Literal Equations in Neural Networks