Yeah this idea has been tried before, see for example the cycleGAN architecture (not for language, but image-to-image translation) https://bellchen.me/wp-conte

Yeah this idea has been tried before, see for example the cycleGAN architecture (not for language, but image-to-image translation) https://bellchen.me/wp-content/uploads/2018/01/LUOHAO_1513256736.8965938.png https://bellchen.me/wp-content/uploads/2018/01/LUOHAO_151325...