EN | FR
LABEL is an open source text annotation software created by the french supreme court 'Cour de cassation'. It is currently used for annotating court decisions for the french justice open data project.
This tool allows annotators to proofread and review decisions that have been pre-annotated by an NLP algorithm.
We have designed an easy-to-use administration interface that integrates many features. Here are the features available in the different tabs of the administration interface: consultation of statistics, lists of documents (to be processed, processed, documents requiring double review), list of reports, user management, management of document assignments.
For each document, you can view the anonymized document, assign the document to yourself or to a chosen annotator and even remove the assignation (if for instance the document needs to be treated by another annotator).
You can also see that you have a search bar. You can type a document number to find it instantly. The refresh button updates the document list with the last available data.
You can filter the list of documents. There are many fields you can filter. For instance, you can filter by treatment date, by import date, by source database and more. The fields are grayed out if the filters cannot find data. This prevents empty selections.
Note that you can also sort the table by clicking on the columns' header.
By default, LABEL allows annotations of various types: phone numbers, dates of birth or bank account numbers for example. You can customize everything and the "additional annotations" field allows the addition of custom annotations that do not fit into a predefined type. This is very useful for non-generic annotation requests such as flight numbers, horse names, etc. for example. This field also allows you to flag specific terms that should not be annotated.
You can browse and edit annotations from the left panel, but also directly in the text. The labels are interactive: you can change annotation type, link or unlink the annotation or even delete it. Obviously, you can label other stuff by selecting them in the text. The software checks the consistency of your labeling to avoid mistakes. You can still go back or reset your modifications at any time.
LABEL allow you to link annotations. Sometimes names are written lowercase or in capitals. By linking them, the replacement terms will be the same (whether the names are in lowercase or not). This increases readability of anonymized documents.
This is a non-exhaustive review of LABEL's features. Contact us if you want a demo or a detailed presentation. And yes, we do i18n!
- October 2016: French law for a Digital Republic.
- November 2017: Repport of Cadiet mission.
- 23 march 2019: Programmatic justice law 2018-2022.
- 29 june 2020: Decree about the publishing of court decision.
- September 2020: The project is initiated by the Entrepeneur d'Intérêt Général (EIG) program.
- June 2021: Production launch with supreme court decisions (Jurinet database).
- April 2022: Addition of appeal court's decisions (JuriCA database).
- December 2023: Addition of 9 first judicial courts' decisions.
- During 2024-2025: Addition of other judicial courts' decisions.
Next steps:
- December 2024: Addition of industrial courts' decisions.
- Septembre 2025: Addition of labour courts' decisions and all first instance.
Learn more with the roadmap.
Follow the installation guide.
LABEL has been designed to be reused whatever the annotation context. There are two different kinds of packages in LABEL
specific: what is specific to the Cour de cassation (machine learning engine API, database connector, etc.)generic: what is not linked to the specific needs of the Cour de cassation
Learn more in the reuser guide.