Posteditors corrected English-to-Chinese machine translations at about 37.8 characters per minute
Source
Turner (2015). Machine Translation of Public Health Materials From English to Chinese: A Feasibility Study. JMIR Public Health Surveill.
Description #
Across the 25 machine-translated Chinese public-health documents, posteditors corrected the raw Google Translate output at a mean rate of 37.8 characters per minute (SD 10.2), i.e., approximately 38 characters per minute on average (Table 3). Per-document speed varied widely (18.5–79.6 CPM), and postediting time showed no relationship to document length or document type.
"The total mean CPM across all documents was 37.8 (SD 10.2). Thus, on average a posteditor corrected approximately 38 CPM, with a variation of around 10 CPMs." (Turner, 2015, p. 5)
Methods Context #
What? #ⓘ
The observable: postediting effort, operationalized as the number of characters per minute (CPM) a posteditor produced while correcting a machine-translated document.
"To determine and analyze the amount of time required for postediting, we calculated the number of characters per minute (CPM) for each document and then computed means and standard deviations (SDs) in CPM for each document" (Turner, 2015, p. 5)
How? #ⓘ
A proprietary postediting tool timed each editing session: when a posteditor began editing the machine-translated text a timer started, and the tool recorded total editing time (minus pauses), keystrokes, and the postedited output; each of the 25 documents was corrected by at least two posteditors.
"The proprietary postediting tool recorded the time taken to postedit each machine-translated document." (Turner, 2015, p. 5)
"The 25 machine-translated documents were each corrected by at least 2 posteditors in order to permit consistency checks across posteditors and computation of average time, adequacy, and fluency ratings per document." (Turner, 2015, p. 4)
Who? #ⓘ
Six native-Chinese-speaking translators recruited from local Chinese cultural organizations, screened for language ability and health experience, with varying levels of translation experience (Table 1).
"From the memberships of local Chinese cultural organizations, 6 Chinese translators were recruited for postediting and screened for language ability and health experience." (Turner, 2015, p. 4)
Other Notes #
The abstract reports the rate as "approximately 41 characters per minute," whereas the Results text and Table 3 give the total mean as 37.8 CPM (Table 3 average 37.4); this EVD uses the body/table figure. Postediting speed is a cost/effort proxy, not a quality measure.