Inherent Biases of Recurrent Neural Networks for Phonological Assimilation and Dissimilation

TL;DR

This paper introduces a recurrent neural network model that captures human-like biases in learning phonological patterns, demonstrating faster learning of simpler and more homogeneous patterns without complex feature representations.

Contribution

The study shows that a simple recurrent neural network inherently exhibits biases similar to human phonological learning, reducing the need for elaborate feature encoding.

Findings

Recurrent neural networks learn single-feature patterns faster than multi-feature ones.

Vowel-only or consonant-only patterns are learned more quickly than mixed patterns.

Recurrent models naturally display biases without additional feature elaborations.

Abstract

A recurrent neural network model of phonological pattern learning is proposed. The model is a relatively simple neural network with one recurrent layer, and displays biases in learning that mimic observed biases in human learning. Single-feature patterns are learned faster than two-feature patterns, and vowel or consonant-only patterns are learned faster than patterns involving vowels and consonants, mimicking the results of laboratory learning experiments. In non-recurrent models, capturing these biases requires the use of alpha features or some other representation of repeated features, but with a recurrent neural network, these elaborations are not necessary.