Conversational interfaces are becoming more and more important in the current period, with many different mathematical models aiming to improve quality and syntax structure of a generated text. In this work we propose an approach using a Sequence to Sequence generator model which generates sentences starting from an acquired knowledge, which are then evaluated from a discriminator and a feedback on the quality of the generated ones is given to the generator, allowing it to adjust and improve quality. In this work we show the approach and experimental results obtained.