Project Details
Projekt
TruQALIA: Trustworthy Question Answering to Enable Legal Information Access for Laypeople
Subject Area
Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing
Term
since 2025
Project identifier
Deutsche Forschungsgemeinschaft (DFG) - Project number 564339494
The rule of law represents one of the most important pillars of German democracy. In everyday life, citizens are confronted with situations in which legal advice is essential, such as work, family, or administrative affairs. However, even though the civil law system codified in statutes is accessible to everyone online, understanding German legal texts by laypeople is considered hard to almost impossible.We see four major limitations in existing research on legal question answering, namely (1) almost exclusive focus on case law and English, (2) solving question of legal experts only while ignoring laypersons and their incapability to write and understand the legal jargon, (3) the lack of holistic evaluation of generated legal answers and the reliance on closed-source models, and (4) the lack of understanding of the broader ecosystem related to answering legal questions using NLP in the German context including legal compliance both for LLM's training data as well as for using LLMs in such a setup, and the overall regulatory framework for legal LLMs.TruQALIA (Trustworthy Question Answering to Enable Legal Information Access for Laypeople) aims at conducting foundational research in the area of legal NLP and question answering, whichwill pave the way towards barrier-free access to legal information for laypeople. We envision new NLP approaches to answering laypeople’s legal questions in language they would understand and trust, which we frame as trustworthy legal question answering.We formulate the following three overarching research questions:RQ1 How to address question-answering with domain and language mismatch between questions and expert legal knowledge? (NLP)RQ2 How to provide trustworthy answers and evaluate automatic systems holistically? (NLP)RQ3 Which regulatory frameworks, data compliance, and model compliance requirements are fundamental to legal question answering in the German context? (Law)From the legal perspective, we will investigate the legal requirements and implications of operating a legal LLM, with a specific focus on data protection, regulatory compliance, and professional law requirements. From the NLP perspective, our main research objectives are advances in NLP methods for efficient legal model adaptation, legal dense retrieval, legal answer generation, and legal answering evaluation.
DFG Programme
Research Grants
