Attention is the gateway to information processing and memory. Information to which we attend enters short-term or working memory, our "workbench" where consciousness is experienced. This "workbench" is where we "think"--attending to external stimuli, internal thought processes, or both--reason, problem solve, make decisions, read, perform mathematical computations, etc. It is also where we consciously direct memory recall and initiate responses and actions.
In short-term/working memory, we encode information from the sensory registers for transfer into long-term memory. Sub-vocalization is part of this. (Yes, it is OK to talk to yourself--silently--otherwise, you would not be able to think!)
Short-term/ working memory characteristics, important for the design of human-to-system interfaces as well as training/learning programs, are:
Capacity - Very limited and in some models considered a "bottleneck" in human information processing. The classic work of Miller (1956) determined the number of units that can be processed at any one time as 7 + 2. Subsequent studies have indicated that 5 + 2 may apply to most of the items we wish to remember.
Duration - About 15 to 30 seconds, however, it can be indefinite if one continues to concentrate on, attend to, and rehearse the information in its store.