POLAR: A Corpus of Questions, Responses and Argumentation in Polish Political Radio Discourse

Abstract

In this paper, we present POLAR: an experimental dataset designed to investigate question–answer structures in political interviews. The study also aims to integrate this level of annotation with the identification of argumentative structures. The dataset comprises orthographic transcriptions of Polish political radio interviews conducted between December 2023 and March 2024, with a total duration of nearly 10 hours of recordings (94,015 tokens). Manual annotation was performed on three levels: (a) identification of questions as speech acts, (b) classification of responses to questions, and (c) argumentative structures in which interrogative sentences function as premises or conclusions. The results show that not all interrogative sentences function as questions in the sense of requesting information — 23% do not serve this function, while 13% were identified as components of argumentative structures. We also introduce a gold-standard corpus, together with baseline experiments and LLM-based evaluations, demonstrating the usefulness of the resource for both theoretical research and NLP applications.

Resources

Details