‘Just because you are right, doesn’t mean I am wrong’: Overcoming a bottleneck in development and evaluation of Open-Ended VQA tasks