HTER

HTER stands for Human-targeted Translation Edit Rate. It is a metric used in machine translation evaluation to measure the amount of editing effort a human post-editor must exert to transform a machine-translated sentence into a post-edited reference translation. The score is typically computed as the minimum number of edit operations—insertions, deletions, substitutions, and sometimes shifts—divided by the length of the reference translation, usually measured in words. The resulting value ranges from 0 to 1 (or 0 to 100%), with lower scores indicating less editing effort and higher quality MT output relative to the post-edited reference.

HTER is a variant of the Translation Edit Rate (TER) that uses human post-edits as the reference instead of an automatic or synthetic reference. By using actual corrections made by human translators, HTER aims to reflect the real editing workload involved in producing a publishable translation. It is commonly employed to evaluate MT systems and to assess changes in post-editing workload across system iterations or configurations. Analyses may report HTER for individual sentences or across larger corpora.

Limitations of HTER include dependence on the quality and style of the post-edits, as well as the need for a corpus of post-edited translations. It measures editing distance rather than time, cognitive effort, or economic cost, and can be sensitive to editors’ conventions and languages involved. Consequently, HTER is typically used alongside other MT evaluation metrics, such as BLEU or standard TER, to provide a more comprehensive view of translation quality and post-editing workload.