Partecipazione di J. Bogdani e L. Cardarelli al convegno internazionale CAA 2026
Pubblicato il 1 aprile 2026
Partecipazione di J. Bogdani e L. Cardarelli al convegno internazionale CAA 2026
Il primo aprilwe 2026, i ricercatori del LAD (Laboratorio di Archeologia Digitale) della Sapienza University of Rome, Julian Bogdani e Lorenzo Cardarelli, parteciperanno al convegno internazionale CAA 2026 da tenersi a Vienna. Il convegno, che si svolgerà dal 31 marzo al 4 aprile 2026, rappresenta un’importante occasione per presentare i risultati della ricerca condotta dal team del LAD nel campo dell’archeologia digitale e dell’applicazione dell’intelligenza artificiale (AI) nello studio dei materiali ceramici.
L’intervento verrà presentato durante la sessione S20: Digital Archaeological Collections as AI Training Data, organizzata da Vera Moitinho de Almeida, Nevio Dubbini, Aurore Mathys, Gabriele Gattiglia, e sarà intitolato “Open Archaeological Workflows: the example of pottery”.
Abstract in inglese:
Ceramic materials are considered to be among the most fundamental sources of archaeological knowledge, as they provide key information about chronology, function, technology, and cultural interactions (Hunt 2019). However, the study of ceramics remains an arduous process, entailing nu- merous manual stages—from cleaning and drawing to classification and pub- lication. This fact has had a significant impact on the percentage of items published compared to those excavated. Moreover, this fact has had a considerable impact on the percentage of items published in comparison to those excavated. Moreover, a significant proportion of valuable ceramic information remains confined within legacy data sources, including printed catalogues, excavation reports, unpublished archives, and drawings. These materials are seldom digitised or standardised into machine-readable formats, hindering effective access and utilisation. In parallel with this, archaeology has undergone a transformative shift towards computational approaches and the large-scale use of artificial intelligence (AI) and machine learning (ML) systems for over a decade (Gattiglia 2025). These technologies are increasingly reliant on digital archaeological collections as a source of training data. However, the integration of ceramic datasets into AI workflows gives rise to significant methodological and theoretical challenges in terms of data quality, interoperability, standardisation, and semantic consistency. The critical question is not simply whether AI can be applied to the documentation and analysis processes of archaeological ceramics, but how legacy documentation can be transformed into high-quality, ethically curated datasets that preserve archaeological context while enabling computational analysis. This contribution presents an open-source and open-weights modular framework designed to transform raw and diverse ceramic documentation into structured, reusable, and interoperable data suitable both for classical archaeological analysis and data-driven research. The still under active development framework already provides a reproducible computational environment for processing and analysing ceramic corpora. Its workflow is organised into three main stages: 1) Digitisation and Preprocessing of Legacy Data. Using a combination of computer vision models such as YOLO and the Segment Anything Model (SAM), the framework automates the extraction of ceramic profile drawings from both unpublished archaeological documentation (drawings table) and published materials (PDFs). This process converts static 2D illustrations into structured digital representations enriched with metadata, including contextual information, dimensional data, and other descriptions where available. The automated detection pipeline significantly reduces processing time up to 20x compared to manual methods. 2) Inking, Semantic Enrichment, and Vectorisation. Raw drawings are processed through an automated inking pipeline that standardises line weights and visual conventions, making them publication ready. The extracted data are then transformed into semantic vector representations that integrate morphological components of each vessel—profile, decorations, handles—as discrete, machine-readable entities. This semantic segmentation ensures postprocessing and interoperability across different archaeological databases and repositories, allowing comparative analysis at the component level rather than treating each vessel as a raster indivisible unit. 3) Similarity Analysis. Through ML algorithms, including unsupervised clustering and metric learning approaches, a specific module identifies morphological similarities and functional patterns within and across collections. The framework employs both unsupervised contrastive learning to discover latent typological structures and supervised methods that incorporate archaeological domain knowledge through expert feedback. These methods facilitate the creation of new typological groupings based on quantifiable morphological features, while highlighting previously unnoticed relationships between large ceramic assemblages from different contexts or periods. Furthermore, this approach allows for a theoretical comparison with the traditional classification of ce- ramics, offering new information and reflections on the concepts of ‘similar- ity’ and ‘typology,’ which are widely used in the study of ceramics and material culture. Importantly, this process transforms digitised ceramic data into high-quality training datasets that can be reused for future AI research in archaeology.
The framework, currently under development at LAD (Laboratorio di Archeologia Digitale) at Sapienza University of Rome which has been tested on 1) Raw archaeological drawings from different excavations of the Sapienza University 2) Published ceramic catalogues from monographs and articles. 3) Standardised ceramic profiles from the AIPPA, a comprehensive dataset for quantitative analysis (Cardarelli 2024). The findings illustrate the efficacy of the computational pipeline in reducing the time required for the publication process, enhancing the layout, and ensuring a data-driven approach to material analysis at the analytical level. More significantly, the ML-based similarity analysis enables quantitative comparison across large ceramic assemblages, revealing morphological patterns and relationships that would be difficult to identify through traditional autoptic analysis. This data-driven methodology provides archaeologists with computational tools that complement traditional typological expertise, operating at scales and with levels of consistency that enhance the interpretative process without replacing human judgment. All components of the framework will be released as open-source software under permissive licensing, and the project encourages transparent data workflows through public documentation of preprocessing steps, model parameters, and uncertainty levels. Each stage of the pipeline is designed to be inspectable and reproducible, allowing other researchers to adapt the framework to their own ceramic assemblages or to critically evaluate the computational methods employed.
References
- Cardarelli L. 2024. Morphological Variability and Standardisation of Vessel Shapes in the 2nd and First Half of the First Millennium BC in Continental Italy. Adrias 18. Edipuglia.
- Hunt, A. M. W. 2019. The Oxford Handbook of Archaeological Ceramic Analysis. Oxford University Press.
- Gattiglia, G. 2025. ‘Managing Artificial Intelligence in Archeology. An overview’. Journal of Cultural Heritage 71: 225–33. DOI: 10.1016/j.culher.2024.11.020.
Risorse: