Abstract
Check-worthiness detection is the task of identifying claims, worthy to beinvestigated by fact-checkers. Resource scarcity for non-world languages andmodel learning costs remain major challenges for the creation of modelssupporting multilingual check-worthiness detection. This paper proposescross-training adapters on a subset of world languages, combined by adapterfusion, to detect claims emerging globally in multiple languages. (1) With avast number of annotators available for world languages and thestorage-efficient adapter models, this approach is more cost efficient. Modelscan be updated more frequently and thus stay up-to-date. (2) Adapter fusionprovides insights and allows for interpretation regarding the influence of eachadapter model on a particular language. The proposed solution oftenoutperformed the top multilingual approaches in our benchmark tasks.