The various components of the short term memory window in the ARL system do not have a modular property. For example, the system could observe three gestures in a certain order: gesture A, then gesture B, then gesture C. Even though all three are in the system's short term memory, it will not parse these into 3 separate components and recognize them independently. Therefore, there is no higher order abstraction about the evolution through event A, B then C. This abstraction and modularity of signals is reminiscent of linguistic processing. Therefore, the rearrangement of different chunks or modular components can not be understood by the system. Grouping lower order signals into sophisticated hierarchies and dependencies involves the ability to segment and rearrange them. These mechanisms do not exist in the system as is. There might be possibilities to reformulate the architecture such that this type of learning can be encompassed.