Quantifying stylistic distance between Athenian vase-paintings
Adrian Ryan (University of Kwazulu-Natal, South Africa)
Digital Classicist London & Institute of Classical Studies seminar 2013
Friday June 28th at 16:30, in Room S264, Senate House, Malet Street, London WC1E 7HU
Athenian painted pottery is a valuable archaeological resource, not only because of the relative abundance of these ceramics, but also because the stories told by their figured paintings can often provide a valuable supplement to the literary record. Getting the most of these artefacts requires more precise dating than trapped charge methods can provide, and with most of the relevant archaeological contexts unrecorded, stratigraphy is often of little help. Style analysis is perhaps the most useful tool in this regard, not only as it allows scholars to identify when different pots have been painted by the same hand (thus placing limits on how far apart in time they could have been produced) but also because it allows the relationships between different painters and schools to be mapped out.
Unfortunately, not only has the century old approach of style analysis come under attack for being too subjective, but very few young scholars are trained in the art, and there is a danger that this skill may be lost within a generation or two. On the other hand, computational approaches may both lend some objectivity to this art and also allow that the practice of attribution survive its ageing exponents. However, while there exists a considerable body of literature describing the use of machine learning to aid in the analysis of artistic style very little of this research is applied to Greek pottery and even less has been done to quantify stylistic distance, a necessary step towards mapping the relationships between the various painters and schools.
Fitting into this gap are two distance functions that are proposed in this paper, both of which take as their starting point the way in which vase-painters render the heads of their male figures. These two functions are applied to a selection of male figures rendered by various vase-painters, and the results are evaluated against predictions made on the basis of art historical domain knowledge. The results show that these functions match art historical intuition to a far greater degree than can be explained by chance. Together with a report on the most recent findings using this method, this paper also provides a detailed exposition of the way in which the functions are constructed and evaluated, justifying the methods on the basis of both mathematical and art historical criteria. Finally, the paper draws attention to questions that still remain unanswered and plots a possible trajectory for future research in the application of machine learning to style analysis of Greek vase-paintings.
The seminar will be followed by wine and refreshments.