evidoria

← Back to browse

Good practice Imported

Anoppi — Finland's AI Tool for Anonymising Court and Authority Decisions Before Publication

Finland · Helsinki · See the Finland profile · See the Helsinki profile

Evidence: Descriptive / self-reported Top 46% 60/100 · Ask Evidence Copilot about this practice

Finland's Ministry of Justice, with the University of Helsinki, Aalto University and Statistics Finland, built Anoppi — an open-source tool that automatically pseudonymises names and personal data in court and authority decisions before they are published as open data.

Anoppi — Finland's AI Tool for Anonymising Court and Authority Decisions Before Publication

Details

Maturity
Established
Promoter
Finnish Ministry of Justice, with the University of Helsinki (HELDIG), Aalto University and Statistics Finland
Period
2018–2020 (project); tool remains available as open-source software
Keywords
justice, data protection, open data, public administration

Context

Under Finland's GDPR obligations, court decisions and other official rulings published for research and public scrutiny must first have personal data removed — traditionally a slow, manual redaction task. The Anoppi project (2018-2020), led by the Ministry of Justice with the University of Helsinki (HELDIG), Aalto University, Statistics Finland and publisher Edita, built a language-technology tool to automate this for Finnish-language documents.

Activities

Anoppi combines statistical and rule-based named-entity recognition with morphological analysis to detect and replace names and identifiers with neutral placeholders (e.g. 'Person A') while preserving document structure, offered as a web application and REST API; it was demonstrated at the Council of Europe's AI conference in Helsinki in February 2019 and published as open-source code.

Results

The project's own peer-reviewed evaluation found Anoppi 'performs well with different types of documents'.

Conclusions

The evaluation also flagged that named-entity recognition and disambiguation still produce errors, an honest acknowledgment that automated pseudonymisation is not yet error-free — publishers using it should keep human review in the loop for highly sensitive documents.

Implementation
Indicative cost
Low (< €50k)
Time to results
Medium (1–3 years)
Staffing & skills
Ministry of Justice, University of Helsinki (HELDIG), Aalto University Semantic Computing Research Group, Statistics Finland

Conditions for success

  • Combining statistical and rule-based NER with morphological analysis for a morphologically rich language
  • Open-source release for reuse by other authorities

Common failure modes

  • Named-entity recognition and disambiguation still produce errors; human review recommended for highly sensitive documents

Commonly funded by

Digital Europe Programme CERV — Citizens, Equality, Rights and Values

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful