LSR: Linguistic Safety Robustness Benchmark for Low-Resource West African Languages

1. 1Identify low-resource languages for evaluation
2. 2Develop a dual-probe evaluation protocol
3. 3Measure cross-lingual refusal degradation in large language models
4. 4Analyze results to improve safety alignment

NLP researchers and AI engineers working on low-resource languages can benefit from LSR to improve safety alignment in their models, while product managers can use LSR to assess the robustness of their language models

💡 LSR highlights the need for cross-lingual evaluation of safety alignment in large language models

🚨 Introducing LSR: a benchmark for measuring cross-lingual refusal degradation in West African languages 🌍