source texts. All of our unencumbered lexica, encyclopedias, commentaries, and other reference materials will follow suit and be released under the same license. Likewise, all components of the new digital library system that underlies Perseus are being written for open source distribution and will, we hope, be integrated into the next generation of digital library systems.
As with access, hypertextual documents depend upon policy – even web links, primitive though they may be, provide a starting point. Classicist Christopher Blackwell has produced what may be the best example of a publication that bridges the gap between traditional print and densely hypertextual web publication. He produced an electronic publication as his tenure book, a website that surveys Athenian democracy.43 Figure 2.2 illustrates a snapshot of this site. The site includes not only PDF visualizations of the text optimized for print but also HTML representations of the same documents. The HTML documents contain a dense set of primary source citations that are filtered out of the print-oriented PDF publications. Blackwell has striven to provide the primary source evidence behind every significant assertion. The secondary scholarship on this subject has grown so tangled that many publications simply cite other secondary scholarship, leaving readers to dig through multiple sources before they can assess the underlying evidence. Blackwell’s publication assumes the presence of a stable, comprehensive digital library to make the citations actionable links.
Figure 2.2 Hypertextual writing from Christopher Blackwell’s Demos, which illustrates a genuine step beyond print monographs.
Hypertextual writing builds on ubiquitous access to source materials. We can create hypertextual documents with links to subscription-based resources, but in so doing we implicitly define an audience of academics and a handful of committed non-professionals with access to good libraries. Hypertextual writing hidden from the outside world behind subscription barriers cannot, of course, reach beyond academic elites. Dense hypertextual links that are in open-access publications but that point to academic subscription-based sources have no more impact on society as a whole than citations to print-only resources. Only open access publications with links to open-access sources can increase the transparency of what we in the humanities do and engage a broader audience in the intellectual discourse that we pursue.44
Aside from the content, Blackwell’s work demonstrates the potential of the form and exhibits a scholarly leadership badly needed within the humanities. Had he worked with a conventional academic publisher he might have earned greater conventional prestige, but he would have reached a smaller audience and would probably not have had the freedom to create expository texts so well adapted to the digital environment.
We need compound documents, dynamically generated to serve particular users at particular times, that draw upon materials from a range of sources to create a new, unified whole.45 Such documents have two requirements:
Figure 2.3 illustrates an entry from the Liddell Scott Jones Greek English Lexicon (LSJ 9)48 Notice that the mention of “Pi. Pae.” has not been expanded to a textual form but has been linked instead to an authority list (in this case, the numeration of the TLG Canon49) unambiguously stating that “Pi. Pae.” denotes Pindar’s Paean odes. Such links are fundamental as collections grow larger and increasingly ambiguous. The beginnings and ends, not only of the article as a whole but of each sense within it, are clearly marked and each has a unique identifier with which other documents can cite it.
Third parties can dynamically extract well-formed fragments of XML from the Perseus Digital Library, including canonical chunks of source texts, articles from various reference works, as well as the entire contents or individual senses from lexica. Figure 2.4 shows the same article as it appears in <http://www.dendrea.org/>, a third party site separate from the Perseus source collection: because it has access to the XML source, this site has been able to generate services (such as a browser for etymologically related terms, synonyms and antonyms) not available at Perseus.
The artificial intelligence pioneer Marvin Minsky suggested that the time would come when no one will imagine that the books in a library did not talk with one another. While Minsky may have envisioned very powerful artificial intelligence spawning conversations between books far beyond what is currently possible, our books are already beginning to converse in simple but substantive ways.50 Put another way, so much material is already online that only machines can scan more than a tiny fraction of what is available. Smart books are already beginning to appear to provide knowledge-intensive services and offer up more information about themselves than any reader might have thought to ask.
Figure 2.3 XML entry from LSJ 9 on the Perseus website.
Figure 2.4 LSJ entry from Dendrea website.
Figures 2.5, 2.6, 2.7, and 2.8 illustrate four dynamically generated views based on the interaction of different books within the Perseus digital collection.
Figure 2.5 is a “basic report” from the Perseus website that lists various translations, editions, commentaries, and other resources about a particular passage of classical Greek – Thucydides’ History of the Peloponnesian War, Book 1, chapter 86. While it resembles the page of a book, it reflects the fact that many books have been analyzed and relevant sections extracted to create a dynamic view that would be not feasible in print. Different works represent Thucydides as “Th.,” “Thuc.,” “T.,” “Thucyd.,” etc., the history as “Hist.,” “H.,” “Pel. War,” etc., and the citation as “I, 86,” “I.86,” “1,86,” “1.86.” All of these representations are mapped onto a single canonical reference around which we can then cluster a range of information. When the user calls up one translation, the translation calls out to the library for other translations, Greek editions, commentaries, lexica, grammars, and other reference works which cite words in this passage. The text in focus thus interacts with a range of other related resources, which align themselves in real time, ready to provide background information or to become themselves the focus of attention.
Figure 2.5 Basic report: A user has called up a translation of Thucydides, History of the Peloponnesian War, Book 1, chapter 86.
Figure 2.6 Parallel text analysis: word clusters associated with uses of the Greek word arche in Thucydides’ History of the Peloponnesian War (c. 150,000 words) and five English translations. Translation equivalents are underlined.
Figure 2.6 displays the word clusters associated with uses of the Greek word arche in Thucydides’ History of the Peloponnesian War (c. 150,000 words) and five English translations. By comparing the English translations with the source text, the automatic process identified clusters of meaning associated with various Greek words – in effect, creating a rough English/Greek lexicon and semantic network. The clusters capture the senses “empire,” “government,” “political office,” and “beginning.” The cluster headed “ancient” (marked in bold) captures a distinct word that happens to share the stem arch. Such parallel text analysis can update its results as new translations and source texts appear within the system, providing dynamic conclusions based on interaction of books within the digital library.
Likewise, Figure 2.7 shows the results of automatic named entity identification. In this case, a translation of Thucydides compares its vocabulary to authority lists such as encyclopedias and gazetteers to determine possible names and then uses the context in other books to resolve ambiguous references in actual text51 (e.g., does “Salamis” designate the island near Athens, a place in Cyprus or some other location?).
Figure 2.7 Named entity tagging: an XML fragment of Thucydides with all named entities automatically extracted and disambiguated.
Figure 2.8 A prototype of a basic report of Tacitus’ Annales where users have the option to see automatically generated syntactic parses of the sentences. Users can contribute to the system by correcting the automatic parse (e.g., Romam should not be in apposition to Vrbem) and transforming the partial parse into a complete one (here, by assigning tags to Vrbem and habuere).
Figure 2.8 shows the results of automatic syntactic parsing. Here a parser assigns tags to words by comparing the current text to other texts that have been syntactically analyzed by hand. By communicating with other texts in this way, the parser can determine the likelihood that a given morphological sequence (e.g., accusative noun, preposition, ablative noun) has a given syntactic parse. In the prototype shown in the figure, only tags with a reasonably high probability are assigned (allowing the system to have higher precision at the expense of greater coverage). If errors arise (as shown at the bottom of the figure, where Romam should not modify Vrbem as an apposition), users can correct the syntactic dependencies to improve the overall system, providing a valuable feedback mechanism whereby both the user and the text can productively learn from each other.
The figures above thus provide initial examples of books interacting with each other to create new forms of publication. These examples point the way toward increasingly intelligent collections which become more powerful and sophisticated as their size and internal structure improve – the more books communicate with each other, the more information about themselves they can provide.
Documents can learn from each other and drive automated processes to identify people and places in full text, analyze the contents of collections to provide integrated reports drawing on multiple information sources and perform similar tasks to apply existing classification or mine new potential knowledge.52 But even when such processes address questions with discrete, decidable answers, users will want to refine the results and these user-contributed refinements are important not only for other users but for improving the quality of subsequent automated analysis.53 Thus, an automated system may incorrectly identify “Washington” in one passage as Washington, DC, when it is in fact Washington state. Or it may simply fail because its gazetteer does not include an entry for the right Washington in a given passage (e.g., Washington, NC). Thus, even when working with very simple conceptual systems, users should be able to correct system conclusions whether by selecting a different existing answer or by adding a new possible answer to the existing set. Figure 2.9 shows an existing feedback mechanism whereby users can vote against a machinegenerated analysis.
As machines perform more sophisticated analyses where there is no single right answer, user feedback may be even more important: lexicographers do not always agree on how to describe the senses of a word.54 Machines can infer possible senses by studying the contexts in which a word appears but we still want to be able to modify the suggested word senses, even if experienced lexicographers would not agree on any one final configuration of senses.
Figure 2.9 A morphological analysis system: this system has calculated the possible analyses for a given form. A simple machine learning system has ranked the possibilities of each analysis in the given context. Users can now vote for the analysis which they see as correct.
Customization and personalization constitute two other methods by which machines respond dynamically to user behavior. In customization, users explicitly set parameters to shape subsequent system behavior. Personalization generally implies that the system takes action on its own, comparing the behavior of a new user to that of other users that it has encountered in the past.55 Some of us create our own customized versions of internet portals (e.g., “My Yahoo!”). Most humanists had, by 2006, encountered the personalization on sites such as Amazon, which inform us that people who bought the book that we just chose also bought books X, Y, and Z.56
Both customization and personalization have great potential within the humanities.57
Figure 2.10 illustrates how a user profile can help filter information, showing readers what terms they have and have not encountered. A reader has informed the system that she has studied Latin from Wheelock’s fifth edition. The system has then compared a passage from Suetonius against the vocabulary in the textbook (drawing upon the morphological analysis system which can match inflected words to their dictionary entries). Of the 115 possible dictionary words in this passage, the reader has probably encountered 54 and will find 61 that are new. These new words are then listed according to their frequency in the given passage. Alternate sorting orders could stress words that would be important in readings that have been assigned for the rest of the semester, for Suetonius in general or for some particular topic (e.g., military events) of interest to the reader. The technology can be based on straightforward principles of ranking and filtering from information retrieval but have a significant impact. The example given addresses language learning but the same techniques are applicable to technical terms. The key to this approach would be the development of learning profiles which track the contents of many textbooks, handouts, and assigned readings over different learning which we pursue throughout our lives.58
Figure 2.10 Customization in the Perseus Digital Library. This work was done by David Mimno and Gabriel Weaver, Perseus Project, Tufts University.
Figure 2.11 illustrates an example of personalization from the Perseus Digital Library. Once a user has asked for information on four or five words in a 300-word passage of Ovid, we can then predict two-thirds of the subsequent words that will elicit queries. This recommender system is similar in principle to the systems that Amazon and other e-commerce systems use to show consumers new products based on the products purchased by people who also bought product X. The application, however, reduces the search space of a language passage, suggesting words for study rather than products for purchase.59
Customization and personalization are fundamental technologies. While the examples given above address the needs of intermediate language learning, the same techniques would support professional researchers working with source materials outside of their own areas of specialization (e.g., an English professor with a background in classical Latin working through sixteenth-century English Latin prose). Customization and personalization have potential for filtering and structuring information for experts within their own field of expertise. They are core services for any advanced digital infrastructure underlying ePhilology.
Figure 2.11 Personalization in the Perseus Digital Library. This work was done by D. Sculley, PhD candidate in Computer Science at Tufts University under the supervision of Professor Carla Brodley, with help from Gabriel Weaver of the Perseus Project.
The examples in the preceding section illustrate current steps toward future possibilities. This section describes an infrastructure to move the field forward. On the one hand, we need to exploit emerging technologies. This not only includes downloading applications and compiling source code but reading research publications and implementing suitable algorithms. At the same time, in the long run we in classics and in the humanities may primarily contribute the knowledge sources whereby developed tools can analyze historical materials. Thus, named entity systems applied to texts about the Greco-Roman world will perform much better if they have access to information about the people and places of the Greco-Roman world than if they must rely wholly on resources which describe the contemporary world.60
Primary sources and reference materials that evolve in real time should include the following features:
These features have at least one profound implication. Once documents become dynamic and can evolve over time, we must evaluate them according to their potential for growth – their state at any one time constitutes only a single data point. In classics, editions and reference works more than a century old but which are in the public domain and can be freely updated may thus prove more valuable in an electronic environment than the best current resources if these are either static or even updated according to a traditional editorial process.
A range of community-driven reference works has emerged in recent years. The most famous, Wikipedia, arguably constitutes the most important intellectual development of the early twenty-first century: a new form of intellectual production, community driven and dynamic, has produced more than 1,000,000 general articles in five years.63 If and when the need for new articles diminishes, it will be interesting to see whether this vast resource enters a phase of refinement, thus suggesting a two-fold model: an open phase of development to bootstrap the system, followed by a period of revision. Criticizing this work is important, but only insofar as such criticism helps us to draw upon and contribute to this flood of intellectual energy.64 Other community-driven systems with more centralized editorial control have appeared for math and physics.65 A 2005 grant from the National Endowment for the Humanities has even provided support for Pleiades, a community-driven project on Greco-Roman geography.66
An infrastructure for ePhilology would contain two fundamental components: the primary sources and a network of reference works, linked to and constructed from the sources. Dynamic and intelligent links should connect all components of the infrastructure. When changes are suggested to a text, the effects of these changes upon associated reference works should be tracked and all affected places in all reference works should automatically report the change. Conversely, work based on analysis of a particular reference work should be noted in the text (e.g., a new study of a particular person that suggests reading one name vs. another).
Technically, this environment needs two things: a set of data structures and data. The Text Encoding Initiative (TEI) provides serviceable structures for texts themselves.67 Text mining can identify many patterns latent within these texts,68 but once we have ways of identifying people, places, organizations, and other entities within texts we need methods to reason, at least in rudimentary fashion, about them. Knowledge bases differ from databases in that they are designed to support inferencing: thus, if the system knows that no events in Herodotus postdate 400 bce, that Alexander the Great was born after 400 bce and that Alexander the Great was a king of Macedon, then it can avoid identifying the Alexander, king of Macedon, in Herodotus as Alexander the Great. Fortunately, the slowly emerging Semantic Web is designed to support such reasoning. Promising formats exist for geographic information69 and for museum objects,70 and we now have a well-developed set of guidelines for ontology production in OWL (Web Ontology Language).71 Ontologies, however, rapidly grow idiosyncratic and their development is as much a social as a technical process.72 To drive that development, however, we need enough data for serious experimentation – data structures and data will need to evolve, however cautiously, in tandem. We need services of interest to attract long-term user communities and enough data to raise issues of scale if we are to engineer solutions that will support intellectual life over time.
The Google Library and especially the Open Content Alliance, which has an open source policy, will help provide access to image books of virtually all useful public domain materials. These will provide immediate access to Latin and Roman script publications, with searchable OCR (optical character recognition) for classical Greek probably not far behind. These texts will provide the foundation on which we can build a dynamic knowledge base that evolves and grows more intelligent.
Moving from print to knowledge involves three steps:
Automated methods can address all three of the above phases but all methods are imperfect and print sources differ just enough that methods still need to be tuned for most reference materials. The three steps above constitute the most important and probably the most difficult work that we face, but they are essential and foundational to any serious infrastructure.
Classicists are fortunate in having a well-developed set of public domain print resources with which to begin their work.
The digital world makes possible a new kind of editor: the corpus editor occupies a middle ground between the algorithm-heavy, knowledge-light approaches of computer science and the wholly manual practices of traditional editing. The corpus editor works with thematically coherent bodies of text that are too big to be processed and checked by hand and that therefore demand automated methods. The corpus editor combines knowledge bases and automated methods to apply automated markup and/or extract information. The corpus editor cannot check every automated decision but is able both to document how the automated decisions were made and to provide statistical measures for the accuracy of those decisions.87
The role of the traditional editor also changes in an electronic environment. The traditional editor becomes responsible for preparing documents for use not only by people but by machines. The ePhilologist reviews a high percentage – and ideally all – of the automated decisions that link a particular text to knowledge sources such as those listed above: the editor manages the automated processes and reviews the results. The editor checks the morphological analyses and parse trees, comments on passages where the identification of a person or place is ambiguous, etc. The edited documents in the digital library provide crucial training sets that improve the performance of automated methods generally: thus, careful work on a few lives of Plutarch should improve results on the other lives and on similar Greek prose generally.
Digital culture already dominates serious intellectual life, even if its dominance still subordinates itself to the superficial – and, to a classicist, quite recent – forms of print culture. The previous section described one partial survey of what form classics might take as a digital culture matures and intellectual practice begins to exploit this digital world for its own strengths. The examples given reflect substantive work with existing technologies applied to questions common to all students of historical languages. All of the examples above either are, or could become, general services.88 Nevertheless, they constitute a few first steps in a much larger process.
Much of the above work was possible because the National Science Foundation and the National Endowment for the Humanities collaborated on the Digital Library Initiative Phase II, a program which supported a range of humanities projects. We cannot expect such levels of support in the future.89 If we are to move forward as a field, we must use what we have learned from what worked and what did not work in the past to develop a strategy to help us move forward in the future. Classics may or may not pursue the particular directions suggested in the previous section, but passively drifting along a broader current of academic practice is a dangerous course. The Mellon Foundation and American Council on Learned Societies recently funded a “Commission on Cyberinfrastructure for Humanities and Social Sciences.”90 A PhD in English (John Unsworth) chaired the commission, which included five humanists, including another person from English literature (Jerome McGann), an American historian (Roy Rosenzweig), an art historian (Sarah Fraser), and the director of an archaeological research collection (Bruce Zuckerman). The draft report available in May 2006 makes cursory mention of classics. Classicists cannot expect colleagues who work primarily in English and with relatively recent sources to anticipate the problems of working with historical languages. Classics – and all disciplines which draw upon languages of the past – must tirelessly engage in larger conversations and be prepared to defend the significance of language.
One effective solution is the creation of a new area of informatics designed to bridge the gap between a discipline and current research in computer science – a demanding task, if performed well, because it requires a command of emerging, as well as established, issues in two radically different disciplines. The field of biology, confronted with overwhelming amounts of raw data, produced the field of bioinformatics, thus creating an intellectual space, primarily grounded in biology, to connect research in computer science with biology research.
Classics probably cannot command a hundredth part of the resources on which biological research depends. We cannot call forth a major new discipline with the funding to attract the attention of grant-driven computer scientists. Nevertheless, we can accomplish a great deal.
Some emerging technologies could, if applied to classics and to other philological disciplines, have a swift and dramatic impact upon the questions that we pursue: machine translation, parallel text analysis, named entity identification, syntactic analysis, cross-language information retrieval and a range of text mining methods are well suited to a range of needs. The impact of digital technology will, however, be far broader and more pervasive than any particular tools we can deploy in the immediate future. The future of classics depends less upon particular tools than upon an emerging digital environment that integrates an increasing number of tools together into a dynamic world, constantly evolving to answer our questions and support the life of the mind. From the nineteenth century through the twentieth, we were able to take our scholarly infrastructure for granted: we had our publishers and libraries, our editions, commentaries, lexica, journals, monographs, and encyclopedias. We now have the merging of print, broadcast media, and gaming, new commercial entities planning universal access to a better library than the wealthiest academic institution on earth could provide to its faculty; we have new forms of intellectual production such as blogs and wikis; we have ontologies and knowledge bases at the core of reference materials; we have a world of dynamic information – books that read and learn from each other and from their human readers. The challenge now – and it is perhaps the greatest challenge classicists have faced since they found themselves pushed out of the center of the academy – is to shape this world and negotiate a new place for classical studies within it.
NOTES
1 The work described here builds on support from a variety of sources, including the Digital Library Initiative, Phase 2, the National Endowment for the Humanities, the National Science Foundation, and the Institute for Museum and Library Services. Many individuals have contributed. We mention in particular Carla Brodley, Lisa Cerrato, David Mimno, Adrian Packel, D. Sculley, and Gabriel Weaver.
2 For some reviews of how technology has been used within classics, please see: Crane 2004, McManus 2003, Latousek 2001, and Hardwick 2000.
3 Classicists were quick to embrace the Bryn Mawr Classical Review (<http://ccat.sas.upenn.edu/bmcr/>), which began publication in 1990 as a mailing list. BMCR was successful for three reasons: first, it used email to speed up the pace of scholarly communication, thus addressing a single, nagging problem; second, the electronic form allowed BMCR greater flexibility than its print counterparts, allowing it to accept a greater range of reviews, thus encouraging a wider range of submissions; third, its articles were, and remain, electronic analogues of print: they do not challenge their authors to rethink the substantive form of their work. The Stoa publishing consortium, by contrast, began in 1997 and has supported a range of more innovative projects (including the Demos project described below).
4 For some recent overviews of the issues with scholarly publishing, please see Unsworth 2003.
5 Derrida 1972.
6 For a discussion of fears that Google and the digitization of libraries will lead to serious decontextualization of learning, see Garrett 2006.
7 Dino Franco Felluga discusses this issue as well in regards to literary studies; please see Felluga 2005.
8 The idea of a permanent personal digital archive or storehouse of lifetime memories and knowledge has been well articulated by the creators of MyLifeBits (Gemmell 2006). Neil Beagrie has also explored this concept (Beagrie 2005).
9 A wealth of research has been conducted into how systems can best automatically adapt themselves to the needs of different readers, such as Russell 2003, Dolog 2004, Niederee 2004, Rouane 2003, Wang 2004, and Terras 2005.
10 Text mining is increasingly being used in humanities applications; see, for example, Kirschenbaum 2006 and Xiang 2006.
12 For one exploration of the impact of the TLG on classical scholarship, please see Ruhleder 1995.
13 For a discussion of some of this work, see Packard 1973.
14 Crane 2004.
15 <http://www.neh.gov/news/awards/preservation2000.html>.
16 <http://www.tlg.uci.edu/CDROME.html>.
17 Biblioteca Teubneriana Latina 2004.
18 Maria Pantelia, the director of the TLG, reported (private communication, September 2006) that lemmatized searching was in active development and would become part of the core TLG functionality.
19 There is a growing body of research into the need for more complex linguistic querying capabilities, particularly with historical language materials, please see de Jong 2005, Egan 2005, and Gerlach 2006.
20 See, for example, (Church 1989 and Justeson 1995).
21 The Perseus Digital Library has done extensive research in terms of the importance of named entity recognition and searching; please see Crane and Jones 2006a, Smith 2001.
22 For an example of a prototype system that supports many of these features, please see Ignat 2005.
23 A search for –pemp- turns up “(4.) OI)KH/TORAS A)POPE/MPEIN. OI( DE\ *) EPIDA/MNIOI OU)DE\N AU)TW=N U(P-” with the label Thucydides “Book 1 chapter 26 section 4 line 1.” In fact, the word is part of section 3, with section four beginning in the middle of the print line after the period. Simple programming can capture most of these section breaks, although some lines have more than one full stop and editors may use commas – or nothing – to mark the divisions of established units.
24 In the late 1990s, while Theodore Brunner was director of the TLG, David Smith of the Perseus Project created an SGML version of the TLG that validated against the TEI DTD. Mark Olsen of ARTFL also created a similar experimental version at the University of Chicago. In both cases, understanding the idiosyncratic reference encoding of the TLG proved the major barrier.
25 The largest Greek dramas are, with extensive XML markup, just over 120,000 bytes and would cost $120 to $180 to enter, depending on the vendor.
26 Research into variant editions and how best represent this information digitally has received a growing amount of attention, for example see Dekhytar 2005, Pierazzo 2006, Schmidt 2005, Audenaert 2005, and Riva 2005, and for an example in classics Bodard 2006.
27 <http://www.tlg.uci.edu/CDEworks.html#supp>.
28 The online TLG does not seem to provide any information about the texts that have been “suppressed,” in effect consigning these editions to an electronic damnatio memoriae. A print copy of the second edition of the TLG Canon preserves the fact that the TLG had originally contained the Murray edition of Euripides. The online TLG canon simply lists the Diggle edition of Euripides now included in the TLG.
29 <http://www.openarchives.org/>.
30 Blackwell 2005. For some examples of how the CTS protocols are being used, please see Porter et al. 2006.
31 According to at least one participant at the international gathering of Hellenists which launched the TLG in the early 1970s, the experts in the field assumed that the texts of ancient authors, as published in editions, were not copyrightable. We need automated methods with which not only to compare but to quantify the differences between various electronic editions of the same text. Preliminary analysis suggests that changes from one edition to another are comparable to copy-editing. The best model for editors employed by academic institutions may thus be a work-for-hire, with the rights holders more properly being institutions who paid their salary.
32 The representative of one UK publisher stopped at Perseus years ago en route, as he informed us, to assert rights to electronic versions of texts that a third project had entered. We paid $7,000 for rights to two editions – only to discover that those editions had unambiguously gone into the public domain by UK law and had never been under copyright in the US. Another US publisher that had knowingly published materials in the public domain reportedly charges permissions fees for these materials for which it has no legal rights.
33 For more on the issue of the public domain and copyright issues in the face of mass digitization, please see Thatcher 2006 and Travis 2005.
34 Classicists define their own conventions of what does and does not count, and we can accept monographs published in emerging institutional repositories – in effect, we would return to a scholarly publication model, separate from university and commercial presses, that has served us well in the past.
35 For further discussion of Perseus examples, please see Crane et al. 2006a.
36 <http://www.gutenberg.org/wiki/Main_Page>.
37 For discussion of the Google Library project, please see MacColl 2006, for the Open Content Alliance please see Tennant 2005.
38 For a comprehensive look at the open access movement, please see Willinsky 2005.
39 <http://www.thelatinlibrary.com/>.
40 <http://www.thelatinlibrary.com/readme>.
41 <http://www.perseus.tufts.edu/hopper/>.
42 <http://creativecommons.org/>.
43 <http://www.stoa.org/projects/demos/home>.
44 The work of the Public Knowledge Project attempts to link scholarship to freely available sources in order to support reading by a broader audience; see Willinsky 2003.
45 This need for reusable digital objects that can draw upon a range of services is a major theme of the recent Mellon-funded study to support interoperability between digital repositories (Bekaert 2006).
46 For a good overview of the possibilities inherent in better exploiting the semantic content of digital objects, please see Bearman and Trant, 2005.
47 A similar issue is often raised by those researchers who wish to analyze Wikipedia, but find its unstructured data requires a great deal of work to support automated processing. See Volkel 2006.
48 Liddell et al. 1940.
49 Berkowitz and Squitier 1990.
50 For more on the potential of what can happen when the knowledge within digitized books interacts, please see Kelly 2006, Crane 2005a, Crane 2005b.
51 Smith 2001. For more on the technical details of this system, see Crane and Jones 2005.
52 A variety of work is beginning to explore how best to exploit both the structured and unstructured knowledge already present in digital library collections to train other systems with document analysis and machine learning; see for example Nagy and Lopresti 2005 and Esposito et al. 2005.
53 Research into how to capture the knowledge of users to drive both machine learning processes and personalization is growing rapidly, see for example Chklovski 2005, Carrera 2005, Gilardoni 2005, Kruk 2005.
54 Some initial work in having user contributions assist in automated word sense disambiguation has been reported in Navigli and Velardi 2005.
55 For an expansion of these definitions see Russell 2003, and for a particular application see Bowen and Fantoni 2004.
56 For more on the Amazon system, please see Linden et al. 2003.
57 There is growing body of literature as to how these technologies might be applied within the humanities, most often digital libraries, for an overview please see Smeaton and Callan 2005,
58 Developing accurate user models and profiles to support and track learning is a topic of significant study, for some recent work please see Brusilovsky 2005 and Kavcic 2004.
59 Work on how personalization, particularly recommender systems, might be used within humanities environments has been explored by Bia 2004, Kim 2004, to name only a few.
60 On the need for historical knowledge sources, see Crane and Jones 2006b and also Siemens 2006.
61 For some recent work on creating reference works that allow users to both edit and create materials, please see Witte 2005 and Kolbistch 2005.
62 For an intriguing exploration of the potential of “machines as readers”, see Shamos 2005.
63 As of May 23, 2006, the count for English articles on <http://www.wikipedia.org> stands at 1,145,000.
64 For example, see Rosenzweig 2006.
65 <http://planetmath.org/>; <http://planetphysics.org/>.
66 <http://www.unc.edu/awmc/pleiades.html>.
68 This is the approach of the Nora text mining project: <http://nora.lis.uiuc.edu/description.php>; Plaisant et al. 2006.
69 <http://www.alexandria.ucsb.edu/gazetteer/ContentStandard/version3.2/GCS3.2-guide.htm>.
70 <http://cidoc.ics.forth.gr/>.
71 <http://www.w3.org/TR/owl-features/>.
72 For some particular applications of ontologies in the humanities, see Nagypal 2004, 2005, Mirzaee 2005, and for the merging of various efforts, see Eide 2006 and Doerr 2003.
73 For some lengthier discussion of these issues see Bearman and Trant 2005 and Sankar 2006.
74 Named entity recognition and semantic classification have large bodies of literature, but the use of theses applications in the humanities is receiving more examination; see Hoekstra 2005 and Shoemaker 2005.
75 For interesting work in this area, see Barzilay 2005.
76 For some previous work in this vein see Spencer 2004.
77 If we have “arma virumque cano Troiae qui primus ab oris” tagged in one text as Aen. 1.1, then we locate other instances of this line and apply the same markup. This strategy draws upon the fact that runs of repeated words are surprisingly uncommon, even in large corpora.
78 For a recent exploration of the uses of parallel texts, see Mihalcea 2005, and their use in machine translation Smith 2006.
79 Smith 1873: ix: “Some difficulty has been experienced respecting the admission or rejection of certain names, but the following is the general principle which has been adopted. The names of all persons are inserted, who are mentioned in more than one passage of an ancient writer: but where a name occurs in only a single passage, and nothing more is known of the person than that passage contains, that name is in general omitted. On the other hand, the names of such persons are inserted when they are intimately connected with some great historical event, or there are other persons of the same name with whom they might be confounded”; (Smith 1854: viii: “Separate articles are given to the geographical names which occur in the chief classical authors, as well as to those which are found in the Geographers and Itineraries, wherever the latter are of importance in consequence of their connection with more celebrated names, or of their representing modern towns,–or from other causes. But it has been considered worse than useless to load the work with a barren list of names, many of them corrupt, and of which absolutely nothing is known. The reader, however, is not to conclude that a name is altogether omitted till he has consulted the Index; since in some cases an account is given, under other articles, of names which did not deserve a separate notice.”
80 For more on machine translation and WSD see Smith 2006, Marcu 2005, and Carpuat 2005.
81 Liddell et al. 1940, Andrews et al. 1879.
82 Kühner et al. 1890.
83 Smyth 1920.
84 Allen et al. 1904.
85 Slater 1969.
86 The use of HEML Historical Event Markup Language could be applicable in this area; see Robertson 2006.
87 For more on the role of corpus editors, see Crane 2000.
88 For examples of potential services, please see Patton 2004 and Crane et al. 2006b.
89 For a discussion for the future of digital library funding, see Griffin 2006.
90 <http://www.acls.org/cyberinfrastructure/cyber.htm>.
91 For more on the needs of new library services and infrastructures, see Dempsey 2006.
BIBLIOGRAPHY
Allen, J. H., J. E. Greenough, et al. (1904). Allen and Greenough’s New Latin Grammar for Schools and Colleges, Founded on Comparative Grammar. Boston, MA, and London: Ginn & Company.
Andrews, E. A., W. Freund, et al. (1879). A Latin Dictionary Founded on Andrews’ Edition of Freund’s Latin Dictionary. Oxford and New York: Clarendon Press.
Audenaert, N., et al. (2005). “Integrating Collections at the Cervantes Project.” Proceedings of the 5th ACM/IEEE-CS Joint Conference on Digital Libraries (Denver, CO, USA, June 7–11, 2005). JCDL ’05. New York: ACM Press, pp. 287–8.
Barzilay, R., and N. Elhadad (2003). “Sentence Alignment for Monolingual Comparable Corpora.” Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP 2003), pp. 25–32.
Beagrie, N. (2005). “Plenty of Room at the Bottom? Personal Digital Libraries and Collections.” D-Lib Magazine 11 (6, June). <http://dlib.anu.edu.au/dlib/june05/beagrie/06beagrie.html>.
Bearman D., and J. Trant (2005). “Converting Scanned Images of the Print History of the World to Knowledge: A Reference Model and Research Strategy.” RDLP 8 (5). <http://www.elbib.ru/index.phtml?page=elbib/eng/journal/2005/part5/BT>.
Bekaert, J., and H. Van de Sompel (2006). “Augmenting Interoperability across Scholarly Repositories.” <http://eprints.rclis.org/archive/00006924/>.
Berkowitz, L., and K. A. Squitier (1990). Thesaurus Linguae Graecae Canon of Greek Authors and Works. New York: Oxford University Press.
Bia, A. C., I. Garrigós, and J. Gómez (2004). “Personalizing Digital Libraries at Design Time: The Miguel De Cervantes Digital Library Case Study.” ICWE 2004, pp. 225–9.
Bibliotheca Teubneriana Latina, BTL-3 (2004). Turnhout: Brepols; Munich: K. G. Saur.
Blackwell, C., and N. Smith (2005). A Guide to Version 1.1 of the Classical Text Services Protocol. Digital Incunabula: a CHS site Devoted to the Cultivation of Digital Arts and Letters. <http://chs75.harvard.edu/projects/diginc/techpub/cts-overview>.
Bodard, G. (2006). “Inscriptions of Aphrodisias: Paradigm of an electronic publication.” CLiP 2006. <http://www.cch.kcl.ac.uk/clip2006/content/abstracts/paper33.html>.
Bowen, J. P., and S. F. Fantoni (2004). “Personalization and the Web from a Museum Perspective.” Museums and the Web 2004. <http://conference.archimuse.com/biblio/personalization_and_the_web_from_a_museum_perspective>.
Brusilovsky, P., S. Sosnovksy, and O. Shcherbinina (2005). “User Modeling in a Distributed E-Learning Architecture.” User Modeling 2005, Lecture Notes in Computer Science 3538, pp. 387–91.
Carpuat, M., and D. Wu (2005). “Word Sense Disambiguation vs. Statistical Machine Translation.” Proceedings of the Association for Computational Linguistics 2005, pp. 387–94.
Carrera, F. (2005). “Making History: An Emergent System for the Systematic Accrual of Transcriptions of Historic Manuscripts.” Eighth International Conference on Document Analysis and Recognition (ICDAR ’05), pp. 543–9.
Chklovski, T., and Y. Gil (2005). “Improving the Design of Intelligent Acquisition Interfaces for Collecting World Knowledge From Web Contributors.” Proceedings of the 3rd International Conference on Knowledge Capture, K-CAP ’05, pp. 35–42.
Church, K., and P. Hanks (1989). “Word Association Norms, Mutual Information, and Lexicography.” Proceedings of the 27th Annual Meeting of the Association for Computational Linguistics, pp. 76–83.
Crane, G. (2004). “Classics and the Computer. An End of the History.” In Susan Schreibman, Ray Siemens, and John Unsworth (Eds.). A Companion to the Digital Humanities. Oxford: Blackwell Publishing, pp. 46–55.
Crane, G. (2005a). “No Book is an Island: Designing Electronic Primary Sources and Reference works for the Humanities.” In H. van Oostendorp, Leen Breure, and Andrew Dillon (Eds.). Creation, Use, and Deployment of Digital Information. Hillsdale, NJ: Erlbaum, pp. 11–26.
Crane, G. (2005b). “Reading in the Age of Google: Contemplating the Future with Books That Talk to One Another.” Humanities 26 (5, September/October). <http://www.neh.gov/news/humanities/2005–09/readingintheage.html>.
Crane, G., and A. Jones (2005). “The Perseus American Collection 1.0.” <http://www.perseus.tufts.edu/~gcrane/americancoll.12.2005.pdf>.
Crane, G., and A. Jones. (2006a). “The Challenge of Virginia Banks: An Evaluation of Named Entity Analysis in a 19th Century Newspaper Collection.” Proceedings of the 6th ACM/IEEE-CS Joint Conference on Digital Libraries, pp. 31–40.
Crane, G., and A. Jones (2006b). “Text, Information, Knowledge and the Evolving Record of Humanity.” D-Lib Magazine 12 (3, March). <http://purl.pt/302/1/dlib/march06/jones/03jones.html>.
Crane, G., and J. A. Rydberg-Cox (2000). “New Technology and New Roles: the Need for ‘Corpus Editors’.” Proceedings of the Fifth ACM Conference on Digital Libraries, pp. 252–3.
Crane, G. et al. (2006a). “Beyond Digital Incunabula: Modeling the Next Generation of Digital Libraries.” Proceedings of ECDL 2006, pp. 353–66.
Crane, G. et al. (2006b). “Services Make the Repository.” Paper presented at JCDL 2006 Workshop: “Digital Curation and Trusted Repositories.” <http://www.ils.unc.edu/tibbo/JCDL2006/Jones-JCDLWorkshop2006l.pdf>.
de Jong, F. et al. (2005). “Temporal Language Models for the Disclosure of Historical Text.” XVIth International Conference of the Association for History and Computing. <http://eprints.eemcs.utwente.nl/7266/01/db-utwente-433BCEA2.pdf>.
Dekhytar, A., et al. (2005). “Support for XML Markup of Image Based Electronic Editions.” International Journal on Digital Libraries 6 (1): 55–69.
Dempsey, L. (2006). “The (Digital) Library Environment: Ten Years After.” Ariadne <http://www.ariadne.ac.uk/issue46/dempsey/>.
Derrida, J. (1972). “La Pharmacie de Platon.” In La Dissemination. Paris: Éditions du Seuil, pp. 69–196.
Doerr, M., J. Hunter, and C. Lagoze (2003). “Towards a Core Ontology for Information Integration.” Journal of Digital Information 4 (1). <http://jodi.tamu.edu/Articles/v04/i01/Doerr/>.
Dolog, P., et al. (2004). “The Personal Reader: Personalizing and Enriching Learning Resources Using Semantic Web Technologies.” Proceedings of the 3rd International Conference on Adaptive Hypermedia and Adaptive Web-Based Systems, pp. 85–94.
Eide, O., and C. E. Ore (2006). “TEI, CIDOCCRM and a Possible Interface between the Two.” Proceedings of the ALLC-AHC 2006.
Egan, G. (2005). “Impalpable Hits: Indeterminacy in the Searching of Tagged Shakespearian Texts.” Paper delivered on March 17 at the 33rd Annual Meeting of the Shakespeare Association in America, in Bermuda. <http://hdl.handle.net/2134/1294>.
Esposito, F., et al. (2005). “Intelligent Document Processing.” Proceedings of Eighth International Conference on Document Analysis and Recognition, pp. 1100–4.
Felluga, D. F. (2005). “Addressed to the NINES: the Victorian Archive and the Disappearance of the Book.” Victorian Studies 48 (2): 305–19.
Garrett, J. (2006). “KWIC and Dirty? Human Cognition and the Claims of Full Text Searching.” Journal of Electronic Publishing, 9 (1). <http://hdl.handle.net/2027/spo.3336451.0009.106>.
Gemmell, J., G. Bell, and R. Lueder (2006). “MyLifeBits: A Personal Database for Everything.” Communications of the ACM 49 (1, January): 88–95.
Gerlach, A. E., and N. Fuhr (2006). “Generating Search Term Variants for Text Collections with Historic Spellings.” ECIR 2006, pp. 49–60.
Gilardoni, L. et al. (2005). “Machine Learning for the Semantic Web: Putting the User into the Cycle.” Proceedings of the Dagstuhl Seminar on Machine Learning for the Semantic Web, February 13–18, 2005, Dagstuhl, Germany. <http://www.quinary.com/pagine/downloads/files/Resources/QuinaryDagstuhl.pdf>.
Griffin, S. (2005). “Funding for Digital Libraries: Past and Present.” D-Lib Magazine 11 (7/8). <http://www.dlib.org/dlib/july05/griffin/07griffin.html>.
Hall, F. W. (1913). A Companion to Classical Texts. Oxford: Clarendon Press.
Hardwick, L. (2000). “Electrifying the Canon: The Impact of Computing on Classical Studies.” Computers and the Humanities 34: 279–95.
Hinman, C. (1968). The First Folio of Shakespeare: The Norton Facsimile. New York: W. W. Norton.
Hoekstra, R. (2005). “Integrating Structured and Unstructured Searching in Historical Sources.” Proceedings of the XVI International Conference of the Association for History and Computing, pp. 149–54.
Hornblower, S., and A. Spawforth (1996). The Oxford Classical Dictionary. New York: Oxford University Press.
Ignat, C., et al. (2005). “A Tool Set for the Quick and Efficient Exploration of Large Document Collections.” Proceedings of the 27th Annual ESARDA Meeting. <http://arxiv.org/abs/cs.CL/0609067>.
Justeson, J. S., and M. K. Slava (1995). “Technical Terminology: Some Linguistic Properties and an Algorithm for Identification in Text.” Natural Language Engineering 1: 9–27.
Kavcic, A. (2004). “Fuzzy User Modeling for Adaptation in Educational Hypermedia.” IEEE Transactions on Systems, Man and Cybernetics Part C, 34 (4, November): 439–49.
Kelly. K. (2006). “Scan This Book!” New York Times Magazine. <http://www.nytimes.com/2006/05/14/14publishing.html>.
Kim, S., and E. A. Fox (2004). “Interest-Based User Grouping Model for Collaborative Filtering in Digital Libraries.” 7th International Conference of Asian Digital Libraries, pp. 533–42.
Kirschenbaum, M. (2006). “The NORA Project: Text Mining and Literary Interpretation.” Digital Humanities 2006, pp. 255–6.
Kolbitsch, J., and H. Maurer (2005). “Community Building around Encyclopaedic Knowledge.” Journal of Computing and Information Technology 14 (3): 175–90.
Kruk, S. R., S. Decker, and L. Zieborak (2005). “JeromeDL – Adding Semantic Web Technologies to Digital Libraries.” DEXA 2005, Lecture Notes in Computer Science 3588, pp. 716–25.
Kühner, R., F. Blass, et al. (1890). Ausführliche Grammatik der Griechischen Sprache. Hannover: Hahnsche Buchhandlung.
Latousek, R. (2001). “Fifty Years of Classical Computing: A Progress Report.” CALICO Journal 18 (2): 211–22.
Liddell, H. G., R. Scott, et al. (1940). A Greek–English Lexicon. Oxford: The Clarendon Press.
Linden, G., B. Smith, and J. York (2003). “Amazon.com Recommendations: Item-to-Item Collaborative Filtering.” Internet Computing 7 (1): 76–80.
MacColl, J. (2006). “Google Challenges for Academic Libraries.” Ariadne 46. <http://www.ariadne.ac.uk/issue46/maccoll/>.
Marcu, D., and K. Knight (2005). “Machine Translation in the Year 2004.” Proceedings of Acoustics, Speech and Signal Proceedings (ICASSP 2005), vol. 5, pp. 965–8.
McManus, B. F., and C. A. Rubino (2003). “Classics and Internet Technology.” American Journal of Philology 124 (4): 601–8.
Mihalcea, R., and M. Simard (2005). “Parallel Texts.” Natural Language Engineering 11 (3, September): 239–46.
Mirzaee, V., et al. (2005). “Computational Representation of Semantics in Historical Documents.” Proceedings of AHC 2005.
Nagy, G., and D. Lopresti (2006). “Interactive Document Processing and Digital Libraries.” Proceedings of the Second International Conference on Document Images, Analysis for Libraries (DIAL 2006), pp. 2–11.
Nagypal, G. (2004). “Creating an Application-Level Ontology for the Complex Domain of History: Mission Impossible?” In Proceedings of Lernen – Wissensentdeckung – Adaptivitat (LWA 2004), FGWM 2004 Workshop, Berlin, Germany, pp. 287–94.
Nagypal, G., et al. (2005). “Applying the Semantic Web: The VICODI experience in Creating Visual Contextualization for History.” Literary and Linguistic Computing 20 (3): 327–49.
Navigli, R., and P. Velardi (2005). “Structural Semantic Interconnections: A Knowledge-Based Approach to Word Sense Disambiguation.” IEEE Transactions on Pattern Analysis and Machine Intelligence 27(7): 1075–86.
Niederée, C., et al. (2004). “A Multi-Dimensional, Unified User Model for Cross-System Personalization.” Proceedings of Advanced Visual Interfaces International Working Conference (AVI 2004) – Workshop on Environments for Personalized Information Access, pp. 34–54.
Packard, D. W. (1973). “Computer-Assisted Morphological Analysis of Ancient Greek.” Proceedings of the 5th Conference on Computational Linguistics, Pisa, Italy, pp. 343–55.
Patton, M. S., and D. M. Mimno (2004). “Services for a Customizable Authority Linking Environment.” Proceedings of the 4th ACM/IEEE-CS Joint Conference on Digital Libraries, p. 420.
Pierazzo, E. (2006). “Just Different Layers? Stylesheets and Digital Edition Methodology.” Digital Humanities 2006.
Plaisant, C., J. Rose, et al. (2006). “Exploring Erotics in Emily Dickinson’s Correspondence with Text Mining and Visual Interfaces.” Proceedings of the 6th ACM/IEEE-CS Joint Conference on Digital Libraries, pp. 141–50.
Porter, D., et al. (2006). “Creating CTS Collections.” Digital Humanities 2006, pp. 269–74.
Riva, M., and V. Zafrin (2005). “Extending the Text: Digital Editions and the Hypertextual Paradigm.” Proceedings of the Sixteenth ACM Conference on Hypertext and Hypermedia, pp. 205–7.
Robertson, B. (2006). “Visualizing an Historical Semantic Web with HEML.” Proceedings of the WWW 2006, pp. 1051–2.
Rosenzweig, Roy (2006). “Can History Be Open Source: Wikipedia and the Future of the Past.” Journal of American History 93 (1): 37–46.
Rouane, K., C. Frasson, and M. Kaltenbach (2003). “Reading for Understanding: A Framework for Advanced Reading Support.” Proceedings of the 3rd IEEE International Conference on Advanced Learning Technologies, pp. 394–5.
Ruhleder, K. (1995). “Reconstructing Artifacts, Reconstructing Work: From Textual Edition to On-Line Databank.” Science, Technology, & Human Values 20 (1, Winter): 39–64.
Russell, J. (2003). “Making it Personal: Information That Adapts to the Reader.” SIGDOC ’03: Proceedings of the 21st Annual International Conference on Documentation, pp. 160–6.
Sankar, K. P., et al. (2006). “Digitizing a Million Books: Challenges for Document Analysis.” Document Analysis Systems VII, 7th International Workshop, DAS 2006, pp. 225–36.
Schmidt, D., and T. Wyeld (2005). “A Novel User Interface for Online Literary Documents.” Proceedings of the 19th Conference of the Computer–Human Interaction Special Interest Group (CHISIG) of Australia on Computer–Human Interaction, pp. 1–4.
Shamos, M. I. (2005). “Machines as Readers: a Solution to the Copyright Problem.” Journal of Zhejiang University Science 6A, 11: 1179–87.
Shoemaker, R. (2005). “Digital London: Creating a Searchable Web of Interlinked Resources on Eighteenth Century London.” Program: Electronic Library and Information Systems 39 (4): 297–311.
Siemens, R. (2006). “Knowledge Management and Textual Cultures? Work Toward the Renaissance English Knowledgebase (REKn) and its Professional Reading Environment.” Paper delivered at CASTA 2006.
Slater, W. J. (1969). Lexicon to Pindar. Berlin: de Gruyter.
Smeaton, A. F., and J. Callan (2005). “Personalisation and Recommender Systems in Digital Libraries.” International Journal of Digital Libraries 5: 299–308.
Smith, D. A. (2006). “Debabelizing Libraries: Machine Translation by and for Digital Collections.” DLib Magazine 12 (3, March). <http://www.dlib.org/dlib/march06/smith/03smith.html>.
Smith, D. A., and G. Crane (2001). “Disambiguating Geographic Names in a Historical Digital Library.” Proceedings of the 5th European Conference on Research and Advanced Technology for Digital Libraries (ECDL ’01), Lecture Notes in Computer Science, pp. 127–36.
Smith, W. (1854). Dictionary of Greek and Roman Geography. Boston: Little Brown & Co.
Smith, W. (1873). A Dictionary of Greek and Roman Biography and Mythology. London: J. Murray.
Smith, W., W. Wayte, et al. (1890). A Dictionary of Greek and Roman Antiquities. London: J. Murray.
Smyth, H. W. (1920). A Greek Grammar for Colleges. New York, Cincinnati [etc.]: American Book Company.
Spencer, M., and C. Howe (2004). “Collating Texts Using Progressive Multiple Alignment.” Computers and the Humanities 38 (3, August): 253–70.
Tennant, R. (2005). “The Open Content Alliance.” Library Journal December 15. <http://www.libraryjournal.com/article/CA6289918.html>.
Terras, M. (2005). “Reading the Readers: Modelling Complex Humanities Processes to Build Cognitive Systems.” Literary and Linguistic Computing 20 (1): 41–59.
Thatcher, S. G. (2006). “Fair Use in Theory and Practice: Reflections on its History and the Google Case.” Journal of Scholarly Publishing 37 (3): 215–29.
Travis, H. (2005). “Building Universal Digital Libraries: An Agenda for Copyright Reform.” Forthcoming, Pepperdine Law Review. <http://papers.ssm.com/sol3/papers.cfm?abstract_id=793585>.
Unsworth, J. (2003). “The Crisis in Scholarly Publishing in the Humanities.” ARL Bimonthly Report 228 <http://www.arl.org/newsltr/228/crisis.html>.
Volkel, M., et al. (2006). “Semantic Wikipedia.” Proceedings of the 15th International Conference on World Wide Web, pp. 585–94.
Wang, C. Y., and G. D. Chen (2004). “Extending E-Books with Annotation, Online Support and Assessment Mechanisms to Increase Efficiency of Learning.” SIGCSE Bulletin 36 (3): 132–6.
Willinsky, J. (2003). “Opening Access: Reading (Research) in the age of Information.” In C. M. Fairbanks, J. Worthy, B. Maloch, J. V. Hoffman, and D. L. Schallert (Eds.). 51st National Reading Conference Yearbook. Oak Creek, WI: National Reading Conference, pp. 32–6.
Willinsky, J. (2005). The Access Principle: The Case for Open Access to Research Scholarship. Cambridge, MA: MIT Press.
Witte, Rene (2005). “Engineering a Semantic Desktop for Building Historians and Architects.” SemDesk 2005 Workshop Proceeding. <http://www.semanticdesktop.org/xwiki/bin/download/Wiki/EngineeringASemanticDesktopForBuildingHistoriansAndArchitects/34_witte_engineeringsd_final.pdf>.
Xiang, X., and J. Unsworth (2006). “Connecting Text Mining and Natural Language Processing in a Humanistic Context.” Digital Humanities 2006.