Daejeon, July 27–31
Quite a few projects that produce scholarly editions of texts have a neat, linear sequence: following a phase of data collection and curation, a standardised edition is produced. While it is being revised, a workflow to generate outputs — typically an online edition and a printed version — is implemented. Finally, the raw edition with its metadata is put in a long-term archive backed by an institution, and the online edition is kept running for a few years, often at personal cost to researchers.
This talk outlines some strategies for dealing with projects that do not follow this linear model, and discusses their development and implementation around a long-running editorial effort that is designed to contribute to a larger, uniform research library. Projects of this sort might lack a definitive completion point, or simply run for such a long period of time that one cannot expect continuity in staff or technology. Our experience draws on a project that has been running for more than ten years, but has finally resulted in the submission of a printable PDF to an Academic publisher. This sort of long-running editorial project poses specific problems.
In this talk, we discuss these problems in the case of the critical and diplomatic edition of chapter 5 in Jinendrabuddhi’s Viśālāmalavatī Pramāṇasamuccayaṭīkā (ca. eighth cent. CE). This chapter comments on chapter 5 of Dignāga’s Pramāṇasamuccaya, a foundational text for a highly technical yet hugely influential variant of Buddhist philosophy in South Asia from around the fourth to at least the thirteenth cent. CE. The edition is part of the “Sanskrit Texts From the Tibetan Autonomous Region” (STTAR) series, which focuses on the publication of Sanskrit editions of Buddhist philosophical texts that were either not available at all or only in classical Chinese or Tibetan translations. Of Jinendrabuddhi’s commentary, so far, chapters 1,2, and 6 have been published (respectively, Steinkellner, Krasser, and Lasic 2005, Lasic, Krasser, and Steinkellner 2013, and Ono, Muroya, and T. Watanabe 2023). There are various reasons for the extended duration of the editorial process, including the following three. First, the text is simply too large for a single scholar to edit, and was therefore shared between editorial teams active in Austria and Japan. Second, few if any people in each team were able to devote their undivided attention to the text. Third, the base-text, which Jinendrabuddhi is commenting on, is lost in its Sanskrit original and needs to be reconstructed from quotations in order to make sense of Jinendrabuddhi’s comments. The lag caused by these problems led to technological difficulties. The teams agreed on a uniform format for the publication, but did not use compatible software to produce the publications. It is only for chapter 5, the focus of this presentation, that an editorial team switched to the TEI/XML format.
The TEI/XML workflow was chosen also because it is highly compatible with the FAIR principles (Wilkinson et al. 2016; Creamer et al. 2021; Tello et al. 2024). But it is worth remembering that data and the tools to work with the data are not wholly separable. To see this point, one can consider how a TEI project satisfies the last two of the FAIR principles (“interoperable” and “reusable”). The TEI’s mechanism of schema customisation allows an editor to define only the differences of their specific usage of elements versus the general ideas expressed in the overall framework (TEI P5 Guidelines). At the same time, declaring one’s encoding principles as a subclass or variation of a more general set of concepts means that one satisfies the first two sub-principles of the third FAIR principle, “to be interoperable:” the edition (or data) will “use a formal, accessible, shared, and broadly applicable language for knowledge representation”, satisfying sub-principle I1, and it will “use vocabularies that follow FAIR principles” themselves, satisfying sub-principle I2 (Wilkinson et al. 2016: Box 2). As observed in Tello et al. 2024: §§8, 10, the tei:teiHeader is perfectly adequate for recording detailed information that satisfies all the key points making the data (and its metadata) “reusable.”
In order to see how this consideration applies for our edition, the talk first discusses the general background of the edited text. Difficulties to be highligted in this regard include the text’s high degree of intertextuality, inexistent sources, and the different languages of witnesses that have to be used to constitute the text. We present TEI/XML based solutions to all of these problems (see also McAllister 2021).
In a second part, we make some general observations that should be valid for any project that runs for an extended period of time or that has to integrate itself into a larger editorial environment (the presented edition has to accomplish both): while best practices for scholarly editing (based on TEI P5 Guidelines) and data management (see Creamer et al. 2021; Tello et al. 2024; Wilkinson et al. 2016) are established, it is less clear what the FAIR principles, for example, mean for an ongoing editorial project. How can versioning over more than ten years, e.g., be secured without freezing an archive? And which data formats can be edited across that period of time without loss, either because of humans changing their habits or software going out of date? The best answer is certainly plain-text formats like TEI/XML encoded files. But the toolset used to produce the first volume of the STTAR series (published in 2005) varies considerably from the tools used nowadays. Yet, we know that, once all six chapters of Jinendrabuddhi’s massive commentary have been edited, we will be required to produce a single uniform version, at least for an online publication that, ideally, will allow us to read and study the text in ways that the printed editions do not (see, e.g., Stinne Greve Rasmussen 2016).
To this end, we introduce a framework called Vienna Encoding Guidelines for Editing Sanskrit Texts (VEGEST) (VEGEST schemas). VEGEST proposes a holistic approach that integrates open documentation, modular schema development (see VEGEST schemas for these two), and a reproducible software toolchain (see VEGEST library). VEGEST thus tries to apply FAIR principles to all parts of an editorial project: data, documentation, and software (see Barker et al. 2022). The result is that it can support a heterogeneous range of editorial projects, from small individual endeavors to large-scale collaborative editions, while ensuring that both the encoded texts and the processes used to create and transform them remain transparent, verifiable, and adaptable far into the future. VEGEST adopts standard TEI models by using a modular ODD-based structure that allows for the inheritance and extension of TEI schemas. Documentation and schema files are tracked alongside each other in version control systems, increasing the chances that both the rationale and the rules for encoding remain accessible (see Huskey 2023). Second, and crucially, the framework incorporates a reproducible software environment based on GNU Guix (GNU Guix Developers 2026), a functional package manager capable of specifying and replicating entire computational environments down to the level of compiled programs. By using only certifiable open source software and by declaring the exact version of all tools and libraries that are required for each transformation step—whether validating XML, transforming documents with XSLT, or generating PDFs with TEX-based systems—the framework ensures that future researchers can reproduce results even as operating systems and software ecosystems evolve.
For the example of Jinendrabuddhi’s chapter 5, we will demonstrate how this framework helps us to:
The use of a functional package manager (GNU Guix Developers 2026) makes all but the last application fully reproducible, and allows editors to save any particular configuration of software at an arbitrary point. We argue that this extends the FAIR principles, accepted for the data and documentation, to the software stack necessary to work with the former two elements (see Barker et al. 2022; FAIR4RS; Vallet, Michonneau, and Tournier 2022), and that this extension goes beyond current packaging and containerising solutions (see Alser et al. 2024). This solution is also applicable to high-performance computing environments, which will facilitate emerging editorial workflows that involve machine learning elements and cannot be completed in a reasonable amount of time on a personal computer any more.