Counter-Contrastive Learning for Language GANs