The following keymaps are defined by predictive mode:
The various keymaps define key bindings for different situations that arise in predictive mode. The main predictive-map keymap is enabled whenever predictive mode is enabled, and auto-completion-map is enabled whenever auto-completion-mode is. Note: if you find yourself thinking of re-binding printable characters in auto-completion-map to something other than completion-self-insert, don't! (at least not until you've read on a bit). What you probably want to change are the auto-completion-syntax-alist and auto-completion-override-syntax-alist variables. See Syntax.
completion-overlay-map and auto-completion-overlay-map are active when the point is located at a completion. The former is used when auto-completion-mode is disabled, the latter when it is enabled. They are used to provide key bindings for most of the completion features.
The key bindings in the keymaps enabled by the completion-use-* customization variables are effectively added to completion-overlay-map, so they are active when the point is in a completion and the corresponding user-interface is enabled.
If the keymaps are not defined when predictive mode is first loaded
(see Obtaining and Installing), it creates the default keymaps and
loads them into Emacs. Therefore, to completely re-define predictive
mode key bindings, it is simpler to re-define the keymap variables
before loading predictive mode (e.g. before the (require
'predictive) line in your .emacs file, see Obtaining and Installing). (However, doing this for the
auto-completion-mode-map will almost certainly break
auto-completion-mode entirely!) But if you simply want to
modify a few key bindings, as will usually be the case, you can do it in
the usual Emacs way, after loading predictive mode in your
.emacs file.