Knowledge Enhanced Reflection Generation for Counseling Dialogues

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 3096 - 3107 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Knowledge Enhanced Reflection Generation for Counseling Dialogues Siqi Shen Verónica Pérez-Rosas Charles Welch Soujanya Poria DeCLaRe Rada Mihalcea University of Michigan, USA Conversational AI and Social Analytics (CAISA) Lab, Marburg University, Germany DeCLaReDeCLaRe Lab, Singapore University of Technology and Design, Singapore {shensq, vrncapr, cfwelch, mihalcea}@umich.edu sporia@sutd.edu.sg Abstract In this paper, we study the effect of common- sense and domain knowledge while generat- ing responses in counseling conversations us- ing retrieval and generative methods for knowl- edge integration. We propose a pipeline that collects domain knowledge through web min- ing, and show that retrieval from both domain- specific and commonsense knowledge bases improves the quality of generated responses. We also present a model that incorporates knowledge generated by COMET using soft positional encoding and masked self-attention. We show that both retrieved and COMET- generated knowledge improve the system’s performance as measured by automatic met- rics and by human evaluation. Lastly, we present a comparative study on the types of knowledge encoded by our system, showing that causal and intentional relationships ben- efit the generation task more than other types of commonsense relations. 1 Introduction Mental health care has been of great importance as the ongoing COVID-19 pandemic poses a seri- ous negative impact on people’s mental wellbeing (Paredes et al., 2021). Not only there is a larger unmet need for counseling services, the health care workers are also in tremendous physical and mental strain (Huffman et al., 2021). With this in mind, it is natural to consider how the advancement in natural language processing can be leveraged to help counseling. Across different counseling styles, reflective lis- tening has always been a fundamental procedure underlying effective counseling practices (Katz and McNulty, 1994). Reflective listening asks the coun- selor not only to listen to the client carefully, but also to actively make a guess of what the client means. If carried out the right way, it gives the client a sense of being understood and facilitates further self-exploration. However, people do not always say what they mean, which is especially the case for patients seeking mental support. Re- flection, as the response made based on reflective listening, sometimes needs to decode the client’s meaning not explicitly expressed in words. On the other hand, pressing the client to clarify the miss- ing part may hinder them from expressing their own experience (Miller and Rollnick, 2012). Thus, counseling frequently calls for counselors to make inferences based on their prior knowledge. For ex- ample, when the client says I had a really hard time sticking to my diet this week, a plausible reflection may be You’re wondering whether you’ll be able to lose weight this way, which relates diet with los- ing weight as an inference based on commonsense knowledge. Moreover, making a good reflection may sometime require domain knowledge. For ex- ample, to understand the client in …

Content truncated to 500 words