Intention recognition requires reasoning about affordances, or possible goals for an agent in a given environment. Recognizing the intent of the user is a task of critical importance in the context of Human Robot Interaction to allow any kind of predictive cooperation to occur.
In classical plan recognition, affordances and actions usually suffer from poor semantic generalization capabilities due to the high amount of hand-crafted specifications that must be put into a system, especially when interaction with humans is required. We try to overcome this problem by utilizing the semantic generalization offered by the Natural Language Process set of techniques known as word embedding. We propose a machine learning system based on Deep Belief Networks and Generative Adversarial Networks to generate possible affordances from a given textual representation of an object.