A deep learning-based taphonomical approach to distinguish the modifying agent in the Late Pleistocene site of Toll Cave (Barcelona, Spain)

One of the most widely used methods to associate lithic tools and bone assemblage in archaeological sites is the identification of cut-marks. However, the identification of these marks is still problematic in some localities on account of the similarities with the modifications generated by non-human processes, including biostratinomic and post-depositional bone surface modifications. Toll Cave (Barcelona, Spain), with chronologies between 47.310 BP and 69.800 BP, is one of the case studies where the cut-marks could easily be confused with abundant grooves generated by the dragging of sedimentary particles (e.g. trampling), but also with the scores produced by carnivores. In this work, we present the results obtained from applying Deep Learning (DL) models to the taphonomic analysis of the site. This methodological approach has allowed us to distinguish the bone surface modifications with 97.5% reliability. We show the usefulness of this technique to help solve many taphonomic equifinality problems in the archaeological assemblages, as well as the need to implement new approaches to eliminate subjectivity in the descriptions of bone damage and make more accurate inferences about the past.