Machine translation with human post-editing is substantially cheaper and faster than full human translation, but reaches quality parity only under some post-editing conditions

Narrative synthesis #

Machine translation with human post-editing is consistently cheaper and faster than full human translation: post-editing ran 2–10× faster (11.76 vs 3.30 WPM), MT added no per-document cost against $130–$1,220 for human translation, and human turnaround reached up to 35 days end-to-end. Whether it reaches quality parity is contested and left first-class here rather than resolved: bilingual raters preferred post-edited MT and human translation equivalently and revised "make all necessary corrections" instructions moved MT toward equivalence, yet blinded raters preferred human translation for all 20 Chinese documents and an expert post-editor did not close the gap.

Crucially, support and opposition come from the same research group (Turner), so this is not a between-group disagreement. The candidate reconciling variables are target language (English-target materials in Turner 2014 vs Chinese in Turner 2015) and post-editing instructions — the revised-instructions EVD is the direct evidence for the latter, since changing instructions moved the result. Wiring the human-translation-turnaround EVD here also fixes a real gap: it previously had no claim link at all. Merge candidates (proposal only — do not touch): C-0214ClaimC-0214Initial AI draftHuman professional translation is preferred over machine translation with postediting for quality of translated public-health materialsFor translated public-health materials, expert raters prefer professional human translation over machine translation with human postediting (MT+PE) on translation quality, especially for linguistically divergent language…, C-0207ClaimC-0207Initial AI draftMachine translation of public-health materials should be followed by human post-editing to reach acceptable quality and be adoptedMachine translation of public-health materials should never be used in isolation but must be followed by human review and post-editing by a native speaker for translations to reach acceptable quality and for the technolo…, C-0217ClaimC-0217Initial AI draftPostediting instructions to make all corrections needed rather than only minimal edits improve machine-translation qualityInstructing posteditors to make as many corrections as needed for quality — rather than only minimal necessary edits — improves the quality of machine-translated health materials, at the cost of slower postediting., C-0195ClaimC-0195Initial AI draftMachine translation with human postediting is faster to produce than fully human translation of health materialsMachine translation with human postediting takes less human time to produce a translated health document than fully human translation, because postediting an existing machine draft is faster than translating (or postedit…, C-0196ClaimC-0196Initial AI draftMachine translation with human postediting lowers the cost of translating health materials versus fully human translationMachine translation with human postediting lowers the cost of translating health materials compared with fully human (vendor) translation, chiefly by eliminating the vendor's charge for the initial translation while reta….