IACM 2022 Abstract Book »
Building an interpretable NLP system to encourage conversational receptiveness
Political views can often be divisive, causing disagreements to spiral out of control - especially when sharing your viewpoints online. Previous work shows that when individuals use language demonstrating thoughtful consideration to others - “conversational receptiveness”, they can prevent disagreements from escalating into conflict. We develop and test an interpretable, personalised NLP feedback system that can evaluate written text and suggest how to communicate better in real-time (i.e., using more receptive language). Across two pre-registered studies, we show that static instructions on how to communicate receptively has a significant effect, but only in the short term. Over time, this usage rate decays. However, our dynamic, personalised feedback system generated sustained usage rates of receptive language in the longer term. We contribute to the literature on conflict management by showing how small, dynamic, responsive feedback messages can sustain conversational receptiveness for longer.