End-user robot programming has the potential to increase accessibility of general-purpose robot programming. Despite there are many attempts to control robots by using natural language conversations, most of the existing end-user programming systems for robots, assume a computer screen to write and modify programs. This paper explores robot programming by using natural language conversations. We focus on the levels of programming abstraction and the styles of programming from the viewpoint of voice user interface, and present our design along with a prototype implementation. Our design and implementation attempt to take advantages inherent to modular programming approach, namely ease of identification and ease of editing. In this work we present our implementation and rationale for choices regarding programming style, language and level of abstraction. Using the programming approach demonstrated in this paper, one could investigate extending the natural language programming system to support multi-modal interaction, e.g., using gestures, for more natural interaction and present our rationale for level of abstraction, programming style and programming language design.