Towards identifying the optimal datasize for lexically-based Bayesian inference of linguistic phylogenies