Case study

Natural language processing for Land Registry documentation in Sweden

Learn how the Swedish Land Registry used natural language processing to handle land registry requests more efficiently.

This guidance is part of a wider collection about using artificial intelligence (AI) in the public sector.

AI application used

  • natural language processing (NLP)

Objective

The Swedish Land Registry (SLR) needed to increase their efficiency when dealing with land registry requests.

Situation

The SLR’s case handlers receive many requests about properties and property rights from citizens. To make decisions, case handlers need to review historical information for that property, often dating back to the 1850s.

Case handlers spent approximately 48,000 hours a year manually translating and evaluating the handwritten documents. The government charges citizens for the number of hours case handlers spend working on their request.

Action

The handwritten documents were of low quality and resolution which made analysis challenging. To fix this, the SLR carried out pre-processing work on the handwritten documents to improve the quality of the input data. The SLR then used handwritten text recognition (HTR) to extract information from the handwritten documents.

The SLR then used a neural network to apply word corrections and associations to complete any sentences with words not captured by HTR. Once SLR had extracted the text, they used an AI model to highlight key features in the document, for example location, names and summary.

Impact

The model now allows case handlers to rapidly respond to citizen requests and focus more on urgent decisions.

Updates to this page

Published 10 June 2019