This article presents analyses that have been carried out on the data compiled by the Arch-I-Scan Project. The project is ultimately concerned with using Roman tablewares for consumption-oriented analyses, particularly for investigating socio-cultural behaviour around eating and drinking. In addressing this concern, not only are the forms of such tableware vessels important but also their actual and relative sizes. This article, therefore, used rim diameters of different forms of terra sigillata vessels, to analyse the size ranges of each form and to assesses evidence of differences in the sizes of each of these vessel forms across three different locations – Leicester, London and Vindolanda. It also assesses whether there are any evident patterns in the sizes and size ranges of the different traditional morphological, or shape, categories, here called 'use labels' (i.e., so-called bowls, plates and dishes), among these locations. It speculates on what these might tell us about differences in the eating and drinking practices among these locations. Finally, it presents a process for the automation of rim diameter measurement from photographs of these vessel remains, attempting to deal with the evident variation in how recorders have physically measured rim diameters. We also discuss the issues faced during this automated rim diameter measurement process, whether using rim sherds from our image dataset or taking further specific photographs of rim sherds (e.g., blur in the images, lens distortion), and provide suggestions for more reliably automating rim diameter measurement.
This article discusses the process for creating a dataset to train and test the core and the corrector AI classifiers of the Arch-I-Scan Project. This dataset includes the photographs of real sherds that were compiled by the project from the collections of our project partners as well as synthetic sherds that were generated to supplement the low numbers of real sherd training data and to account for the imbalances among the numbers of real sherds per class (i.e., per vessel form). This article presents the simulation process used to create the synthetic sherds, as well as the process we used to decrease the differences between images of real and simulated sherds so that they were useable together for training the machine. Finally, the compilation and size of the resulting overall dataset are outlined.
This article describes the structure and content of the repositories which are available on the project's GitHub site. These repositories contain the code and models that the Arch-I-Scan Project created over the course of the development process towards training the core AI sherd classification machine, and include the trained models and the photographs and simulated data of the complete/near complete vessels used for the project's pilot study.
For the London photographing sessions, the Arch-I-Scan Project photographed the terra sigillata from four of the largest archaeological sites excavated in London in the last ten years by the Museum of London Archaeology (MOLA), as well as 389 complete vessels held in the collections of the London Museum. Thousands of sherds were retrieved and photographed from over 1,900 boxes of pottery by groups of Arch-I-Scan Project members and volunteers for c. 60 days. This article focuses on the decision-making processes and logistics required to implement these sessions. Additionally, it contains observations on the Arch-I-Scan Project's process from the perspective of a Roman pottery specialist working in commercial archaeology, including the benefits and limitations for the project objectives.
Digital transformation is inevitable in archaeological research, from data collection to the formatting of rules that define fundamental research topics, such as that of typological arrangements. Digitising the archaeological record is a situated process that inherits substantial complexity when it comes to typological classification, since traditional material analyses rely strongly on visual inspection, (tacit) expert knowledge, and first-hand impressions, rather than numbers and algorithms. New numerical and digital approaches are needed that incorporate human contributions and capacities, both from a data science and a material specialist perspective. In this article, we propose a visual-analytics workflow to promote a white-box model for typological class definition of archaeological pottery that aims to align the archaeological and the data analytical rationale, and keep humans in the loop. Using technical pottery drawings as input, we build open-source processes for digitising, assessing and maintaining a typological model of this pottery. Three components of visualisation are discussed: shape representation methods, association representation methods, and label representation methods. The workflow is applied to a dataset of Sagalassos Red Slip Ware profile drawings, a tableware produced in Southwest Anatolia, akin to Italian and South Gaulish terra sigillata. In our study, we assess and map variation within five functional classes, revealing potential subclasses and identifying their mode. Our workflow enriches existing digital practices in the archaeological toolkit, providing higher granularity, integration, and replicability.
