A test of Generalized Bayesian dating: A new linguistic dating method

Area of Science:

• Computational linguistics
• Phylogenetics
• Computational phylogenetics

Background:

• Traditional Bayesian phylogenetic methods for language dating require subjective calibration points and family-specific priors.
• The absence of internal calibration points for many language families poses a significant challenge for accurate phylogenetic dating.
• Manual cognate identification in phylogenetic tree inference can introduce subjectivity.

Observation:

• A novel, generalized Bayesian framework for language phylogeny dating is introduced, requiring no family-specific priors or calibration points.
• An automated approach to cognate identification is implemented, reducing subjectivity in tree inference.
• A Gamma regression model is utilized, fitting tree lengths with known time depths from 30 independent calibration points.

Findings:

• The new framework predicts time depths for the root and internal nodes across 116 language families, yielding 1,287 dates.
• Results from the new method are comparable to existing Bayesian studies on individual language families.
• Performance evaluation shows the Automated Similarity Judgment Program (ASJP) outperforms the new method, which in turn is superior to automated glottochronology.

Implications:

• This generalized Bayesian framework offers a more objective and broadly applicable method for dating language family evolution.
• Automated cognate identification enhances the reproducibility and scalability of phylogenetic analyses.
• Comparative analysis highlights the strengths and weaknesses of different computational methods for historical linguistics, guiding future research.