Bridging the Domain Gap for Stance Detection for the Zulu language

TL;DR

This paper introduces a domain adaptation method to transfer stance detection capabilities from English to Zulu, addressing language and domain gaps without requiring human expertise, and provides a new Zulu dataset.

Contribution

It presents a novel black-box domain adaptation approach for stance detection in low-resource languages like Zulu, leveraging English datasets and machine translation.

Findings

Achieved comparable stance detection performance in Zulu as in English

Demonstrated effectiveness of domain adaptation without human language expertise

Provided a new Zulu stance detection dataset

Abstract

Misinformation has become a major concern in recent last years given its spread across our information sources. In the past years, many NLP tasks have been introduced in this area, with some systems reaching good results on English language datasets. Existing AI based approaches for fighting misinformation in literature suggest automatic stance detection as an integral first step to success. Our paper aims at utilizing this progress made for English to transfers that knowledge into other languages, which is a non-trivial task due to the domain gap between English and the target languages. We propose a black-box non-intrusive method that utilizes techniques from Domain Adaptation to reduce the domain gap, without requiring any human expertise in the target language, by leveraging low-quality data in both a supervised and unsupervised manner. This allows us to rapidly achieve similar results for stance detection for the Zulu language, the target language in this work, as are found for English. We also provide a stance detection dataset in the Zulu language. Our experimental results show that by leveraging English datasets and machine translation we can increase performances on both English data along with other languages.