Transfer in Machine Translation by Non-Confluent Term-Rewrite Systems

Abstract

The transfer component of a machine translation (MT) system typically comprises three aspects, namely the transfer structure used to transfer source into target structures, the (set of) transfer rules and the transfer process, which interpretes the set of transfer rules to derive the target from the source structure. The transfer structure can be formally considered as a term and the set of transfer rules as a term-rewrite system, which is terminating if a well-founded ordering on the terms can be defined, but not confluent. Most of the interpreters of rewrite systems require both termination and confluence. This paper is concerned with the transfer process and presents a well-founded ordering on the terms representing the transfer structures and an algorithm for the interpretation of terminating and non-confluent rewrite systems, which is used within the transfer component of the MT system of the project KIT-FAST ¹ at the Technical University of Berlin. The other aspects concerning the transfer component will be dealt with in [We89b].

This work has been developed in the project KIT-FAST (KIT = Künstliche Intelligenz und Textverstehen (Artificial Intelligence and Text Understanding); FAST = Functor-Argument Structure for Translation), which constitutes the Berlin component of the complementary research project of EUROTRA-D. It receives grants by the Federal Minister for Research and Technology under contract 1013211.