So far, it has been assumed that referring items may be non-cohesive for two main reasons: their referent can be found within the same sentence, or they refer exophorically to something outside of the text. In the results of Stage 2 it was already pointed out that further specification of non-cohesive uses of referring items might improve the reliability of the analysis. Although the focus of Stage 2 was the representation of cohesive chains, in the course of the analysis it was found useful to label the items that were not cohesive with explanatory notes on why they did not establish a cohesive tie. For this Stage 3 of the research, notes from the previous analyses were collected and based on their similarities, the five main categories emerged:

These five main categories reflect the functions that non-cohesive items have, or they indicate where the presupposed item is to be found if not in the surrounding text outside of the containing sentence. This section discusses the content of these categories according to the order in which they appear in Table 19. In Table 19 the list is not exhaustive, the listed phenomena are ordered neither according to the importance nor frequency of occurrence in the corpus; nevertheless, all the types of items listed occur in the corpus and would have been problematic if only relying on the sample analysis and procedures provided by Halliday and Hasan (1976). In the examples in Table 19, presupposed items are underlined, referring items are in bold. As noun phrases tend to be very complex, particularly in academic writing, it is not surprising to find several layers of embedding. “The complexity of phrases in academic prose and news reportage reflects the higher lexical density in these registers […], which is associated with a greater potential for the formation of complex phrases (usually built around lexical words […])” (Biber et al., 1991, 117). In the analysis of these complex noun phrases, the referring item whose meaning needs to be specified is analyzed as one that is specified by the closest phrase that immediately follows or precedes it (i.e., usually the complement that will be under the same node in a phrase structure tree).