A rule-driven dynamic programming decoder for statistical MT

Abstract

The paper presents an extension of a dynamic programming (DP) decoder for phrase-based SMT (Koehn, 2004; Och and Ney, 2004) that tightly integrates POS-based re-order rules (Crego and Marino, 2006) into a left-to-right beam-search algorithm, rather than handling them in a pre-processing or re-order graph generation step. The novel decoding algorithm can handle tens of thousands of rules efficiently. An improvement over a standard phrase-based decoder is shown on an Arabic-English translation task with respect to translation accuracy and speed for large re-order window sizes.