FloW is a humanoid robot that is being built and developed by ER2C (Eepis Robotic Research Center). FloW is made to have human-like capabilities and one of the abilities, the robot is able to communicate with humans. Therefore, we need to build speech processing system which consists four sub-systems. There are speech-to-text system, sentence understanding system, sentence generating system and text-to-speech system. This paper proposes how FloW can generate sentences to start and respond conversation. The output of this system is text that will be combined with text-to-speech system. So, the robot's answers are sound. We use template-based method which consists of word list process, templates and answers to start a conversation. While using the pattern-matching method which consists of answer rules, answer extraction and answers to respond a conversations. The experimental result shows generated sentences to start and respond conversations in closed-domain around the ER2C (Eepis Robotic Research Center) has reached 100%.