Mutual predictiveness of sound correspondences for reconstruction and language subgrouping

Abstract This paper uses Gyalrongic languages, a conservative branch of Sino-Tibetan, to illustrate a new method to evaluate proto-language reconstructions in general historical linguistics and to conduct exploratory analyses in language phylogeny. It first reconstructs a proto-system of Gyalrongic preinitials and computes and compares the implicative entropies between reconstructed and modern systems. In a second step, Mutual Implicative Entropy (MIE) is used to measure genetic distances between related languages and to generate Neighbornet networks to visualize the subgrouping of Gyalrongic languages. The resulting networks are in agreement with qualitative historical linguistic analyses and allow adjustments to previous subgroupings obtained by Bayesian phylogenetic inference. Thus, this method can be used to detect nuances in lower sub-branches, which are sometimes neglected by lexicon-based methods. Using MIE in historical linguistics is therefore a quick and efficient means of checking the effectiveness of reconstructions and establishing the accurate preliminary shape of language subgrouping.