Here we illustrate some preferences to help the generator make decisions on where to ``break'' the lexico-semantics of the semantic input into different sentences. For instance, in absence of any other preference, it is not natural to split a sentence on an entity slot, such as in I drive a car. The car has a color and a cost. The color is purple and the cost is unaffordable. It is better to perform an entity split, that is only breaking on the entities, such as in I drive an unaffordable purple car..