The evolution of cultural diversity in Pama-Nyungan Australia

The study investigates how different cultural domains in Pama-Nyungan Australia evolved and whether their present-day diversity reflects historical population change and dispersal. Building on phylogenetic approaches widely used in linguistics and increasingly in cultural analyses, the authors ask if ceremonial practices (initiation, mortuary) and rock art motifs show evolutionary patterns comparable to Pama-Nyungan language diversification (used as a proxy for population history). They note ethnographic evidence for both diffusion (borrowing between neighboring groups) and stability (especially in ritual practice), motivating a comparative framework to quantify vertical (descent) versus horizontal (diffusion) processes. The purpose is to reconstruct and compare cultural histories across domains, assess their coherence over time, and evaluate associations with language phylogeny to shed light on mechanisms behind Pama-Nyungan expansion and broader cultural evolution in hunter-gatherer contexts.

The paper situates its inquiry within two decades of cultural phylogenetics, where methods from biology (e.g., Bayesian phylogenetics) have been applied to languages, manuscripts, folktales, musical instruments, and material culture. Prior work links language family divergence to migrations (Indo-European, Bantu, Austronesian, Pama-Nyungan). Studies suggest both vertical inheritance and horizontal diffusion shape cultural traditions, with varying coherence across domains. For Australia, ethnographers documented diffusion of rock motifs, myths, and decorative items, but noted relative stability in ritual practices. Competing hypotheses for Pama-Nyungan dispersal include technological innovations (multi-purpose stone tools, subsistence changes) and cultural mechanisms (expanded kinship systems, outward-reaching alliances, collaborative ceremonial activity). The authors adopt a framework (Boyd et al., 2005) classifying cultural domains by phylogenetic stability, to test whether initiation, mortuary practices, and rock motifs map onto different positions along a descent–diffusion continuum.

Data: A new dataset of 90 traits across ~100 Pama-Nyungan societies was compiled from historical ethnography and prior syntheses. Three domains were chosen for coverage, importance, and assignability to linguistic groups: (1) Adolescent initiation (27 traits; presence/absence across 109 linguistic groups; sources primarily 1850–1910 ethnographies: Mathews, Spencer & Gillen, Howitt, Roth; traits selected for frequent reporting, independence, and representativeness; sensitive material excluded or anonymized). (2) Mortuary practices (26 traits; from Meehan’s continent-wide analysis of 98 attributes refined for variability/independence; coded across 136 linguistic groups with broad geographic coverage including Western Australia; awareness of potential missionary influence on later practices). (3) Rock art motifs (37 motifs; presence/absence across >100 locations compiled from Layton 1992; recent age of many images permits assignment to ethnographic linguistic groups with adjustments for ancient sites; motifs span figurative and geometric forms; 97 sites mapped overall, with datasets pruned for specific analyses). Assignment to linguistic groups used ethnographic info cross-checked with Austlang.

Analytical scale: To reduce local diffusion noise and missingness, analyses used 15 monophyletic language areas derived from the Pama-Nyungan phylogeny (Bouckaert et al., 2018), balancing area sizes and counts across domains. For initiation and rock motifs, traits present in ≥50% of societies in an area were coded as present; otherwise absent. For mortuary (sparser data), presence in any society within an area counted as present.

Phylogenetic inference: Bayesian character-based inference with MrBayes (MCMC, burn-in, posterior sampling; trees rooted using Wagaya Warluwaric as outgroup for initiation and mortuary; for rock motifs, a topology constraint grouped all Pama-Nyungan taxa with Arnhem Land and Kimberley as outgroups). Majority-rule consensus trees (nodes ≥50% posterior) assessed hierarchical structure and support.

Network analysis: Neighbor-Net distance networks (Bryant & Moulton) visualized conflicting signals indicative of borrowing/diffusion and calculated δ scores (0 = perfect tree). Neighbor-joining trees aided interpretation of ambiguous splits.

Phylogenetic signal at society level: Using the consensus Pama-Nyungan language tree (Bouckaert et al., 2018) as a proxy for descent, the D statistic for binary traits (Fritz & Purvis, 2010) measured phylogenetic signal. Observed D values were compared to a random distribution null (D = 1) via 1,000 simulations; data adjusted to match the language tree (societies not in the tree removed).

Autologistic modeling (controlling for geography): Following Towner et al. (2012), binary trait presence in each society was modeled as a function of linguistic neighbors (phylogenetic graph by subfamily membership) and spatial neighbors (radius-based neighbor graph), estimating parameters λ (linguistic dependence), θ (spatial dependence), and β (levelling). Graphs were calibrated to yield similar counts of spatial and linguistic neighbors; uniqueness of neighbor ties quantified. Model fitting used Gibbs sampling (12,000 MCMC generations; first 600 burn-in; 6,000 realizations, thinning by 2). Traits with extremely low frequencies (≤5 occurrences) were excluded (7 rock motifs). Predictive accuracy and parameter magnitudes were summarized by domain.

• Bayesian trees: All three domains show some hierarchical structure. Initiation exhibits the strongest phylogenetic structure (5 clades ≥70% posterior), rock motifs moderate (4 clades ≥70%), mortuary weakest (2 clades ≥70%). Initiation supports an eastern clade (91%); mortuary supports a southern grouping (88%); rock motifs support a south & west grouping (92%).
• Tree comparisons: Initiation tree closely aligns with the pruned Pama-Nyungan language tree (not identical; western internal structure differs), while mortuary and rock motifs show tangled mismatches.
• Neighbor-Net δ scores: Rock motifs are most tree-like (δ = 0.23), initiation intermediate (0.30), mortuary most reticulate (0.34), consistent with variable influence of diffusion. Networks indicate major splits: initiation east/west; mortuary north/south; rock motifs north & east versus south & west, with noted outliers (e.g., YKW).
• D-statistic phylogenetic signal (p<0.05 vs D=1 null): Initiation 27/27 traits (100%) significant; mortuary 15/26 (58%); rock motifs 9/37 (24%). For initiation, highest signals are genital mutilation traits (e.g., circumcision, sub-incision). Mortuary significant traits span disposal methods, corpse treatment, and grave features. Rock motif significant traits include both geometric and figurative.
• Autologistic modeling (mean positive effects): Initiation shows strongest linguistic dependence (λ≈+0.20) and minimal geographic effect (θ≈+0.04); mortuary weak for both (λ≈+0.05, θ≈+0.04) with large β; rock motifs show notable spatial influence (θ≈+0.10) and moderate linguistic influence (λ≈+0.11). Average predictive accuracy ~78%.
• Confirmed phylogenetic signals (D significant and λ>θ): 41 traits total—26 initiation (of 27), 9 mortuary (of 15), 6 rock motifs (of 9). Initiation’s strongest λ for female genital mutilation (+0.45), circumcision (+0.36), sub-incision (+0.35); tooth avulsion showed equal linguistic and spatial effects (λ=θ=0.14). Several rock motifs (e.g., dots, tridents, lizardHumanForms, mazeForms) showed positive λ and negative θ, suggesting lineage markers that differentiate neighboring groups; arcs showed strong negative λ (−0.85), potentially a boundary marker.
• Compound mortuary disposal: Mapping and trait-level results do not support a general north-to-south diffusion hypothesis from Southeast Asia; three compound disposal traits show phylogenetic rather than diffusion patterns. Overall: Initiation rituals coevolved closely with language (population history), mortuary practices are relatively labile with mixed influences, and rock motifs have an independent inheritance history with stronger geographic diffusion components.

Findings indicate differing evolutionary dynamics across cultural domains. Initiation rituals display high coherence over time with strong vertical transmission aligned to language phylogeny, consistent with a “core tradition” (Boyd et al. type 2) potentially tied to group identity and membership. This supports proposals that complex initiation systems could have facilitated or accompanied Pama-Nyungan dispersal via broader social alliances, though causality is not established. Mortuary practices show weaker phylogenetic structure (type 3/4), suggesting lability and context-dependent variation (e.g., status of the deceased), yet with some deeper nodes and traits retaining descent signals, including elements of compound disposal. Rock motifs best fit a type 3 process with their own inheritance history, minimally linked to language phylogeny; spatial proximity and exchange networks likely influenced their spread, producing regionally coherent yet partly independent motif traditions. The domain-comparative approach thus clarifies that cultural diversification is heterogeneous: some practices remain conservative markers of lineage, while others recombine or diffuse across boundaries. Quantitatively disentangling descent and diffusion offers a scalable framework for reconstructing deep cultural history and testing hypotheses about cultural mechanisms in large-scale language expansions.

The study assembles and analyzes a novel, continent-scale dataset of Australian cultural traits, demonstrating that computational phylogenetic methods can recover meaningful historical structure in some domains. Initiation rituals show strong alignment with Pama-Nyungan language phylogeny, indicating a coherent, conservative cultural core that may have been significant during the language family’s dispersal. Mortuary practices, though partly structured, appear more labile, and rock motifs exhibit an inheritance trajectory relatively independent of language, with stronger geographic diffusion. Methodologically, combining Bayesian trees, network analyses, D-statistics, and autologistic models provides a robust toolkit for disentangling vertical and horizontal transmission. Future work could test causal links between cultural traits and language diversification dynamics (e.g., state-dependent speciation/extinction frameworks) and apply emerging Bayesian models that explicitly incorporate horizontal transfer. Further fine-grained analyses of the north/south divergence in mortuary practices and the role of exchange networks in motif transmission are also warranted.

• Source bias and temporal scope: Most initiation data derive from 1850–1910 ethnographies; later accounts reflect colonial disruptions (relocations, missionization) that complicate assignments to linguistic groups and may bias mortuary practices (e.g., discouragement of compound and tree disposal).
• Geographic coverage gaps: Initiation data are sparse for Western Australia.
• Data curation choices: Trait selection emphasized frequent reporting and independence; rare or interdependent traits were excluded, potentially omitting informative variability. Seven rock motif traits were removed for low frequency.
• Aggregation to language areas: Majority-rule presence thresholds (or any-presence rules for mortuary) reduce local variation and may mask within-area heterogeneity and recent diffusion.
• Proxies and modeling assumptions: Language phylogeny is used as a proxy for population history; spatial radius-based neighbor graphs approximate diffusion potential. Model results depend on calibration choices (radius, neighbor balancing) and tree topology constraints (e.g., rock motif outgroups).
• Unobserved confounds: Environmental, social, or status-related factors (e.g., for mortuary rites) may drive variation outside descent/diffusion mechanisms.