Morphology in Machine Translation Systems: Efficient Integration of Finite State Transducers and Feature Structure Descriptions

Abstract

We present a finite state morphology system augmented with typed feature structures as weights on transitions. This mechanism allows the use of highly efficient finite state approaches for morphological analysis and generation, while providing the rich linguistic descriptions often used in Machine Translation systems. Using a semiring interpretation, the weight of a morphological analysis result represents the possible linguistic interpretations of an input word, while the resulting character string itself represents the lemma of the input. Long-distance phenomena and infixation can be handled in an easy and elegant manner, simultaneously providing a seamless interface to subsequent linguistic processing modules. Two extensions to the basic model are discussed: the incorporation of lexical knowledge into the finite state transducer and a transformation that renders unification-based finite state models as efficient as those employing other weight structures. The model is applied to morphological operations in a Persian--English Machine Translation system.