tailieunhanh - Báo cáo khoa học: "THE DESIGN OF THE KERNEL ARCHITECTURE FOR THE EUROTRA* SOFTWARE"

Starting from the assumption that machine translation (MT) should be based on theoretically s o u n d grounds, we argue that, given the state of the a r t , the only v i a b l e solution for the designer of software tools for MT, is to provide the linguists building the MT system with a generator of highly specialized, problem oriented systems. We propose that such theory sensitive systems be generated automatically by supplying a set of definitions to a kernel software, of which we give an informal description in the paper. We give. | THE DESIGN OF THE KERNEL ARCHITECTURE FOR THE EUROTRA SOFTWARE . Johnson . . Box 88 Manchester M60 1QD . s. Krauwer Rijksuniversiteit Trans 14 3512 JK Utrecht Holland . Rosner ISSCO University of Geneva 1211 Geneve 4 Switzerland . Varile Commission of the European Communities . Box 1907 Luxembourg ABSTRACT Starting from the assumption that machine translation MT should be based on theoretically sound grounds we argue that given the state of the art the only viable solution for the designer of software tools for MT is to provide the linguists building the MT system with a generator of highly specialized problem oriented systems. We propose that such theory sensitive systems be generated automatically by supplying a set of definitions to a kernel software of which we give an informal description in the paper. We give a formal functional definition of its architecture and briefly explain how a prototype system was built. I. INTRODUCTION A. Specialized vs generic software tools for MT Developing the software for a specific task or class of tasks requires that one knows the structure of the tasks involved. In the case of Machine Translation MT this structure is not a priori known. Yet it has been envisaged in the planning of the Eurotra project that the software development takes place before a general MT theory is present. This approach has both advantages and disadvantages. It is an advantage that the presence of a software framework will provide a formal language for expressing the MT theory either explicitly or implicitly. On the other hand this places a heavy responsibility on the shoulders of the software designers since they will have to provide a language without knowing what this language will have to express. We are grateful to the Commission of the European Communities for continuing support for the Eurotra Machine Translation project and for permission to publish this paper and also to our colleagues in Eurotra for many interesting