Error Classification for MT Evaluation
Mary Flanagan · 1994
A classification system for errors in machine translation (MT) output is presented. Translation errors are assigned to categories to provide a systematic basis for comparing the translations produced by competing MT systems. The classification system is designed for use by potential MT users, rather than MT developers. The error categories can be ranked according to criteria which are important to the user, such as improvability and intelligibility. The results of CompuServe's use of the error classification methodology are presented as a case study. 1.