There is an ocean of untapped insight lurking in unstructured legal writing, such as judgments, reports and journal articles. The extraction of insight at scale requires the ability to impose structure at scale. Blackstone is an experiment directed towards the development of an open-source model capable of automatically enriching unstructured legal texts at scale.
Natural language processing; machine-learning; unstructured text data
Automatic enrichment of unstructured legal texts, with an emphasis on judgments and other long-form legal material (e.g. academic articles, reports etc). Enrichment of contracts is already well-served and therefore does not fall within the scope of this project.
There is no shortage of commercial, closed-source software that uses natural language processing and computational linguists to provide insight into unstructured legal texts, such as contracts, credit agreements and leases.
The purpose of Blackstone is to apply similar techniques, technologies and strategies to other types of unstructured legal texts, particularly judgments, in order to generate an open-source model that can be used and extended by others.
The initial focus of the project will be to develop a model that is capable of automatically identifying the following fundamental entity types that are peculiar to legal writing:
Case titles - e.g.
Regina v Smith
Neutral citations - e.g.
 EWCA Crim 345
Regular citations - e.g.
 AC 345or
 2 Cr App R 7
Primary legislation - e.g.
Criminal Justice Act 2003
Secondary legislation - e.g.
The Wine (Amendment) Regulations 2019
Regnal years - e.g.
8 & 9 Geo. 6, c. 4
The second phase of the project will focus on extending the model developed in
Phase 1 to identify the following:
- Instances in which the author appears to postulate an axiom of the law (e.g. It is a well-established principle that...)
- Instances of ratio in judgments
- Instances in which earlier authority is being subjected to "judicial consideration" going beyond mere citation (e.g. where an earlier authority is subject to positive or negative judicial consideration)
Dataset - XML archive of law reports published by ICLR dating back to 1865 and raw text of unreported judgments dating back to 2000.
Runtime - Python 3.6+
Blog + Resources
Interested in getting involved?
Fill out this form and tell us a bit about yourself.