Gappy Translation Units under Left-to-Right SMT Decoding

This paper presents an extension for a bilingual n-gram statistical machine translation (SMT) system based on allowing translation units with gaps. Our gappy translation units can be seen as a first step towards introducing hierarchical units similar to those employed in hierarchical MT systems. Our goal is double. On the one hand we aim at capturing the benefits of the higher generalization power shown by hierarchical systems. On the other hand, we want to avoid the computational burden of decoding based on parsing techniques, which among other drawbacks, make difficult the introduction of the required target language model costs. Our experiments show slight but consistent improvements for Chinese-toEnglish machine translation. Accuracy results are competitive with those achieved by a state-of-the-art phrasebased system.