Speech production

Cascading activation drives incomplete neutralization in acoustic signals

Phonological theories often assume the presence of abstract discrete units (e.g., /d/ and /t/ are distinct categories in English). In speech, distinct sounds can undergo neutralization (e.g., /d/ and /t/ are neutralized in American English flapping, so “writer” and “rider” become homophones). Recent decades of research have revealed a multitude of linguistically conditioned systematic phonetic variability, so previously reported cases of neutralization should be more precisely characterized as incomplete (perceptually indistinguishable but acoustically separable). The presence of incomplete neutralization has fueled intense debates in linguistics and psychology, as it challenges the fundamental assumptions about symbolic representations and categorical phonological operations. I hypothesize that domain-general information cascading in the speech production procedure drives incomplete neutralization, so downstream acoustic signals still contain residual information of the original /d/-/t/ distinction. By manipulating task demands, we successfully observed the predicted more neutralized production when speakers produced speech under cognitive load.

Practically, training/intervention should help speakers achieve high automaticity, similar to the application of Mandarin T2-T3 neutralization. Theoretically, our study shows how embracing cascading activation can inform the long-standing debate between discrete vs. exemplar representations/operations surrounding incomplete neutralization. How cascading activation is compatible with core assumptions of generative phonology is also discussed in the manuscript.