FACTA-PT Claim-Evidence Entailment — XLM-R-large-XNLI

This repository contains the model fine-tuned for Claim-Evidence Entailment in FACTA-PT.

Model Description

The model is derived from joeddav/xlm-roberta-large-xnli and was adapted to classify the factual relation between a claim span and an evidence passage extracted from a document previously classified as Relevant.

Task

The evidence passage is treated as the premise and the claim span as the hypothesis.

The model predicts one of three classes:

• Supported
• Refuted
• NEI (Not Enough Information)

In the experimental formulation, each evidence passage consists of three consecutive sentences.

Base Model

• Base model: joeddav/xlm-roberta-large-xnli
• Architecture: XLM-RoBERTa-large
• Prior specialization: three-class Natural Language Inference
• Base-model license: MIT

Training Data

The model was fine-tuned using evidence passages constructed from the CLEVER external-evidence annotations developed within the associated dissertation.

Intended Use

The model is intended to characterize the factual relation between a claim span and a local evidence passage after document relevance classification.

Limitations

The model performs passage-level classification and does not independently identify the appropriate evidence region within a complete document.

Performance should also be interpreted with particular care for classes represented by a limited number of examples in the evaluation data.

License

This fine-tuned model is released under the MIT License, following the license of the base model joeddav/xlm-roberta-large-xnli.

FACTA-PT

This model is one of the components developed and selected for FACTA-PT.