More than a biomarker: could language be a biosocial marker of psychosis?

Abstract

Automated extraction of quantitative linguistic features has the potential to predict objectively the onset and progression of psychosis. These linguistic variables are often considered to be biomarkers, with a large emphasis placed on the pathological aberrations in the biological processes that underwrite the faculty of language in psychosis. This perspective offers a reminder that human language is primarily a social device that is biologically implemented. As such, linguistic aberrations in patients with psychosis reflect both social and biological processes affecting an individual. Failure to consider the sociolinguistic aspects of NLP measures will limit their usefulness as digital tools in clinical settings. In the context of psychosis, considering language as a biosocial marker could lead to less biased and more accessible tools for patient-specific predictions in the clinic.

Speech disturbances in schizophrenia: assessing cross-linguistic generalizability of NLP automated measures of coherence

“…A second possible explanation is socio-demographic heterogeneity . Gender, age and socio-economic status are known to impact speech production and therefore likely semantic coherence, and relatedly to affect the expression of specific symptoms such as thought disorder (e.g., Palaniyappan, 2021). Indeed, recent studies seem to suggest that NLP algorithms, and coherence measures in particular, can be biased by socio-demographic variables such as racial identity.…”

“…A third possible explanation is cross-linguistic (and relatedly cultural) variation. Different languages present different linguistic structures and usage patterns (Evans & Levinson, 2009), and indeed computational measures of different linguistic aspects, including semantic coherence, have been shown to vary across languages (Palaniyappan, 2021;Sumiyoshi et al, 2004Sumiyoshi et al, , 2014Wydell & Butterworth, 1999;Dideriksen, et al, 2020). Nevertheless, previous literature has implicitly assumed the existence of NLP markers of schizophrenia independent of language and cultural groups analyzed.…”

“…They are particularly evident at the discourse levelranging from reduced syntactic complexity to loss of semantic coherence and cohesion -, and are often associated with specific symptoms (e.g., formal thought disorders). Language disorders can seriously impair the patients' social functioning and communicative ability (e.g., Bliksted et al, 2014;Green et al, 2015;Gallagher & Varga;Parola et al, 2018;2021) and are pervasive: they are an early distinctive feature of schizophrenia, preceding the onset of initial psychosis, and occur in individuals at high clinical risk as well as in patients' relatives (Bedi et al, 2015;Corcoran et al, 2018;Rezaii et al, 2019). Therefore, language disordersand disorganized and incoherent speech in particular -could play a critical role for developing digital phenotyping of schizophrenia de Boer et al, 2018;De Boer et al, 2020;Hitczenko et al, 2021).…”

“…Sample sizes are usually small (median size for participants with schizophrenia = 34.5), and given the heterogeneity of schizophrenia, differences in demographic and clinical characteristics between studies can be quite large, and findings be overfit to the specific sample. Moreover, linguistic and/or cultural specificity may seriously affect coherence patterns, thus leading to differences in language impairments between samples collected in different countries, as well as differences in their specific association with symptomatology (Palaniyappan, 2021;Sumiyoshi et al, 2004Sumiyoshi et al, , 2014Wydell & Butterworth, 1999). For example, native speakers of different languages display differences in word retrieval and word processing strategies and in the use of pauses, hesitations and false starts (Sumiyoshi et al, 2004;Ishkhanyan et al, 2020;Palaniyappan, 2020).…”

Modern Views of Machine Learning for Precision Psychiatry

“…This poses a major challenge in developing ML applications for mental health, making the effect of algorithmic bias possibly worse than other fields of medicine. In addition to biological underpinnings, the domains of data (such as language) also represent social underpinnings [366], and so it is important to consider how socioeconomic factors are influencing measurements. Using training and validation sets that are representative across all demographics including race, gender, and age can help address some of the issues with bias while also uncovering new learning about symptom expressions in various groups and allowing variability in modeling approaches for various groups.…”