A Taxonomy of Social Cues for
When using the taxonomy, please cite as Feine, J., Gnewuch U., Morana S. and Maedche A. (2019): “A Taxonomy of Social Cues for Conversational Agents” International Journal of Human-Computer Studies. To read the paper, please click here.
The appearance of the CA ranges from an extremely photorealistic appearance to a comic appearance.
Real photo of a human being, comic figure.
The degree of anthropomorphism impacts the perceived credibility and likeability of agent (Nowak 2004; Forlizzi et al. 2007), trust resilience, and a higher resistance to breakdowns in trust (Visser et al. 2016). Human visualizations increase the amount of recognized emotions of agent (Beer et al. 2015) and also influences self-awareness which impacts the amount of self-disclosure (Sah, Peng 2015). Speakers used significantly fewer words that referred to physical appearance when speaking to robot looking bot (Brahnam, Angeli 2012) or symbolic looking bot (Knijnenburg et al. 2016). Degree of anthropomorphism of agent should further match the role of agent (Trovato et al. 2015; Keeling et al. 2004). Robot and iconic looking images are more useful to represent AI-based service agent and human images to signal a human service agent (Wuenderlich, Paluch 2017). Furthermore, anthropomorphized agents can create a backfire effect on user engagement (Kim et al. 2016) and not all kind of appearances can be matched to certain voices (Mersiol et al. 2002; Louwerse et al. 2005). Finally, not all studies find differences in social responses between a human or symbolic appearance (Catrambone et al. 2002; Verhagen et al. 2014). Furthermore, users have different preferences for partly or completely embodied agents (McBreen, Jack 2001; Cowell, Stanney al. 2005).