Can we read stones? Quantifying the information loss in flintknapping

Abstract

One of the methods of analysing the manufacturing process of lithic tools is by “reading” the scars of removals visible on the surface of cores or bifaces. The paper aims to review the limitations of this approach, by answering the question as to how much information from the original knapping process is still preserved in the cores and bifaces. To quantify the information loss we used experimentally knapped pieces, that we then refitted to check which removals are still visible on the surface of the analysed forms. We tested different measures of information loss in the context of removals and sequences of removals, but also the chronological relationship between sequences i.e. the number of linear extensions and missing nearest connections. On average, 39.9% of the original removals and 54.1% of the sequences of removals are preserved on the surface, but only 14.1% of the original relationships between sequences can be observed. Finally, we used simple machine learning to find an algorithm for estimating the loss of the nearest chronological relationships between sequences when no refits are available. Due to the extremely high information loss for the linear order of the knapping process we: a) propose to abandon these elements of the diacritic approach which aim to reconstruct the rhythms and schemes of knapping; b) evaluate the quality of the reconstructed knapping scheme (partial order) prior to further comparative analyses. Here we propose an algorithm to estimate the number of missing connections to test the quality of the knapping process while using diacritic schemes.