Curating global datasets of structural linguistic features for independence

Area of Science:

• Linguistics
• Computational Linguistics
• Evolutionary Linguistics

Background:

• Cross-linguistic databases are crucial for studying language evolution and human history.
• Existing databases often lack feature independence, invalidating large-scale statistical analyses.
• This limits the reliable use of linguistic data for comparative studies.

Purpose of the Study:

• To curate existing linguistic data into new, statistically sound global datasets.
• To address the issue of feature independence in cross-linguistic data.
• To enable more robust analyses of global linguistic diversity.

Main Methods:

• Curated data from five major linguistic databases (Grambank, WALS, AUTOTYP, PHOIBLE, Lexibank).
• Generated two global cross-linguistic datasets (GBI and TLI) minimizing feature dependencies.
• Densified datasets to reduce missing data and ensure reusability.

Main Results:

• Developed two global cross-linguistic datasets (GBI and TLI) with minimized feature dependencies.
• The curated datasets reveal robust and comparable global patterns of structural linguistic diversity.
• The methodology ensures reusability for future linguistic data curation.

Conclusions:

• The newly curated datasets provide a statistically valid resource for large-scale linguistic analyses.
• These datasets facilitate deeper insights into language evolution and human history.
• The framework supports the creation of independent feature datasets from various sources.