Calculating adjunction trees is still slow for large grammars. Possible solutions: - randomized generation - goal-directed generation - transform the grammar to a form that is easier to use for generation