Wethen comprehensively study detoxifying LMswithparameter sizesranging from126Mupto530B(3 largerthanGPT3), a scale that has never been studied before. We find thati) large LMs have similar toxicity levels as smaller ones given the same pre-training corpus, and ii) large LMs require more endeavor to unlearn the toxic content seen at pretraining. Wealso explore parameter-efficient training methods fordetoxification.

Toxic language in Bengali remains prevalent, especially in online environments, with few effective precautions against it. Although text detoxification has seen progress in high-resource languages, Bengali remains under-explored due to limited resources. In this paper, we propose a novel pipeline for Bengali text detoxification that combines Pareto class-optimized large language models (LLMs) and Chain-of-Thought (Co T) prompting to generate detoxified sentences. To support this effort, we construct BANGLANIRTOX, an artificially generated parallel corpus of 68,041 toxic Bengali sentences with class-wise toxicity labels, reasonings, and detoxified paraphrases, using Pareto-optimized LLMs evaluated on random samples. The resulting BANGLANIRTOX dataset is used to fine-tune language models to produce better detoxified versions of Bengali sentences. Our findings show that Pareto-optimized LLMs with Co T prompting significantly enhance the quality and consistency of Bengali text detoxification. Warning: This paper contains examples of toxic and offensive language.

Large language models (LLMs) are now ubiquitous in user-facing applications, yet they still generate undesirable toxic outputs, including profanity, vulgarity, and derogatory remarks. Although numerous detoxification methods exist, most apply broad, surface-level fixes and can therefore easily be circumvented by jailbreak attacks. In this paper we leverage sparse autoencoders (SAEs) to identify toxicity-related directions in the residual stream of models and perform targeted activation steering using the corresponding decoder vectors. We introduce three tiers of steering aggressiveness and evaluate them on GPT-2 Small and Gemma-2-2B, revealing trade-offs between toxicity reduction and language fluency. At stronger steering strengths, these causal interventions surpass competitive baselines in reducing toxicity by up to 20%, though fluency can degrade noticeably on GPT-2 Small depending on the aggressiveness. Crucially, standard NLP benchmark scores upon steering remain stable, indicating that the model's knowledge and general abilities are preserved. We further show that feature-splitting in wider SAEs hampers safety interventions, underscoring the importance of disentangled feature learning. Our findings highlight both the promise and the current limitations of SAE-based causal interventions for LLM detoxification, further suggesting practical guidelines for safer language-model deployment.

Detoxification in large language models (LLMs) remains a significant research challenge. Existing decoding detoxification methods are all based on external constraints, which require additional resource overhead and lose generation fluency. This work proposes Detoxification with Self-Constrained Decoding (DSCD), a novel method for LLM detoxification without parameter fine-tuning. DSCD strengthens the inner next-token distribution of the safety layer while weakening that of hallucination and toxic layers during output generation. This effectively diminishes toxicity and enhances output safety. DSCD offers lightweight, high compatibility, and plug-and-play capabilities, readily integrating with existing detoxification methods for further performance improvement. Extensive experiments on representative open-source LLMs and public datasets validate DSCD's effectiveness, demonstrating state-of-the-art (SOTA) performance in both detoxification and generation fluency, with superior efficiency compared to existing methods. These results highlight DSCD's potential as a practical and scalable solution for safer LLM deployments.

As social-media platforms emerge and evolve faster than the regulations meant to oversee them, automated detoxification might serve as a timely tool for moderators to enforce safe discourse at scale. We here describe our submission to the PAN 2025 Multilingual Text Detoxification Challenge, which rewrites toxic single-sentence inputs into neutral paraphrases across 15 typologically diverse languages. Building on a 12B-parameter Gemma-3 multilingual transformer, we apply parameter-efficient LoRA SFT fine-tuning and prompting techniques like few-shot and Chain-of-Thought. Our multilingual training corpus combines 3,600 human-authored parallel pairs, 21,600 machine-translated synthetic pairs, and model-generated pairs filtered by Jaccard thresholds. At inference, inputs are enriched with three LaBSE-retrieved neighbors and explicit toxic-span annotations. Evaluated via Style Transfer Accuracy, LaBSE-based semantic preservation, and xCOMET fluency, our system ranks first on high-resource and low-resource languages. Ablations show +0.081 joint score increase from few-shot examples and +0.088 from basic CoT prompting. ANOVA analysis identifies language resource status as the strongest predictor of performance ($η^2$ = 0.667, p < 0.01).

The evolution of digital communication systems and the designs of online platforms have inadvertently facilitated the subconscious propagation of toxic behavior. Giving rise to reactive responses to toxic behavior. Toxicity in online content and Artificial Intelligence Systems has become a serious challenge to individual and collective well-being around the world. It is more detrimental to society than we realize. Toxicity, expressed in language, image, and video, can be interpreted in various ways depending on the context of usage. Therefore, a comprehensive taxonomy is crucial to detect and mitigate toxicity in online content, Artificial Intelligence systems, and/or Large Language Models in a proactive manner. A comprehensive understanding of toxicity is likely to facilitate the design of practical solutions for toxicity detection and mitigation. The classification in published literature has focused on only a limited number of aspects of this very complex issue, with a pattern of reactive strategies in response to toxicity. This survey attempts to generate a comprehensive taxonomy of toxicity from various perspectives. It presents a holistic approach to explain the toxicity by understanding the context and environment that society is facing in the Artificial Intelligence era. This survey summarizes the toxicity-related datasets and research on toxicity detection and mitigation for Large Language Models, social media platforms, and other online platforms, detailing their attributes in textual mode, focused on the English language. Finally, we suggest the research gaps in toxicity mitigation based on datasets, mitigation strategies, Large Language Models, adaptability, explainability, and evaluation.

Modern Large Language Models (LLMs) are excellent at generating synthetic data. However, their performance in sensitive domains such as text detoxification has not received proper attention from the scientific community. This paper explores the possibility of using LLM-generated synthetic toxic data as an alternative to human-generated data for training models for detoxification. Using Llama 3 and Qwen activation-patched models, we generated synthetic toxic counterparts for neutral texts from ParaDetox and SST-2 datasets. Our experiments show that models fine-tuned on synthetic data consistently perform worse than those trained on human data, with a drop in performance of up to 30% in joint metrics. The root cause is identified as a critical lexical diversity gap: LLMs generate toxic content using a small, repetitive vocabulary of insults that fails to capture the nuances and variety of human toxicity. These findings highlight the limitations of current LLMs in this domain and emphasize the continued importance of diverse, human-annotated data for building robust detoxification systems.

Controllable Text Generation (CTG) is a vital subfield in Natural Language Processing (NLP), aiming to generate text that aligns with desired attributes. However, previous studies commonly focus on the quality of controllable text generation for short sequences, while the generation of long-form text remains largely underexplored. In this paper, we observe that the controllability of texts generated by the powerful prefix-based method Air-Decoding tends to decline with increasing sequence length, which we hypothesize primarily arises from the observed decay in attention to the prefixes. Meanwhile, different types of prefixes including soft and hard prefixes are also key factors influencing performance. Building on these insights, we propose a lightweight and effective framework called Dynamic Token-level Prefix Augmentation (DTPA) based on Air-Decoding for controllable text generation. Specifically, it first selects the optimal prefix type for a given task. Then we dynamically amplify the attention to the prefix for the attribute distribution to enhance controllability, with a scaling factor growing exponentially as the sequence length increases. Moreover, based on the task, we optionally apply a similar augmentation to the original prompt for the raw distribution to balance text quality. After attribute distribution reconstruction, the generated text satisfies the attribute constraints well. Experiments on multiple CTG tasks demonstrate that DTPA generally outperforms other methods in attribute control while maintaining competitive fluency, diversity, and topic relevance. Further analysis highlights DTPA's superior effectiveness in long text generation.

In this work, we introduce our solution for the Multilingual Text Detoxification Task in the PAN-2025 competition for the ylmmcl team: a robust multilingual text detoxification pipeline that integrates lexicon-guided tagging, a fine-tuned sequence-to-sequence model (s-nlp/mt0-xl-detox-orpo) and an iterative classifier-based gatekeeping mechanism. Our approach departs from prior unsupervised or monolingual pipelines by leveraging explicit toxic word annotation via the multilingual_toxic_lexicon to guide detoxification with greater precision and cross-lingual generalization. Our final model achieves the highest STA (0.922) from our previous attempts, and an average official J score of 0.612 for toxic inputs in both the development and test sets. It also achieved xCOMET scores of 0.793 (dev) and 0.787 (test). This performance outperforms baseline and backtranslation methods across multiple languages, and shows strong generalization in high-resource settings (English, Russian, French). Despite some trade-offs in SIM, the model demonstrates consistent improvements in detoxification strength. In the competition, our team achieved ninth place with a score of 0.612.