Duration

11-2024 to 03-2025 

Goal

The goal of this project is to develop an AI-powered language simplification tool that helps editors simplifying content by reducing the time required, improving the quality, ensuring uniform output and improving job satisfaction. 

Motivation

• Helping editors write in a way that’s easier to read and more inclusive, especially for residents who may have limited language skills.

• Showing how AI can support—not replace—public servants. Editors remain in control, while the tool offers suggestions, explanations, and a faster way to produce clearer messages. It's a step toward more transparent, understandable government communication.

Participants

The project was carried out in collaboration with the AI Lab, and initiated by Frank Willemsen. The target audience included 14 DVL editors and around 40 communication advisors.

Process

We followed the double diamond process (page 6):

1. Discover: analysis of exisiting findings, ideation session, and technical exploration;
2. Define: design prototypes, facilitate usability tests, and refine deliverable;
3. Develop: design and build MVP
4. Test: facilitate two pilot weeks, create surveys, and analysis of the results.

Key Insights

• Current tools on the market do not provide an answer to current needs. For example prompt-based tools require a lot of iterations to achieve a good simplification (page 15).
• The human should remain in the loop and content writers should continue in control of what is published. The simplification tool should provide explanations and work as an aid to achieve better and faster simplifications (page 41).
• During this short-term project we focused on developing an interface for this use case, so the model still has some limitations, such as hallucinations and not taking the context of the whole into consideration (page 45).
• Despite some limitations, most users said AlleTaal made their work easier and even gave them new ideas for how to write simpler texts (page 36).

Conclusion & Recommendations

AlleTaal was considered a good starting point for AI-supported text simplification in municipal communication. Further improvements should include (pages 46-51) :

• Enhance model accuracy to reduce hallucinations and improve context handling;

• Preserve formatting and support version control;

• Refine LiNT integration for more accurate readability scoring;

• Collect more training data and explore document-level fine-tuning;

• Implement smarter quality control, such as AI-based verification systems.

Open-source library available at: https://github.com/Amsterdam/alletaal-lint