Abstract
<p>The study of extended cognition encompasses individuals' ability to come together as a group to solve complex problems -- often with greater processing power than is available to any one person alone. Research in this area has pointed to conversation as an integral component of extended cognitive processes. However, if conversation acts as a crucial support for extended cognition it needs to be able to enact certain functions to do so. The current study deploys a hybrid information-theoretic and language modeling framework to analyze a dataset of 5,447 conversations across nine different languages in order to ascertain the computational rules that allow speakers and their partners to generate new information relative to what they have expressed in prior turns -- a key part of extended cognition. We conclude by providing a set of rules governing how speakers form an extended cognitive system, finely tuned for jointly creating new content from old ideas.</p>