AIUCD 2026 conference report

Cagliari, the capital of the Italian island of Sardinia, hosted the AIUCD’s yearly meeting between 3 and 5 June 2026 — in the stunning hilltop historical district that overlooks the Mediterranean Sea and seamlessly blends ancient history with a remarkably preserved 2nd-century Roman amphitheater, alongside lagoons that are home to thousands of pink flamingos. At UNICA (Università degli Studi di Cagliari) we from the UNI Graz DH delivered an oral presentation and two posters:

• Tamás Kovács, Anguelos Nicolaou, Johannes Laroche, Georg Vogeler. “Roman and Justinianic Legal Terminology in Thirteenth-Century Western European Diplomatic Sources” (oral)
• Christian Marcantonio, Giuseppe Consolo. “SEBASTIAN: uno strumento per l’organizzazione, l’analisi spaziale e la visualizzazione dei dati storici” (poster)
• Selina Galka, Cristina Scuderi, Marcella Tambuscio, Rolf Wissmann, Georg Vogeler. “From Violin Lessons to Linked Open Data. Reconstructing Tartini’s Scuola delle Nazioni through Digital Prosopography and Digital Editions” (poster)

Roman and Justinianic Legal Terminology in Thirteenth-Century Western European Diplomatic Sources grew out of our ERC-funded DiDip project (From Digital to Distant Diplomatics), and it asked a simple question: how did the legal language of Justinian’s Corpus Iuris Civilis actually find its way into the everyday charters that notaries and chanceries produced across medieval Central Europe?

The reception of Roman law has been studied for well over a century. But most of that scholarship rests on a relatively small, carefully chosen set of documents. We wanted to know whether the patterns inferred from dozens of charters would still hold when tested against tens of thousands. Our source was Monasterium.net, the largest digital charter archive in Europe, with more than 600,000 documents, from which we built a focused study set spanning forty-five archival collections.

The methodological problem, and why ordinary tools fail. Anyone who has tried to run standard text-analysis tools on medieval Latin knows the frustration. The same word appears in five or six different spellings; word order is free; and scribes constantly interrupted otherwise stable formulas by inserting case-specific clauses in the middle. Exact string matching collapses immediately, and even more flexible “gapped” methods struggle, because legal conditionals can stretch the premise (quod si…) and its consequence (…robur obtineat) across a whole clause. We needed a method that recognizes the structure of a formula even when its surface has been scrambled.

Our approach. The pipeline we presented does this in three moves, and the audience was most interested in the parts that are usually taken for granted. First, an adaptive alphabet: rather than hand-building a normalization table, the system scans the corpus, detects which characters fall outside the target alphabet, and learns their normalizations automatically — so the same code transfers, untouched, to a Greek or vernacular corpus. Second, a data-driven vocabulary size: instead of borrowing the conventional figure used for English-language models, we measure the morphological richness of medieval Latin directly, choosing the subword vocabulary that covers 85% of the corpus’s affix mass. Third, a leave-n-out n-gram engine that deliberately blurs each six-token window by dropping one position at a time, so that an insertion or substitution in any single slot still produces a match. Every resulting fragment is reduced to an integer through polynomial hashing, which turns the whole comparison into fast set overlap — the trick that makes the method viable at the scale of hundreds of thousands of documents.

What we found. The results came in two registers. At the macro scale, the data confirmed the textbook narrative — a clear peak of Justinianic language across the thirteenth century, with the Digesta dominating throughout — but it also delivered two findings that the qualitative literature could only guess at. First, at this scale, institution type matters far more than geography: a Benedictine house in Hungary looks more like a Benedictine house in Austria than like a secular archive in the same town, with a roughly two-and-a-half-fold gap between the most and least “Romanizing” institutional types. Roman legal knowledge travelled along institutional networks — monastic orders, cathedral schools, shared formularies — rather than by simple geographic proximity. Second, adoption was staggered rather than synchronized: individual institutions peak as much as 250 years apart, so what the aggregate curve presents as a single “thirteenth-century wave” is really a sequence of sharp, displaced local peaks.

At the micro scale, the corpus let us prove genuine textual continuity rather than vague familiarity. Among nearly 56,000 scored matches, the long ones — eight words and more — cannot be coincidental, and we could trace technical terms such as per se vel per alios (Digesta 39.4.11) migrating straight from the law books into the charters. The single most striking find was the Digna Vox (Codex 1.14.4) — the constitutional maxim that the ruler professes himself bound by the law — appearing verbatim in a working Göttweig charter. That it surfaces as late as 1480, long after the reception peak, became part of the argument rather than an awkward exception: once acquired, these texts remained a living reference for centuries.

Why it matters. Much of the historical narrative our method recovers confirms what specialists already suspected — and at this scale, confirmation is itself worth something, because it was previously based on a small sample. But the genuinely new contributions are the demonstration that institutional structure, not geography, drove differentiation, the staggered shape of adoption, and — most durably — a reproducible, language-agnostic instrument that can be turned on any corpus built from formulaic language. The tool does not retire the close reader; it tells the close reader where to look next, pointing directly to the documents and scriptoria where the deepest legal learning was concentrated.

The questions afterward especially during the lunch break were exactly the sharp, source-critical ones we had hoped for — about digitization bias in the underlying archive, about distinguishing direct from indirect transmission, and about disentangling institution from region — and they have already shaped our sense of where the project goes next: dating every match precisely, so that we can watch a single legal term spread through a region within a decade or two, and adding a semantic layer so the system grasps not only that two formulas share a shape, but that they share a meaning.

A companion tool from the project: SEBASTIAN. Of the two posters, one speaks directly to the same problem that animates our own work — turning scattered documentary evidence into structured, analyzable data — and it comes from within the DiDip orbit: SEBASTIAN: uno strumento per l’organizzazione, l’analisi spaziale e la visualizzazione dei dati storici, by Christian Marcantonio and Giuseppe Consolo, the latter a member of the DiDip project. Where our presentation tackled the textual side of distant diplomatics, SEBASTIAN tackles the spatial and relational side.

SEBASTIAN (the code is openly available at github.com/xpi-an/SEBASTIAN) is a Python application with a graphical interface that lets a historian collect, manage, and explore historical entities extracted from documentary sources — people, places, and the references attached to them — without writing a single line of code. Each entry captures the things that matter for historical-geographical analysis: the document identifier, the date in several forms, the entity and its type, the toponym as it appears in the source alongside its modern equivalent, geographic coordinates, and a set of landscape categories. Everything is stored in a lightweight relational SQLite database, which keeps a whole project in a single, portable file.

What makes the tool genuinely useful, rather than just another form to fill in, is what it does after data entry. A built-in analysis module — powered by pandas — counts entities by type or by modern place, plots their distribution across centuries, and cross-tabulates landscape categories against places and periods, all behind a few filters in the interface. A small but telling convenience is the smart coordinate reuse: enter a place already in the database, and its latitude and longitude are filled in automatically, quietly removing one of the most tedious and error-prone steps in this kind of work. Results export to CSV, and — the part that drew people to the poster — turn into interactive HTML maps through Folium, with single-record views, aggregation by place, and basic clustering. The roadmap the authors sketched — Excel import from a predefined template, management of multiple project databases, hierarchical controlled vocabularies, explicit handling of spatial uncertainty, and GeoJSON export — reads like a thoughtful answer to the practical frictions that every historian working with messy documentary evidence eventually runs into. Together, the talk and SEBASTIAN make a neat pair: both rest on the conviction that the move from close reading to large-scale analysis depends, first and foremost, on getting the data into a clean, queryable, and shareable form.

And from elsewhere in the institute. The third contribution came from the same Digital Humanities institute in Graz, though from a different research group: From Violin Lessons to Linked Open Data. Reconstructing Tartini’s Scuola delle Nazioni through Digital Prosopography and Digital Editions, by Selina Galka, Cristina Scuderi (Università Statale di Milano), Marcella Tambuscio, Rolf Wissmann (Schola Cantorum Basiliensis), and Georg Vogeler (see the Tartinians project homepage). Selina Galka, a colleague at the University of Graz, works among other things on the digital edition project Die Memoiren der Gräfin Schwerin (1684–1732), and here turned the same distant-reading instincts toward the history of music. The poster reconceives Giuseppe Tartini’s eighteenth-century Scuola delle Nazioni not as a fixed institution but as a dynamic network of pupils, patrons, printers, and other intermediaries, captured through a digital prosopographical dataset — each piece of information modelled as an individual scholarly assertion — and published, together with digital editions, as Linked Open Data.

Standing between these three contributions, each starts from the same premise: that historical knowledge buried in unstructured sources only becomes analyzable at scale once it is modelled, made explicit, and rendered queryable. Whether the object is a charter, a place, or a person, the discipline is converging on a shared craft.

Cagliari and AIUCD 2026 was, in the end, a reminder that the tools we build are only ever in service of that older question: how to read what earlier people left behind, more of it, more carefully, and together. We came away with sharper questions than we arrived with, which is about the best thing a conference can do.

---

---