A novel framework for symbol grounding in artificial agents is presented, which relies on the key idea that concepts "emerge" implicitly at the perceptual level as clusters of points with similar features forming homogeneous regions in multiple perceptual Conceptual Spaces (pCS). Such spaces describe percepts such as color, texture, shape, and position that in turn are the properties of the objects populating the agent's environment. Objects are represented in a suitable object Conceptual Space where all their features are composed together again using clustering in pCSs. Symbols will be learned from such a tensor space. A detailed description of both the framework and its theoretical foundations are reported and discussed in this work.
|Number of pages||6|
|Publication status||Published - 2018|
All Science Journal Classification (ASJC) codes
- Computer Science(all)