Arch-I-Scan: Automated recording and machine learning for collating Roman ceramic tablewares and investigating eating and drinking practices

Lead Research Organisation: University of Leicester
Department Name: Sch of Archaeology and Ancient History


Global challenge. The value of our social and cultural heritage cannot be overestimated as it affects the direction of our development as humans. How to understand and preserve it is a fundamental question in modern science. Centuries of archaeological investigation has produced millions of artefacts that are part of this heritage and are classified and analysed by human experts, presenting a huge challenge to process and interpret all these remains. We aim to meet this challenge using artificial intelligence to create an unprecedented system for automated artefact classification and collation which can be used by non-specialists, allowing experts to focus on analysing these artefacts for greater understandings of the past.

Specific challenge. The Roman period is exceptionally rich archaeologically. The millions of artefacts from across the Roman world are infinitely more informative about people's lives than iconic monuments. However, these artefacts are currently under-utilised in studies of social practices - particularly the wealth of ceramic tablewares used by almost everyone from senator to slave. This under-utilisation is largely due to these artefacts' extensiveness and inherent difficulties in recording them all. For decades ceramics have been recorded and analysed selectively (e.g. only diagnostic sherds of specific vessel types) for chronological sequencing of individual excavations, or for investigating trade patterns between regions, rather than to answer socio-cultural questions, e.g. how were particular vessels used and in what circumstances. More comprehensive recording can facilitate consumption-oriented analyses for new levels of understanding of varying social practices among the diverse communities that made up the Roman world.

This project develops a state-of-the art image-recognition and machine-learning service, Arch-I-Scan, a proof-of-concept experiment of which was successfully carried out for the AHRC network, 'Big Data on the Roman Table'. We will train this service and develop its machine-learning capacity on 100,000s of Roman tableware remains in extensive collections from different social and regional contexts in Roman Britain - London, Colchester and its environs, Vindolanda (on Hadrian's Wall), and Leicestershire. Machine training will move from recording complete/near complete vessels to more fragmentary remains. Used on handheld devices (e.g. mobile phones) by non-specialists and specialists, Arch-I-Scan will automatically recognise and record details of pottery remains and digitally collate and store large quantities of data. Roman tableware remains, often from large-scale production centres (e.g. samian ware from South Gaul), constitute some of the most easily recognisable and extensive bodies of archaeological data with high levels of similarity, in ranges of forms and fabric types, across a wide geographical area. Thus, besides being crucial evidence for Roman food- and drink-consumption practices in different social contexts in Britain, the selected material comprises an excellent body of artefacts to ensure wider application of Arch-I-Scan at other Roman sites, in Britain and beyond. Once Arch-I-Scan is sufficiently trained and has recorded and 'learned' to classify the artefacts from these collections, the resulting datasets will be made freely available for other archaeologists to use as comparanda in their own analyses. This can lead to more comprehensive analyses across Roman archaeology for more socially-oriented questions. Arch-I-Scan can continue to 'learn' from these and other types of pottery as well as other archaeological artefacts. Greater knowledge of how the micro-histories of objects - and the 'human-thing entanglements' of their micro-archaeological contexts - play important roles in our understandings of socio-cultural practice in the context of Roman history can also transform material-cultural approaches to social practice in global history.

Planned Impact

With dwindling public funds, government, charitable and professional cultural heritage and archaeological organisations have inadequate resources for detailed, comprehensive classification, and digital collation, management, and analyses of the full artefact datasets in their care. This project sets agendas for more comprehensive, accurate, and cost-effective artefact collation (from past, recent and future excavations) to build more robust, analysable datasets that can be more easily digitally recorded, managed and disseminated for more effective interpretation. It can influence the practices of a range of commercial and government archaeologists, heritage managers, and museum professionals. Enthusiasm for Arch-I-Scan's proof-of-concept experiment from museum professionals and archaeologists working in developer-funded archaeology across the world who participated in the AHRC-funded network, 'Big Data on the Roman Table' (BDRT), and requests by some for involvement in this current project, signify its potential impact among these sectors.
This project has the potential to influence heritage management policies. By facilitating more comprehensive artefact recording and data sharing cultural-heritage organisations will be able to improve on guidelines for effective collation and management of artefactual data.
The custodians of artefact collections used as case studies - the Univ. of Leicester Archaeology Service and the partner organisations Museum of London, Museum of London Archaeology, The Vindolanda Trust, the Colchester and Ipswich Museums; - will benefit from Arch-I-Scan's machine-learning expertise in recording Roman tablewares in their collections and from the resulting annotated datasets. The project outputs will also be useful for other professional archaeology and heritage organisations, providing an AI service for efficient, accurate and automated recording of, and comparanda for, Roman ceramics in their collections.
The possibilities Arch-I-Scan provides for greater non-specialist involvement in artefact recording will also be of economic benefit, particularly in developer-funded archaeology, freeing up specialists' time to focus on more comprehensive analyses and interpretation of their respective organisations' datasets, and facilitating the sharing of information among these organisations. Roman pottery specialists are more usually employed in museum and professional archaeology organisations rather than HEI organisations.
The project can benefit community groups and members of the public interested in archaeology and museum collections. Museum volunteers, community archaeology groups, and students involved in the recording processes can engage first-hand with digital, state-of-the-art archaeological recording processes and develop skills in this area. More comprehensive analyses of such large datasets of Roman tablewares also means they can be used more effectively, both by HEI and non-HEI archaeologists and museum specialists, for interpreting Roman ways of life and particularly differing eating and drinking practices in the various parts of the Roman world. Such interpretations can enhance museum display, improve visitor experience, and provide better understandings of the role material culture can play in informing on Roman foodways.
Government and professional organisations are also responsible for the excavation and management of extensive datasets of other types of pottery, beyond Roman tablewares, and other types of artefacts. Arch-I-Scan can potentially provide a service to record other types and fabrics of pottery (e.g. other Roman wares, Medieval and post-Medieval) to further facilitate more effective, comprehensive and consistent digital collation and management of many of their datasets.
The system also opens avenues for the AI technology in other areas where AI's mistakes can be rectified immediately once detected, e.g. health sciences and high-security applications.


10 25 50