coastalcph/populism-trump-2016

Dataset Card

Dataset Summary

We curate and release a dataset of 70 presidential campaign speeches by Donald Trump during the Republican Party primaries and the 2016 presidential campaign, dubbed TRUMP-2016. The speech transcripts, collected from UC Santa Barbara’s American Presidency Project, range from June 2016 to January 2017. All transcripts were curated manually to fix transcription errors, remove chants and other interventions from the audience in these rallies, and eliminate remarks by guest speakers or third parties.

Labeling / Annotation

The speeches in this set were split into separate sentences, which were annotated based on the 3-class labeling by four undergraduate students of the Department of Political Science at Yale University, under the supervision of our domain expert. The students underwent extensive training based on a detailed coding scheme that identifies the core markers of populist discourse and disambiguates populism from other social science concepts such as nationalism, nativism, or socialism. Having reached an inter-annotator agreement of Krippendorff's α=0.751 in a subset of the data, the students annotated a total of 15,025 sentences with approximately 15 words per sentence on average. We use this dataset for training and evaluation purposes, splitting it chronologically into training (56 speeches) and test (14 speeches) subsets.

Related Task

The goal of this task is to develop a method to identify fine-grained forms of populist discourse. Under our working definition, populism has two main components: people-centrism and anti-elitism that need to operate in tandem before we can label a discursive body (e.g. a political speech or a social media post) as a populist one. To enable a fine-grained analysis of political content and take advantage of higher-frequency data for examining variations in populist discourse, we move beyond full speeches or paragraphs as units of analysis and instead focus on a lower-level discursive structure: the sentence. We frame the task as a three-way multi-label classification, where each sentence can be labeled as neutral, anti-elitist, or people-centric. The co-occurrence of the latter two labels indicates a fully populist sentence.

Statistics

Subset	Samples
Train	11,428
Test	3,597

Category	Samples
Neutral (N)	13,910 (92.6%)
Anti-elitism (AE)	826 (5.5%)
People-centrism (PC)	517 (3.4%)

Data Fields

• speech_id: a string with the speech identifier.
• sentence: a string with a sentence from a speech.
• pop_code: an integer [0,1,2,3] identifying the category of the sentence.
• speaker: a string with the full name of the speaker.
• location: a string with the name of the location of the event.
• date: a string with the full date (YYYY-MM-DD) of the speech.

Data Instance / Example

{
'speech_id': '2016_06_22_trump.txt',
'sentence': "Here is my promise to the American voter: If I am elected President, I will end the special interest monopoly in Washington, D.C.",
'pop_code': 1,
'speaker': 'Donald Trump'
'location': 'New York, NY'
'date': '2016-06-22',
}

How to use

from datasets import load_dataset
train_dataset = load_dataset('coastalcph/populism-trump-2016', split='train')

Intended Usage

The dataset is intended for academic research for training and/or evaluating language models.

Discussion of Biases

The dataset includes speeches of Donald Trump from presidential campaign rallies. It may contain derogatory language against protected groups based on race, gender, or other characteristics. The authors do not endorse the use of such language.

Citation Information

Identifying Fine-grained Forms of Populism in Political Discourse: A Case Study on Donald Trump’s Presidential Campaigns. Ilias Chalkidis, Stephanie Brandl, and Paris Aslanidis. Arxiv Preprint, 2025.

@misc{chalkidis-et-al-2025-populism,
    title = "Identifying Fine-grained Forms of Populism in Political Discourse: A Case Study on Donald Trump’s Presidential Campaigns",
    author = "Chalkidis, Ilias and Brandl, Stephanie and Aslanidis, Paris",
    year = "2025",
    archivePrefix={arXiv},
    primaryClass={cs.CL},
    url={https://arxiv.org/abs/2507.19303}
}