We have been developing a spoken dialogue system of information retrieval on academic documents with a special focus on reply speech generation. In order to realize speech reply with its prosodic features properly controlled to express the dialogue focus, we had developed a concept-to-speech conversion scheme where the reply concept was directly converted to a sequence of phone and prosodic symbols. In our original system, however, a priority was given to the automatic processing, and the method for prosodic focus control was rather simplified. Aiming at improving the reply speech quality, new rules were constructed for prosodic focus control. Through the listening experiment, the new rules were evaluated to be revised further. The validity of the revised rules was verified through an evaluation experiment of the system.