Typically, they can answer two kinds of question: ``how'' and ``why''. `How'' means ``how did you conclude that fact?''; ``why'' means ``why are you asking me this question?''. For examples, see the Mycin dialogue or the plant-cure one in Goodall.
To answer a ``how did you prove X'' question, a backward chainer would keep note of all the rules that were proven true en route to proving X. It would translate each into pseudo-English, and display the results.
To answer a ``why are you asking this'' question, the system looks to see what rule it's currently trying to prove true. (Remember that questions ensue from trying to prove the conditions of rules). Suppose the rule is, schematically, ``If Q1 and Q2 and Q3 then C''. It translates that rule into pseudo-English, and says ``I am asking you Q1 because I want to prove C, and rule N says that C if Q1 and Q2 and Q3''.