Querying Pre-Modern Cultures with LLMs
St Edmund Hall Oxford / Online ; 20-22 November 2026
Call for Papers & Workshops
Following the first Ars Inquirendi conference in December 2025, we invite proposals for a second hybrid exploration of the impact of Large Language Models (LLMs) on the study of cultures before the dominance of movable-type printing (for example, Western Europe up to the sixteenth century, Russia the eighteenth, or Central Asia the late nineteenth).
Even in the few months since that first conference, the ground has shifted dramatically. New LLM techniques – more powerful sometimes by the week – are allowing individual scholars, without programming experience, to bring new objects of study to light within the pre-print-era archive: objects previously beyond the reach of even lavishly endowed projects. The inferential power of LLMs is especially promising for pre-print cultures, since their archives are overwhelmingly one of loss, dispersal and uneven survival. Even the old problem of machine transcription of handwriting from any era, far harder than for print, is now advancing rapidly.
Hearteningly, we are also learning more about the grounding of these systems in the core interests of the humanities. LLMs are language-shaped machines, trained on human expression and responsive to all the breadth and detail of natural language. Some of the most striking new techniques for eliciting powerful behaviour from models depend not on esoteric programming, but on skills such as rhetoric, philology, hermeneutics, poetics, psychology, and the art of asking. This is renewing perceptions of the humanities’ importance across our economies, institutions and shared intellectual life.
However, pre-print-era scholarship is also exceptionally vulnerable to the floods of synthetic text, shallow automation, and dangerously plausible AI nonsense that we all see, inside and outside the academy. Not only are its extant materials just a fraction of what existed; only a fraction of those materials fall within the data of current models, which train overwhelmingly on post-print production, above all twenty-first-century internet culture, to build their ever more convincing worlds.
Ars Inquirendi 2026 therefore invites contributions, from papers to workshops, that explore not only how scholars should use LLMs, but also shape the entire means by which the pre-print past becomes machine-readable. Topics might include, but are not limited to:
From Dream to Actuality. We especially welcome demonstrations of projects that use LLMs to make real what had previously been technically unreachable, or even unthinkable. These might include: reconstructions of lost texts, libraries, subcultures and other forms of lost knowledge from across the pre-print-era world, along with the relations between them; re-imaginings of history itself as a field of possibility, from individual motivations and landmark historical events to manuscript traditions; scholastic disputation engines; historical route, map, or landscape systems; prosopographical workbenches; catalogue-enrichment tools; and any projects that transform scholarly argument into working systems.
The Pre-Print AI Ecosystem. What would it mean to build a genuinely pre-print AI ecosystem – and what are the challenges? Should we train sovereign LLMs excluding later materials, whether distilled from frontier models or built from scratch, or concentrate instead on boosting the volume of pre-print material available for frontier-model training? How should we address copyright, public-domain distortions, machine-transcription errors, and low-resource languages? How should LLMs, frontier or local, work within the wider ecology of corpora, other computational methods, and older scholarly practice – from HTR / OCR, Retrieval-Augmented Generation (RAG), metadata enrichment, entity recognition, or knowledge graphs, to corpus analysis? What are the strengths and weaknesses of LLMs relative to these other methods ?
Workshops. We encourage proposals for workshops of one to two hours teaching techniques that humanities researchers can use themselves. These might include the bread and butter of using LLMs well, whether in a chat window or through more elaborate interfaces; LLMs in teaching; the ethics and pitfalls of LLM usage; comparative use of different models and systems; coding with LLMs (daunting at first glance but overwhelmingly driven by natural language, even among expert programmers); and introductions to how LLMs work under the hood, including how scholars might make, adapt, fine-tune, or distil their own. We are particularly interested in humanistic practice itself as a source of method and standards: wiki-like approaches to explanation, structure, and staged understanding as ways of eliciting stronger model behaviour; prompt design as rhetoric; and the transferability of such skills beyond the academy, from cultural institutions to commercial AI labs and government.
Conference format. Ars Inquirendi 2026 will be a hybrid event taking place on 20 to 22 November 2026, online and in-person at St Edmund Hall, Oxford. Lunch will be provided on 20 and 21 November, with a conference dinner that evening. Following the format of Ars Inquirendi 2025, all talks (except keynotes) will be pre-recorded and released a week beforehand, allowing us to devote live conference time to discussion. Proposals for papers and roundtables should be circa 250–300 words. Workshop proposals should be c. 500–700 words and specify the intended audience, duration, and technical requirements; workshops should be followable both in person and online. Please include a short biography of no more than 100 words, along with your preferred format and whether you wish to participate online, in person, or either.
Deadline for proposals: 31 August 2026 ; notification of acceptance by 15 September 2026. Please send submissions to arsinquirendi@gmail.com
Organising committee. Matthew Barber (Aga Khan University), Rubén González Cuerva (Instituto de Historia-Centro Superior de Investigaciones Científicas-Madrid), Stephen Pink, Yasmin Faghihi (University of Cambridge), Estelle Guéville (Yale University & Bibliothèque nationale de France), Huw Jones (University of Cambridge), Anthony John Lappin (University of Stockholm), Henrike Lähnemann (University of Oxford), Roger Martínez-Dávila (University of Colorado & Plus Ultra Collective),
About Ars Inquirendi 2025. Held online and in Oxford on 4–7 December 2025, Ars Inquirendi convened pre-modernists, archaeologists, digital humanists, and AI researchers to test – rather than merely celebrate – what large multimodal models can do for pre-print-era evidence. Keynotes by Maurizio Forte (Duke University) and Roger Martínez-Dávila (University of Colorado) framed AI both as an epistemic partner in archaeology and as a tool for historically disciplined simulation. Panels then evaluated LLM performance on Old English NLP benchmarks (Mark Faulkner and Elisabetta Magnanti, Trinity College Dublin), the Graphom project’s attempt to model the pre-print “graphosphere” (Stephen Pink), AI-assisted restructuring of manuscript catalogues (Madeline Rose, Trinity College Dublin), visual language models and traditional HTR for Church Slavonic, Glagolitic, and Ukrainian materials (Achim Rabus, Freiburg), reconstruction of fragmented birchbark letters (Dmitri Sitchinava, Potsdam), AI and homoiconic coding, where code itself uses the medium of natural language (Damon Wischik, Cambridge), and models for transcription and exploration of historical maps through MapReader (Peter Broadwell, Simon Wiles, and Katherine McDonough). Practical workshops by Anthony Harris (Cambridge) and Ben Kiessling (PSL) examined prompt design, confidentiality, citation, hallucination control, tokenization, Unicode representation, and the structural mismatch between current LLM design and historical corpora. Roundtables also included Laura Morreale (Harvard), Sarah Bowen Savant (Aga Khan University), Tom Revell (Oxford), and Daniele Nardi (Sapienza). The conference’s central achievement was to shift discussion from general claims about AI to concrete scholarly evaluation: which tasks can now be responsibly accelerated, which still require specialized models or human judgment, and how future infrastructure for the pre-modern archive should be built. See the OMS webpage for conference recordings, and the synthetic proceedings at the Plus Ultra Collective website.
