Revisiting the Exit from Nuclear Energy in Germany with NLP

TL;DR

This paper investigates the potential of current NLP techniques, including unsupervised, zero-, and few-shot learning, to automate the annotation of political discourse, reducing reliance on resource-intensive manual annotation.

Contribution

It demonstrates how modern NLP methods can replicate manually annotated datasets with minimal or no additional manual labeling, advancing automated political discourse analysis.

Findings

Transformer models outperform human annotators in some tasks

Unsupervised and few-shot methods can replicate annotations with limited data

Potential to significantly reduce manual annotation efforts in political NLP

Abstract

Annotation of political discourse is resource-intensive, but recent developments in NLP promise to automate complex annotation tasks. Fine-tuned transformer-based models outperform human annotators in some annotation tasks, but they require large manually annotated training datasets. In our contribution, we explore to which degree a manually annotated dataset can be automatically replicated with today's NLP methods, using unsupervised machine learning and zero- and few-shot learning.