<p>We describe a novel approach to machine reading of the primary scientific literature. We treat a description of an experiment as a discourse, viewing a scientific corpus not merely into a collection of documents, but also an extended conversation formed by the collective set of experiments, their introductions and interpretations. This paper introduces this approach as a methodology called ‘Cycles of Scientific Investigation in Discourse’ (CoSID). In CoSID, we capture the central conceptual structure of a paper as a series of nested reasoning loops, composed of passages in results sections, which describe individual research findings. We ground our work with a number of worked examples based on data from the MINTACT and Pathway Logic databases, and illustrate the idea in the context of machine-enable biocuration.</p>