Dialogue systems have been widely studied by the NLP community in the last decades. This kind of systems generally consist of three main modules: natural language understanding, dialogue manager and natural language generation. Lately, motivated by machine comprehension, conversational QA has emerged as a intermediate task between QA and dialogue, where machines have to assist users in information gathering by conversational means.

In this work, we merge traditional dialogue systems and conversational QA models for task oriented dialogues in the cooking domain. STS and IR systems are used as an intermediate layer between the dialogue system and the ConvQA model, where the impact of simple dialogue history modeling is also analyzed.