A Bayesian Solution to the Controversy over the Identification of Bone Surface Modification in Paleoanthropology

Summary

Bone surface modification (BSM) remains a primary source of taphonomic inference in paleontological and archaeological contexts. However long-standing debates in BSM studies have undermined the utility of this approach. We use an objective machine-based learning algorithm rooted in Bayesian probability theory designed to quantify the level of uncertainty associated with a formal assignment of agent to individual BSM. Our multivariate Bayesian model, trained on large assemblages of experimentally generated BSM, accurately assigned agent of modification to an out-of-sample test with an average of 86.5% accuracy. Here we present the results of our updated probability model. We improved upon our existing model with the introduction of novel variables and an increased sample of experimentally generated training data produced by Nile crocodile feeding, spotted hyena feeding, large ungulate trampling, and human unmodified rock butchery. Here we also present the posterior distributions of model parameters associated with the morphology of individual BSM. Certain BSM attributes frequently associated with stone tool butchery marks (e.g. perpendicular orientation relative to the long axis of the bone, occur as a result of non-human agents of modification. Here we quantify the probability that key attributes are likely to occur with each respective agent of modification.

This Resource is Part of the Following Collections

• Rethinking Methods of Faunal Analysis   •
• Society for American Archaeology 82nd Annual Meeting, Vancouver, BC (2017)

Keywords

Spatial Coverage

Record Identifiers

Abstract Id(s): 15458