Representing Part-Whole Relations

Part-whole relations appear throughout human knowledge. Their ubiq-uity suggests that they play an important role in intelligence. We conjecture that the cognitive value of partonomic knowledge lies in its ability to support translation between representations at diierent levels of granu-larity. This permits more eeective problem solving by allowing reasoners to integrate the unique beneets that diierent representations can ooer. We describe a representation for part-whole relations that supports this perspective. Our representation captures two types of partonomic knowledge: decompositions and knowledge translations. Decompositions identify the parts that compose objects. Knowledge translations relate knowledge about objects to knowledge about their parts. Our approach organizes partonomic knowledge around a taxonomy of decomposition functions. This allows us to use inheritance to share partonomic knowledge among similar decompositions.