A POS-based preordering approach for english-to-Arabic statistical machine translation

Abstract

In this work, we present a POS-based preordering approach that tackles both long- and short-distance reordering phenomena. Syntactic unlexicalized reordering rules are automatically extracted from a parallel corpus using only word alignment and a source-side language tagging. The reordering rules are used in a deterministic manner; this prevents the decoding speed from being bottlenecked in the reordering procedure. A new approach for both rule filtering and rule application is used to ensure a fast and efficient reordering. The tests performed on the IWSLT2016 English-to-Arabic evaluation benchmark show a noticeable increase in the overall Blue Score for our system over the baseline PSMT system.