"I'm going to need about 600 bits of entropy for this. Can you go the store and pick up some playing cards for me? Let's see, six hundred divided by log-base-two fifty-two-factorial—yes, three packs should be enough."
(Later, opening them ...)
"... and when we want to look at just a subset of the variables in a joint distribution, we have to sum over all the other variables: the probability that X is blue, is equal to the probability that both X is blue and Y is blue, plus the probability that X is blue and Y is red, plus ... and so on for all the values Y could take. We call this marginalizing over Y to get the marginal distribution for X. Note that you can think about this as taking an expected value. Does that make sense?"
"Ummm ... ye-es?"
"You don't sound very confident."
"The exact referent of the word that in 'Does that make sense?' was ambiguous, because it was preceded by a long, multi-part explanation. Most of the potential referents made perfect sense, but my response had to average over all of them, hence the hesitation and uncertain tone."