TransDocument
Takeo Igarashi · 2014
Machine translation systems are effective for translating relatively short sentences, but they often return incomprehensible results when applied to long sentences with complicated dependency structures. We therefore propose to expose sentence structure to writers and readers to facilitate translingual communication via imperfect machine translation systems. The writer encodes his or her message into a structured text, which is a collection of sentence fragments organized in a hierarchical structure. The system then translates these fragments individually, using a standard machine translation system. The structure itself is preserved. The reader reads the structured text in the target language. In this paper, we present an interactive text-editing system that supports the writing and reading of such structured texts. We also run a small scale user study to investigate the potential of the proposed method, and the result suggests that the structure information can be helpful in improving the understandability of machine translation results.