Neuron-Level Interventions for Gendered and Gender-Neutral Generation in Language Models

2026-05-28

AI Summary

This work addresses the persistent issue of gendered language generation in current language models, which often produce binary-gendered outputs even under neutral prompts and largely overlook gender-neutral expressions. The study presents the first systematic identification of neurons strongly associated with feminine, masculine, and gender-neutral linguistic forms. Building on this discovery, the authors propose an intervention mechanism that enables precise and controllable gender expression through targeted neuron activation or suppression, while preserving semantic fidelity. Comprehensive analyses—including neuron probing, controlled text generation, human evaluation, and cross-layer distribution studies—reveal that gender-related neurons are predominantly concentrated in the shallower layers of the model. Evaluated on two open-source language models, the proposed method significantly outperforms existing approaches by reducing unintended gender leakage, enhancing control accuracy, and improving overall output quality.

Abstract

Language models (LMs) can produce gendered language and stereotypes even when given neutral prompts. Most prior work on gender bias in LMs primarily examines gender through a binary lens (feminine vs. masculine), with limited attention to gender-neutral forms, such as they/them pronouns or neutrally phrased job titles. How gender-related signals are encoded in the internal representations of LMs remains an open question. In this work, we study gender-specific neurons in LMs across three categories: feminine, masculine, and gender-neutral. We propose a neuron-level intervention method to identify neurons that are strongly tied to each gender category. We then test these neurons through controlled generation, showing that activating or masking gender-related neurons can steer a sentence toward a target gender form while preserving its original meaning. To evaluate the effectiveness of our gender-intervention approach, we curate two datasets with controlled sentences labeled across all three gender categories and validate the data quality through human evaluation. Experiments on two open-source LMs show that gender-specific neurons are not evenly distributed across model layers; instead, they concentrate heavily in the earliest layers with smaller contributions from later layers. Compared to existing methods, our method achieves more precise gender control, with less leakage into non-target gender categories and stable output quality through two evaluation criteria. Overall, our work examines how gender is encoded in LMs and provides a simple yet effective approach toward controlled gender intervention for both neuron intervention evaluation and gender bias mitigation. Code and datasets are available at: https://github.com/zhiwenyou103/Gender-Neuron-Intervention

Problem

• gender bias
• language models
• gender-neutral
• neuron-level intervention
• internal representations

Innovation

• neuron-level intervention
• gender-neutral generation
• language model bias
• controlled text generation
• gender-specific neurons