Legal Information Retrieval and Entailment Based on BM25, Transformer and Semantic Thesaurus Methods.

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Legal Informatics

Background:

• The Competition on Legal Information Extraction and Entailment (COLIEE) is a key event for evaluating AI systems in legal domains.
• Accurate legal information retrieval and entailment are crucial for legal professionals and research.

Purpose of the Study:

• To detail the methodologies employed by the University of Alberta (UA) team in COLIEE 2021.
• To evaluate the performance of specific AI techniques in legal information processing tasks.
• To identify areas for future research in legal AI.

Main Methods:

• Transformer-based models for case law entailment.
• BM25 information retrieval technique for legal document retrieval.
• Natural language inference with semantic knowledge for statute law analysis.

Main Results:

• The transformer approach for case law entailment ranked 4th (Task 2).
• BM25 achieved 3rd place in legal information retrieval (Task 3).
• Semantic natural language inference for statutes ranked 4th (Task 4).
• The combined BM25 and semantic inference approach secured 2nd place (Task 5).

Conclusions:

• The UA team's methods demonstrated strong performance in COLIEE 2021.
• Transformer and semantic-based approaches show significant promise for legal AI.
• Error analysis provides valuable insights for advancing state-of-the-art legal NLP.