Narratives in the EU Parliament

This dataset contains speeches given in the European Parliament between 2006 and 2022. The speeches are annotated for their use of EU narratives of sovereignty, nationalism (nation states should have power) and supranationalism (EU institutions should have the power). Annotations are yes-or-no answers to the question "does the given speech express the given narrative". 

The file contains on each line a paragraph of a speech in the json format. Each object has the following properties:
- id: str -- A uuid4 ID of the paragraph
- text: str -- The text of the paragraph
- narrative: str -- The narrative that was annotated, supranational or national
- score: int -- The combined annotation, 0 for negative and 1 for positive
- annotation: list[int] -- The original decisions of the annotators
- sample: str -- The sampling strategy by which the speech was selected to be annotated. random or query
- pro: dict[str:str] -- An LLM generated argument for why the speech expresses the narrative. Three models were used for generation: {gpt, qwen, llama}. Arguments were only generated for unanimous anntations, otherwise these dicts are just empty. 
- con: dict[str:str] -- An LLM generated argument for why the speech does not expresses the narrative. Three models were used for generation: {gpt, qwen, llama}