Effects of Integrating Multiple Bilingually-Trained Segmentation Schemes for Japanese-English SMT

This paper proposes a method to integrate multiple segmentation schemes into a single statistical machine translation (SMT) system by characterizing the source language side and merging identical translation pairs of differently segmented SMT models. Experimental results translating Japanese into English revealed that the proposed method of integrating multiple segmentation schemes outperforms SMT models trained on any of the utilized word segmentations and performs comparably to available state-of-the-art monolingually built segmentation tools.