<p>While many methods for end-to-end task-oriented dialogue systems have been recently proposed to achieve specific tasks through human-machine interactions, these methods are still sub-optimal for at least two reasons. First, the dialogue history that generates response templates often ignores certain entity tags. Secondly, external knowledge about entities is sometimes incorporated into the system, even though this information could be easily retrieved from the dialogue history. Such redundancy deteriorates prediction accuracy. To address these issues, we propose a parallel model that uses dialogue history templates to improve the quality of generated entity tags. Our model uses reasoning about the entities extracted from the dialogue history to determine consistent, contextual semantics within the current dialogue entity instead of retrieving knowledge about the current dialogue before the encoding phase. Our experiment on existing benchmarks demonstrates the effectiveness of our approach in generating comprehensive response content and accurate entity prediction.</p>