Assessing the potential of latent class modelling for classifying stone artefacts and the quantification of technological diversity

Abstract

Archaeological typologies are used to determine the number and diversity of artefact forms within a prehistoric toolkit. However, objective classification is challenging, complicating cross-assemblage comparisons. We explore the potential of latent class modelling (LCM) for grouping stone tools based on their morphological and technological attributes. LCM identifies unobserved (‘latent’) subgroups that share certain observed characteristics, producing posterior probabilities of artefact membership to latent classes. Applied to a large dataset of Middle Stone Age and Middle Palaeolithic lithics from northern Africa and Arabia, we compare LCM results with the original typological assessment of each artefact as well as hierarchical clustering, another non-model based unsupervised technique of group classification. Our results show that, although both methods are equally (in)coherent with the original typology, LCM can group artefacts with important technological and morphological characteristics, such as diverse bifacially worked pieces and different types of unretouched Levallois products. We further evaluate LCM performance using permutation tests, which highlight that our model fits the observed data substantially better than any randomly generated structure. Using latent class proportions, we then quantify technological diversity robustly across varying sample sizes. Assemblage-level diversity patterns indicate that northern African MSA toolkits are generally variable, with only a limited number of assemblages departing significantly from null expectations. Overall, LCM offers a transparent, probabilistic framework for capturing the polythetic nature of stone tool assemblages and provides an objective basis for refining lithic typologies grounded in measurable morphological and technological criteria.