Testing the validity and adequacy of linguistic phylogenetic analyses

Area of Science:

• Computational linguistics
• Evolutionary biology
• Historical linguistics

Background:

• Bayesian phylogenetics is widely used in historical linguistics, often employing models from evolutionary biology.
• Standard phylolinguistic analyses use binary data with ascertainment bias and specific substitution models, but lack thorough validation.

Purpose of the Study:

• To validate standard phylolinguistic analysis setups using simulation-based calibration.
• To assess the impact of ascertainment bias and data partition characteristics on phylogenetic accuracy.
• To evaluate the suitability of the covarion substitution model for lexical evolution.

Main Methods:

• Simulation-based calibration studies using BEAST2 software.
• Testing standard phylolinguistic models with binary data, ascertainment bias, and correlated partitions.
• Posterior predictive simulations to assess model adequacy.

Main Results:

• Standard analysis with default reweighting of partition rates by cognate count leads to poorly calibrated posteriors.
• An alternative approach assuming equal cognate set sizes per meaning performs better in simulations and fits empirical data.
• The covarion substitution model inadequately approximates lexical evolution due to semantic shift and non-independent cognate substitutions.

Conclusions:

• Thorough testing of phylogenetic models and their implementations in phylolinguistics is crucial.
• Adjusting for ascertainment bias and cognate set size is necessary for accurate phylogenetic inference.
• Improved models are needed to better represent the complexities of lexical evolution.