The E-framework: a formalism for natural language processing

This paper presents the most important characteristic of the new formalism used in Eurotra, the E-Framework. It is a formalism for natural language processing within a stratificational model.In the E-Framework, mapping between levels of representation is performed on the basis of transitions between trees and partial descriptions of objects, called descriptors. These descriptors are completed using the definition of the target level.The tree to descriptor strategy simplifies the expression of complex two-way relations between text and abstract representations when major structural changes have to be performed on the input. This is illustrated by way of a detailed example showing the interaction of the two formal devices of the E-Framework, the translator and the generator, the basic ideas of which are also briefly described.The E-Framework has been implemented and forms the basis of the development of Eurotra's pre-industrial prototype of a transfer-based, multi-lingual machine translation system.