Multilingual Learning for Mild Cognitive Impairment Screening from a Clinical Speech Task

Proceedings of Recent Advances in Natural Language Processing, pages 830–838 Sep 1–3, 2021. https://doi.org/10.26615/978-954-452-072-4_095 830 Multilingual Learning for Mild Cognitive Impairment Screening from a Clinical Speech Task Hali Lindsay1, Philipp M¨uller1, Insa Kr¨oger1, Johannes Tr¨oger1 2, Nicklas Linz2, Alexandra K¨onig3 4, Radia Zeghari3, Frans RJ Verhey5, Inez HGB Ramakers5 1German Research Center for Artificial Intelligence (DFKI), Saarbr¨ucken, Germany 2ki-elements, Saarbr¨ucken, Germany, 66111 3CoBTeK (Cognition Behaviour Technology) Research Lab, University Cˆote d’Azur, France 4 National Institute for Research in Digital Science and Technology (INRIA), Nice, France 5Maastricht University, Department of Psychiatry & Neuropsychology, School for Mental Health and Neuroscience, Alzheimer Center Limburg, Maastricht, Netherlands, 6200 hali.lindsay@dfki.de, philipp.mueller@dfki.de, insak@coli.uni-saarland.de, johannes.troeger@dfki.de, nicklaslinz@googlemail.com, akonig03@gmail.com, Radia.zeghari@univ-cotedazur.fr, f.verhey@maastrichtuniversity.nl, i.ramakers@maastrichtuniversity.nl Abstract The Semantic Verbal Fluency Task (SVF) is an efficient and minimally invasive speech- based screening tool for Mild Cognitive Im- pairment (MCI). In the SVF, testees have to produce as many words for a given semantic category as possible within 60 seconds. State- of-the-art approaches for automatic evaluation of the SVF employ word embeddings to an- alyze semantic similarities in these word se- quences. While these approaches have proven promising in a variety of test languages, the small amount of data available for any given language limits the performance. In this pa- per, we for the first time investigate multilin- gual learning approaches for MCI classifica- tion from the SVF in order to combat data scarcity. To allow for cross-language general- isation, these approaches either rely on trans- lation to a shared language, or make use of several distinct word embeddings. In evalua- tions on a multilingual corpus of older French, Dutch, and German participants (Controls=66, MCI=66), we show that our multilingual ap- proaches clearly improve over single-language baselines. 1 Introduction Mild Cognitive Impairment (MCI) is a medical con- dition (Petersen et al., 2014) that often precedes Alzheimer’s Disease (AD). The development of cost-effective and scaleable screening approaches for MCI is crucial for the early treatment and man- agement of AD (Dubois et al., 2016). The Semantic Verbal Fluency Task (SVF) is a promising screen- ing approach as it combines a time-efficient testing procedure with the possibility of remote and au- tomatic evaluation (Tr¨oger et al., 2018). In this task, the testee is asked to name as many words as possible from a given semantic category (e.g. ani- mals) in a given time (e.g. 60-seconds). Tradition- ally, the number of named within-category items is used to detect cognitive impairment. However, recent research has shown that in-depth analysis of the underlying cognitive strategies used for the SVF (e.g. semantic memory retrieval, executive control) enables a more fine-grained differential diagnosis (Tr¨oger et al., 2019). To harness the diagnostic power of the SVF, cur- rent automatic evaluation approaches identify se- mantic clusters in the participants’ word sequences, based on semantic word embeddings (Woods et al., 2016; Linz et al., 2017b; Paula et al., 2018). As the word embeddings used in these approaches are language-specific, training diagnostic machine learning approaches for target languages with small available datasets of SVF tests …

Content truncated to 500 words