Technology Category
Natural Language Processing: Safety and RobustnessMachine Learning: Large Multimodal Models (LMMs)Philosophy and Ethics of AI: Safety, Robustness & Trustworthiness
Toxicity-Aware Few-Shot Prompting for Low-Resource Singlish Translation
🤖 AI Summary
Translating toxic content embedded in low-resource Singlish—characterized by slang, code-switching, and culturally grounded harmful expressions—poses significant challenges: existing systems fail to preserve sociolinguistic nuance and toxicity, while suffering from scarce parallel data and insufficient toxicity awareness. Method: We propose a two-stage toxicity-aware translation framework. It integrates human-verified few-shot prompting with model-prompt co-optimization to enable controlled retention of toxic expressions under extreme scarcity of safety-annotated data. Toxicity preservation is rigorously validated via forward/back-translation semantic similarity scoring, expert curation, and cross-LLM comparative verification. Results: Experiments demonstrate substantial improvements in preserving sociolinguistic detail and toxic register, with quantitative metrics and human evaluation confirming effectiveness, robustness, and reproducibility. This work establishes a novel paradigm for socioculturally informed safety governance of multilingual large language models.
Technology Category
Application Category
📝 Abstract
As online communication increasingly incorporates under-represented languages and colloquial dialects, standard translation systems often fail to preserve local slang, code-mixing, and culturally embedded markers of harmful speech. Translating toxic content between low-resource language pairs poses additional challenges due to scarce parallel data and safety filters that sanitize offensive expressions. In this work, we propose a reproducible, two-stage framework for toxicity-preserving translation, demonstrated on a code-mixed Singlish safety corpus. First, we perform human-verified few-shot prompt engineering: we iteratively curate and rank annotator-selected Singlish-target examples to capture nuanced slang, tone, and toxicity. Second, we optimize model-prompt pairs by benchmarking several large language models using semantic similarity via direct and back-translation. Quantitative human evaluation confirms the effectiveness and efficiency of our pipeline. Beyond improving translation quality, our framework contributes to the safety of multicultural LLMs by supporting culturally sensitive moderation and benchmarking in low-resource contexts. By positioning Singlish as a testbed for inclusive NLP, we underscore the importance of preserving sociolinguistic nuance in real-world applications such as content moderation and regional platform governance.
Problem
Research questions and friction points this paper is trying to address.
Translate low-resource Singlish with slang and toxicity
Address scarce data and safety filters in translation
Preserve sociolinguistic nuance in multicultural content moderation
Innovation
Methods, ideas, or system contributions that make the work stand out.
Human-verified few-shot prompt engineering
Optimize model-prompt pairs via semantic similarity
Toxicity-preserving translation for low-resource languages
🔎 Similar Papers
No similar papers found.
