In China's national "863" MT evaluations, BLEU/NIST are used to evaluation Chinese and Japanese translations. The result are computed in a character-based way, rather than word-based. Since "word" is not a well-defined concept in Chinese and Japanese. Different applications may have different specification for word segmentation.
As a result, the BLEU/NIST scores of Chinese and Japanese translations are some what higher than that of English translations, since the granularity of character is finer than that of word. For more information in "863" evaluation, please visit: http://www.863data.org.cn http://www.863data.org.cn/english (English version) In the website, some information are only available in Chinese, for example, the results of 2004 evaluation. Liu Qun On Thu, 14 Apr 2005 02:19:16 +0100 "Somers, Harold" <[EMAIL PROTECTED]> wrote: > Does anyone have any experience of applying BLEU/NIST scores to non-spacing languages like Japanese, Chinese, Korean? Since they are word-based metrics, I presume output has to be segmented somehow. -- liuqun <[EMAIL PROTECTED]> _______________________________________________ Mt-list mailing list
