Testing methods of linguistic homeland detection using synthetic data.

Area of Science:

• Computational linguistics
• Historical linguistics
• Phylogeography

Background:

• Quantitative methods like Bayesian phylogeography and the diversity method are used to infer language family homelands.
• Bayesian methods model population movements on phylogenetic trees.
• The diversity method posits that areas of high linguistic diversity indicate homelands.

Purpose of the Study:

• To systematically test and compare the performance of different quantitative methods for inferring language family homelands.
• To establish a performance hierarchy for these methods in a linguistic context.

Main Methods:

• Simulated language families with branching structures and word lists.
• Simulated speaker populations moving in space.
• Tested six methods: BayesTraits (two versions), BEAST 2 (relaxed random walk), RevBayes (fixed and variable rates random walk), and the diversity method.

Main Results:

• A performance hierarchy for the tested methods was established.
• Geographical idiosyncrasies, incomplete sampling, tree imbalance, and small family sizes negatively impacted performance.
• The general performance hierarchy remained largely consistent despite these factors.

Conclusions:

• The study provides the first systematic performance comparison of homeland inference methods.
• The established hierarchy guides the selection of appropriate methods for reconstructing prehistoric languages.
• Understanding the impact of various factors is essential for accurate homeland inference.