EAA2020: Abstract

Traditionally, the study of Bone Surface Modifications (BSM) has been done by classifying them systematically, which has produce very different results if done by one team of scientists or another. Tooth marks are a big part of BSM. They can be found in any archaeological assemblage and have been studied thoroughly during the past two decades. The importance of their well recognition is knowing what taphonomy calls “agency”, so that further interpretations about the origin of the archaeological site can be made. In previous studies, pits have proved themselves to be good tools to distinguish some carnivores from another’s (crocodile from felids, for example), but they fail when comparing animals from the same family order, which are much more similar. In this sense, what has been never done is trying to discern among carnivores just by looking at scores. They have always been considered to be more variable, hence less useful. In the present work, Machine Learning (ML) algorithms (CNNs) are used to compare crocodile and wolves tooth scores. The main goal is to spread the use of ML in Archaeology, as well as to test the limits of this new method, in an attempt to make our science much more objective, reducing to minimum the personal bias introduced by the investigator.

Taphonomy, Machine Learning, Tooth Scores, Carnivores, Convolutional Neural Networks

 Abstract (PDF)

Natalia Abellán Beltrán^1,2