ABSTRACT
Statistical approaches for supervised machine learning were not invented in the 1990s: the first supervised learning models developed in the late 1950s for checkers or triangles already operated on this principle. These algorithms calculate an association between the numeric values of the inputs and the expected result. The greater difficulty of using statistical machine learning methods was, thus, not worth it. Over the decades, advancements in computer science have helped produce much more numeric data and increased the number of variables people can work with, surpassing the limits of symbolic models. Vapnik and his colleagues had a great idea: add imaginary features to the kernel to force the algorithm to work in a space with more dimensions. Processing natural language is a difficult problem for Artificial Intelligence and, yet, one of the most ambitious problems to solve. Indeed, language is a very complex phenomenon that is impossible to describe using symbolic rules alone.
