A robot's ability to assist humans in a variety of tasks, e.g. in search and rescue or in a household, heavily depends on the robot's reliable recognition of the objects in the environment. Numerous approaches attempt to recognize objects based only on the robot's vision. However, the same type of object can have very different visual appearances, such as shape, size, pose, and color. Although such approaches are widely studied with relative success, the general object recognition task still remains very challenging. We build our work upon the fact that robots can observe humans interacting with the objects in their environment, and thus providing numerous non-visual cues to those objects' identities. We research on a flexible object recognition approach which can use any multiple cues, whether they are visual cues intrinsic to the object or provided by observation of a human. We realize the challenging issue that multiple cues can have different weight in their association with an object definition and need to be taken into account during recognition. In this paper, we contribute a probabilistic relational representation of the cue weights and an object recognition algorithm that can flexibly combine multiple cues of any type to robustly recognize objects. We show illustrative results of our implemented approach using visual, activity, gesture, and speech cues, provided by machine or human, to recognize objects more robustly than when using only a single cue.