EU funding helps develop an application for the rehabilitation of aphasia patients and more sustainable generative AI

Grants worth €150,000 awarded by the European Research Council help identify the innovation potential of top-level research.

University of Helsinki researchers Teppo Särkämö and Jörg Tiedemann have received Proof of Concept grants from the European Research Council (ERC). This will allow them to investigate the possibilities of turning the findings of their previous ERC-funded research into practical applications and innovations benefitting society.

A research group headed by Professor Särkämö is developing a music-based remote application for the rehabilitation of people with aphasia. 

Aphasia, a condition caused by strokes or other damage to the brain, impairs speech production. However, people with aphasia are often able to produce words by singing. For the time being, there are no music-based remote rehabilitation tools widely available and used in healthcare. Developing such solutions would help in treating aphasia patients increasingly comprehensively and cost-effectively.

With the Proof of Concept grant awarded by the ERC, Professor Särkämö aims to develop a new music-based rehabilitation application and investigate its usefulness in aphasia rehabilitation. The goal is to make the application publicly available and part of the rehabilitation of cerebrovascular disorders.

The application is based on a study that previously received €1.5 million in funding from the ERC, where Särkämö’s research group advanced understanding of the structural and functional connections between singing, speech and music in the ageing, recovering and degenerating brain. The study investigated how music can support healthy ageing as well as improve wellbeing and functional capacity in aphasia and Alzheimer’s disease.

Read more about the research group.

More sustainable and accessible generative AI

In recent years, we have seen the fast development of large language models i.e. generative AI, that become available in products such as Chat GPT. 

Due to their massive size and lack of transparency, these systems cause a growing carbon footprint and create a dependence on monolithic commercial services. Building such models comes with immense costs and cannot be done from scratch repeatedly for new applications and additional languages.

Professor Jörg Tiedemann’s research group intends to develop modular architectures that create components, which can easily be reused and combined in order to adjust an existing system to new tasks and to scale across the thousands of languages spoken in the world. 

The goal is to produce a repository of light-weight modules that can be plugged into end-user applications such as automatic translation tools without the overhead of a massive language model. Such modular systems have the potential of reducing the growing carbon footprint of generative AI and can grow without unnecessary replication of existing solutions.

The work builds upon the findings of Tiedemann’s earlier project that the ERC funded with €2 million. The project looked at the development of massively multilingual translation models to study the ability of neural networks to pick up the intended meaning from raw text and translations.

Read more about the Language technology research group.

Read more about research conducted at the University of Helsinki funded by the European Research Council.