From Stigma to Support: A Parallel Monolingual Corpus and NLP Approach for Neutralizing Mental Illness Bias

Proceedings of Recent Advances in Natural Language Processing, pages 249–254 Varna, Sep 4–6, 2023 https://doi.org/10.26615/978-954-452-092-2_028 249 Abstract Negative attitudes and perceptions towards mental illness continue to be pervasive in our society. One of the factors contributing to and reinforcing this stigma is the usage of language that is biased against mental illness. Identifying biased language and replacing it with person-first, neutralized language is a first step towards eliminating harmful stereotypes and creating a supportive and inclusive environment for those living with mental illness. This paper presents a novel Natural Language Processing (NLP) system that aims to automatically identify biased text related to mental illness and suggest neutral language replacements without altering the original text’s meaning. Building on previous work in the field, this paper presents the Mental Illness Neutrality Corpus (MINC) comprising over 5500 mental illness- biased text and neutralized sentence pairs (in English), which is used to fine-tune a CONCURRENT model system developed by Pryzant et al. (2020). After evaluation, the model demonstrates high proficiency in neutralizing mental illness bias with an accuracy of 98.7%. This work contributes a valuable resource for reducing mental illness bias in text and has the potential for further research in tackling more complex nuances and multilingual biases. 1 Introduction Globally, 970 million people are currently living with mental illness. Each year, 14.3% of deaths, or approximately 8 million people, are caused by mental disorders. Depression affects over 300 million people worldwide, affecting people of all demographic and socioeconomic backgrounds. Anxiety disorders are almost as common, affecting 284 million people worldwide (Children’s Hopechest, 2022). During the first year of the COVID-19 pandemic, incidences of depression and anxiety increased by an astounding 25% (WHO, 2022). In addition to depression and anxiety disorders, suicidal ideation, bipolar disorder, autism spectrum disorder (ASD), substance use disorder, and eating disorders such as bulimia and anorexia nervosa fall under the umbrella of mental illness (Zhang et al., 2022). Mental illnesses have traditionally been one of the most stigmatized health conditions. Stigma and derogatory language about mental illness can be found everywhere, from casual conversations to media, even among medical professionals. Language can be used to reinforce stigma or fight it. Guidelines for responsible reporting by the media emphasize how language used in media can either encourage help-seeking behavior or inadvertently lead to suicide contagion (Reporting on Suicide, 2023). As an example, the phrase “committed suicide” is commonly used but has been replaced by the preferred “died by suicide” to avoid an association with “committing (a murder)” and demonstrate compassion through word choice (Mental Health Coalition, 2021). In casual conversation, mental health diagnostic terms are frequently used to describe a non- medical event, such as the weather as “bipolar” or a situation as “insane”. Although seemingly harmless, misusing diagnostic terms can lead to misunderstandings about mental illness by undermining the severity of mental illness and reinforcing negative stereotypes (Volkow, 2021). Biased language affects how those with mental illnesses view themselves, how others treat them, and whether they seek help (Rose et …

Content truncated to 500 words