Style
The style of language relates to how we express ourselves, rather than the content we express. Our lab recently conducted a big data text analysis of 64 different online mental health forums, examining over 6,400 members. “Absolutist words” – which convey absolute magnitudes or probabilities, such as “always”, “nothing” or “completely” – were found to be better markers for mental health forums than either pronouns or negative emotion words.
From the outset, we predicted that those with depression will have a more black and white view of the world, and that this would manifest in their style of language. Compared to 19 different control forums (for example, Mumsnet and StudentRoom), the prevalence of absolutist words is approximately 50% greater in anxiety and depression forums, and approximately 80% greater for suicidal ideation forums.
Pronouns produced a similar distributional pattern as absolutist words across the forums, but the effect was smaller. By contrast, negative emotion words were paradoxically less prevalent in suicidal ideation forums than in anxiety and depression forums.
Our research also included recovery forums, where members who feel they have recovered from a depressive episode write positive and encouraging posts about their recovery. Here we found that negative emotion words were used at comparable levels to control forums, while positive emotion words were elevated by approximately 70%. Nevertheless, the prevalence of absolutist words remained significantly greater than that of controls, but slightly lower than in anxiety and depression forums.
Crucially, those who have previously had depressive symptoms are more likely to have them again. Therefore, their greater tendency for absolutist thinking, even when there are currently no symptoms of depression, is a sign that it may play a role in causing depressive episodes. The same effect is seen in use of pronouns, but not for negative emotion words.
Practical implications
Understanding the language of depression can help us understand the way those with symptoms of depression think, but it also has practical implications. Researchers are combining automated text analysis with machine learning (computers that can learn from experience without being programmed) to classify a variety of mental health conditions from natural language text samples such as blog posts.
Such classification is already outperforming that made by trained therapists. Importantly, machine learning classification will only improve as more data is provided and more sophisticated algorithms are developed. This goes beyond looking at the broad patterns of absolutism, negativity and pronouns already discussed. Work has begun on using computers to accurately identify increasingly specific subcategories of mental health problems – such as perfectionism, self-esteem problems and social anxiety.
That said, it is of course possible to use a language associated with depression without actually being depressed. Ultimately, it is how you feel over time that determines whether you are suffering. But as the World Health Organisation estimates that more than 300 million people worldwide are now living with depression, an increase of more than 18% since 2005, having more tools available to spot the condition is certainly important to improve health and prevent tragic suicides.
****
Mohammed Al-Mosaiwi is a PhD Candidate in Psychology, University of Reading