Understanding or Memorizing? A Case Study of German Definite Articles in Language Models

TL;DR

This study investigates whether language models understand German definite articles through rules or memorization, finding evidence that they rely partly on memorized associations rather than purely rule-based encoding.

Contribution

The paper introduces a gradient-based interpretability method to analyze parameter updates, revealing that models rely on memorization rather than strict rule-based encoding for German articles.

Findings

Parameter updates for specific gender-case transitions affect unrelated settings.

Neurons involved in article transitions overlap across different gender-case settings.

Results suggest reliance on memorization over rule-based generalization.

Abstract

Language models perform well on grammatical agreement, but it is unclear whether this reflects rule-based generalization or memorization. We study this question for German definite singular articles, whose forms depend on gender and case. Using GRADIEND, a gradient-based interpretability method, we learn parameter update directions for gender-case specific article transitions. We find that updates learned for a specific gender-case article transition frequently affect unrelated gender-case settings, with substantial overlap among the most affected neurons across settings. These results argue against a strictly rule-based encoding of German definite articles, indicating that models at least partly rely on memorized associations rather than abstract grammatical rules.