Here’s a tough (and rather philosophical) question: what does it look like to develop or “learn” a behavior—and how could we measure this development? As a Ph.D. student in the Information Science department at Cornell University, I spent the past year collaborating with Crisis Text Line to analyze conversations in order to understand the skill and expertise Crisis Counselors need to navigate helping a texter. In taking conversations, I developed a central research question: How do Crisis Counselors develop or “learn” conversational behaviors as they gain experience?
In this post, I’ll share some of my findings on how the language that Crisis Counselors use can change over time. These findings are the subject of a paper we’ve published at the Association for Computational Linguistics Conference. These findings only scratch the surface of the broader question of how Crisis Counselors can learn to have conversations.
There are many possible ways to characterize language use in a conversation. Additionally, many linguistic qualities might develop with experience. My analysis focused on the diversity of words used by Crisis Counselors over time.
Every texter is different, so rather than learning a magic formula for taking conversations, Crisis Counselors have to adapt to each new conversation and learn from experience. But constantly having to try new things is difficult too. It might not always be clear whether the language a Crisis Counselor uses improves texter outcomes.
For Crisis Counselors, change could also seem risky. After all, Crisis Counselors are talking to a texter in an intense, critical moment of their life. Maybe it feels safer to conservatively stick to the types of language that became familiar during training. In my own conversations as a Crisis Counselor, I kept going back to the textbook for reference. So, in theory, these limitations could stall Crisis Counselors’ linguistic development and make the language they use sound a bit robotic.
Experience might make Crisis Counselors better equipped to deal with these forces in two ways. First, Crisis Counselors might become more adaptable to each individual texter they talk with. This means that the types of words that you choose between one conversation and another will be more and more different, as Crisis Counselors gain experience. This means that there is diversity within the conversations Crisis Counselors take. Second, Crisis Counselors might gradually relax into a middle ground beyond the language of the training material that all Crisis Counselors go through, and their own unique voice. This means that as Crisis Counselors gain experience, the types of words they choose become uniquely their own. So, there’s also diversity between Crisis Counselors when it comes to the language they use.
Drawing on these hypotheses, we looked at two types of linguistic diversity: within-diversity (how much Crisis Counselors mix things up from one conversation to the next) and between-diversity—(how distinctive Crisis Counselor language is from one to the next). So, we developed a method to measure both of these diversity types across a Crisis Counselor’s conversations.
For this study, we compared language used by Crisis Counselors at two extreme stages of their services: Crisis Counselors in their first 20 conversations and Crisis Counselors in their 100th to 120th conversations.
We found that 73% of Crisis Counselors increase the diversity of the words they use when compared to each other (between-diversity) as they progress in their service. We also found that 58% of Crisis Counselors increase the diversity of words they use from texter to texter (within-diversity) as they progress in their service. Essentially, Crisis Counselors grow into their own unique voices the longer they spend on the platform.
My analysis has shown that Crisis Counselors tend to diversify in their language use as they take more conversations. But that’s where my story ends, and your story begins. For one, it would be interesting to hear if these results resonate with you, or if there are parts that don’t seem to fit. After all, we’ve shown that the majority of Crisis Counselors develop in a certain way, but any automated analysis necessarily misses many nuances of each of your individual experiences. Plus, our measures of diversity capture only a narrow slice of the rich ways you use language in conversations. Are there other ways in which you’ve felt yourself develop—or struggle to develop—in volunteering with Crisis Text Line?
It’s also worth noting that we need to do more work to show that linguistic diversity is a good characteristic. Intuitively, it’s better to sound natural and adaptive than robotic and inflexible, but sounding too distinctive, or deviating too far from the training, might also be detrimental. Managing this tradeoff is hard (even beyond crisis counseling conversations!)—and I’d love to hear your thoughts on how you pull this off.
To wrap up, I’d like to pose another guiding question: how could we use analysis like this to help support Crisis Counselors throughout their experience with Crisis Text Line—during training, during their first shift, and 100 conversations in? After all, learning to be a good conversationalist is a difficult and meaningful accomplishment. Beyond understanding how this learning process plays out, it would be valuable to understand what could help along the way.