One of the success stories of formal semantics is the identification of universals in the meaning of lexicalized quantifiers (Barwise & Cooper, 1981). A popular explanatory strategy for linguistic universals sees them as the result of pressures that shape language (Martinet, 1952), especially towards languages that lead to successful communication and languages that can be learned more easily (Kirby et al., 2014). Explanations of this kind have been given for universals of quantification like monotonicity and quantity (Carcassi et al 2021, Steinert-Threlkeld & Szymanik 2020), but conservativity has proved resistant (van de Pol et al 2023). I identify a theoretical gap in evolutionary semantics that prevents an explanation of conservativity, namely a clear way for defining semantic space. I propose a method to fill this gap based on the probabilistic Language of Thought (Piantadosi et al 2016) building on previous work (Mollica & Piantadosi 2021). I present work in progress on modelling the evolution of conservativity, and discuss some promising directions for expanding this method to other semantic phenomena.