Symbolic Distributed Representations

Is it possible to perform symbolic computations in a neural network? Such a capability would enable implementation of discrete logic in the massively parallel computing architecture of neural networks.