SemEval-2021 Task 6: Persuasion Techniques in Memes
SemEval-2021 Task 6 labels persuasion techniques in 950 English memes across text, spans, and the full image. This Potato config reproduces the multiselect technique labeling and span marking.
About this dataset
SemEval-2021 Task 6, "Detection of Persuasion Techniques in Texts and Images", was organized by Dimitrov, Bin Ali, Shaar, Alam, Silvestri, Firooz, Nakov, and Da San Martino at the 15th International Workshop on Semantic Evaluation (SemEval-2021), held with ACL. It targets the rhetorical and propaganda techniques used in internet memes.
The shared task released a corpus of 950 English memes split into 687 for training, 63 for development, and 200 for testing. The memes were collected during 2020 from 26 public Facebook groups covering politics, COVID-19, vaccines, and gender equality. The task drew 71 registrations and 22 teams that submitted official runs on the test set.
There are three subtasks: detecting techniques in the text, marking the text spans where each technique appears, and detecting techniques across the whole meme (text plus image). Subtask 1 uses an inventory of 20 techniques such as Loaded Language, Name Calling, Appeal to fear/prejudice, Flag-Waving, and Black-and-White Fallacy. Subtask 3 adds two image-only techniques for a total of 22.
The Potato config below reproduces the text-side annotation: a multiselect scheme lets annotators tag which persuasion techniques are present, and a span scheme lets them highlight the exact text where each technique is employed.
- Total memes
- 950
- Train / dev / test
- 687 / 63 / 200
- Techniques (text)
- 20
- Techniques (full meme)
- 22
- Subtasks
- 3
- Teams submitting
- 22
Configuration Fileconfig.yaml
This Potato config reproduces the annotation task. Save it as config.yaml and run potato start config.yaml to try it.
# Persuasive Techniques in Multimodal Content
# Based on Dimitrov et al., SemEval 2021
# Paper: https://aclanthology.org/2021.semeval-1.7/
# Dataset: https://propaganda.qcri.org/semeval2021-task6/
#
# Annotators identify persuasive techniques used in memes and news articles.
# The task involves both selecting which techniques are present and marking
# the text spans where each technique is employed.
annotation_task_name: "Persuasive Techniques in Multimodal Content"
task_dir: "."
data_files:
- sample-data.json
item_properties:
id_key: "id"
text_key: "text"
output_annotation_dir: "annotation_output/"
output_annotation_format: "json"
port: 8000
server_name: localhost
annotation_schemes:
- annotation_type: multiselect
name: persuasion_techniques
description: "Select all persuasive techniques present in the text."
labels:
- "Loaded Language"
- "Name Calling"
- "Bandwagon"
- "Appeal to Fear"
- "Exaggeration"
- "Flag Waving"
- "Black and White"
- "Thought-Terminating Cliche"
- "Appeal to Authority"
- "Slogans"
tooltips:
"Loaded Language": "Using words or phrases with strong emotional implications to influence the audience"
"Name Calling": "Labeling the object of the campaign as something the audience fears or loathes"
"Bandwagon": "Attempting to persuade by claiming that many people support the idea"
"Appeal to Fear": "Building support by instilling fear or anxiety in the audience"
"Exaggeration": "Representing something in an excessive manner beyond what the facts warrant"
"Flag Waving": "Playing on strong national or group feelings to justify an action"
"Black and White": "Presenting only two choices when more options exist"
"Thought-Terminating Cliche": "Using a commonly used phrase to quell cognitive dissonance"
"Appeal to Authority": "Citing an authority figure to support a claim without relevant evidence"
"Slogans": "Using a brief, striking phrase to provoke desired responses"
- annotation_type: span
name: technique_span
description: "Highlight the text span where a persuasive technique is employed."
labels:
- "Technique Span"
annotation_instructions: |
You will see a text excerpt from a meme or news article. Your task is to:
1. Read the text carefully and identify any persuasive or propaganda techniques.
2. Select all techniques that appear in the text from the checklist.
3. Highlight the specific text span where each technique is used.
A single text may contain multiple techniques, and techniques may overlap in spans.
html_layout: |
<div style="padding: 15px; max-width: 800px; margin: auto;">
<div style="background: #fef2f2; border: 1px solid #fecaca; border-radius: 8px; padding: 16px; margin-bottom: 16px;">
<strong style="color: #991b1b;">Source Type:</strong>
<span style="font-size: 14px; background: #fee2e2; padding: 2px 8px; border-radius: 4px;">{{source_type}}</span>
</div>
<div style="background: #f0f9ff; border: 1px solid #bae6fd; border-radius: 8px; padding: 16px; margin-bottom: 16px;">
<strong style="color: #0369a1;">Text:</strong>
<p style="font-size: 16px; line-height: 1.7; margin: 8px 0 0 0;">{{text}}</p>
</div>
</div>
allow_all_users: true
instances_per_annotator: 50
annotation_per_instance: 2
allow_skip: true
skip_reason_required: false
Sample Datasample-data.json
[
{
"id": "pers_001",
"text": "Wake up, people! If we don't act now, our great nation will be overrun by those who want to destroy everything we stand for. It's us or them.",
"source_type": "meme"
},
{
"id": "pers_002",
"text": "Everyone knows that vaccines are the greatest medical achievement of all time. Only a fool would question what millions of scientists have proven beyond doubt.",
"source_type": "news_article"
}
]
// ... and 8 more itemsGet This Design
Clone or download from the repository
Quick start:
git clone https://github.com/davidjurgens/potato-showcase.git cd potato-showcase/semeval/2021/task06-persuasion-multimodal potato start config.yaml
Dataset & paper
Dimitrov et al., SemEval 2021
Citation (BibTeX)
@inproceedings{dimitrov-etal-2021-semeval,
title = "{S}em{E}val-2021 Task 6: Detection of Persuasion Techniques in Texts and Images",
author = "Dimitrov, Dimitar and Bin Ali, Bishr and Shaar, Shaden and Alam, Firoj and Silvestri, Fabrizio and Firooz, Hamed and Nakov, Preslav and Da San Martino, Giovanni",
booktitle = "Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021)",
pages = "70--98",
year = "2021",
publisher = "Association for Computational Linguistics",
doi = "10.18653/v1/2021.semeval-1.7",
url = "https://aclanthology.org/2021.semeval-1.7"
}Details
Annotation Types
Domain
Use Cases
Tags
Found an issue or want to improve this design?
Open an IssueRelated Designs
Detecting Persuasion Techniques in News
Identification of propaganda and persuasion techniques in news articles through both multi-label classification and span-level detection. Based on SemEval-2023 Task 3 (Piskorski et al.).
Food Hazard Detection
Food safety hazard detection task requiring annotators to identify hazards, products, and risk levels in food incident reports, and classify the type of contamination. Based on SemEval-2025 Task 9.
Multilingual Narrative Extraction
Multilingual narrative extraction task requiring annotators to identify narrative elements such as events, actors, and causal relations in news texts, and classify the narrative themes. Based on SemEval-2025 Task 10.