Abstract
In this paper we consider the crucial problem of providing ‘coherence relation’ analyses for natural texts. Among the several accounts currently being pursued, claims of broad compatibility usually combine simultaneously with a lack of detailed consensus. To improve on this situation, we take a selection of approaches from the current state of the art—ranging from theories rooted in functional linguistics to formal semantic discourse theory—and apply them to the concrete task of text analysis. Contrasting these analysis styles discloses a range of information assumed by the distinct approaches. We organize this information according to principles of linguistic “stratification,” “metafunction,” and “paradigmatic/syntagmatic axiality” in order to provide a bi‐stratal, three‐way classification system from which individual discourse structure relations can be motivated. This provides a more effective decomposition of the space of “discourse structure information” which at the same time synthesizes most previous views. We illustrate some of the benefits of the specification for linguistic analysis and point out problematic areas for future formal and computational specification.