Spanish Datasets for Sensitive Entity Detection in the Legal Domain

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 3751–3760 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 3751 Spanish Datasets for Sensitive Entity Detection in the Legal Domain Ona de Gibert1, Aitor Garc´ıa-Pablos2, Montse Cuadros2, Maite Melero1 1Barcelona Super Computing Center (BSC), 2SNLT group at Vicomtech Foundation, Basque Research and Technology Alliance (BRTA) Plac¸a Eusebi G¨uell 1-3, Barcelona 08034, Spain Mikeletegi Pasealekua 57, Donostia/San-Sebasti´an, 20009, Spain ona.degibert@bsc.es, agarciap@vicomtech.org, mcuadros@vicomtech.org, maite.melero@bsc.es Abstract The de-identification of sensible data, also known as automatic textual anonymisation, is essential for data sharing and reuse, both for research and commercial purposes. The first step for data anonymisation is the detection of sensible entities. In this work, we present four new datasets for named entity detection in Spanish in the legal domain. These datasets have been generated in the framework of the MAPA project, three smaller datasets have been manually annotated and one large dataset has been automatically annotated, with an estimated error rate of around 14%. In order to assess the quality of the generated datasets, we have used them to fine-tune a battery of entity-detection models, using as foundation different pre-trained language models: one multilingual, two general-domain monolingual and one in-domain monolingual. We compare the results obtained, which validate the datasets as a valuable resource to fine-tune models for the task of named entity detection. We further explore the proposed methodology by applying it to a real use case scenario. Keywords: Anonymisation, Named Entity Recognition, Sensitive Data Detection 1. Introduction The de-identification of sensible data, also known as automatic textual anonymisation, is essential for data sharing and reuse both for research and commercial purposes, in order to comply with the General Data Protection Regulation (GDPR) law1, which protects data privacy. Anonymisation helps tackle the issue of protecting personal information, by identifying and subsequently processing (for example, by ofuscating) personal data such as names, emails, addresses, etc. from a text, while preserving its formal structure. In the context of the European Union, where Public Ad- ministrations are encouraged to re-use and open pub- lic sector information2, there is a need to find effec- tive ways of sharing large amounts of collected infor- mation, while complying with the GDPR. The project MAPA (Multilingual Anonymisation for Public Ad- ministrations)3, an INEA-funded Action for the Euro- pean Commission under the Connecting Europe Facil- ity (CEF) program, aims at addressing this issue by targeting domain-specific anonymisation in the 24 lan- guages of the European Union (Gianola et al., 2020). The first step for data anonymisation is the detection of sensible entities that need to be anonymised. In this work, we present several datasets annotated for named entity detection in Spanish, with a view to train mod- els for anonymisation of personal information in the le- gal domain. These datasets have been developed in the framework of the MAPA project and are available to the community. We have used them to fine-tune dif- ferent Transformer-based pre-trained language models, resulting in a …

Content truncated to 500 words