This study was designed to identify whether advanced nonnative speakers of English rely on recurrent sequences to produce fluent speech in conformance with neural network theories and symbolic network theories; participants were 6 advanced, speaking and listening university students, aged 18-37 years (their native countries being Korea, Japan, Taiwan, Hong Kong, and the People's Republic of China). Subjects' introspections on how their words come to mind in relation to speech production are also investigated. Subjects were taped repeatedly while practicing speeches on familiar and newly learned topics; they were interviewed, and they completed questionnaires on background and speech preparation strategies. Results indicated that a mean of 66.4% used recurrent sequences (RSs) that were consistent across topics and tellings as predicted by semantic and neural network theories and that all subjects reported use of partial phrases, phrases, or strings of phrases; subjects who approached the task as if memorizing a script or piano piece, used immediate repetition of RSs and lengthy RSs. Findings confirm that recurrent speeches are an essential element of planning and executing speeches in real time by advanced nonnative English-speakers. (Contains 44 references, 1 figure, and 9 tables.) (NAV)