Back to search

FRIHUMSAM-Fri prosj.st. hum og sam

Morphosyntactic Production in Stroke-induced Agrammatic Aphasia: A Cross-linguistic Machine Learning Approach

Alternative title: Morfosyntaktisk produksjon i agrammatisk afasi som følge av hjerneslag: En tverrspråklig maskinlæringstilnærming.

Awarded: NOK 10.8 mill.

The aim of the Machine Learning Aphasia project is to gain better insight into aphasia, a condition characterized by language and communication difficulties resulting from brain damage. Moreover, Verb-related morphosyntactic impairment is one of the main features of agrammatic aphasia. Although there have been many studies on morphosyntactic deterioration in agrammatic aphasia, we currently know little about which factors govern the preservation or impairment of a given verb-related morphosyntactic category in a person with agrammatic aphasia. With inspiration from the development of methods within machine learning, this project will adopt an original and innovative approach to this phenomenon by using machine learning techniques. The Machine Learning Aphasia project will focus on aphasia resulting from stroke and will investigate grammatical (morphosyntactic) aspects of sentence production in linking verbs, including subject-verb context (e.g., 'Every morning John goes to work'), inflection/tense (e.g., 'Yesterday John went to the cinema'), and sentence negations (e.g., 'George doesn't like chocolate'). During the project’s data collection, Norwegian, Italian, Greek, English and Russian-speaking people will be tested in this project to increase understanding of agrammatic aphasia as a result of stroke. The study will lead to new knowledge, which in turn can be used for treatment programs for people with agrammatic aphasia. The project started at the Center for Multilingualism at the University of Oslo on 1 September 2019 and will last for four years. During the first year, project manager Valantis Fyndanis and his collaborators have developed language experiments and test batteries for the languages that are part of the project (e.g. Norwegian, Greek, Russian, Italian and English). Despite the challenges posed by the Covid-19 pandemic, Dr. Fyndanis and collaborators conducted tests between 2020 and 2022. They tested Greek-speaking PWAA, Russian-speaking PWAA, as well as neurologically healthy individuals from Greece, Norway, and Russia. Testing in countries such as the USA has been almost impossible during this period, due to the pandemic. During the second year, PI Valantis Fyndanis has further established a new collaboration with the Cyprus University of Technology and a Greek Cypriot linguist, Dr. Natalia Pavlou, who has adapted the experimental battery to Cypriot Greek. Data collection in Cyprus started in 2021. At the project's partner San Camillo Hospital (Lido, Venice, Italy); a pilot study has been completed; participant recruitment and data collection also have started here. Postdoctoral fellow Dr. Qingyuan Gardner was hired in autumn 2021 and she is based in Oslo (MultiLing, University of Oslo). She has completed the development of the English test battery in 2022 and visited Boston University in the USA on an exchange stay from January 2023. Together with Boston University there have been collected data from English speaking persons with aphasia and healthy control persons. The data collection in USA, Norway, Cyprus, Greece and Italy will continue in 2023 and 2024.The final stages of the data collection is scheduled to finish by the first half of 2024. These data will be used to address questions relevant to the Machine Learning Aphasia project, such as the effect of demographic variables and cognitive capacities on production of time reference/tense and grammatical aspect. The Machine Learning Aphasia project is scheduled to conclude on December 31, 2024, and is currently on track. The dissemination activities for the Machine Learning Aphasia project for 2023 included two research articles submitted in peer-reviewed journals and several oral and poster conference presentations.

The aim of the proposed project is to improve state-of-the-art knowledge on aphasia, focusing on verb-related (morpho)syntactic production. This deficit is considered to be the hallmark of agrammatic aphasia, which usually occurs following damage to Broca’s area and neighbouring areas in the left hemisphere. Although many studies on (morpho)syntactic production in agrammatic aphasia have been conducted thus far, little is known on the factors that determine the relative preservation or impairment of a given verb-related morphosyntactic category (e.g., Tense, subject-verb Agreement, grammatical Mood) in a given person with agrammatic aphasia (PWAA) in a given language. Inspired by the developments in machine learning, the proposed project aims at filling this gap by taking an original and innovative methodological approach. Machine Learning Aphasia addresses two important, yet unanswered questions: (1) Which factors determine the performance accuracy of a given PWAA, native speaker of a given language, on verb-related morphosyntactic production? (2) What is the hierarchy of factors/predictors of successful verb-related morphosyntactic production in agrammatic aphasia? Addressing questions 1 and 2 will advance our understanding of the complexities underlying morphosyntactic production in agrammatic aphasia, which will be a significant contribution to cognitive science (in particular to psycholinguistics, neurolinguistics, theoretical linguistics, and cognitive neuropsychology). Importantly, achieving goals 1 and 2 will also have significant clinical implications, as the findings about the best predictors of morphosyntactic production in agrammatic aphasia will inform and improve treatment programmes for PWAA. For instance, if the proposed study finds that verbal working memory capacity is one of the best predictors of performance on (morpho)syntactic production, treatment programmes should also include cognitive training targeting verbal working memory.

Publications from Cristin

No publications found

No publications found

Funding scheme:

FRIHUMSAM-Fri prosj.st. hum og sam