During collaborative writing each author works on a copy of the shared document. These copies are then merged to produce the final document. This asynchronous work is supported by several collaborative writing tools. While these tools are excellent at merging and detecting syntactic conflicts, they are not able to easily recognise semantic inconsistencies. This hinders the coherence of the document because while each individual copy might be well constructed, they may not be after the merge. To address this, we investigate the combination of the Rhetorical Structure Theory with Operational Transformation approach. In this paper, we define a data model, a set of operations to manipulate the RST structures and a set of transformation functions. A validity checker alerts the authors to areas in the text with possible semantic lapses in the merged documents.