Trait-Based Hypothesis Selection For Machine Translation

Jacob Devlin and Spyros Matsoukas
Raytheon BBN Technologies


Abstract

In the area of machine translation (MT) system combination, previous work on generating input hypotheses has focused on varying a core aspect of the MT system, such as the decoding algorithm or alignment algorithm. In this paper, we propose a new method for generating diverse hypotheses from a single MT system using "traits". These traits are simple properties of the MT output such as "average output length" and "average rule length." Our method is designed to select hypotheses which vary in trait value but do not significantly degrade in BLEU score. These hypotheses can be combined using standard system combination techniques to produce a 1.2-1.5 BLEU gain on the Arabic-English NIST MT06/MT08 translation task.