Report on AIUCD2023, Siena, July 5-7

Our team has been present at this year’s AIUCD conference from July 5^th to 7^th, where we presented a poster titled “Unsupervised Information Extraction from Medieval Charters”. The 12th annual conference of the Associazione per l’Informatica Umanistica e la Cultura Digitale (AIUCD) took place at the university of the historical city of Siena, nestled in the scenic Tuscan hills. With its rich history dating back to 1240, the prestigious institution stands as a testament to the spirit of learning, tradition, and innovation.

The conference was a rich and varied event that brought together scholars from multiple disciplines – digital humanities, history, computer science, philology, and many more, promoting interdisciplinary dialogue and collaboration. Under the topic of “Digital Memory: Forms of Text and Knowledge Organization”, experts mostly from across Italy and beyond gathered to share their latest research findings, engage in thought-provoking discussions, and explore future prospects of digital humanities. From workshops on TEI encoding and digital edition visualization to panel sessions on infrastructures for research and open science, the conference offered a comprehensive insight into the state of digital humanities in the contemporary academic context.

Out of the myriad of topics presented, some stood out due to their connection to the study of medieval charters that lie in our focus of research.

Chiara De Bastiani’s presentation examined the process of integrating imperial diplomas from Germanic rulers related to Veneto into the OntoVE knowledge base. Thescope of the speaker’s project is to display the relation between the Germanic population and the Veneto territories via Germanic cultural heritage items in that zone. The bibliographic and diplomatic information (issuing date and place, the content of the documents and the contained actors and places) about these diplomas are taken from the resources of the dMGH (Monumenta Germaniae Historia digital), and modelled into the knowledge base by using an “OntoVE_WrittenRecords module”. 

The project’s unified bibliographic ontology combines elements from FRBRoo and Bibframe, where e.g. the contents of the acts are modelled as performative events (instances of frbroo:F8_Event) and information on falsifications/interpolations is modelled as events too, in the introductory section of the record. For the future, it is planned to further test and add an user interface that makes it possible to use the database also as user without experience with SPARQL queries.

Timo Korkiakangas, who also spoke at our very own Digital Diplomatics 2022 conference, presented his work on a digital diplomatic edition of 72 10th-century charters from Lucca. These charters preserved in the Archivio storico diocesano have previously been transcribed only in an incomplete and imprecise way by Domenico Barsocchini in the 19th century. The principles of Korkiakangas’ edition imitate the Chartae Latinae Antiquiores series based on the Leiden Conventions for epigraphic and papyrological editions. Through the use of regular expressions, the Leiden markup was converted into a markup conform with the ALIM (Archivio della Latinità Italiana del Medioevo) guidelines subset of TEI. The tagset is largely compatible with the CEI guidelines as well.

The speaker stressed how mainstream corpus linguistics largely overlooks the textual-scriptural characteristics of its historical texts, a shortcoming he proposed to address by integrating diplomatic information into linguistic annotation. He achieves this by creating 1. a diplomatically enriched treebank, and 2. a linguistically enriched diplomatic edition, the latter still work in progress. The linguistic annotation in the CoNLL-U format allows for the addition of knowledge crucial for the identification of historically relevant key terms, paving the way for more precise diplomatic and historiographical research.

Highlight of the first conference day was an extensive poster session with a total of 32 posters in the rectorate aula in the city centre, in combination with a prior poster slam where all the posters were presented in a very condensed form.

Our own poster dealt with the question of how historical important information could be computationally extracted from huge amounts of charters with reduced time and resource requirements. Our approach employs a prototypical network which calculates class prototypes from the embeddings of a small support set with only a few annotated examples. New data fed to the network will then be classified by comparing it to these prototypes, computing the Euclidean distance of the means of embeddings. The training of the network relies on Adam optimizer, a learning rate scheduler and the use of a cross-entropy loss function.

On entity extraction, our approach reached a 94,5% F1 score (n=50), compared to our “1000 NLP charters” evaluation set which is constantly improved by annotations for different task-specific purposes. The proposed few-shot classification has the advantage of relying only on a minimal amount of annotated data for training while allowing applicability on big unlabelled datasets, rendering it a valuable approach not only for medieval charters but also other types of historical documents.

Our poster met the interest of several scholars at the conference, which we appreciated much. It is worth mentioning how many students and early stage researchers have been present at the event, demonstrating its inclusiveness for all kinds of career stages. We therefore encourage other young scholars to participate at this conference in the future and very much look forward to next year’s edition.

Franziska Decker, Tamás Kovács

---

---