Robust Pronoun Fidelity with English LLMs: Are they Reasoning, Repeating, or Just Biased?

TL;DR

This paper introduces RUFF, a large dataset to evaluate the robustness of pronoun use in English language models, revealing significant weaknesses in pronoun fidelity especially with certain pronouns and distractors.

Contribution

The paper presents RUFF, a new dataset of over 5 million instances for measuring pronoun fidelity in language models, and provides a comprehensive evaluation across multiple model architectures and sizes.

Findings

Models struggle with pronouns like she/her, singular they, and neopronouns.

Pronoun fidelity drops significantly with distractor sentences.

Humans achieve nearly 100% accuracy in pronoun reuse.

Abstract

Robust, faithful and harm-free pronoun use for individuals is an important goal for language model development as their use increases, but prior work tends to study only one or two of these characteristics at a time. To measure progress towards the combined goal, we introduce the task of pronoun fidelity: given a context introducing a co-referring entity and pronoun, the task is to reuse the correct pronoun later. We present RUFF, a carefully-designed dataset of over 5 million instances to measure robust pronoun fidelity in English, and we evaluate 37 model variants from nine popular families, across architectures (encoder-only, decoder-only and encoder-decoder) and scales (11M-70B parameters). When an individual is introduced with a pronoun, models can mostly faithfully reuse this pronoun in the next sentence, but they are significantly worse with she/her/her, singular they and neopronouns. Moreover, models are easily distracted by non-adversarial sentences discussing other people; even one sentence with a distractor pronoun causes accuracy to drop on average by 34 percentage points. Our results show that pronoun fidelity is not robust, in a simple, naturalistic setting where humans achieve nearly 100% accuracy. We encourage researchers to bridge the gaps we find and to carefully evaluate reasoning in settings where superficial repetition might inflate perceptions of model performance.