The Automatic Extraction of Linguistic Biomarkers as a Viable Solution for the Early Diagnosis of Mental Disorders

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 5234–5242 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 5234 The Automatic Extraction of Linguistic Biomarkers as a Viable Solution for the Early Diagnosis of Mental Disorders Gloria Gagliardi, Fabio Tamburini FICLIT-University of Bologna, Italy {gloria.gagliardi, fabio.tamburini}@unibo.it Abstract Digital Linguistic Biomarkers extracted from spontaneous language productions proved to be very useful for the early detection of various mental disorders. This paper presents a computational pipeline for the automatic processing of oral and written texts: the tool enables the computation of a rich set of linguistic features at the acoustic, rhythmic, lexical, and morphosyntactic levels. Several applications of the instrument - for the detection of Mild Cognitive Impairments, Anorexia Nervosa, and Developmental Language Disorders - are also briefly discussed. Keywords: Mental Disorders, Automatic Speech and Language Analysis, Digital Linguistic Biomarkers 1. Digital Linguistic Biomarkers: a novel NLP tool for diagnostic purposes Natural Language Processing methods and tools are becoming increasingly crucial in the medical domain (Wang et al., 2020), with a broad range of applications ranging from direct patient care, to diagnostics, clin- ical coding, and patient-facing services (Locke et al., 2021). In particular, a growing interest surrounded the possibility to exploit the automatic analysis of speech and language as a sensible early clue of pathological processes. Language is a complex cognitive function, which re- lies on broadly distributed brain networks (Catani et al., 2012). Its neurobiological basis is not limited to the left-dominant perisylvian language network (i.e., the neural loop located around the lateral sulcus, composed by the arcuate tract connecting Broca’s, Wernicke’s and Geschwind’s territories, cf. Catani et al. (2005)), but includes a complex of cortical and subcortical brain structures related to motor and sensory-related repre- sentations, non-verbal memory skills, emotional pro- cessing, and executive functions (Hagoort, 2017; Her- trich et al., 2020). Consequently, even minor brain changes due to mental health issues (e.g., reversible or progressive cerebral atrophy) can result in subtle lan- guage alterations. The term “Digital Linguistic Biomarkers” (henceforth referred to as “DLBs”) indicates the possibility to au- tomatically extract objective indications of the medical state of the patients directly from their verbal produc- tions. These highly accurate and reproducible measure- ments should allow “for a low-cost pathology detec- tion, classification, and monitoring” (Gagliardi et al., 2021). In the last ten years, this approach has gained popu- larity among researchers and clinicians seeking fast, replicable, and objective proxy-measures of mental dis- orders (Gagliardi, under revision). Beside Dementia, which has represented the earliest area of application, this approach has been applied to several clinical devel- opmental and acquired pathologies (e.g., Autism Spec- trum Disorder, Parkinson’s Disease, and Progressive Supranuclear Palsy) with encouraging results. In the following pages, we will present our NLP pipeline for DLBs extraction from Italian oral and writ- ten texts. The paper is structured as follows: §2 presents the overall structure of the pipeline (i.e., pre- processing steps and DLBs computation), §3 illustrates successful case studies …

Content truncated to 500 words