Researchers have developed CrisiLangs, a new multilingual dataset designed to improve crisis warning generation by language models. This resource extends the existing CrisiText dataset to include warning messages for crisis events in Italian, Bulgarian, German, and Greek, in addition to English. The dataset comprises over 5,000 event-message pairs, translated by native speakers, and has been used to fine-tune a multilingual model. Evaluation showed that the fine-tuned models significantly outperformed a zero-shot baseline across all languages and configurations, with human preference studies indicating that lower decoding temperatures yield the best results. AI
IMPACT Enhances AI's ability to generate timely and accurate crisis warnings across multiple languages, improving public safety communication during emergencies.
RANK_REASON The cluster describes a new academic paper introducing a dataset and evaluation of language models for a specific task. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
