Die Generierung natürlicher Sprache aus einer formalen Repräsentation

Since the 50’s there have been many attempts to design computer programs which are able to translate texts from one natural language into another. Despite 40 years of intensive research the systems developed so far are still far from being able to translate texts correctly. In this thesis we present a different approach to the machine generation of texts into several languages, namely their generation from the expressions of a formal language. The idea underlying this is that an author formulates his text in a formal language. This language contains all the information which the computer needs to produce a natural language text in the desired target languages. The problem of the analysis of a natural language source text, which has, to a great extent, not yet been mastered in machine generated translation, does not arise with this approach. The first part of the thesis deals with the question of which form a formal language must take if it is to produce texts of a sufficiently large fragment of speech. In the second part we then cite algorithms which produce German, English and French sentences from the expressions of the formal language. In addition, a more user-friendly version of our formal language is part of our research project. Its syntax is discussed in a separate paper. It allows the user to work with a disambiguated fragment of his mother tongue, which can be translated unambiguously into the formal version. This way he does not have to work with the actual formal language, for it is very machine-oriented and therefore unwieldy. This abstract, for example, has been written in the user-friendly version of the formal language. The possible practical applications of our method are obvious: For example, an author who wishes to publish a text in various natural languages can write his text in the formal language. The computer then takes over the task of formulating the text in the desired target languages. The same applies when an author wishes to formulate a text in a language with which he is not familiar. This procedure has definite advantages over translations by a person or a machine: When a formal language is used as source language, there is no need for the analysis of the source text in the natural language. This is a problem which has to date been insufficiently resolved in machine translation. In the case of non-specialist persons doing the translation there is the additional problem of specialist texts being misunderstood.