Information Extraction in Legal Documents

Legal data in general, and contracts in particular, are diverse and complex due to an important number of contract types that can cover different fields and topics. The difference in writing style and background of lawyers and paralegals can lead to different ways of covering the same topic in a contract.

These key elements for contract analysis come in various forms. They can be entire paragraphs covering sensitive topics such as the scope of the agreement, responsibilities and obligations. They can be phrases representing valuable information (a key amount or date) such as term related information (termination, renewal, …).

This part of our research focuses on question answering based approach to legal documents analysis to extract such key elements. Additionally, we search for procedures to leverage our non annotated documents to improve our results for both our retriever and reader models.