Keynotes

Abstract: Understanding the underlying mechanisms that drive change within a system is a fundamental challenge. Often, these mechanisms cannot be directly observed and must instead be inferred from aggregated data. This challenge is not unique to cultural evolution, and significant progress has been made in addressing similar inverse problems in other disciplines. One promising approach to tackle this challenge is generative inference, which employs a generative model—a mathematical representation of the system—to establish causal links between the evolutionary mechanisms in question and observable data, which are then evaluated for statistical consistency. Naturally, the accuracy of this method is influenced by two primary factors: the appropriateness of the generative model in reflecting the key cultural and demographic properties of the system, and the quality of the available data—such as its level of aggregation (e.g., population-level or individual-level data), its sparsity, and its spatial and/or temporal resolution.

In this talk, I will outline how the framework of generative inference can be applied to one of the main questions in cultural evolution: understanding why and how various forms of social learning are used in human populations, both in the present and past. In particular, we will address the issues associated with choosing the appropriate model and imperfect data, highlighting the potential inferential consequences of these issues. First, we revisit a long-standing question: under what circumstances might we expect the ability to learn socially to be favored by selection? Our findings suggest that incorporating human cognitive capacities, such as memory and forgetting processes, can significantly alter predictions regarding the usefulness of social learning. This underscores the importance of considering specific model assumptions when developing frameworks for social learning. Second, using baby name statistics—one of the best-documented cultural datasets—we demonstrate that having "a lot of data" does not necessarily lead to accurate inference results. Our findings indicate that the presence or absence of rare variants, along with their spread behaviour, may provide a stronger signature of underlying processes than the dynamics of high-frequency variants. Given that this data is often unavailable due to privacy considerations, we illustrate how inference results can vary depending on whether the data is complete or incomplete.

Abstract: Protolanguages are hypothesized intermediate stages in language evolution. Although the term "protolanguage" was introduced by Gordon Hewes in 1973, it was popularized for a particular model of language evolution by Derek Bickerton. For many scientists, Bickerton's model of a protolanguage with words, but without syntax, became the canonical usage of this term, but other very different models of protolanguage have been explored both before and after Bickerton's. I argue that any non-instantaneous model of language evolution needs to have some notion of an intermediate stage or stages, and that the general term "protolanguage" can be used for any of these possible hypothesized intermediates. In previous work I proposed three different plausible hypotheses: musical, lexical, or gestural protolanguage. I suggest that empirical work in language evolution should, whenever possible, test specific predictions of differing models of protolanguage. In this talk I will review models of protolanguage, methods of testing among them, and conclude by suggesting that, given current data, models of language evolution that involve multiple protolanguages are the most plausible option.