Summary: A recent study has demonstrated that analyzing a person’s word selection can identify worsening signs of major depressive disorder. Researchers compared written responses to people assessors and ChatGPT, finding that both helped to accurately predict melancholy severity days later.
While traditional speech analysis tools like LIWC failed to capture the emotional resonance through phrase placement and phrase meaning, ChatGPT did a good job of capturing the emotion. This finding might help to create AI-assisted mental health assessments, giving doctors innovative tools for identifying and predicting mental health results.
Important Facts:
- Coming depression symptoms were accurately predicted by ChatGPT and human raters.
- Traditional word-count devices like LIWC were less efficient at prediction.
- AI language study may improve the ability of doctors to determine mental health.
Origin: Yale
A new Yale research shows that a child’s choice of words may affect how much worse their symptoms of major depressive disorder progress.
Researchers demonstrated that created responses to open-ended questions could be used to determine who would have worse depressive symptoms weeks afterwards using both human evaluators and the huge language model ChatGPT.
The findings,  , reported Sept. 16 in the Proceedings of the National Academy of Sciences, recommend automated procedures that you examine speech use may enhance and improve psychological evaluations.
A growing body of research has found a link between sadness and the speech a person uses. For example, people with depression use more offensive language on social internet and in text messages. And word selection is related to how well people respond to treatment.
For this review, Yale researchers wanted to know if speech might even reveal information about a person’s potential symptoms. To better understand this, they asked 467 participants to complete nine open-ended, neutral short-answer questions and the Patient Health Questionnaire ( PHQ-9), which assesses depression severity. Three weeks later, all respondents completed the PHQ-9 quiz again.
The researchers determined how many words in the participants ‘ written responses to the short-answer questions had a positive or negative emotional tone using a tool called Linguistic Inquiry and Word Count ( LIWC), which can determine how many words belong to a particular category.
Scientists discovered that while LIWC results were associated with depression intensity when respondents responded to the issues, they did not, according to their findings, predict melancholy severity three weeks later.
Sentiment ratings given by people raters, on the other hand, did identify upcoming depression symptoms.
This revealed to us that people were picking up on information that only counting emotion-based words had not, according to Robb Rutledge, an associate professor of psychology at Yale’s Faculty of Arts and Sciences and senior author of the study.
LIWC treats each word independently, which may be why it falls short in this specific program, said the experts.
According to lead creator Jihyun Hur, a Ph. D.,” we wanted to appear at word order and the comprehensive feature of language that are crucial to shaping emotional voice.” D. scholar in Rutledge’s test and the facility of coauthor , Jutta Joormann, the Richard Ely Foundation Professor of Psychology.
” That’s when we got engaged in ChatGPT”.
An artificial intelligence program called ChatGPT is designed to emulate human verbal speech. In a way that regular language-related tools like LIWC do not, term buy and the meaning within and between statements are taken into account.
When the researchers instructed ChatGPT versions 3. 5 and 4. 0 to assess the participants ‘ responses ‘ positive and negative tone, the scores predicted future changes in depression severity much like the results of the human assessors.
The analysis, according to experts, serves as a point of entry and provides the foundation for further study. For instance, Rutledge and his team are interested in how this technique may be applied to other medical conditions over a longer period of time.
Anyone can participate in the research by playing the games in the firm’s completely smartphone app, Happiness Quest, which is a part of the agency’s ongoing investigation into the connection between mood and decision-making.
Reynolds said he can see this kind of speech assessment serving as a valuable improvement to the tool for clinicians in the future.
” Our approach could be one way clinicians evaluate their patients,” Rutledge said, “because an analysis of the language people use provides additional information that clinicians currently do n’t have.”
You want a set of tools that can be used by a large number of people, all of whom can give you a preview of an individual. If some of those resources are automated in this way, the doctor has more time to devote to assisting the person.
And finally, a better understanding of symptoms and how to identify them may be helpful.
” The tremendous amount of language data already available in the clinical setting can be used in a new way to better understand mental health,” Hur said of artificial intelligence tools like ChatGPT.
About this research on AI and depression
Author: Bess Connolly
Source: Yale
Contact: Bess Connolly – Yale
Image: The image is credited to Neuroscience News
Original Research: Open access.
Jihyun Hur and colleagues ‘ study” Depressive symptoms change as a function of language sentiment..” PNAS
Abstract
Depressive symptoms change as a function of language sentiment.
There is a pressing need to develop tools that can predict when people will become depressed, and the prevalence of depression is a major societal health issue. Although some research suggests that depression affects how we communicate, it’s not clear whether language affects how we express our feelings.
Here, we test whether the sentiment of brief, succinct linguistic responses can be used to make up for changes in depression.
Across two studies ( N , = 467 ), participants provided responses to neutral open-ended questions, narrating aspects of their lives relevant to depression ( e. g., mood, motivation, sleep).
Additionally, participants completed the Patient Health Questionnaire ( PHQ-9 ) to evaluate depressive symptoms and a risky decision-making task that involved periodic measurements of momentary happiness to assess mood dynamics.
The sentiment of written responses was evaluated by human raters ( N , = 470 ), Large Language Models ( LLMs, ChatGPT 3.5 and 4.0), and the Linguistic Inquiry and Word Count ( LIWC ) tool.
At a three-week follow-up, we discovered that language sentiments that were influenced by human raters and LLMs, but not LIWC, were inversely related to changes in depressive symptoms.
We discovered that while language sentiment was related to current mood, it still predicted symptom changes even after considering the current mood.
In summary, we demonstrate a scalable tool that combines sentiment analysis from AI tools to predict upcoming psychiatric symptoms using brief written responses.
