Diagnostic tests of segmental duration models

Abstract

Duration models map combinations of segmental identities and multi‐factorial prosodic contexts onto the temporal domain. Two new diagnostic tests that focus on interactions between contextual factors were applied to vowel durations measured in contexts varying in several prosodic factors and speaking rate. The first test concerns a distinction that cuts across a wide array of models (e.g., additive, multiplicative): Which groups of factors are functionally combined? When factors are functionally combined, constellations of values that yield identical durations for one segment should also yield identical durations for other segments. The data‐showed that speaking rate cannot be combined with other prosodic factors, contradicting any generalized version of the incompressibility model [D. H. Klatt, J. Acoust. Soc. Am. 59, 1208–1221 (1976)] that would count speaking rate among its prosodic factors. The second test concerns modes that would count speaking rate among its prosodic factors. The second test concerns models that express duration as the sum of additive and multiplicative terms, and can diagnose which terms are needed. Results included that speaking rate is combined with segmental identity into a single multiplicative term. These results show the advantages of diagnostic tests over the standard approach of model fitting.