This thesis invites you to work at the intersection of AI/NLP and law: You will explore how generative AI can be used to convert court decisions — often written in dense legal jargon — into plain, understandable language for non-experts (laypersons). The aim is to improve access to justice and transparency by making legal decisions more comprehensible. The thesis will be supervised at TU Dresden (AI / NLP) in collaboration with a professor of law. Together we aim for a result that is robust and potentially publishable in a peer-reviewed venue.

What are the tasks?

• Analyse legal texts & identify simplification challenges
  • Study a collection of (anonymized) court decisions. Identify common sources of complexity: specialized legal vocabulary, long sentences, nested references, subtle logical/legal structure.
  • Review existing “plain-language” or “legal simplicity” guidelines to understand what counts as legally faithful but accessible text.
• Develop & apply AI-based simplification methods
  • Work with modern NLP / generative-AI models (e.g., encoder-decoder or decoder-only models) to produce simplified versions of legal texts. You may build on methods from recent research (e.g., https://link.springer.com/article/10.1007/s10462-025-11392-7)
  • Alternatively or additionally, fine-tune pre-trained multilingual or domain-adapted models.
  • Explore retrieval-augmented generation (RAG) pipelines: retrieve relevant statutes, case-law or explanatory text to ground the generative simplification, to mitigate hallucinations — a major risk in legal text generation. (https://arxiv.org/abs/2510.06999)
  • Explore LLMs agents to come up with simplified texts.
• Evaluate simplified texts via user study
  • Design a user study in which participants (e.g., students or other laypersons) read both original and simplified versions of decisions and answer comprehension questions or rate readability/clarity.
  • Collect participant responses, analyze results (e.g., readability, comprehension, perceived clarity).
  • Evaluate also based on other metrics e.g., LLM-based and embedding-based metrics.

What prerequisites do you need?

• Strong motivation for combining AI/NLP and legal communication.
• Very good programming skills (Python) and experience with data processing. Practical familiarity with transformer-based NLP libraries (e.g., Hugging Face) and working with Large Language Models (LLMs) or other modern generative models is highly desirable.
• Willingness to engage with both technical aspects and empirical evaluation (via user study).
• Very good English; German is highly helpful (in particular if working with German court decisions or legal texts).
• Ambition to deliver a high-quality thesis that could serve as the basis for a peer-reviewed publication.

Why this thesis is special

• Legal texts are often opaque to non-experts — simplifying them can significantly enhance access to justice and civic transparency.
• You will combine state-of-the-art NLP / generative AI techniques with real-world legal domain challenges, bridging technical depth and societal impact.
• The project is designed to be achievable within the scope of a Master thesis, but at the same time ambitious enough for scientific publication and real-world relevance.