Exploring the impact of inference method, character coding scheme, and cognacy criterion on phylolinguistic inference: The case of the Japonic language family

<p>The application of phylogenetic methods to linguistic data requires several critical methodological choices, yet their relative influence on the resulting trees remains poorly understood. This study conducts a systematic sensitivity analysis to evaluate the impact of three key factors: (1) the phylogenetic inference method (neighbor-joining, maximum parsimony, and Bayesian inference), (2) the character coding scheme (binary vs. multistate), and (3) the cognacy (i.e. homology) criterion (root-based vs. stem-based cognacy). Using the Japonic language family as a model, we generated a diverse set of phylogenetic trees and compared them using multiple tree metrics and distance measures between trees. Variance partitioning through distance-based redundancy analysis reveals a clear hierarchy of influence: the choice of inference method is the primary driver of variation between trees, followed by the character coding scheme, while the cognacy criterion exerts a comparatively minor effect. Specifically, we show that the choice of inference method fundamentally shapes tree topology, whereas the character coding scheme impacts branch lengths and evolutionary tempo. While maintaining a consistent and transparent definition of cognacy remains good practice, our results indicate that the choice of one definition over another does not always lead to a significant loss of phylogenetic signal. Therefore, phylolinguistic research should prioritize first and foremost the rigorous selection of inference method.</p>