A quantitative analysis of Final Palaeolithic/earliest Mesolithic cultural taxonomy and evolution in Europe

Abstract

Archaeological systematics, together with spatial and chronological information, are commonly used to infer cultural evolutionary dynamics in the past. For the study of the Palaeolithic, and particularly the European Final Palaeolithic and earliest Mesolithic, proposed changes in material culture are often interpreted as reflecting historical processes, migration, or cultural adaptation to climate change and resource availability. Yet, cultural taxonomic practice is known to be variable across research history and academic traditions, and few large-scale replicable analyses across such traditions have been undertaken. Drawing on recent developments in computational archaeology, we here present a data-driven assessment of the existing Final Palaeolithic/earliest Mesolithic cultural taxonomy in Europe. Our dataset consists of a large expert-sourced compendium of key sites, lithic toolkit composition, blade and bladelet production technology, as well as lithic armatures. The dataset comprises 16 regions and 86 individually named archaeological taxa ('cultures'), covering the period between ca. 15,000 and 11,000 years ago (cal BP). Using these data, we use geometric morphometric and multivariate statistical techniques to explore to what extent the dynamics observed in different lithic data domains (toolkits, technologies, armature shapes) correspond to each other and to the culture-historical relations of taxonomic units implied by traditional naming practice. Our analyses support the widespread conception that some dimensions of material culture became more diverse towards the end of the Pleistocene and the very beginning of the Holocene. At the same time, cultural taxonomic unit coherence and efficacy appear variable, leading us to explore potential biases introduced by regional research traditions, inter-analyst variation, and the role of disjunct macroevolutionary processes. In discussing the implications of these findings for narratives of cultural change and diversification across the Pleistocene-Holocene transition, we emphasize the increasing need for cooperative research and systematic archaeological analyses that reach across research traditions.

Fig 1. Pyramid of evidence in archaeological research and the place of higher-order analyses and collaborative practice.

Modified from Hussain et al. [113].

Fig 2. Flowchart of data-sourcing, analytical and interpretative process.

Fig 3. Regional units as used in this study (boxed large numbers) and the distribution of associated key sites (small numbers).

Large numbers represent regional centroids (mean site coordinates). 1: Southern Scandinavia, 2: Lithuania, 3: Northern Germany, 4: Britain, 5: Poland, 6: Belgium and Southern Netherlands, 7: Bohemia and Moravia, 8: Southern Germany, 9: Northern France, 10: Austria, Slovakia, Hungary, 11: Switzerland, 12: Northern Italy, 13: Western France, 14: Cantabrian Spain, 15: Mediterranean Iberia, 16: Atlantic Iberia. A detailed breakdown of key sites can be found in SI Table 2 in S1 Data and associated data.

Fig 4. Within-group mean distances (coloured) of higher-order macro-units and the mean distances of the associated null-distributions (grey bars) for each data domain.

The SES-value informs about the validity of the higher-order groupings. SES-values between -2 and 2 (-2>SES<2) suggest that the mean distance within a given higher-order grouping is not significantly different from the mean distance of groupings consisting of randomly drawn NACs. SES-values of -2 or lower (SES≤-2) instead suggest that the observations within a higher-order grouping are significantly more similar to one another than if they had been grouped together by chance. The opposite is the case for SES-values of two or more (SES≥2), pointing to substantial internal variability. In general, the further the coloured line is away from the randomized distance measures obtained (grey bars) the better does the proposed classification perform. The difference in SES values between the Outlines domain on the one hand and the Technology and Tools domains on the other is driven by sample size.

Fig 5

A. Bootstrapped dendrogram of Tools using Ward’s [147] method based on the Gower distance, conducted at the level of named archaeological cultures (NACs). Clades below a threshold of 50% were collapsed. The coloured grids show the tips’ associations with higher-order macro-units. B. Bootstrapped dendrogram of Technology using Ward’s [147] method based on the Gower distance, conducted at the level of named archaeological cultures (NACs). Clades below a threshold of 50% were collapsed. The coloured grid shows the tips’ associations with higher-order macro-units. C. Bootstrapped dendrogram of Outlines using Ward’s [147] method based on the Euclidean Distance of the obtained PCA data obtained, conducted at the level of individual sites. Clades below a threshold of 50% were collapsed. The coloured grid shows the tips’ associations with higher-order macro-units. A stratified subset has been drawn to include two random outlines for each individual NAC.

Fig 6. Tanglegram of Tools (left) and Technology (right) non-bootstrapped dendrograms, constructed using the cophyloplot function in ape [149].

The topography of the two dendrograms differs and only few cultural taxonomic units occur in the same clades. Those NACs which occur in the same clade are coloured.

Fig 7. Machine-learning classification and regression trees (CART) for Tools (A) and Technology (B) vis-à-vis higher-order taxonomic groupings.

The two decision trees show the most important predictor variables and their character states in our dataset for the higher-order archaeological groupings in each lithic data-domain. Each step in the tree splits the remaining units from the previous subset according to the identified key variable. Colour-coded treemaps show the precision and importance of the any given variable’s character state for classifying the seven higher-order taxonomic groupings in the dataset. CART was performed with ‘Partial data’ and ‘Minimum error’ pruning in R and then redrawn from DisplayR.

Fig 8

A. Mantel tests for the three analytical domains testing the null hypothesis that between-data similarity (NAC-level for Tools and Technology and site-level for Outlines) decreases with geographical distance. B. Mantel tests for the three analytical domains testing the null hypothesis that between-data similarity (NAC-level for Tools and Technology and site-level for Outlines) decreases with time-slice distance.

Fig 9. Disparity of lithic armature outlines across time (time-slices I to IV).