SP-BERT: A Language Model for Political Text in Scandinavian Languages

Abstract

Language models are at the core of modern Natural Language Processing. We present a new BERT-style language model dedicated to political texts in Scandinavian languages. Concretely, we introduce SP-BERT, a model trained with parliamentary speeches in Norwegian, Swedish, Danish, and Icelandic. To show its utility, we evaluate its ability to predict the speakers' party affiliation and explore language shifts of politicians transitioning between Cabinet and Opposition.

Citation

@InProceedings{10.1007/978-3-031-35320-8_34,
author="Doan, Tu My
and Kille, Benjamin
and Gulla, Jon Atle",
title="SP-BERT: A Language Model for Political Text in Scandinavian Languages",
booktitle="Natural Language Processing and Information Systems",
year="2023",
publisher="Springer Nature Switzerland",
address="Cham",
pages="467--477",
abstract="Language models are at the core of modern Natural Language Processing. We present a new BERT-style language model dedicated to political texts in Scandinavian languages. Concretely, we introduce SP-BERT, a model trained with parliamentary speeches in Norwegian, Swedish, Danish, and Icelandic. To show its utility, we evaluate its ability to predict the speakers' party affiliation and explore language shifts of politicians transitioning between Cabinet and Opposition.",
isbn="978-3-031-35320-8"
}