Perspective-Taking in Conversational Agents

Vanderbilt University, Nashville TN

Investigators

Abstract

This proposal aims to unlock a deeper understanding of how intelligent minds communicate, delivering potentially transformational insights into how we talk and listen in day-to-day life, including when communicating with artificial intelligence (AI) systems. By studying how people remember and use language in everyday conversations, the proposal explores the distinct cognitive processes of speakers and listeners. Through a combination of empirical data collection and computational modeling, this work provides critical insights into when and why communication fails and how we can change the way we communicate in order to achieve success. These insights can support efforts to define new goals for advancing artificial intelligence systems, and critical insights needed to develop computer dialog systems that mimic human interaction more closely. These novel insights can also be leveraged to improve educational outcomes in the classroom by offering evidence-based insights to improve communication. Multiple students will gain hands-on research experience, inspiring the next generation of scientists. The project investigates cognitive processes governing language use and memory during conversation. The aim of the work is to test predictions of a proposal which argues that the way you use language shapes how you remember it, in turn guiding future language use. Data from thirteen behavioral experiments are analyzed using advanced statistical methods. In addition, computational modeling of key predictions using artificial neural networks make explicit precisely which theoretical assumptions are made and test the implications of these assumptions. This rare combination of the study of unscripted language use with computational models provides insights into real-time language processing which can be leveraged to advance AI dialog systems and offer evidence-based approaches for pedagogy and mediation. This project is jointly funded by the Perception, Action and Cognition program and the Linguistics program. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.