Word-order biases in deep-agent emergent communication - View it on GitHub