The Text Anonymization Benchmark (TAB) is a new, open-source corpus for text anonymization. It comprises 1,268 English-language court cases from the European Court of Human Rights (ECHR) manually annotated with:
- semantic categories for personal identifiers,
- masking decisions (in regard to the re-identification risk for the person to protect),
- confidential attributes,
- co-reference relations.
Details about the annotation process employed to develop this corpus can be found in the following paper:
Ildikó Pilán, Pierre Lison, Lilja Øvrelid, Anthi Papadopoulou, David Sánchez, Montserrat Batet.
The Text Anonymization Benchmark (TAB): A Dedicated Corpus and Evaluation Framework for Text Anonymization,
arXiv:2202.00443.
This repository contains the v1.0 release of the Text Anonymization Benchmark, a corpus for text anonymization. The corpus comprises 1,268 English-language court cases from the European Court for Human Rights (ECHR). The documents were manually annotated with information about personal identifiers (including their semantic category and need for masking), confidential attributes and co-reference relations.
The data is distributed in a standoff JSON format consisting of a list of document object with the following information:
| Variable name | Description |
|---|---|
| annotations | an object with document annotations, each containing an object with entity mention annotations |
| dataset_type | which data split the court case belongs to (train /dev / test) |
| doc_id | the ID of the court case (e.g. “001-61807”) |
| meta | an object with metadata for each case (year, countries and legal articles involved etc.) |
| quality_checked | whether the document was revised by another annotator |
| task | the target of the anonymisation task (i.g. who to anonymise) |
| text | the text of the court case used during the annotation |
Each entity mention object under 'annotations' has the following attributes:
| Variable name | Description |
|---|---|
| entity_type | the semantic category of the entity (e.g. PERSON) |
| entity_mention_id | ID of the entity mention |
| start_offset | start character offset of the annotated span |
| end_offset | end character offset of the annotated span |
| span_text | the text of the annotated span |
| edit_type | type of annotator action for the mention (check / insert / correct) |
| identifier_type | the need for masking, masked if 'DIRECT' or 'QUASI', 'NO_MASK' otherwise |
| entity_id | ID of the entity the entity mention is related to in meaning |
| confidential_status | category of a potential source of discrimination (e.g. beliefs, sexual orientation etc.) |
TAB is released under an MIT License.
The MIT License is a short and simple permissive license allowing both commercial and non-commercial use of the software. The only requirement is to preserve the copyright and license notices (see file License). Licensed works, modifications, and larger works may be distributed under different terms and without source code.