On the accuracy of language trees.

Area of Science:

• Historical linguistics
• Computational linguistics
• Phylogenetics

Background:

• Language evolution is studied by inferring phylogenetic trees from linguistic features.
• Reconstructing these trees is an inverse problem using present data to infer past history.
• Evaluating reconstruction accuracy is crucial, but artificial data models have limitations with real datasets.

Purpose of the Study:

• To survey and assess the accuracy of phylogenetic reconstruction methods in historical linguistics.
• To compare algorithmic inference with expert classifications, specifically Ethnologue.
• To introduce new metrics for evaluating tree reconstruction accuracy.

Main Methods:

• Focus on state-of-the-art distance-based phylogeny reconstruction methods.
• Utilized worldwide linguistic databases for analysis.
• Introduced and characterized two generalizations of tree distance metrics.
• Quantified the impact of database completeness and coverage on reconstruction accuracy.

Main Results:

• Quantified the relative performances of different distance-based algorithms.
• Assessed the accuracy of inferred language trees against Ethnologue classifications.
• Demonstrated how data completeness and coverage influence reconstruction accuracy.

Conclusions:

• Provided an assessment of the current accuracy of reconstruction methods in historical linguistics.
• Identified key factors affecting accuracy, such as database quality.
• Suggested directions for improving language tree reconstruction accuracy.