Virtual knapping (and refitting) with neural networks: proofs of concept

Abstract

Recent advances in neural networks have brought about new opportunities for application in archaeological research. Stone tools, due to their longevity and prevalence across most of prehistory, are a valuable source of evidence for archaeologists. For most of prehistory, stone tools were made by striking a core – often with a stone hammer – to produce flakes with sharp cutting edges. Modern experimental replication of such stone tools, as well as the refitting of prehistoric lithic material are both important methods for understanding in greater detail how prehistoric stone tools were manufactured, and by extension, what insights they can bring to our knowledge of human evolution. However, replication experiments can require considerable time and raw materials. Lithic experiments themselves are also difficult to control and replicate, e.g., it is difficult to control many knapping variables. Refitting can be an even more time-consuming task, as archaeologists must find two matching pieces of stone amongst an entire assemblage of them. Here we discuss the development of a toy model and a recently published proof of concept for a virtual knapping framework capable of accurately predicting the shape of computer-generated flake removals from the surface information of the intact core. In addition, we present an early prototype for a virtual refitter as an extension of our virtual knapping framework. Both models, after additional development and validation, could become important tools for lithic experimentation and analysis and provide more robust results with which to understand prehistoric stone tool production and, thus, human evolution.