Enhancing function word translation with syntax-based statistical post-editing
John T. E. Richardson, Toshiaki Nakazawa, Sadao Kurohashi · 2015
The generation of precise and comprehensible translations is still a challenge in the patent and scientific domain. In particular, function words are often poorly translated in standard machine translation systems, particularly across language pairs with greatly differing syntax. In this paper we exploit the target-side structure in tree-to-tree machine translation to post-edit function words automatically using a tree-based function word language model. We show that a significant improvement in human evaluation can be achieved with our proposed method. 1