The U.S. is in the midst of a mental health epidemic: Suicide is the second-leading cause of death among individuals between the ages of 10 and 34, according to the most recent data from the Centers for Disease Control and Prevention.
To fight this, platforms like Crisis Text Line – a free, 24/7 text-message line – connect individuals in distress with volunteer crisis counselors. Language becomes critical in these sensitive, high-stakes conversations. The counselors’ ability to listen, articulate and – most importantly – adapt their language to soothe people in crisis can be a matter of life and death.
Together with Crisis Text Line, Cornell researchers analyzed more than 1 million anonymized texts from nearly 3,500 counselors to better understand how counselor language use develops with experience. Their findings are included in “Finding Your Voice: the Linguistic Development of Mental Health Counselors,” which was presented at the Annual Meeting of the Association for Computational Linguistics, July 28-Aug. 2 in Florence, Italy.
Over time, the researchers found, crisis counselors change their language in systematic ways, drawing on their training and personal empathy to develop unique voices to calm those in distress. These methods and findings will help inform volunteer training and development at Crisis Text Line.
To perform their analysis, the researchers used cutting-edge computational techniques in natural-language processing (NLP), a kind of artificial intelligence that can process large amounts of human language data.
“Exploring this setting has reinforced my view that having conversations can be really hard, and we’ve only scratched the surface of our central question, which is how can we use NLP techniques to help inform the development of conversational skills?” Zhang said.
Since Crisis Text Line launched in 2013, volunteer crisis counselors and users have exchanged more than 105 million messages on the platform, on subjects including relationships, depression, suicide and loneliness. These anonymized, text-based exchanges offer a rich dataset for computational analysis through natural language processing, providing researchers with a bird’s-eye view of lingual trends across counselor conversations, Zhang said.
When interacting with texters on the Crisis Text Line, less-experienced volunteers were more prone to stick to words emphasized during their initial 35-hour training, such as “elaborate,” “prompted,” “normal” and “understandable,” the researchers found. More experienced counselors tended to shift to less standardized, more colloquial words like “hey,” “nice,” “unfair” and “gotcha.”
This shift toward a more personable approach played out throughout the text exchange, too. Experienced crisis counselors were far more likely to mention specific problem-solving approaches like “meditation,” “therapist” and “apps” rather than generic ones like “brainstorming” and “activities.”
Even if you’ve never counseled before,” said a CTL coauthor, “this research shows people are capable of achieving mastery, with training and practice.
Most datasets in mental health only contain information about a problem; for instance, data that shows the climbing suicide rate in the U.S., said Crisis Text Line’s Robert Filbin, a paper co-author. This research offers a solution, helping reveal the words and strategies that can move distressed texters to a healthier place.
“These are living insights,” Filbin said. “As our data corpus grows, and we expand our service internationally, we can continue to use Zhang’s models to see what experienced counseling looks like in different cultures, and over time.”
He said Zhang’s research will help Crisis Text Line – accessible anytime and anywhere in the U.S. by texting 741741 – train and develop its crisis counselors and serve as motivation to would-be volunteers.
“Even if you’ve never counseled before,” he said, “this research shows people are capable of achieving mastery, with training and practice.”
Continuing this research collaboration with Crisis Text Line, Zhang hopes to better understand how these findings can help train and guide counselors by examining how people grow into more complex conversational practices, such as pacing the conversation.
“This is an example of how natural language processing techniques can assist the development of skills in conversation-heavy professions,” said co-author Cristian Danescu-Niculescu-Mizil, assistant professor of information science. “These methods can inform training and assist supervisors in providing more personalized feedback.”
Paper co-authors also include Christine Morrison and Jaclyn Weiser of Crisis Text Line.
This story, written by Louis DiPietro, originally appeared at the Cornell Chronicle on August 21, 2019.