Imbalanced Multilingual Data Distribution
多語言資料分佈失衡
Design & Development
Risk Description
When an organization serves multiple languages with a single model; due to unmitigated control gaps, performance in the primary language is strong, but responses in local languages frequently show semantic drift and culturally inappropriate phrasing, triggering external stakeholder impacts and causing service quality for local users is noticeably poorer, and complaints concentrate among speakers of minority languages.
Framework Mappings
EU AI ActArt.10
NIST AI 600-1Harmful Bias and Homogenization
NIST AI RMFMEASURE 2.11
ISO/IEC TR 24027
ISO/IEC 5338設計與開發
MIT AI Risk RepositoryDomain 1
Risk Treatment & Implementation Guidance
Balance multilingual data so local languages meet minimum quality and volume thresholds; Benchmark quality across languages against uniform standards; Invest in data augmentation and local review for lagging languages