Cascading activation in spoken word production drives incomplete neutralization: An internet-based study of Mandarin 3rd tone sandhi

Incomplete neutralization occurs when two underlying contrastive sounds are phonologically neutralized but remain phonetically distinct (e.g., “latter” and “ladder” become homophonous when the intervocalic stops are flapped in American English). Its proper understanding is foundational to phonology and speech production. Using the incomplete neutralization of the Mandarin 3rd tone sandhi as a test case (T3 + T3 → T2 + T3), we confirmed the presence of this incomplete neutralization with generalized additive modeling (GAMM) and growth curve analysis (GCA). Crucially, we found that the two tones (T2 and T3) became more neutralized when speakers were additionally required to perform a concurrent verbal working memory task while speaking; similar patterns were found when pseudowords were tested, although the overall effects were weaker. Since the concurrent verbal working memory task is expected to add processing load and decrease cascading activation in the spoken word production process, our results suggest that cascading activation, which permits upstream distinctions to surface in downstream acoustics, drives incomplete neutralization. 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 the core assumptions of generative phonology is also discussed.