You can't tell. Perhaps for most people, the predominance of the Subject-Verb-Object construction would bias it towards the world illuminating the face. I assume you are asking about situations when set expressions also don't give a clue about what governs what.