Language production is known to operate on different levels of representation. ‘Flow’ in writing results from parallelism in the coordination of subsequent planning units. In this article we discuss three points arising from Roeser et al. (2025b): (1) parallel processing results in non-additive effects; (2) study of the production of multisentence texts permits testing of questions around how language production is co-ordinated in real time, and, more generally; and (3) statistical models must closely align with what we know about the cognitive process of what is being studied.
Roeser et al. (Mon,) studied this question.