@article{ART003381645},
author={Lee Jin-Ho},
title={Predicting the Size of Consonant Inventories through Machine Learning},
journal={Journal of Humanities, Seoul National University},
issn={1598-3021},
year={2026},
volume={83},
number={3},
pages={297-325},
doi={10.17326/jhsnu.83.3.202608.297}
TY - JOUR
AU - Lee Jin-Ho
TI - Predicting the Size of Consonant Inventories through Machine Learning
JO - Journal of Humanities, Seoul National University
PY - 2026
VL - 83
IS - 3
PB - Institute of Humanities, Seoul National University
SP - 297
EP - 325
SN - 1598-3021
AB - Thi s study investigates how accurately machine learning can estimate consonant inventory sizes, or the number of consonant phonemes, across the world’s languages. A typologically balanced dataset of 3,035 languages was divided into training (2,124, 70%), validation (455, 15%), and test (456, 15%) sets. Predictive models were built using 47 binary consonant variables alongside genealogical and geographical data. Among five algorithms compared, LightGBM performed best. On the test set, this model achieved a 13.7% error rate and an R 2 of 0.770. Furthermore, 24.8% of the languages had an error rate below 5.0%, and the predictions showed 61.2% agreement with the W ALS-based five-level classification. Compared to baseline models, LightGBM attained high accuracy using only binary segmental features and limited extra-linguistic variables. Analysis revealed that prediction accuracy varied significantly across several factors but could not be explained by isolated consonants or simple co-occurrence patterns. The results suggest that the model’s predictions were more likely driven by complex, multidimensional relationships among variables than by isolated predictors.
KW - Consonant Inventory Size;Machine Learning;LightGBM;Predictor Variable; Feature Importance;Linguistic Typology
DO - 10.17326/jhsnu.83.3.202608.297
ER -
Lee Jin-Ho. (2026). Predicting the Size of Consonant Inventories through Machine Learning. Journal of Humanities, Seoul National University, 83(3), 297-325.
Lee Jin-Ho. 2026, "Predicting the Size of Consonant Inventories through Machine Learning", Journal of Humanities, Seoul National University, vol.83, no.3 pp.297-325. Available from: doi:10.17326/jhsnu.83.3.202608.297
Lee Jin-Ho "Predicting the Size of Consonant Inventories through Machine Learning" Journal of Humanities, Seoul National University 83.3 pp.297-325 (2026) : 297.
Lee Jin-Ho. Predicting the Size of Consonant Inventories through Machine Learning. 2026; 83(3), 297-325. Available from: doi:10.17326/jhsnu.83.3.202608.297
Lee Jin-Ho. "Predicting the Size of Consonant Inventories through Machine Learning" Journal of Humanities, Seoul National University 83, no.3 (2026) : 297-325.doi: 10.17326/jhsnu.83.3.202608.297
Lee Jin-Ho. Predicting the Size of Consonant Inventories through Machine Learning. Journal of Humanities, Seoul National University, 83(3), 297-325. doi: 10.17326/jhsnu.83.3.202608.297
Lee Jin-Ho. Predicting the Size of Consonant Inventories through Machine Learning. Journal of Humanities, Seoul National University. 2026; 83(3) 297-325. doi: 10.17326/jhsnu.83.3.202608.297
Lee Jin-Ho. Predicting the Size of Consonant Inventories through Machine Learning. 2026; 83(3), 297-325. Available from: doi:10.17326/jhsnu.83.3.202608.297
Lee Jin-Ho. "Predicting the Size of Consonant Inventories through Machine Learning" Journal of Humanities, Seoul National University 83, no.3 (2026) : 297-325.doi: 10.17326/jhsnu.83.3.202608.297
