In this paper we introduce a variant of Memory Networks that needs significantly less supervision to perform question and answering tasks. The original model requires that the sentences supporting the answer be explicitly indicated during training. In contrast, our approach only requires the answer to the question during training. We apply the model to the s