Universally, human language changes over time. One fundamental assumption about language change is that it originates at the level of an individual as a shift in frequencies with which a particular variant of linguistic expression is chosen over another. Conceiving of language change as the loss of an equilibrium in a system of stochastically used grammatical options, a change presupposes an emerging disturbing factor or a cause. A change can then be viewed as an adaptation process whereby in a linguistic environment affected by a disturbing factor a hitherto marginal grammatical option begins to create a certain communicative advantage and grows in frequency. We might then in principle expect a fast transition to the new variant in the speech of an individual, given the general learning abilities of humans. However, another fundamental property of language change is that it proceeds gradually over generations. Therefore, there must be powerful conservative factors at play pushing against the communicative advantage presented by the new variant. It has been suggested that the contrast between the speed of learning projected based on an individual’s cognitive abilities and the attested pace of language change may be due to the social aspect of language, whereby the communicative advantage of an innovation is balanced off by the necessity of group synchronization. Since many social coordination tasks can be completed within individual’s lifetime, the question stands open which properties of language are responsible for a relatively slow group coordination when it comes to grammatical shifts. This project aims at understanding, via game-theoretic and reinforcement learning modeling, the interplay between the causes of language change and conservatism. The models will be evaluated against the empirical material from three West Germanic languages, which includes creating a treebank of historical Dutch.