Learning with Properties: Restrictiveness and Typological Structure

Authors

  • Natalie DelBusso Rutgers University

DOI:

https://doi.org/10.3765/amp.v8i0.4641

Keywords:

Learnability, Property Theory, Output-Driven Learner

Abstract

A learner's task is to find the most restrictive grammar consistent with the data of their language. This paper develops an OT learning algorithm that incorporates typological-level information from Property Analysis to increase restrictiveness and successfully learn subset languages. Based on Tesar's (2014) Output-Driven Learner (ODL), Property-ODL (PODL) uses ERCs taken from property values encoding specific markedness > faithfulness rankings. PODL was tested in a learning simulation for the phonological system in Tesar (2014), Paka, which presents the challenging case of languages in paradigmatic subset relations. In ODL, these require additional methods to be learned. PODL eliminates the need for these in learning the paradigmatic subsets and overall reduces the use of less-tested methods in learning the grammars of the typology.

 

Downloads

Published

2020-05-02

Issue

Section

Supplemental Proceedings