Transfer and architecture : views from chart parsing

Abstract

The objective of this report is to describe the embedding of a transfer module within an alternative architectural approach for machine translation of spontaneous spoken language. The approach is cognitively oriented, i.e. it adapts some of the assumed properties of human language comprehension and production. The aspects to be modeled will include incrementality and robustness with respect to disturbances caused by the environment and performance phenomena of speech. Interaction between software modules is used to reduce ambiguity. The transfer stage of a translation system clearly has to obey these requirements to be an integral part of such a system. This paper outlines the kind of demands to be placed on the transfer module. Relations between the basic formalisms representing linguistic knowledge on the one hand and transfer on the other hand are demonstrated as well as the consequences for algorithms and data structures.