Global-scale phylogenetic linguistic inference from lexical resources.

Area of Science:

• Computational historical linguistics
• Phylogenetic inference
• Computational linguistics

Background:

• Phylogenetic inference in linguistics traditionally relies on expert cognate judgments.
• This reliance limits studies to a small subset of well-documented language families.
• Automated methods are needed to analyze linguistic diversity more broadly.

Purpose of the Study:

• To develop machine learning techniques for compiling data suitable for phylogenetic inference.
• To expand the scope of phylogenetic inference to a larger number of language families.
• To create new character types for phylogenetic analysis from linguistic data.

Main Methods:

• Utilized the ASJP (Automated Sources of Javanese Phonology) database comprising nearly 7,000 word lists.
• Estimated Pointwise Mutual Information scores between sound classes using weighted sequence alignment.
• Applied Support Vector Machine (SVM) classification for cognate clustering and defined binary characters based on inferred cognates and sound occurrences.

Main Results:

• Generated a dissimilarity matrix from ASJP data suitable for distance-based phylogenetic inference.
• Successfully applied cognate clustering using supervised machine learning.
• Defined novel binary characters for character-based phylogenetic inference, demonstrating their suitability.

Conclusions:

• Machine learning techniques can effectively compile linguistic data for phylogenetic inference.
• The developed methods expand the potential for analyzing global linguistic diversity.
• This approach offers new avenues for computational historical linguistics research.