Weak evidence for word-order universals