In my last post, I suggested centering and tightly packing buttons in order to promote muscle memory for repeated navigations or combinations of actions.

This sketch shows a collection of such buttons, which are obviously too packed to label.
So how will the user know what button does what? It seems like I should go back to interface school, no?
Well…no, thank you very much. There is a logical solution to this problem: create a textbox, which I'll call "hotbox" just to be perverse. When the user rolls over the button, you can throw label text into the hotbox:

Before you knuckle-dragging traditionalists protest that I am trying to perfect perfection (the button, the label, a marriage appointed on high!), let's just consider the advantages of this scheme:
1. You can throw more descriptive text into the hotbox than can fit comfortably in a label.
2. Writing and graphics are cleanly separated.
3) It is easier to remember buttons based on location and performed action rather than text.
It's like a very easy game of concentration. The first time, you flip the card over (roll over). Afterwards, it is easier to remember "there, where I rolled" than home>company>info, especially it is "there, where I clicked."
Remember when you were in college and you were trying to find that line to quote? You knew it was on the upper left hand corner of the page, near the end of the first paragraph…but you had no idea what chapter it was in.
I'd argue that remembering points on a two-dimensional grid is more more natural–and faster–than climbing down the "string" tree: this category>that category, unless you're a descendent of Immanuel Kant, of course.
Category hierarchies are more convenient for people placing information than for people extracting information, so don't mistake your convenience for the convenience of the user. (As Yahoo did, to its detriment.)
4) Getting rid of labels frees up space for more buttons, so it prevents tedious drilling down.
5) The hotbox can be multifunctional. It could be the voice not only of the buttons, but tool tips, help, etc. That way, the user always know where to look for reaction, if she wants it.
(Perhaps optional text should be in a ligher, low-contrast color, so users can ignore it.