Originally, translation was the main aim of the NLP technology development.
From the late 1940s to the 1960s, there was widespread belief that translation was not so difficult to obtain. Nonetheless, in 1966 the US government ALPAC report shut down all the work in this field: "There has been no machine translation of general scientific text, and none is in immediate prospect."
Having in mind the idea that all of the text's meaning is in the words directly on the surface or only slightly below, scientists only reluctantly realized that there is more to it than mere 1-to-1 mapping of words or phrases. Instead, you have to consider many layers that can possibly hold information.
You can have interdependencies between parts of the texts, allusions to external objects, ambiguities and metaphors. So they need to extract virtually everything about the context from the text itself, what in turn brings us to the first constraint on automatic translation or every other subfield of NLP: The text needs to be restricted to a limited range of subjects. There is a concept of controlled or restricted languages, also called "Caterpillar English", which allows only a limited range of syntactic constructs and operates on a somewhat smaller vocabulary than everyday language.

<---