Abstract
This article presents the aims, technical processes, and initial results of the Arch-I-Scan Project, which is using artificial intelligence and machine learning to enhance the collection of Roman ceramic data so that these data can contribute more effectively to improved understandings of Roman foodways. The project is developing a system for the automated identification of ceramic types (fabrics, forms and sizes), and potentially the automated collation of the resulting datasets, to facilitate more holistic recording of these big archaeological data, and avoiding the current time-consuming and costly specialist process for classifying these artefacts. The particular focus of the project is to develop datasets that are suitable for inter- and intra-site analyses of eating and drinking behaviours in the Roman world which require more comprehensive recording of these remains than the current sampling practices used to date sites or to investigate production and trade practices. The article includes a brief overview of approaches to material culture, particularly ceramics, for improving understandings of cultural patterns in past food-consumption practices. We then outline the project’s rationale and planned approaches to harnessing the potential of artificial intelligence and machine learning for artefact recording, specifically of Roman terra sigillata tablewares, and the processes used to develop a sufficiently large dataset to develop and test the AI system. The important aspect of this article is the changes made to these processes to mitigate the impact of the Covid pandemic on our ability to record large datasets of real ceramics. These changes involved the development of simulated datasets that substantially enhance our original real dataset and the accuracy of identification. Here we present our results to date, contextualised within the overall aims of the project and briefly discuss the steps we are taking to improve these.