When Irregularity Helps: A Subclass Analysis of Inductive Bias in Neural Morphology

TL;DR

This paper reveals that a small, specific irregular subclass in Japanese verb inflection causes disproportionate errors in neural models, highlighting the importance of subclass analysis for better morphological evaluation.

Contribution

It introduces a subclass analysis approach to identify how rare irregularities impact neural morphological generation and suggests more granular evaluation methods.

Findings

A tiny irregular subclass (<1%) accounts for most errors.

Removing this subclass improves model generalization more than removing all irregulars.

Error concentration is linked to low-frequency patterns and specific morphophonological processes.

Abstract

Neural morphological generation systems often achieve high aggregate accuracy on benchmark datasets, yet such performance can conceal systematic errors concentrated in rare morphological subclasses. We examine Japanese past-tense verb inflection and show that a very small, structurally specific irregular subtype (<1% of data) accounts for a disproportionate share of model errors. Controlled ablation experiments demonstrate that removing this subtype yields larger improvements in generalization than removing all irregular verbs, indicating that not all irregularity contributes equally to model instability. These findings suggest that error concentration is driven by the interaction between extreme low-frequency morphological patterns and specific morphophonological processes, particularly gemination. We argue that morphological evaluation should incorporate finer-grained subclass analysis beyond standard conjugation categories.