The explanatory value of multidimensional measures of linguistic complexity for aphasia severity: evidence from AphasiaBank speech data

Abstract

Aphasia is an acquired language disorder marked by substantial individual variability. Quantitative measures of connected speech may clarify how linguistic performance relates to aphasia severity. Semi-spontaneous speech samples from 22 English-speaking adults with aphasia in the APROCSA subcorpus of AphasiaBank were analyzed. Eight measures represented three domains: Mean Length of Utterance (MLU), clause density, and embedded clause ratio for syntax; type–token ratio, verb diversity, and content-word ratio for lexis; and verb marking and noun marking for morphology. Pearson correlations examined associations with Quick Aphasia Battery Overall scores. Three domain-specific regression models and one comprehensive model containing all eight measures were compared using adjusted R², AIC, and BIC.MLU (r = 0.734), clause density (r = 0.592), embedded clause ratio (r = 0.488), and verb marking (r = 0.753) were significantly positively associated with QAB Overall scores, with higher scores indicating milder aphasia. The lexical measures and noun marking were not significantly associated with QAB Overall. Among the domain-specific models, the morphological model showed the greatest explanatory power (adjusted R² = 0.523), followed by the syntactic model (adjusted R² = 0.494); the lexical model was nonsignificant. The comprehensive model provided the strongest explanatory power and best fit (adjusted R² = 0.805; AIC = 64.82; BIC = 75.73). Syntactic and morphological measures, particularly MLU and verb marking, were more sensitive to variation in aphasia severity than the lexical measures examined. Combining measures across domains provided complementary information, supporting a multidimensional approach to connected-speech assessment in aphasia.