Modelpublic

anaskhan1/urdu-factcheck-xlmr-counterfactual-fever

sourceHugging Facemitupdated 4mo agoView on Hugging Face

0likes8downloads

Model Card

Urdu Fact Verification XLM-R Counterfactual FEVER

This model is an XLM-RoBERTa Large sequence-classification model fine-tuned on machine-translated Urdu FEVER with counterfactual augmentation.

Labels:

—0: SUPPORTS
—1: REFUTES
—2: NOTENOUGHINFO

This checkpoint is intended as the source-domain model for research on why machine-translated Urdu FEVER does not generalize well to real Urdu fact-checking.

Main reported result:

—MT Urdu-FEVER test accuracy: 91.92%
—MT Urdu-FEVER macro F1: 0.9144
—UrduFactCheck zero-shot accuracy: 55.41%
—UrduFactCheck macro F1: 0.4747

For code, diagnostics, and the research paper summary, see the GitHub repository:

https://github.com/Anas-Khaan/CrossLingual-Hallucination-NLI