WinterJet2021/Fallacy-Hunter

FallacyHunter RoBERTa Fallacy Classifier

This model is a RoBERTa-based fallacy classifier fine-tuned for the FallacyHunter project. It predicts one of 14 fallacy labels for a given argument or statement.

Model Details

• Base model: RoBERTa checkpoint
• Task: fallacy classification
• Output labels: ad hominem, ad populum, appeal to emotion, circular reasoning, equivocation, fallacy of credibility, fallacy of extension, fallacy of logic, fallacy of relevance, false causality, false dilemma, faulty generalization, intentional, no_fallacy
• Repository artifact: local checkpoint directory used for upload

Intended Use

Use this model to label argumentative text for FallacyHunter experiments and related analysis. It is suited for offline evaluation, fairness testing, and research workflows.

Limitations

• The model is trained on the FallacyHunter label set and should not be treated as a general-purpose reasoning system.
• Predictions are only as reliable as the text distribution seen during fine-tuning.
• Multi-label style outputs should be interpreted according to the checkpoint configuration and downstream decoding logic.

Labels

0. ad hominem
1. ad populum
2. appeal to emotion
3. circular reasoning
4. equivocation
5. fallacy of credibility
6. fallacy of extension
7. fallacy of logic
8. fallacy of relevance
9. false causality
10. false dilemma
11. faulty generalization
12. intentional
13. no_fallacy

Files

This repository folder contains the full local checkpoint used for upload:

• config.json
• label_map.json
• model.safetensors
• tokenizer.json
• tokenizer_config.json

Example

from transformers import pipeline

classifier = pipeline("text-classification", model="<username>/<repo_name>")
classifier("That argument ignores the evidence and attacks the person instead.")