Scientists from HSE University-St Petersburg Present Their Research at the Largest Conference on Computational Linguistics in Morocco

At the international conference on computational linguistics (EACL), the team from the Linguistic Convergence Laboratory at HSE University-St Petersburg presented a unique dataset for neurorehabilitation of patients with the speech disorder—aphasia.

Photo courtesy of Anastasia Kolmogorova

EACL is the largest international conference for experts in computational linguistics. In 2026, it was held in Morocco and brought together approximately 2,600 specialists. The researchers' discussion focused on large language models and the development of AI agents.

The researchers from the Linguistic Convergence Laboratory, supervised by Anastasia Kolmogorova, presented the collected and validated dataset of audio recordings of patients with aphasia. The illness is characterised by damage to speech centres in the cerebral cortex, which entails full or partial loss of speech ability.

Anastasia Kolmogorova

In our work, we used not large language models but embeddings—vector representations of sounding speech features. For the automatic classification of the recordings by the types of aphasias, our colleague Anastasia Margolina offered to focus both on the analysis of acoustic characteristics and lexico-grammatical speech patterns. The approach turned out to be effective: we tested it on our unique dataset. It includes the recordings of informants with different types of aphasias, of varying severity, in different speech situations—monologues, dialogues, when reading or retelling. As a result, we managed to classify the patients according to the types of aphasias and level of severity pretty well. Many colleagues appreciated our practice-oriented approach. We're planning to test the research results at clinical sites. It will help to take some of the load off logopedists: they can hold sessions with patients in semi-automatic mode and track the clinical dynamics more accurately.

Photo courtesy of Anastasia Kolmogorova

The research results of the Linguistic Convergence Laboratory are published in the largest research library in the sphere of computational linguistics. In the future, the dataset will help to train artificial intelligence and create multilingual models for more accurate diagnosis of diseases.

Anastasia Kolmogorova highlighted that the conference demonstrated the world trend to use large language models in medicine and cognitive sciences. The experts emphasise that in the near future, the sets of different AI agents, supplementing each other, will analyse patients' clinical pictures.

'We work with artificial intelligence a lot to complete various tasks: for medicine, for detecting emotions, for cognitive research. The conference assured us that such an approach was correct. We drew out a lot for generating texts, including the ones used for the rehabilitation of patients with aphasia. For the next conference, we are planning to prepare research in this sphere', summed up Anastasia Kolmogorova.