Annotating Reflections for Health Behavior Change Therapy

Annotating Reflections for Health Behavior Change Therapy Nishitha Guntakandla and Rodney Nielsen Human Intelligence and Language Technologies University of North Texas Denton, Texas 76201 nishithaguntakandla@my.unt.edu and Rodney.Nielsen@unt.edu Abstract We present our work on annotating reflections, an essential counselor behavioral code in motivational interviewing for psychotherapy on conversations that are a combination of casual and therapeutic dialogue. We annotated all the therapists’ utterances from ten transcripts spanning more than five hours of conversation with Complex Reflection, Simple Reflection, or No Reflection. We also provide insights into corpus quality and code distributions. The corpus that we constructed and annotated in this effort is a vital resource for automated health behavior change therapy via a dialogue system. As the on-going work, additional conversations are being annotated at least by one annotator. Keywords: Reflections, Motivational Interview, Behavioral codes, Companion robot conversations, Virtual health agents, Social- therapeutic agents 1. Introduction Recent research has focused on developing companionable robots (companion bots) (Sarma et al., 2014) and relational agents (Vardoulakis et al., 2012) who build trust, rapport, and therapeutic alliance over time by engaging in both casual, and therapeutic dialogue. The goal of such social- therapeutic agents1 is to play a key role in health care at home. They do so by incorporating therapeutic dialogue into friendly conversations with elderly individuals suf- fering from depression (Nielsen et al., 2010; Ring et al., 2016; Bickmore et al., 2005). Older adults those who lack social support can have negative effects on the health and well-being (Vardoulakis et al., 2012). Furthermore, it is not possible to have a therapist available to every individual at each instance of their life to provide assistance. Situations like these enhance the need for social-therapeutic agents. As psychology studies (Keeley et al., 2014; Arkowitz et al., 2015) show, depression can be treated by Motivational Interviewing (MI) (Miller and Rollnick, 2012), a goal- oriented and client-centered style of psychotherapy. In MI, the therapist is directive in aiding the client to elicit the change talk, resolve ambivalence, and evoke motivation to make positive behavioral changes. The proficiency of the counselor in MI delivery is measured using behavioral coding systems like the MI Skills Code (MISC)(Miller et al., 2003) or the Motivational Interviewing Treatment Integrity (MITI)(Moyers et al., 2014). These systems are designed to manually annotate the counselor and the client utterances in a session with behavioral codes, which is an extremely labor intensive and costly task. Existing work (Can et al., 2012; P´erez-Rosas et al., 2016; Tanana et al., 2015) annotated behavioral codes exclusively on MI sessions, which predominantly focus 1social-therapeutic agents: This term is used to indicate any computer-based agent or bot that provide both social support and therapy. on therapeutic dialogue. MI sessions are conducted to help clients change behaviors that they identify or agree to being very important to improve their health. Unlike MI sessions, social-therapeutic agent conversations are a combination of casual (non-therapeutic) and therapeutic dialogue. They commence the conversation with rapport building and identify a behavior that should be discussed. Hence, there …

Content truncated to 500 words