I had the same reaction, but you have to keep reading. He's building a stack. He's not saying that thoughts and mental images don't actually exist, it's that they don't exist before the game of language is played. And the game of language is turning sensations of physical objects into auditory representations of them. There's no reason to start constructing the abstract objects yet, we're not at that layer of the 'stack'.
It's kind of like saying that when you dive down to the network layer, video doesn't exist. It's up to the application running the network to assemble all the packets to turn it into video.
Thoughts are constructed from the stream of physical experiences that get formed into words, not the other way around. It's perfectly possible for the brain to still operate without thoughts the way we experience them, in fact, deaf people think visually rather than auditorially like most of us do.