Fooling neural networks is a main concern in the process of Artificial Intelligence optimization. Character perturbation make part of a text unnoticeable for some systems, even for human observers.