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

Finished · Speech Production

Summary

This study conducted an online experiment with speakers of Mandarin to test the hypothesis that increased task demand while speaking will overload working memory to the extent that incomplete neutralization of T3-T3 to T2-T3 will decrease in the effect size as task demands increase. The results show that, when presented with a task in which they had to memorize a set of 6 Chinese characters, then speak, then determine if their six random characters matched a new set after speaking, participants exhibit a smaller effect size in the incomplete neutralization process compared to simply speaking without holding memorized characters in working memory. The authors attribute these results to cascading activation, in which multiple lexical items are activated in the lexical selection stage of speech production (stage 1: conceptualization, stage 2: lexical selection, stage 3: form preparation, stage 4: articulation). These multiple activations are activated to varying degrees depending on the planned production, and the activation of a related item without tone sandhi taking place can cause incomplete neutralization. Nevertheless, when other tasks demand attention and take up our working memory capacity, the secondarily activated lexical items can be weakened, which results in a smaller effect of incomplete neutralization.

The authors claim that increased task demand is what results in incomplete neutralization, but I think this is not the whole picture. Specifically, this explanation of neutralization makes the strong prediction that we should only observe complete neutralization when task demand is high, but this is not consistent with the facts on the ground. For instance, in Mankiyali, an experiment I ran showed complete neutralization of vowel nasality for a simple task of reading words into an oral-nasal airflow mask. No high demand additional task was conducted concurrently. Moreover, the experiment included base words and inflected forms of the base as well, which has been argued to increase contrastive intent as speakers become aware of the need to distinguish between oral and nasal forms. It could also be argued that the presence of base forms earlier in the experiment might prime stronger base form activation when producing inflected forms. All of these experimental effects lend themselves to a scenario in which incomplete neutralization should be observed if planning were the only explanation. However, this was not the case, as complete neutralization was observed.

As a counterproposal to Zeng et al's (2025) proposal, I am of the opinion that speakers exhibit planned target pronunciations (in regard to neutralization) that can differ across languages. These targets can either be phonetically complete or phonetically incomplete, but the baseline targets must be distinct in some way. In other words, there are language-specific phonetics. You can't chalk it all up to planning or else you would find the same types of neutralization under the same conditions across studies, which is simply not the case. Moreover, on top of these planned targets, planning effects like cascading activation (Zeng et al, 2025), incremental lexical planning (Du & Durvasula, 2025), and/or contrastive intent (Nelson & Heinz, 2025) can manipulate the end result of the planned target outputs. However, they cannot explain everything.

One final note is that the proposal here ostensibly saves exemplar theoretic frameworks in their inability to model both phonetically complete and phonetically incomplete neutralization. Specifically, if planning/task load can affect the degree of neutralization online, you could argue that increased task load weakens the activation of semantically and phonetically similar exemplars, leading to complete neutralization. However, the same reasons that planning doesn't explain the Mankiyali data are true here. The Mankiyali experiment had no special increased task load. Rather, the task was simple and bases were included in the experiment along with neutralizing forms, which - if anything - should lead to primed activation of similar exemplars.