Multilingual QA

Multilingual question answering studies whether models can retrieve, combine, and explain information when questions and evidence appear in different languages, including in spoken, visual, and specialized settings. Common approaches include multilingual retrieval and retrieval-augmented generation, translation and language routing, task-specific fine-tuning, and decomposition for multi-hop reasoning; benchmarks span factual, medical, legal, financial, geospatial, and speech tasks. Performance often varies across languages and cultural contexts: English evidence can help some systems but may also displace local sources, while low-resource languages often remain challenging. Open issues include building representative evaluations and reliable evidence checks, and balancing consistent answers across languages with respect for legitimate cultural variation.