Objective 1:
Develop a new Slovene language dataset, supporting research in cognitive impairment detection, combining this with existing datasets to form a new multi-lingual collection with associated software tools for feature extraction and/or analysis;
Objective 2:
Develop multi-lingual methods for cognitive impairment detection, capable of being applied to a range
of languages including less-resourced languages such as Slovene, by analysing data in a range of languages (including Slovene);
Objective 3:
Develop cross-lingual models for cognitive impairment detection, capable of performance significantly
better than random on data in a new target language unseen in training.
By using linguistically-informed features, as well as the power of neural network-based large language models combined with recent explanation and visualisation methods, we investigate how our models can be made interpretable so that users can understand the factors that influenced their decisions.