NAP at SemEval-2023 Task 3: Is Less Really More? (Back-)Translation as Data Augmentation Strategies for Detecting Persuasion Techniques

TL;DR

This paper explores the effectiveness of (back-)translation data augmentation strategies using multi-lingual transformers for detecting persuasion techniques in news, highlighting the importance of balancing human and machine-generated data.

Contribution

It demonstrates that (back-)translation enhances persuasion detection performance and provides insights into optimal data balancing for multi-lingual models.

Findings

Both data augmentation strategies improve detection accuracy.

Balancing human and machine-generated data is crucial.

Augmented data benefits are confirmed through automatic and human evaluations.

Abstract

Persuasion techniques detection in news in a multi-lingual setup is non-trivial and comes with challenges, including little training data. Our system successfully leverages (back-)translation as data augmentation strategies with multi-lingual transformer models for the task of detecting persuasion techniques. The automatic and human evaluation of our augmented data allows us to explore whether (back-)translation aid or hinder performance. Our in-depth analyses indicate that both data augmentation strategies boost performance; however, balancing human-produced and machine-generated data seems to be crucial.