Text this: Analysing factorizations of action-value networks for cooperative multi-agent reinforcement learning