Good question! I'm very surprised no one answered you, especially since this has been majorly documented in recent research. This is my attempt to answer it.. let's say that, before you actually start stuttering that you already choose to substitute the word (that you were about to stutter on). In this example, you haven't stuttered yet, but you still "reacted" to something.. right? Well, think about it, what is that 'something' that we react to prior to stuttered speech production? Obviously the answer is different for each person, for some it's a reaction to an auditory stimuli, for others it's a reaction to sensory, environmental or social triggers, .. heck, for some it's a sensation in the throat that they have associated with the idea that stuttering is likely to occur. For others it's a sensation of loss of control. For others it's anxiety or authoritive stress or whatnot. Even though everyone reacts to 'something else', what they all have in common is, that they 'react', right? Another thing what they have in common is, that they react to something that they (subconsciously) perceive as an error.. there is something they perceive or associate or view as an error - in their internal representation (as explained above). So they all have in common: - reaction - to an perceived error As long as we keep reacting to this 'imagined error', then I think we keep staying stuck in the vicious cycle of the monitoring-response-inhibition cycle. Conclusion is, that this will then likely disrupt incentive learning or reinforcement learning. that's just my own take on it