Research Group Neuroinformatics and Rehabilitation
We bring together basic neuroscience, clinical neurorehabilitation and computer science. Our central question is how humans and machines learn – and how digital systems can support rehabilitation after neurological disease.
One focus is the development of digital language and cognitive training systems. Adaptive learning control and motivational mechanisms are intended to support regular, independent practice tailored to the individual.
Research focus
- Learning in humans and machines
- Digital language and cognitive training systems
- Motivational mechanisms
Projects
DailyWords
DailyWords is a free, multilingual and AI-supported tool for learning and relearning everyday words. It addresses how people with aphasia can practise frequently, independently and with personally relevant language material outside conventional therapy sessions.
The platform connects words, images and spoken language. Personally relevant material and low-threshold access are designed to encourage regular practice – including in languages and regions for which few digital resources are currently available.
The US-based National Aphasia Association lists DailyWords among its resources.
Further information and access: DailyWords project website.
RehaLingo
RehaLingo was developed in a project funded by the German Federal Ministry of Education and Research as a playful and adaptive language training system for people with aphasia.
RehaLingo combined speech recognition, automatic difficulty adjustment and game-based motivational mechanisms. The system was designed to enable frequent independent practice and to adapt as closely as possible to each user’s abilities.
Further information: RehaLingo project website.
FishSpotting
FishSpotting is a digital cognitive training system with tasks targeting focused, selective and divided attention. The exercises are embedded in an accessible underwater world and adapt over time to individual performance.
FishSpotting records training activity and performance trajectories in a structured way. The project also examines which forms of feedback, goal setting and social support promote motivation and sustained use. With appropriate consent, the information can be made available to treating professionals or used for scientific analyses.
Work on motivational mechanisms builds on earlier research into the theory, neurobiology and applications of motivation. An overview is provided in the volume edited by Bettina Studer and Stefan Knecht, Motivation: Theory, Neurobiology and Applications (Progress in Brain Research, Volume 229, 2016).
Further information: FishSpotting project website.
Neuroinformatics and artificial intelligence
Artificial intelligence helps us compare learning in humans and machines, generate training materials, analyse language and images, and adapt learning trajectories and task difficulty.
Our work addresses both the capabilities and the limitations of generative models. We therefore evaluate AI-generated content transparently: linguistic plausibility alone does not guarantee therapeutic suitability.
A current focus is the systematic generation and evaluation of image and language materials for aphasia therapy. This work is conducted in cooperation with the Chair of Databases and Information Systems at Heinrich Heine University Düsseldorf, particularly with Prof. Dr. Stefan Conrad and Mihir Mulye.
Under the working title “Artificial Brains”, the research group investigates whether learning AI models can serve as controllable models of language disorders and their rehabilitation. Recent studies show that targeted perturbations of individual components of language models can produce aphasia-like error patterns. Vision-language models also make it possible to investigate such changes in picture naming using tasks with clinical relevance.
The aim is to develop this approach into a synthetic model of disease and therapy.
Three questions are being investigated:
- How reliably can aphasia-like error patterns be produced in different vision-language models and in picture naming?
- Can these artificially induced functional impairments be partially reversed by retraining intact model components?
- Which training methods improve the efficiency, specificity and stability of recovery – for example curriculum learning, error-guided selection of training examples, repetition or parameter-efficient fine-tuning?
These “AI aphasias” are not diseases in the medical sense, but experimental model perturbations. They allow controlled investigation of hypotheses concerning loss of function, plasticity, relearning and possible adverse effects of rehabilitation.
Selected recent conference contributions and publications on our neuroinformatics research
- Mulye M, Conrad S, Knecht S · CVPR Workshop 2026. Does Visual Grounding Produce More Reliable Aphasia Therapy Items Than Caption-Only Generation? Abstract and full text in the CVPR Open Access Repository
- Mulye M, Conrad S, Knecht S · HC@AIxIA/HYDRA 2025, published 2026. Using Multi-modal Generative Models for Creating Speech Therapy Material. Abstract and bibliographic information at Springer
- Mulye M, Conrad S, Knecht S · International Conference on Artificial Intelligence in Healthcare 2025. Exploring Applicability of Text-to-Image Models for Generating Aphasia Rehabilitation Material. Abstract and bibliographic information at Springer
- Mulye M, Conrad S, Knecht S · 35th GI Workshop on Foundations of Databases, 2024. Proof of Clinical Feasibility of Natural Language Learning with Machines. Open-access full text at CEUR Workshop Proceedings
Further publications: Google Scholar Other archives: PubMed · ORCID · ResearchGate
Professional exchange
Interdisciplinary exchange among neurology, neuropsychology, speech and language therapy, computer science and health services research is an integral part of our work. The German Society for Interdisciplinary Neurorehabilitation provides an important professional framework.
Contact
Prof. Dr. med. Stefan Knecht
Institute of Clinical Neuroscience and Medical Psychology
University Hospital Düsseldorf
Moorenstraße 5
40225 Düsseldorf
Germany