# Leashing the Inner Demons: Self-Detoxification for Language Models

arXiv: https://arxiv.org/abs/2203.03072 · HTML: https://mlsanity.com/paper/2203.03072/

## Key facts

- Cited by 19 papers in the map (data through 14 Oct 2026).
- Builds on 24 of its 30 references that are in the map.
- First posted on arXiv 6 Mar 2022.

## Builds on (most cited first, 24 of 24)

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://mlsanity.com/paper/1910.10683/) (2019)
- [Sequence to Sequence Learning with Neural Networks](https://mlsanity.com/paper/1409.3215/) (2014)
- [The Curious Case of Neural Text Degeneration](https://mlsanity.com/paper/1904.09751/) (2019)
- [Don't Stop Pretraining: Adapt Language Models to Domains and Tasks](https://mlsanity.com/paper/2004.10964/) (2020)
- [RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models](https://mlsanity.com/paper/2009.11462/) (2020)
- [Hierarchical Neural Story Generation](https://mlsanity.com/paper/1805.04833/) (2018)
- [A Diversity-Promoting Objective Function for Neural Conversation Models](https://mlsanity.com/paper/1510.03055/) (2015)
- [The Secret Sharer: Evaluating and Testing Unintended Memorization in Neural Networks](https://mlsanity.com/paper/1802.08232/) (2018)
- [Plug and Play Language Models: A Simple Approach to Controlled Text Generation](https://mlsanity.com/paper/1912.02164/) (2019)
- [CTRL: A Conditional Transformer Language Model for Controllable Generation](https://mlsanity.com/paper/1909.05858/) (2019)
- [Universal Adversarial Triggers for Attacking and Analyzing NLP](https://mlsanity.com/paper/1908.07125/) (2019)
- [Is BERT Really Robust? A Strong Baseline for Natural Language Attack on Text Classification and Entailment](https://mlsanity.com/paper/1907.11932/) (2019)
- [The Woman Worked as a Babysitter: On Biases in Language Generation](https://mlsanity.com/paper/1909.01326/) (2019)
- [Be Your Own Teacher: Improve the Performance of Convolutional Neural Networks via Self Distillation](https://mlsanity.com/paper/1905.08094/) (2019)
- [BERT-ATTACK: Adversarial Attack Against BERT Using BERT](https://mlsanity.com/paper/2004.09984/) (2020)
- [DExperts: Decoding-Time Controlled Text Generation with Experts and Anti-Experts](https://mlsanity.com/paper/2105.03023/) (2021)
- [GeDi: Generative Discriminator Guided Sequence Generation](https://mlsanity.com/paper/2009.06367/) (2020)
- [A Neural Conversational Model](https://mlsanity.com/paper/1506.05869/) (2015)
- [Detoxifying Language Models Risks Marginalizing Minority Voices](https://mlsanity.com/paper/2104.06390/) (2021)
- [Learning a Deep Listwise Context Model for Ranking Refinement](https://mlsanity.com/paper/1804.05936/) (2018)
- [Feature-Wise Bias Amplification](https://mlsanity.com/paper/1812.08999/) (2018)
- [Pre-train and Plug-in: Flexible Conditional Text Generation with Variational Auto-Encoders](https://mlsanity.com/paper/1911.03882/) (2019)
- [Bias Amplification in Artificial Intelligence Systems](https://mlsanity.com/paper/1809.07842/) (2018)

## Built on by (most cited first, 19 of 19)

- [Toxicity in ChatGPT: Analyzing Persona-assigned Language Models](https://mlsanity.com/paper/2304.05335/) (2023)
- [Large Language Model Alignment: A Survey](https://mlsanity.com/paper/2309.15025/) (2023)
- [Attack Prompt Generation for Red Teaming and Defending Large Language Models](https://mlsanity.com/paper/2310.12505/) (2023)
- [Self-Detoxifying Language Models via Toxification Reversal](https://mlsanity.com/paper/2310.09573/) (2023)
- [Why So Toxic? Measuring and Triggering Toxic Behavior in Open-Domain Chatbots](https://mlsanity.com/paper/2209.03463/) (2022)
- [MedEval: A Multi-Level, Multi-Task, and Multi-Domain Medical Benchmark for Language Model Evaluation](https://mlsanity.com/paper/2310.14088/) (2023)
- [Controlling Bias Exposure for Fair Interpretable Predictions](https://mlsanity.com/paper/2210.07455/) (2022)
- [LLM-Personalize: Aligning LLM Planners with Human Preferences via Reinforced Self-Training for Housekeeping Robots](https://mlsanity.com/paper/2404.14285/) (2024)
- [Walking in Others' Shoes: How Perspective-Taking Guides Large Language Models in Reducing Toxicity and Bias](https://mlsanity.com/paper/2407.15366/) (2024)
- [Synthetic Pre-Training Tasks for Neural Machine Translation](https://mlsanity.com/paper/2212.09864/) (2022)
- [UniDetox: Universal Detoxification of Large Language Models via Dataset Distillation](https://mlsanity.com/paper/2504.20500/) (2025)
- [IF-GUIDE: Influence Function-Guided Detoxification of LLMs](https://mlsanity.com/paper/2506.01790/) (2025)
- [CMD: a framework for Context-aware Model self-Detoxification](https://mlsanity.com/paper/2308.08295/) (2023)
- [Tiny Refinements Elicit Resilience: Toward Efficient Prefix-Model Against LLM Red-Teaming](https://mlsanity.com/paper/2405.12604/) (2024)
- [FrenchToxicityPrompts: a Large Benchmark for Evaluating and Mitigating Toxicity in French Texts](https://mlsanity.com/paper/2406.17566/) (2024)
- [Interactive Text Generation](https://mlsanity.com/paper/2303.00908/) (2023)
- [IRepair: An Intent-Aware Approach to Repair Data-Driven Errors in Large Language Models](https://mlsanity.com/paper/2502.07072/) (2025)
- [Combating Toxic Language: A Review of LLM-Based Strategies for Software Engineering](https://mlsanity.com/paper/2504.15439/) (2025)
- [Detoxification for LLM: From Dataset Itself](https://mlsanity.com/paper/2604.19124/) (2026)

## Abstract (arXiv)

Language models (LMs) can reproduce (or amplify) toxic language seen during training, which poses a risk to their practical application. In this paper, we conduct extensive experiments to study this phenomenon. We analyze the impact of prompts, decoding strategies and training corpora on the output toxicity. Based on our findings, we propose a simple yet effective method for language models to "detoxify" themselves without an additional large corpus or external discriminator. Compared to a supervised baseline, our proposed method shows better toxicity reduction with good generation quality in the generated content under multiple settings. Warning: some examples shown in the paper may contain uncensored offensive content.