(42) Automatic Evaluation of Machine Translation based on Recursive Acquisition of an Intuitive Common Parts Continuum
Hiroshi Echizen-ya and Kenji Araki
Proceedings of the Eleventh Machine Translation Summit (MT Summit XI), pp.151-158，2007-9
In this paper, we propose a new automatic evaluation method of machine translation. Our method specifically examines the length and position of the common parts between two sequences. First, the common parts continuum is determined using the length and position information of common parts in two sequences. That is, the most intuitive common parts continuum is obtained using this process. Moreover, our method recursively repeats this process to control the difference of the common part order. In this repetition process, a greater penalty is given to the intuitive common parts continuum as the number of repetition increases. We call this method Recursive Acquisition of Intuitive comMon PArts ConTinuum (IMPACT). The evaluation results show that the IMPACT score better correlates with human judgment in both adequacy and fluency than some other automatic evaluation methods.
>>
NEXT