Can we define anxiety and other mental disorders by the words we use?

Esperanza (not his real name) is in the psychologist’s office for the first session. This is no ordinary reception area with a low glass table in the center and a few scattered fashion magazines.

It has laptops for patients who, honoring their condition, wait patiently for their turn. In the meantime, they write a short story about their relationships with friends, family, or colleagues.

Moments after she has finished writing hers, Esperanza is called in for a consultation and a psychologist comes to her holding a detailed account of her case.

“How is it that I came here for the first time and have not yet opened my mouth!” Esperanza thinks in surprise.

The report was created using natural language processing techniques that carefully analyzed the linguistic patterns contained in the newly written story and provided a preliminary diagnosis that would serve as a starting point for the psychologist’s work.

language models

The above is an imaginary description of something that seems like science fiction today but may become reality in the not too distant future. This is true? Can artificial intelligence interpret our behavior to that extent?

To answer these questions, we first need to consider the following: Are there language patterns that show a correlation with various mental disorders or behavioral problems?

As a preface, perhaps it is worth recalling a recent news that may have gone unnoticed by many: the discovery of the work of Lope de Vega thanks to artificial intelligence. In the study that led to this conclusion, a machine learning system was trained to recognize the lexical usages of up to 350 playwrights, and it turned out that a play called french laura shows lexical usage that is closely related to the “witty phoenix” style.

Aside from the literary value of the find, the study brings us to the trail of a really interesting concept: there are linguistic patterns that can be associated with specific people, and that can be detected automatically.

Symptoms and language

In this regard, the experienced reader may ask new questions: are there patterns that can be associated with personality traits? And patterns that can be associated with generalized anxiety disorders? Is it possible to detect anxiety using some kind of language model?

There is now evidence of a statistically significant association between anxiety-related symptoms and language characteristics.

A striking example is the predominance of first-person pronouns and negative words in various mental or psychological pathologies. The authors of the study we linked start with less than 500-word English texts pulled from online mental health forums and manage to find significant differences in the use of such pronouns.

Automatic classification allows you to train a neural network on a set of correctly labeled samples to recognize patterns due to which the text receives one or another label.

This pattern classification method based on deep learning method (an architectural model known as transformersthe same architecture as the already known ChatGPT) has a very high predictive ability.

On the other hand, the lack of explainability of this technique is also high. Given the forecast, the system does not offer any information about why it made such a decision. It goes without saying how important the explanation that must accompany a mental health diagnosis is.

Types of words and emotions

On the other hand, if instead of pattern classification we train in feature extraction, this has lower predictive value but better explainability.

Given the text, one can quantify such elements as the complexity of the sentences formulated or the words used, the frequency of use of certain types of words (pronouns, adverbs, adjectives), the style of narration (passive or active voice) or even one can analyze the primary emotion prevailing in the analyzed text, or the semantic field to which it belongs.

Use in research and discovery

There are many challenges to be solved in this area. The first of these involves more accurate detection of various violations. In other words, right now it is possible to determine whether a patient has a mental health disorder, but it is not currently possible to distinguish which one.

We do not yet know if such accurate detection is possible or not. In any case, the investigation will solve another of the unsolved problems: the collection of complete and reliable data arrays.

Most of the works that exist today use texts extracted from various Internet sources, be it social networks, specialized forums or more specific services. It is not always clear who is the author of each text, and therefore it is difficult (rather impossible) to know the mental reality of the indicated person.

Without a reliable source of data (and social networks do not), the validity of data and results can always be called into question. Therefore, we must work on reliable methods of data collection that meet the research needs of each case.

Explainability problem

Although there are more or less useful approximations, modern automatic classification methods do not provide a list of reasons why each case is labeled one way or another. Without a good set of arguments, it’s hard for any doctor to feel comfortable with such a delicate diagnosis.

Therefore, it is crucial to solve this problem of explainability by providing artificial intelligence tools that can explain the decisions made.

It is quite possible that by combining classification and feature extraction techniques we can solve these problems, and who knows, perhaps the imaginary history of the waiting room will become a reality in the coming years.

Luis de la Fuente Valentin, professor with a master’s degree in analysis and visualization of massive data, UNIR – La Rioja International University and Joaquin Manuel González Cabrera, teacher and researcher. University professor (level 1). Department of School, Family and Society. Faculty of Education. Chief Investigator of the Cyberpsychology Group (UNIR), UNIR – La Rioja International University

This article was originally published on The Conversation. Read the original.