Multi‐level discourse analysis: A feasible approach

Abstract

Background: The analysis of discourse has now become commonplace but the focus continues to be on discrete aspects or levels of discourse processing. Although this has provided the necessary groundwork, investigating the relationships and interconnections between these levels continues to be stressed. Recently, some studies have formulated multi‐level discourse‐processing theories and models that explain these inter‐relationships and that identify the sub‐processes involved in producing discourse. This study has used one such model to analyse different levels of discourse and investigate the interconnections between them. A preliminary version of this article was presented at the 26th World Congress of the International Association of Logopedics and Phoniatrics Conference, Brisbane, Australia, August 2004. Aims: To assess the applicability and utility of using a multi‐level discourse‐processing model to examine the interaction between levels of discourse produced by individuals without brain damage. Methods & Procedures: A total of 14 narrative and procedural discourse samples were elicited from 32 non‐brain‐damaged males of different ages and socioeconomic status groups, yielding a total of 394 samples. These samples were analysed in terms of seven broad features (comprising 23 measures), relating to the levels of the multi‐level discourse model. The correlations between these measures were determined using the Spearman's rho correlation test. Outcomes & Results: From the clustering of correlations, a number of fruitful relationships were revealed. Greater relevance was related to more appropriate discourse grammar as well as greater cohesion and syntactic complexity. Longer samples were correlated with an increased proportion of cohesive ties, cohesive errors, and syntactic complexity. An increase in non‐specific elements was related to reduced syntactic complexity and cohesion. A higher occurrence of left‐branching clauses was associated with increased dysfluency. These correlations are explained in terms of the multi‐level discourse model. Conclusions: Three conclusions can be drawn. First, using a multi‐level discourse‐processing model can offer a more realistic perspective of discourse than the analysis of individual aspects. The differential diagnosis of relatively similar discourse impairments (e.g., following head injury, dementia, right brain damage) may ultimately rely on a comparison of the relative deficits at different levels. Second, certain discourse features that can be assessed more objectively (e.g., the number and type of conjunctions) can signal a breakdown at a more conceptual discourse level (e.g., the linking of propositions to each other at a semantic level). Third, these correlations can provide explanatory information regarding more subjective concepts that are difficult to define and measure (e.g., the perception of “relevance” relates to more structured discourse at the macro and micro level). Although this approach to discourse is challenging, it can provide a starting point for more productive investigations of discourse.