Skip to main content
Paper

BMRS-Net: Learning BMRS Predicates as Decision Trees

Author
  • Yifan Hu

Abstract

This paper investigates the use of Boolean Monadic Recursive Scheme (BMRS) feature predicates, drawing an analogy to binary Decision Trees for large-scale phonological analysis. The study introduces an automated approach for generating BMRS predicates through the Classification and Regression Trees (CART) algorithm, resulting in the creation of BMRS-Trees. These trees form the basis for the BMRS-Net, a network-like structure that outputs a phonological feature matrix. By connecting multiple BMRS-Trees in parallel, this system automates the learning of phonological transformations, offering a promising alternative to traditional rule-based phonology models. Through two case studies, the paper demonstrates the efficiency of this approach in fitting datasets, maintaining high transparency and interpretability, and adapting dynamically to new linguistic contexts. The successful application of BMRS-Trees and BMRS-Net to datasets involving segmental alternation, deletion, and long-distance shifts highlights their potential as a tool for automated phonological research. The results suggest that this decision-based approach, grounded in machine learning techniques, could significantly advance computational phonology, providing an automated framework that minimizes human input while offering meaningful insights into phonological processes.

Keywords: Boolean Monadic Recursive Scheme (BMRS), Decision Tree, computational phonology, model interpretability

How to Cite:

Hu, Y., (2025) “BMRS-Net: Learning BMRS Predicates as Decision Trees”, Society for Computation in Linguistics 8(1): 20. doi: https://doi.org/10.7275/scil.3157

Downloads:
Download PDF

43 Views

12 Downloads

Published on
2025-06-13

Peer Reviewed