ValueEval - Human Values behind Arguments
Identification of human values expressed in arguments, classifying which of the Schwartz basic values an argument appeals to and whether it attains or constrains those values. Based on SemEval-2023 Task 4 (Kiesel et al.).
About this dataset
SemEval-2023 Task 4, called ValueEval, asked systems to detect the human values behind an argument. It was organized by Johannes Kiesel and colleagues and described in the task paper at the 17th International Workshop on Semantic Evaluation. The task builds on Schwartz's theory of basic human values, where an argument's persuasive force often rests on an implicit value such as tradition or universalism.
The shared task released a multi-sourced dataset of arguments, each given as a premise together with a conclusion and a stance of in favor or against. Annotators mark which value categories the argument appeals to, drawn from a taxonomy of 20 value categories grouped under Schwartz's higher-order values. The task is framed as multi-label classification, so each of the 20 categories is judged present or absent for a given argument.
The dataset holds over 9,000 arguments, and 39 teams submitted runs. The best systems reached 0.87 F1 on the universalism: nature category and above 0.70 on three more, while the hardest categories peaked at 0.39 F1, showing that values such as stimulation and hedonism stay difficult to detect.
The Potato config below reproduces a simplified version of this task with one multiselect and one radio: a multiselect over the ten higher-order Schwartz values (self-direction, stimulation, hedonism, achievement, power, security, conformity, tradition, benevolence, universalism) and a radio marking whether the argument attains or constrains those values. It fits argument-mining and ethics research that needs value labels tied to a stance.
- Arguments
- over 9,000
- Value categories (level 2)
- 20
- Participating teams
- 39
- Best category F1
- 0.87 (universalism: nature)
- Lowest peak F1
- 0.39
- Venue / pages
- SemEval-2023, pp. 2287-2303
Configuration Fileconfig.yaml
This Potato config reproduces the annotation task. Save it as config.yaml and run potato start config.yaml to try it.
# ValueEval - Human Values behind Arguments
# Based on Kiesel et al., SemEval 2023
# Paper: https://aclanthology.org/2023.semeval-1.313/
# Dataset: https://touche.webis.de/semeval23/touche23-web/
#
# This task asks annotators to identify which human values (based on
# Schwartz's theory of basic values) are expressed in a given argument,
# and whether the argument attains or constrains those values.
#
# Annotation Guidelines:
# 1. Read the argument and note its topic and stance
# 2. Select all human values the argument appeals to
# 3. Determine whether the argument attains or constrains the selected values
# 4. An argument may appeal to multiple values simultaneously
# 5. Consider the implicit as well as explicit value appeals
annotation_task_name: "ValueEval - Human Values behind Arguments"
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: human_values
description: "Select all human values this argument appeals to"
labels:
- "Self-direction"
- "Stimulation"
- "Hedonism"
- "Achievement"
- "Power"
- "Security"
- "Conformity"
- "Tradition"
- "Benevolence"
- "Universalism"
tooltips:
"Self-direction": "Independent thought and action, freedom, creativity, curiosity"
"Stimulation": "Excitement, novelty, challenge in life"
"Hedonism": "Pleasure, sensuous gratification, enjoyment of life"
"Achievement": "Personal success through demonstrating competence"
"Power": "Social status, prestige, control over people and resources"
"Security": "Safety, harmony, stability of society, relationships, and self"
"Conformity": "Restraint of actions that might harm others or violate norms"
"Tradition": "Respect for customs and ideas from culture or religion"
"Benevolence": "Preserving and enhancing welfare of close others"
"Universalism": "Understanding, tolerance, and protection of all people and nature"
- annotation_type: radio
name: value_stance
description: "Does the argument attain or constrain the selected values?"
labels:
- "Attains Value"
- "Constrains Value"
keyboard_shortcuts:
"Attains Value": "1"
"Constrains Value": "2"
tooltips:
"Attains Value": "The argument promotes or supports the realization of the value"
"Constrains Value": "The argument limits or restricts the realization of the value"
annotation_instructions: |
You will see an argument related to a specific topic and stance. Your task is to:
1. Identify which human values the argument appeals to (select all that apply).
2. Determine whether the argument attains (promotes) or constrains (limits) those values.
An argument may appeal to multiple values -- select all relevant ones.
html_layout: |
<div style="padding: 15px; max-width: 800px; margin: auto;">
<div style="display: flex; gap: 12px; margin-bottom: 16px;">
<div style="background: #ede9fe; border: 1px solid #c4b5fd; border-radius: 8px; padding: 12px; flex: 1;">
<strong style="color: #6d28d9;">Topic:</strong>
<span style="font-size: 15px;">{{topic}}</span>
</div>
<div style="background: #fce7f3; border: 1px solid #f9a8d4; border-radius: 8px; padding: 12px; flex: 1;">
<strong style="color: #be185d;">Stance:</strong>
<span style="font-size: 15px;">{{stance}}</span>
</div>
</div>
<div style="background: #f0f9ff; border: 1px solid #bae6fd; border-radius: 8px; padding: 16px;">
<strong style="color: #0369a1;">Argument:</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": "valueeval_001",
"text": "Universal healthcare ensures that every citizen has access to medical treatment regardless of their economic status, reducing inequality and preventing needless suffering.",
"topic": "Universal Healthcare",
"stance": "In Favor"
},
{
"id": "valueeval_002",
"text": "Mandatory military service builds discipline and national unity, teaching young people the value of sacrifice and duty to their country.",
"topic": "Mandatory Military Service",
"stance": "In Favor"
}
]
// ... 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/2023/task04-valueeval potato start config.yaml
Dataset & paper
Kiesel et al., SemEval-2023
Citation (BibTeX)
@inproceedings{kiesel-etal-2023-semeval,
title = "{S}em{E}val-2023 {T}ask 4: {V}alue{E}val: Identification of Human Values Behind Arguments",
author = "Kiesel, Johannes and Alshomary, Milad and Mirzakhmedova, Nailia and Heinrich, Maximilian and Handke, Nicolas and Wachsmuth, Henning and Stein, Benno",
booktitle = "Proceedings of the 17th International Workshop on Semantic Evaluation (SemEval-2023)",
month = jul,
year = "2023",
address = "Toronto, Canada",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.semeval-1.313",
pages = "2287--2303"
}Details
Annotation Types
Domain
Use Cases
Tags
Found an issue or want to improve this design?
Open an IssueRelated Designs
Food Hazard Detection
Food safety hazard detection task requiring annotators to classify short food-incident report texts by hazard-category and product-category (coarse labels), and to identify the specific hazard and product mentions (fine-grained labels). Based on SemEval-2025 Task 9: The Food Hazard Detection Challenge.
iSarcasmEval: Intended Sarcasm Detection
SemEval-2022 Task 6 (iSarcasmEval): detecting intended sarcasm in English and Arabic. Unlike prior datasets labeled by third parties, iSarcasmEval texts are labeled by their own authors for intended sarcasm. Annotators decide whether a text is sarcastic and, if so, which category of ironic speech it expresses.
LLMs4Subjects - Subject Tagging for a Technical Library
Subject indexing of technical bibliographic records for the TIB Leibniz Information Centre's open-access TIBKAT catalog. Annotators assign subject headings and record types to English/German technical records, mirroring the library workflow SemEval-2025 Task 5 automates with LLMs. This showcase is a simplified adaptation of that task.