A Multilingual Benchmark for Probing Negation-Awareness with Minimal Pairs

“It’s our fault!”: Insights Into Users’ Understanding and Interaction With an Explanatory Collaborative Dialog System Katharina Weitz* University of Augsburg katharina.weitz@uni-a.de Lindsey Vanderlyn* University of Stuttgart vanderly@ims.uni-stuttgart.de Ngoc Thang Vu University of Stuttgart thangvu.ims.uni-stuttgart.de Elisabeth André University of Augsburg elisabeth.andre@uni-a.de Abstract Human-AI collaboration, a long standing goal in AI, refers to a partnership where a hu- man and artificial intelligence work together towards a shared goal. Collaborative dialog allows human-AI teams to communicate and leverage strengths from both partners. To design collaborative dialog systems, it is im- portant to understand what mental models users form about their AI-dialog partners, how- ever, how users perceive these systems is not fully understood. In this study, we designed a novel, collaborative, communication-based puzzle game and explanatory dialog system. We created a public corpus from 117 conver- sations and post-surveys and used this to ana- lyze what mental models users formed. Key takeaways include: Even when users were not engaged in the game, they perceived the AI- dialog partner as intelligent and likeable, im- plying they saw it as a partner separate from the game. This was further supported by users often overestimating the system’s abilities and projecting human-like attributes which led to miscommunications. We conclude that creat- ing shared mental models between users and AI systems is important to achieving success- ful dialogs. We propose that our insights on mental models and miscommunication, the game, and our corpus provide useful tools for designing collaborative dialog systems. 1 Introduction Task-oriented dialog systems are designed to help users accomplish a specific goal, such as getting a recommendation or booking an appointment. How- ever rather than interacting as equals, the user main- tains control in the dialog, all decisions passing through them. While appropriate for tasks, such as recommender systems, this form of communi- cation leaves the labor of decisions solely on the human and discounts advantages in an Artificial *Both authors contributed equally to this research. Intelligence’s (AI) reasoning ability (Grosz, 1996) which can limit the potential of human-AI part- nerships (Sklar and Azhar, 2015). Collaborative dialog offers an alternative (Nothdurft et al., 2017; Narayan-Chen et al., 2019) where both parties con- tribute to the final strategy. It has been a long standing goal in AI to con- struct computer systems which can effectively act as partners in collaborative activity (Grosz et al., 1999). As an example, if a human and robot are working together, collaborative dialog could allow the robot to communicate the scope of its abili- ties, enabling both partners to work together to de- velop an effective strategy (Nikolaidis et al., 2018). In more complex scenarios, collaborative dialog could be useful for applications, such as search and rescue (Burke et al., 2004) or medical diagnosis recommendations (Lundberg et al., 2018). One important aspect of human-AI collaboration is how well users can form sound mental models of their partner. Put simply, a mental model is a person’s cognitive representation for how/why a complex system (e.g., a dialog system) works (Johnson-Laird, 1980; Norman, 2014; …

Content truncated to 500 words