The use of Self-Organizing Map requires that the input is presented as numerical vectors and a metric for comparing the vectors is available. Each input vector may correspond to one ``case''. In this example such a case is one language. The ``cases'' or the classes of the inputs can, after the learning, be used to label the resulting map. It is to be noted, however, that the organization of the map emerges autonomously and the labels are used only to make the result more readable for the human audience.
The relative frequencies of characters in 11 different languages in a small corpus were used as an input. The resulting map corresponds reasonably well with the linguistic groups. It is notable, for example, that Estonian and Finnish are located in a ``valley'' of their own. Bright areas on the map mean that the map unit vectors are close to each other, dark zones mean a jump in the vector values.
The advantage of using SOM is still more obvious when the number of parameters is large.