Then each word thinks alone

Attention, as described, is entirely about communication. Words look at each other, decide what is relevant, and pull in what they need. it collects a little trophy. bank leans toward river. Every word ends up holding a blend of what it gathered from its neighbours.

But gathering is not the same as doing something with what you gathered.

So after the attention step, each floor does a second thing. Every word, now carrying its blend of context, goes off and is processed on its own, with no reference to any other word. Whatever bank collected about rivers gets worked over and passed up to the next floor.

Two steps, on every floor, always in that order. Mix, then think. Gather from your neighbours, then go away and process what you gathered.

An analogy, with its limits. Attention is a meeting where everyone shares what they know. The second step is everyone going back to their desk to work through what they heard. The meeting moves information around. The desk work is where something is done with it. Where the analogy stops is that nobody is deciding anything, and every word gets the same desk.