Computer understanding of natural language (NL) is commonly considered as extracting "e;sense"e; from NL texts. This deprives the problem of determinity, since the notion of "e;sense"e; lacks a formal definition. However, with a man-to-computer NL dialogue involved in some working process, any address by humans is nothing but a task for the computer to fulfil. Then it is immaterial whether the computer understands the address text sense and even the very notion of "e;sense"e;. Only the address task is to be accomplished. This means not revealing morphologic, syntactic and other text structures, but only extracting data on the task. As a result, a new theory of NL dialogue understanding has been created, called "e;orientated linguistics"e;. This theory has been brought to life as several practical systems, which have demonstrated an extremely reliable and correct understanding of a quite free and easy NL, a tiny resource consumption and simple readjustment between various subject areas and, what is more, national languages.