Martin (1992: 263-4):
Summarising these observations as discourse maxims one might come up with the following:
i. Sort out what you've said (worry about what's next as it comes).ii. Keep surfacing (don't get too embedded or too deeply dependent).
Whatever the moral drawn or how framed, there is abundant evidence that texts are produced progressively by speakers who constantly monitor what they say as they say it and for the most part make retrospective connections between new messages and what has already been said. The deep regressive structures proposed by Rhetorical Structure Theory are thus inappropriate as a model of text production and need to be complemented by a more process oriented account. The flatter, covariate structures proposed by English Text on the other hand seem more promising as a first step in the development of dynamic representations.
This is the final section of a critique of Rhetorical Structure Theory.
 This stage of the argument provides some advice on how to organise spoken texts, in terms of a moral to be drawn from observing two texts.
 This stage of the argument presents a commonsense observation as an investigative finding.
 Retrospective connections are deemed appropriate, but regressive structures inappropriate, for a model of text production, on the basis of the discourse maxims and commonsense observation.
 This confuses the semantic relations in a text with the unfolding of the text (logogenesis).
 The relative "flatness" of the English Text model reflects the fact that it is less able to account for the depth of dependencies in a text.
As argued previously, covariate "structures" are not structures. They are units (messages) conjoined by logical relations, but the conjoined units do not form an integrated whole. (A reticulum is a representational device, not a linguistic structure.)