A mathematical model for the CAT framework of Eurotra

Abstract

Eurotra is the machine translation project of the EEC. The basic ideas for the design of the system are given by the CAT framework, which is, together with various relaxations, presented in several publications (e.g. Arnold 1985, 1986 e.a., des Tombes e.a. 1985, Arnold & des Tombes 1987). In the present paper a mathematical model for the CAT framework will be developed. This will be a model of the structural aspects of the framework, such as the structure of the grammars and of the translation steps. The model uses notions and results from universal algebra; a branch of mathematics which deals with structures and their relations. The model is in a certain sense the same as the CAT framework, but it is build with different tools. Eurotra is a project of ongoing research with continuous practical experience, and this might cause changes in the original framework. Since the present paper is mainly based upon the publications mentioned above, it does not necessarily describe the present situation correctly (for your information, the author is not personally involved in the project). The aim of this paper is, however, not to present some version of Eurotra, but to argue for a more abstract and more mathematically based approach to Eurotra (and other machine translation systems). It will be shown that there are several advantages of such a mathematical approach. It brings new insights in the framework (see sections 4 and 5), and gives us a new appreciation of certain Eurotra proposals (see sections 6 and 7). Furthermore, the mathematical model for Eurotra will, I expect, be a good starting point for investigating later stages of the Eurotra system.