Claim Perspectives (Perspectrum)
Annotate diverse perspectives on claims with stance and evidence. Based on Chen et al., NAACL 2019. Identify supporting and opposing perspectives for controversial claims.
text annotation
Configuration Fileconfig.yaml
# Claim Perspectives (Perspectrum)
# Based on Chen et al., NAACL 2019
# Paper: https://aclanthology.org/N19-1053/
# Dataset: https://github.com/CogComp/perspectrum
#
# Perspectrum captures diverse viewpoints on controversial claims.
# For each claim, multiple perspectives are collected that either
# support or oppose the claim, along with evidence.
#
# Key Concepts:
# - Claim: A controversial statement that can be debated
# - Perspective: A specific viewpoint on the claim
# - Evidence: Documents that support the perspective
# - Stance: Whether the perspective supports or opposes the claim
#
# Annotation Task:
# Given a claim and a perspective, determine:
# 1. Does the perspective SUPPORT or OPPOSE the claim?
# 2. Is the perspective relevant to the claim?
# 3. How strong is this perspective?
#
# Guidelines:
# 1. Read the claim carefully first
# 2. A perspective SUPPORTS if accepting it would support the claim
# 3. A perspective OPPOSES if accepting it would undermine the claim
# 4. Perspectives can be factual or opinion-based
port: 8000
server_name: localhost
task_name: "Claim Perspectives"
data_files:
- sample-data.json
id_key: id
text_key: pair
output_file: annotations.json
annotation_schemes:
# Step 1: Stance classification
- annotation_type: radio
name: stance
description: "Does the PERSPECTIVE support or oppose the CLAIM?"
labels:
- "Supports"
- "Opposes"
- "Neutral/Unclear"
tooltips:
"Supports": "The perspective, if true, would support the claim"
"Opposes": "The perspective, if true, would oppose or undermine the claim"
"Neutral/Unclear": "The perspective doesn't clearly support or oppose"
# Step 2: Relevance
- annotation_type: radio
name: relevance
description: "How relevant is this perspective to the claim?"
labels:
- "Highly relevant"
- "Somewhat relevant"
- "Barely relevant"
- "Not relevant"
tooltips:
"Highly relevant": "Directly addresses the core of the claim"
"Somewhat relevant": "Related but not central to the claim"
"Barely relevant": "Only tangentially related"
"Not relevant": "Not related to the claim at all"
# Step 3: Perspective strength
- annotation_type: likert
name: strength
description: "How strong/convincing is this perspective?"
min_value: 1
max_value: 5
labels:
1: "Very weak"
2: "Weak"
3: "Moderate"
4: "Strong"
5: "Very strong"
tooltips:
1: "Unconvincing or poorly reasoned"
2: "Has some merit but not compelling"
3: "Reasonable argument"
4: "Compelling argument"
5: "Highly convincing argument"
# Step 4: Confidence
- annotation_type: likert
name: confidence
description: "How confident are you in your annotation?"
min_value: 1
max_value: 5
labels:
1: "Very uncertain"
2: "Somewhat uncertain"
3: "Moderately confident"
4: "Confident"
5: "Very confident"
allow_all_users: true
instances_per_annotator: 50
annotation_per_instance: 3
allow_skip: true
skip_reason_required: false
Sample Datasample-data.json
[
{
"id": "persp_001",
"claim": "College education should be free for all students.",
"perspective": "Free college would increase social mobility by allowing students from low-income families to access higher education.",
"pair": "CLAIM: College education should be free for all students.\n\nPERSPECTIVE: Free college would increase social mobility by allowing students from low-income families to access higher education."
},
{
"id": "persp_002",
"claim": "College education should be free for all students.",
"perspective": "Making college free would devalue degrees and reduce the incentive for academic excellence.",
"pair": "CLAIM: College education should be free for all students.\n\nPERSPECTIVE: Making college free would devalue degrees and reduce the incentive for academic excellence."
}
]
// ... and 6 more itemsGet This Design
Clone or download from the repository
Quick start:
git clone https://github.com/davidjurgens/potato-showcase.git cd potato-showcase/claim-perspectives potato start config.yaml
Details
Annotation Types
Domain
Use Cases
Tags
Found an issue or want to improve this design?
Open an IssueRelated Designs
Argument Reasoning Comprehension (ARCT)
Identify implicit warrants in arguments. Based on Habernal et al., NAACL 2018 / SemEval 2018 Task 12. Given a claim and premise, choose the correct warrant that connects them.
Scientific Claim Verification (SciFact)
Verify scientific claims against evidence from research abstracts. Based on SciFact (Wadden et al., EMNLP 2020). Classify claims as supported, refuted, or having insufficient evidence, and identify rationale sentences.
Argument Quality Assessment
Multi-dimensional argument quality annotation based on the Wachsmuth et al. (2017) taxonomy. Rates arguments on three dimensions: Cogency (logical validity), Effectiveness (persuasive power), and Reasonableness (contribution to resolution). Used in Dagstuhl-ArgQuality and GAQCorpus datasets.