A corpus of texts for restoring speech after stroke created in Russia

Texts for adult patients instead of children's workbooks

Researchers at HSE University — Saint Petersburg, together with colleagues from Krasnoyarsk, have developed a corpus of texts for therapy sessions with people with aphasia — a speech disorder that often occurs after a stroke and can affect speech, comprehension, reading and writing. This was reported by the university's press service to Gazeta.Ru; the study was published in the journal Aphasiology. The problem the work addresses is that rehabilitation sessions usually involve reading and retelling, but children's texts from ready-made workbooks do not suit all adults and can cause rejection.

How the corpus was created

The collection includes 134 text units created with the help of artificial intelligence: neural networks were given detailed instructions and examples of original–simplified pairs. Topics were chosen based on relatives' answers about what patients liked to read before the illness — news, materials on medicine and gardening, historical and literary texts. The method provides three levels of simplification for mild, moderate and severe aphasia: the algorithm can shorten sentences, remove complex constructions and replace rare words with more familiar ones.

Testing and limitations

The final clinical testing involved 12 patients with mild and moderate aphasia. They read adapted materials on interesting topics at the same speed as simpler standard texts; specialists also noted greater willingness to answer questions and fewer refusals to participate in sessions. At the same time, as the authors emphasize, the testing does not yet prove an acceleration of long-term speech recovery, and severe aphasia was not studied in this group. Next, the team plans to expand the sample and create a chatbot for speech therapists that would allow materials to be adapted to a specific patient.