The description of the bursts of written language is subject to discussion, and is still waiting for a methodological frame (cf. Cislaru & Olive, 2018). The bursts are segmented by pauses whose threshold may be fixed in advance (2 seconds in our case, Chenoweth & Hayes, 2001). This identification method segments data into bursts of varying length that are linguistically heterogeneous (see also Medimorec & Risko, 2017). Linguistic heterogeneity is however natural inasmuch as texts are composed using a great variety of linguistic forms. Thus, the question is not to fix a threshold able to generate more homogenous bursts, but to develop descriptive categories able to grasp their levels of complexity and to offer cognitive linguistic explanations of i) the bursts' unity and of ii) pausing constraints. In order to answer this question, we propose a detailed annotation of the bursts of writing. We first consider the nature and the function/dependency relation of all the elements of a burst. We then describe the frontiers of each burst and identify i) segmentation preferences according to the type of linguistic units and ii) regularities in bursts' interconnectiion both in case of linear concatenation and in case of revision.