In a regular game, the player entity, and the NPC entities, and the environment, all have controllers. For the player, the controller listens to the keyboard/mouse, and determines what the player character should do. The NPC listens to the AI input and stimuly, and determines what happens to the NPC characters. The environment listens to triggers and stimuly, and determines what changes. The camera also has a controller, which will typically be in a mode where it chases the player, but tries to keep a clear line of sight (for third person views).

Now, to do the scripted scenes, you switch out the controllers of things, to controllers whose output data (animations to play, locations of objects) are all driven by the scene script. Sometimes, you''ll animate all of this in your tool, and export as one big animations. Othertimes, you''ll script at a higher level, with things like "YOU need to be HERE at THAT TIME" and the controller figures out how to make it so; walk the guy over to the door; etc. Similarly for environment, instead of waiting for the "key" trigger for a door, it''ll be temporarily wired to a controller that waits for time X, given by the scene script.