One of the main challenges in building hybrid systems that integrate symbolic and sub-symbolic levels of computations is elaboration of methods that would allow expressing symbolic manipulations in a form of neural networks dynamics. In particular, it is extremely important to be able to process symbolic structures in a distributed format and perform analysis over this tensor representation that would be equivalent to a sequence of symbolic operations over the original structure. In this paper, the novel method is proposed for processing and analysis of semantics tree in a form of generated neural network. This method is built upon the Tensor Product Representations (TPRs) framework, scales well for an arbitrary structure and allows expressing simple symbolic operations in a form of neural network dynamics.