Echo state networks and the sustained dynamics of natural language performance

This seminar took place through Zoom on November 5th, 2020 from 1:00 pm to 2:00 pm.
Presented by: Dr. Rick Dale, Professor, Communication, UCLA



Click Here to Access Seminar Recording


Follow these steps to gain access to the recording:

  1. Click on the link above and login to Zoom with SSO using UCF as your company domain
  2. Enter your NID username and password




Echo state networks and the sustained dynamics of natural language performance
I will summarize an emerging new project for our lab that uses the echo state network (ESN) as a model for language and communication. The primary task we explore is generation - the capacity of these networks to spontaneously generate symbolic behaviors associated with complex human communication. We have recently explored the ESN as an intriguing dynamic system to model layered structure, with "symbols" at several time scales (Dale & Kello, 2018). This is a central characteristic of human communication - a kind of hierarchically nested set of symbol sequences. In an empirical study of the ESN reported here, we use recurrence quantification analysis (Bianchi et al., 2016; Dale et al., 2011) and find that the ESN’s feature spaces exhibit trainable paths in its dynamics. These paths expand and become more entrenched as the hidden layer increases in size. We describe ways in which adjusting reservoir updates may permit more "creative" generation at the localist output, illustrating that the ESN may be handy for the emerging problems of creative trains of thought, chatbots, and more. This is a very new project, and I look forward to interacting with UCF's SMST seminar series group to discuss it!

Dr. Rick Dale is a cognitive scientist of communication. He is currently a Full Professor in the Department of Communication at UCLA. His bachelor’s degree is in Linguistics from the University of Toronto (Scarborough Campus), and his doctorate in Experimental Psychology from Cornell University. In his research, Rick and collaborators use conceptual and quantitative tools from the study of complex dynamic systems to investigate human communicative behavior. Prior research has explored the dynamic patterning of verbal and non-verbal behavior during interaction in a variety of contexts. With collaborators, he has studied bodily dynamics of deception, the non-verbal dynamics of debate, the synchrony of seeing during interaction, and multimodal coordination during conversation. Recently, he has been exploring various computational models to facilitate analysis and simulation of interacting systems. He has also held Associate Editor positions at the journals Cognitive Science, Behavior Research Methods, and currently, Discourse Processes. Rick runs the Communicative Mind Laboratory (Co-Mind Lab) at UCLA. For more information and downloadable tools and articles, visit http://co-mind.org.