Detection of Contradictions in Legal Text Using Various NLP Techniques

This project tackles contradiction detection in legal documents via Natural Language Inference (NLI)

Given a premise (e.g. a clause or finding) and a hypothesis (e.g. a claim or conclusion), the system classifies the relation as Entailment, Contradiction, or Not Mentioned. We combine classical NLP (tokenization, lemmatization, TF‑IDF, logistic regression) with transformer-based models: a fine-tuned Legal English RoBERTa for sequence classification and a sentence-transformers setup for contradiction-oriented semantic similarity. All implemented in a single Jupyter pipeline (EDA → preprocessing → baselines → BERT NLI → sentence embeddings).

---

Overview

• Task: 3-class NLI on legal documents (premise vs hypothesis).
• Labels: 0 = Entailment, 1 = Contradiction, 2 = Not Mentioned.
• Pipeline: EDA → text preprocessing (NLTK, TF-IDF) → baseline ML (e.g. logistic regression) → fine-tuned Legal English RoBERTa for sequence classification → sentence-transformers setup for contradiction-focused similarity.

---

Project structure

.
├── main.ipynb          # Full pipeline: EDA, preprocessing, ML baselines, BERT NLI, sentence-transformers
├── requirements.txt   # Python dependencies
└── README.md

Note: Model checkpoints (saved_nli_model/, sentence_transformer_contradiction_model/, bert_contradiction/) and data/ are not in the repo (see Setup).

---