Structure without semantics: no evidence for bonobo compositionality in Berthet et al. (2025)

Understanding the evolutionary origins of combinatorial communication has motivated studies that analyse the semantic capacities of great ape vocal repertoires. A prominent example is the study by Berthet et al. (2025), which used Multiple Correspondence Analysis (MCA) on a large multidimensional dataset of bonobo calls and concluded that certain call combinations exhibit compositional meaning. We applied the published analytical pipeline to randomized datasets lacking any genuine association between call types and contextual features. Depending on implementation choices, the procedure identified significant relationships in 34.5-84.3% of simulations, suggesting substantial inflation of false-positive results. We therefore re-examined the statistical assumptions underlying the analysis itself. Because the MCA was constructed from repeated measures and uneven sampling across call types, its geometry reflects sampling frequencies and hierarchical dependencies in addition to any semantic structure. As such, any subsequent linear modelling of distances within this fixed space further violates assumptions of independence. We introduced a full permutation test in which a weighted MCA is recomputed under random reassignment of call-type labels, thereby generating a null distribution that respects the dependence structure of the data. Applying this procedure to the original dataset yielded a non-significant Monte Carlo p-value (p = 0.2597), indicating that the observed between-call-type separation does not exceed chance expectations. We argue that interpreting MCA distances as semantic relations requires theoretically motivated feature selection and dependence-aware inference, both of which are lacking in the original analysis. We conclude that Berthet et al. (2025) do not provide any statistically significant evidence for bonobo compositionality and highlight the broader limitations of relying on exploratory post-hoc statistical inference to identify semantic structure in animal communication.