Multilingual Detection of Check-Worthy Claims using World Languages and Adapter Fusion

TL;DR

This paper introduces a cost-effective multilingual claim detection method using adapter fusion, which leverages world languages and provides interpretability, outperforming existing approaches in benchmark tests.

Contribution

It proposes a novel approach combining cross-trained adapters and adapter fusion for multilingual claim detection, addressing resource scarcity and interpretability.

Findings

Outperforms top multilingual methods in benchmarks

Cost-efficient due to adapter models and cross-training

Provides interpretability through adapter fusion

Abstract

Check-worthiness detection is the task of identifying claims, worthy to be investigated by fact-checkers. Resource scarcity for non-world languages and model learning costs remain major challenges for the creation of models supporting multilingual check-worthiness detection. This paper proposes cross-training adapters on a subset of world languages, combined by adapter fusion, to detect claims emerging globally in multiple languages. (1) With a vast number of annotators available for world languages and the storage-efficient adapter models, this approach is more cost efficient. Models can be updated more frequently and thus stay up-to-date. (2) Adapter fusion provides insights and allows for interpretation regarding the influence of each adapter model on a particular language. The proposed solution often outperformed the top multilingual approaches in our benchmark tasks.