I fail to see what you're saying. Are you saying that because a simple neural network could behave the same, such behaviour can't thus be evidence of inner speech?

A simple neural network can accomplish symbol association by merely categorizing things, without any deep knowledge or understanding of them.

Neural networks like this are trained with pavlovian methods, and are not self aware or conscious. So such behavior is not at all evidence of any kind of self awareness or consciousness.