Competing complexity metrics and adults' production of complex sentences

Abstract

The adequacy of 11 metrics for measuring linguistic complexity was evaluated by applying each metric to language samples obtained from 30 different adult speakers, aged 60-90 years. The analysis then determined how well each metric indexed age-group differences in complexity. In addition, individual differences in the complexity of adults' language were examined as a function of these complexity metrics using structural equation modeling techniques. In a follow-up study, judges listened to sentences in noise, rated their comprehensibility, and attempted to recall each sentence verbatim. Hierarchical multiple regression was used to evaluate the structural equation model, derived from the language samples, with respect to sentence comprehensibility and recall. While most of the metrics provided an adequate account of age-group and individual differences in complexity, the amount of embedding and the type of embedding proved to predict how easily sentences are understood and how accurately they are recalled.