During a human-robot interaction by dialogue/voice, the robot cannot extract semantic meaning from the words used, limiting the intervention itself. Semantic knowledge could be a solution by structuring information according to its meaning and its semantic associations. Applied to social robotics, it could lead to a natural and fluid human-robot interaction. Ontologies are useful representations of semantic knowledge, as they capture the relationships between objects and entities. This paper presents new ideas for ontology generation using already generated ontologies as feedback in an iterative way to do it dynamically. This paper also collects and describes the concepts applied in the proposed methodology and discusses the challenges to be overcome.