Automatic rule induction in Arabic to English machine translation framework

Abstract

This chapter addresses the exploitation of a supervised machine learning technique to automatically induce Arabic-to-English transfer rules from chunks of parallel aligned linguistic resources. The induced structural transfer rules encode the linguistic translation knowledge for converting an Arabic syntactic structure into a target English syntactic structure. These rules are going to be an integral part of an Arabic-English transfer-based machine translation. Nevertheless, a novel morphological rule induction method is employed for learning Arabic morphological rules that are applied in our Arabic morphological analyzer. To demonstrate the capability of the automated rule induction technique, we conducted rule-based translation experiments that use induced rules from a relatively small data set. The translation quality of the hybrid translation experiments achieved good results. in terms of WER.