Enable Fact Verification across Languages via Parallel Training with Regularizations

Title

Enable Fact Verification across Languages via Parallel Training with Regularizations

Abstract

Fact verification is part of the fact checking task. In this study, we want to enable the fact verification task across language. We will introduce a new cross-lingual fact verification dataset XFEVER, which is constructed by extending the examples of the processed FEVER dataset to 6 languages, including isolated language such as Japanese. Moreover, we apply two translation method, auto-translation and human-translation, to observe that our models’ performance on different resources. For evaluation, we will provide baselines in 2 scenarios: zero-shot transfer learning and translate-train learning. For the second scenario, we will introduce different consistency regularizations.

Location: NII 1810

Time: 2023-03-01 11:00 ~ 12:00

Other information:

Please email us to get the meeting link if you are intersted!