Word Frequency Does Not Predict Grammatical Knowledge in Language Models

TL;DR

This study reveals that in neural language models, noun frequency in the corpus does not predict grammatical understanding, which varies systematically across nouns and can be learned with few-shot training, highlighting a paradox in model behavior.

Contribution

The paper demonstrates that corpus frequency does not correlate with grammatical accuracy and introduces few-shot learning for grammatical properties of novel nouns.

Findings

Nouns are systematically better understood than others across models.

Corpus frequency is unrelated to grammatical performance.

Few-shot learning can impart grammatical properties to new nouns.

Abstract

Neural language models learn, to varying degrees of accuracy, the grammatical properties of natural languages. In this work, we investigate whether there are systematic sources of variation in the language models' accuracy. Focusing on subject-verb agreement and reflexive anaphora, we find that certain nouns are systematically understood better than others, an effect which is robust across grammatical tasks and different language models. Surprisingly, we find that across four orders of magnitude, corpus frequency is unrelated to a noun's performance on grammatical tasks. Finally, we find that a novel noun's grammatical properties can be few-shot learned from various types of training data. The results present a paradox: there should be less variation in grammatical performance than is actually observed.