Applying Main Concept Analysis (MCA) to analyze spoken discourse by Cantonese speakers with aphasia and unimpaired individuals

Abstract

Introduction Main Concept Analysis (MCA; Nicholas & Brookshire, 1995) is a content-based analytic approach that focuses on the quantification of presence, accuracy, and completeness of essential information in oral discourse by persons with aphasia (PWAs). A main concept should contain only one main verb and provide an outline of the gist depicted in a picture, or an outline of the essential steps in a procedure. Recently modified by Kong (2009; 2011) and Richardson and Dalton (2016) to allow a more comprehensive and multilevel coding of PWAs’ spoken output, MCA was reported to be able to distinguish PWAs from controls using single and sequenced pictures, in English (Kong et al., 2015; Nicholas & Brookshire, 1993) and Chinese (Kong, 2009; Kong & Yeh, 2015). Aims This investigation aims to (1) conduct MCA analysis on transcripts in Cantonese AphasiaBank (Kong & Law, 2010-14) and establish the main concept lists for tasks of storytelling, procedural description, single and sequential picture description, (2) examine the effects of age, gender, educational level, and genre type on discourse performance in unimpaired speakers, (3) determine how well MCA can differentiate between fluent and non-fluent PWAs, and (4) investigate how factors of fluency status, semantic processing integrity, and naming ability would predict PWAs’ MCA performance. Method and Results Language samples of 150 neurologically-unimpaired native Cantonese speakers and 105 PWAs were extracted from Cantonese AphasiaBank. Relevant concepts (RCs), referring to utterances related to the discourse tasks that contained one main verb, but may or may not provide an outline of the gist depicted in the picture, or an outline of the essential steps in procedural description, produced by all unimpaired participants were tallied. RCs that were produced by at least 25% of the speakers, together with the acceptable vocabulary specific for each task, subsequently formed the target main concepts. Four-way mixed ANOVAs were used to analyze tasks, gender, age, and education effects, and their interaction effects on overall MCA performance (MC-score) and efficiency of producing accurate and complete main concepts (AC/min). Significant main effects of age and education were found for standardized MC-score. Younger age and higher education resulted in better MC-scores. A similar significant age effect was also found in AC/min. Results of one-way ANOVAs indicated the fluent PWA performed significantly better than the non-fluent counterparts on all tasks and in their overall performance as measured by both standardized MC-score and AC/min. Finally, results of stepwise multiple regression analyses indicated that naming (of action/object) was a significant predictor for both MC-score and AC/min; and fluency status significantly predicted AC/min for all tasks but MC-score for picture description tasks only. In short, PWA who had a higher word retrieval integrity would perform better in MCA, and fluent PWAs was more efficient in producing accurate and complete main concepts than non-fluent PWAs. Intra- and inter-rater reliabilities, based on calculation of 10% of randomly selected PWA and control samples, were high. There was 83.2% or better point-by-point agreement and at least 0.92 Pearson correlations of all MC scoring.