[At Perseus Updates]
Initial Research Plan (April 2013)
Alexander von Humboldt Chair of Digital Humanities
The University of Leipzig
Abstract: The Humboldt Chair of Digital Humanities at the University of Leipzig sees in the rise of Digital Technologies an opportunity to re-assess and re-establish how the humanities can advance the understanding of the past and to support a dialogue among civilizations. Philology, which uses surviving linguistic sources to understand the past as deeply and broadly as possible, is central to these tasks, because languages, present and historical, are central to human culture. To advance this larger effort, the Humboldt Chair focuses upon enabling Greco-Roman culture to realize the fullest possible role in intellectual life. Greco-Roman culture is particularly significant because it contributed to both Europe and the Islamic world and the study of Greco-Roman culture and its influence thus entails Classical Arabic as well as Ancient Greek and Latin. The Humboldt Chair inaugurates an Open Philology Project with three complementary efforts that produce open philological data, educate a wide audience about historical languages, and integrate open philological data from many sources: the Open Greek and Latin Project organizes content (including translations into Classical Arabic and modern languages); the Historical Language e-Learning Project explores ways to support learning across barriers of language and culture as well as space and time; the Scaife Digital Library focuses on integrating cultural heritage sources available under open licenses.The Humboldt Chair of Digital Humanities at Leipzig will create the Open Philology Project. In this we advance a digital successor to that philology which sees in language a source for what Augustus Boeckh in 1822 termed “the understanding of all antiquity, including the events of both the physical and intellectual world.”[1] Philology brings the past to life as deeply and as broadly as possible through the use of surviving linguistic sources. From the human perspective philology constitutes a set of language-based critical scholarly skills — not only annotating (annotation is the basic genre), but also comparing, connecting, interpreting, proving or rejecting hypotheses, finding evidence; critical apparatuses and commentaries often preserve condensed fruits of such reasoning, and Open Philology doesn’t let the scholarly heritage of manuscript and print culture vanish, converting it into digital form and using it as a training field for next generations.
The Open Philology Project will initially focus particularly upon pre-modern society but its methods and goals apply to any society for whom traces of their languages survive. Philology provides an opportunity to advance the intellectual life of individual societies and, equally important, dialogue across civilizations, transcending not only barriers of space and time but of language and culture. Digital technology plays a critical role as a catalyst because — and only because — it allows us to re-imagine how we can more fully achieve, and indeed transform our ability to achieve, these ancient goals of philology. This is not a digital philology or digital humanities project. The Open Philology Project is about philology.
To address the vast challenge of an Open Philology that embraces all historical languages, the Humboldt Chair begins by advancing within a European and a global space the role of that Greco-Roman culture out of which Europe largely emerged. Greco-Roman culture has also contributed significantly to the Islamic world and Europe depended upon Arabic sources. Our goal in this activity is not only to increase the intellectual accessibility of European cultural heritage but also to foster exchange of cultural heritage sources such as Persian, Sanskrit, Classical Chinese, Egyptian from the earliest forms through Coptic, and the Cuneiform Languages of the Ancient Near East, and Classical Mayan from the Western Hemisphere. As a platform for this activity, the Open Philology Project builds upon, and helps develop, the Perseus Digital Library, working with colleagues in Europe, North America and elsewhere to expand open collections and services and to reach an increasingly global audience.
The greatest challenge of humanistic scholarship lies, in our view, in making available the human cultural heritage to the global community. Digitization is a necessary but, by itself, insufficient step in this process. Human cultural heritage must be represented in a way that supports intellectual access across barriers of language and culture. This requirement in turn has implications for the technologies but also for the rights regime that we choose. Open data provides the best strategy by which to promote the circulation of sources within a global context. Collections that are protected behind subscription barriers may serve the interests of specialist communities. Collections that cannot be freely modified and re-circulated may be useful for reference. But scholarship in general and philology in particular must build upon open data if it is to realize its intellectual and social obligations to advance the common understanding of human culture. The Humboldt Chair is therefore committed to open source publication, with machine-actionable Creative Commons licenses requiring attribution and sharing of data and allowing commercial reuse (CC-BY-SA) as the preferred mode of distribution.
The larger Open Philology Project begins with three specific, complementary activities, addressing the challenge of creating comprehensive open resources, providing the education needed to understand and to contribute to those resources, and integrating open resources from many different sources into an integrated computational framework for analysis, annotation, and preservation.
First, the Open Greek and Latin Project makes Greek and Latin sources freely accessible, both digitally and intellectually, to a global public. Second, the Historical Language e-Learning Project provides distributed e-learning of historical languages such as Greek and Latin so that as many as possible may penetrate as deeply as they choose into the sources from which the present has been fashioned. Third, support from the Humboldt Foundation allows us to contribute, after years of planning, to the Scaife Digital Library. The SDL develops methods to aggregate and integrate from various sources open data, textual and archaeological alike, in any medium, about human cultural heritage, including, but not limited to, the Greco-Roman world.
All three of these projects focus on the production, analysis, and preservation of machine-actionable annotations. All data about historical records is based upon transcriptions, whether from text-bearing objects or from sound recordings, which are themselves annotations that describe the textual content from a region of a written surface or a time interval in a recording. We will continue to make arguments in the digital successors to notes, articles and monographs but we should increasingly integrate into, and use as the foundation for, those arguments machine actionable links to the sources upon which they are based. These links include not only citations to particular sources (e.g., a machine actionable link to a particular reading in a particular edition of Aeschylus) but also to aggregate data (e.g., the results of a search posed as they appeared at a particular time). In the end, born-digital notes, articles and monographs — if they preserve labels inherited for the form of a book — may preserve a family resemblance to their predecessors but they will surely evolve into something qualitatively different as the adapt to the different gravity, if not fundamentally different physics, of a digital space.
1. The Open Greek and Latin Project.
The ultimate goal is to represent every source text produced in Classical Greek or Latin from antiquity through the present, including texts preserved in manuscript tradition as well as on inscriptions, papyri, ostraca and other written artifacts. Over the course of the next five years, we will focus upon converting as much Greek and Latin, available as scanned printed books, into an open, dynamic corpus, continuously augmented and improved by a combination of automated processes and human contributions of many kinds. The focus upon Greek and Latin reflects both the belief that we have an obligation to disseminate European cultural heritage and the observation that recent advances in OCR technology for Greek and Latin make these intertwined languages ready for large-scale work.
The Open Greek and Latin Project aims at providing at least one version for all Greek and Latin sources produced during antiquity (through c. 600 CE) and a growing collection from the vast body of post-classical Greek and Latin that still survives. Perhaps 150 million words of Greek and Latin, preserved in manuscripts, on stone, on papyrus or other writing surface, survive from antiquity. Analysis of 10,000 books in Latin, downloaded from Archive.org, identified more than 200 million words of post-classical Latin. With 70,000 public domain books listed in the Hathi Trust as being in Ancient Greek or Latin, the amount of Greek and Latin already available will almost certainly exceed 1 billion words.
Where existing corpora of Greek and Latin have generally included one edition of a work, Open Greek and Latin Corpus is designed to manage multiple versions of, and to represent the complete textual history of, a work: every manuscript, every papyrus fragment, and every printed edition are all versions within the history of a text. In the short run, this involves using OCR-technology optimized for Classical Greek and Latin to create an open corpus that is reasonably comprehensive for the c. 100 million words produced through c. 600 CE and that begins to make available the billions of words produced after 600 CE in Greek and Latin that survive.
The Open Greek and Latin Project assumes the following modules:
A. The Philological Workflow Module enables a digital representation of a written source, available in a 2D or 3D form, to be converted into machine actionable text, corrected, and annotated with an increasing range of information (named entities, morphology, syntax, and other linguistic features, alignments between different versions of the same text, whether in the same language or translated across multiple languages, text re-use detection, including quotation, paraphrase and citation). Automated methods include Optical Character Recognition, Text Alignment, Syntactic Parsing, etc. In each case, human annotation can augment automated annotations or substitute for them altogether where automated methods are not yet able to produce adequate initial results (e.g, manual transcription of inscriptions and medieval manuscripts).
B. The Distributed Review Module provides a range of options by which to assess and represent the reliability produced, whether by automated systems or by human contributors, as part of the Philological work flow. In many cases annotations can be released even when their reliability is not necessarily high (e.g., noisy OCR-generated text). The point is to identify annotations that most require subsequent attention, whether manual correction or action of some other kind (e.g., poor OCR data may reflect the need to create a new scan of a printed book). The Distributed Review Module assumes that multiple annotations may be equally trustworthy (i.e., experts back different interpretations) and can track inter-annotator disagreement among experts. The Distributed Review Module provides default values but also allows for different weights to be placed upon different validations (e.g., include all readings in a particular version of a text, whether these are readings in a particular manuscript or the readings chosen and emendations proposed by a particular editor, include all prosopographical identifications proposed by one particular scholar). The Distributed Review Module should support searching by both text characteristics (specific passages, authors), annotator characteristics (expert, novice, native language etc.), and annotation characteristics (emendations, grammatical or interpretive comments, degree of inter-annotator disagreement, etc.). But it should also permit browsing the history of annotation by passage, annotator, magnitude of disagreement etc.
C. The Philological Repository Module can preserve all published philological data, including the transcriptions and all subsequent annotations (e.g., identifying a transcribed word as being in Latin, a place name, in the accusative case etc.) as well as the provenance of each annotation (e.g., the annotation is born-digital and was published by a particular individual at a given time or the annotation was extracted from a print book by a particular author and published at a given time, with or without human verification, and with an estimated accuracy). The repository is based upon the Canonical Text Services/CITE Architecture for textual sources developed by researchers at the Center for Hellenic Studies within the larger framework developed by the DataConservancy.org.
D. The e-Portfolio Module aggregates and distributes particular subsets of user contributions for particular audiences. The e-Porfolio Module can identify any published contributions according to type, date, and author (e.g., all syntactic analyses published by a particular person during a particular time interval). The e-Portfolio Module can also make selected materials that are not yet published available to selected audiences (e.g., an editorial board or the admission committee for a degree program). The Perseids Project from Tufts University provides a starting point for this work.
2. The Historical Language e-Learning Project.
Anyone, anywhere, regardless of their linguistic or cultural background, whether they are a student in a formal curriculum or not, should be able to learn as much of a historical language as they need to work directly in original-language primary materials. Work in this context entails not only learning but contributing early and in increasingly sophisticated ways: students can add new, or correct existing, data as they learn to type in an unfamiliar language, while they can, in the language of gaming, “level up” to tasks such as linguistic annotation of new materials and the production of aligned, modern language translations, and see their growing proficiency concretely visualized in a way that permits them to compare it to that of others and documents it for use in e-portfolios and other records of their achievement.
In the short run, building upon existing collections and services, we will support students working with Greek, Latin and Classical Arabic texts in a system readily localized for speakers of multiple modern languages (with Croatian, English, German and French emerging as initial languages of interest). The Historical Language e-Learning Project is based upon the existence of extensible richly annotated corpora. Learners draw from the start on existing richly annotated corpora and on images of sources such as manuscripts and inscriptions. They use morpho-syntactic annotation, dictionary links, and aligned modern language translations, so that they immediately work with primary sources in the original. They learn grammar by comparing their morpho-syntactic analyses with vetted analyses already available, by creating their own aligned translations, and by using annotations and alignments to develop active as well as passive mastery of morphology, syntax, and vocabulary. They demonstrate advanced ability by expanding the corpus of richly annotated materials, proposing new annotations of their own and reviewing annotations proposed by others.
Ancient Greek, Latin and Classical Arabic Large collections such as Gallica, Google Books, and the Internet Archive have already made billions of words in Greek and Latin available to a global audience — a far larger collection than the small handful of advanced researchers can document and a far broader collection in terms of genre and style than the classical corpora on which current programs in Greek and Latin still focus. While the amount of openly licensed Classical Arabic is not yet as extensive, more than enough sources are available and require documentation and analysis. We need to train a new generation of students, who can directly analyze sources in the original languages and make substantive contributions earlier and on a wider range of sources than has previously been feasible.
Traditional programs of Ancient Greek and Latin are not designed to support students who first develop an interest in these languages during their undergraduate careers — by the time students are able to begin interacting proficiently with the primary sources, they are ready to graduate. Traditional class schedules are rigid and rarely can an institution offer more than one section of an ancient language. As for Classical Arabic, few institutions offer any formal instruction at all — Modern Language Association statistics report only 285 students enrolled in Classical Arabic in the United States in 2009.
At the same time, Ancient Greek, Latin, and Classical Arabic must also compete for students with fields where students regularly contribute as members of laboratory teams and can often expect to develop their own research projects as undergraduates. The strongest academic programs not only demand that students master complex disciplinary knowledge but also provide students with an opportunity to use that knowledge to make substantive contributions and to develop significant research projects of their own.
Open Greek and Latin creates an inexhaustible range of substantive activity to which any student of these languages can aspire — whether working on manuscripts of well-known authors (e.g., the Homer Multitext Project), creating the first modern language translations of Greek and Latin sources (e.g., Tufts’ Medieval Latin), or adding critical linguistic annotation (e.g., the Perseus Greek and Latin Treebanks).
The Historical Language e-Learning project depends upon the following:
A. Global Editions of Historical Languages include all features of a traditional edition (including textual notes) but are designed to make primary sources available to the widest possible audience. Global editions are richly encoded source materials that include enough annotation so that readers with a general understanding of grammar and of language are able to work directly with primary sources in a historical language that they have not studied. Core elements to this infrastructure include morphological and syntactic analyses, links to machine readable dictionaries (ideally with data about word sense of a given word in a given context), and one or more aligned modern language translations that themselves have substantial annotation and are designed to facilitate machine translation into many other modern languages.
B. Preliminary Source Texts are digital texts that do not yet have the mature annotations needed for global editions. Students of a language can aspire to begin adding new annotations within the opening weeks of study, working at first with each other and with their instructors but ultimately working to level up to roles with more trust and responsibility as they demonstrate their increasing skills. It is both a goal and a necessity to engage students as collaborators, because we believe that this is a good thing in itself, because we believe that this increases learning, and because so many historical sources are already available that we cannot depend upon a handful of professionals to analyze and annotate them all.
C. Machine Actionable Models of Language Competence provide methods by which to assess knowledge of historical languages at every level, from introductory exposure to the language through standardized examinations (e.g., the US-based National Latin exam, the German Graecum and Latinum) to the various PhD level examinations (e.g., US PhD programs in Greek and Latin commonly have combined reading lists of between 500,000 and 1,000,000 words of Greek and Latin). Where the city of Arpino holds the Certamen Ciceronianum Arpinas, a multinational competition for students from various nations — each of whom can compete in their own national language — we can create an on-going contest, where students from around the world and from widely disparate backgrounds can meet to compare their skills and compete to shed light upon Greco-Roman culture. Machine Actionable Models of Language Competence can be configured for various purposes and pedagogical perspectives. The Competence Models also provide mechanisms for evaluation of competence across national languages — examinations on morphology and syntax provide a powerful measure of competence and can be effectively localized in various national languages — whether the student speaks Arabic or Croatian, English or Lithuanian. The Distributed Review Module provides an environment for assessment of language competence as well as for advanced publications.
D. Localized Learning Materials include grammars, lexica, and translations in a national language. Localized Learning Materials need to be able to be shared across, and customized for, many different languages. Within Europe alone, Greek and Latin, for example, are taught in more than thirty different national languages.[2] We need not only to maintain learning materials in dozens of languages but also to provide learning materials in languages where Greek and Latin are not part of formal academic curricula. To accomplish this we must represent as much information about the language in machine actionable form that can be efficiently represented in many languages. We also need to provide an architecture that supports customization for particular languages, especially the creation of aligned translations that contain from the start links between the source text and the modern language translation.
E. Dynamic Syllabi can be analyzed to track the linguistic phenomena that students have encountered (e.g., vocabulary, grammar) and the content that they have covered. As students pursue different dynamic syllabi at different times, they can track their overall background and necessary background information needed to pursue subsequent courses. Instructors in structured classes can generate personalized background readings and examinations that reflect both what students brought with them to, and what they covered in, the class. The e-Portfolio Module uses Dynamic Syllabi to accomplish these goals.
F. Personalized E-learning Tools analyze individual behaviors of particular learners and provide personalized analyses and suggestions, reflecting the strengths and learning styles of particular students. Personalized e-learning tools allow learners not only to track their progress towards target proficiencies and but also to personalize the target proficiencies as well: students of Homer will have different targets than those of the New Testament or of Plato, while students aspiring to fluent comprehension will have different needs than intellectual historians who wish to explore word usage or linguists interested in syntactic phenomena. The goal is to provide as much feedback as possible, as quickly as possible, and as closely adapted to the needs and interests of each learner as possible.
3. The Scaife Digital Library (SDL)
The Scaife Digital Library (SDL) commemorates Ross Scaife (March 31, 1960 – March 15, 2008) who did pioneering work for the study of Greco-Roman culture in a digital age, who was committed to collaborative scholarship and who was a champion of open data. The SDL is designed as a service, as an experiment, and as a space for research. An increasing amount of Ancient Greek, Latin, Classical Arabic and other sources are available under appropriate open licenses. The SDL builds upon the services and collections listed above. The SDL provides a mechanism by which to compare these services and collections with those available elsewhere, allowing research at the Humboldt Chair to explore new methods while making its own work more visible.
As a service, the SDL will aggregate as much content in these languages as possible, converting, where necessary and feasible, into interoperable formats. The goal of this service is to provide a single space to represent all published Ancient Greek, Latin, and Classical Arabic. In this context, publication entails release under an open license. Proprietary collections are neither public nor published.
As an experiment, the SDL will track how many sources in how many historical languages and of how many types it can identify and integrate and then track this data over time. This experiment attempts to measure both our ability to find materials and the change in what is available. It is our hope that the growth in available resources in languages beyond Greek, Latin, and Arabic will greatly outstrip the ability of the SDL to aggregate and analyze them.
As a research space, the SDL collects not only metadata but also source materials into a single environment. Within this space researchers can explore customized collections (e.g., all available versions and translations of the Odes of Horace or the aligned corpus of Classical Arabic translations and Greek sources) or simply analyze all available Greek. While the SDL may collect as widely as possible from open collections representing cultures from around the world, the SDL intends to provide the most comprehensive possible coverage and services for students of Ancient Greek, Latin, and Classical Arabic.
[1] Augustus Boeck, “Oratio nataliciis Friderici Guilelmi III.” (1822): “Itaque ubi, quae et qualis philologia meo iudicio sit, quaeritis, simplicissima ratione respondeo, si non latiore, quae in ipso vocabulo inest, potestate accipitur, sed ut solet ad antiquas litteras refertur, universae antiquitatis cognitionem historicam et philosophicam.”
[2] http://www.eduhi.at/gegenstand/EuroClassica/?TITEL=EuroClassica+in+Europe&modul=europamap