© 1989 by Association for Literary & Linguistic Computing
Articles |
Long-term Data for an MT Policy
Pan American Health Organization Washington, DC
Chief, Translation and Terminology, Pan American Health Organization, 525 Twenty-third Street, NW, Washington, DC 20037
With machine translation (MT) at the Pan American Health Organization (PAHO) fully operational from Spanish to English (SPANAM TM) and English Spanish (ENGSPAN TM), four approaches were being considered for meeting the Organization's translation needs in-house and contract MT, and in-house and contract traditional human translation (HT) An eleven-month experiment was undertaken to help PAHO managers decide on the appropriate mix of resources and to establish whether MT in the in-house PAHO environment is in fact cost-effective, fast in turnaround, and as serviceable as HT when postedited by a professional translator.
Two temporary in-house translators were contracted for the period of the experiment, one for each target language. All incoming translations were screened with a view to capturing a large enough volume to measure full-time utilization of MT After 8.5 months, 530 jobs had been received for translation, of which 382 (72 1%) were done using MT A total of 839,635 words had been translated into English and 427,310 into Spanish With in-house postediting of English it was possible to achieve daily outputs of 7,000 words and higher in the short term, while average long-term daily productivity, including periods without incoming translations, was 2,546 words
Most of the receiving units did not know whether their job had been processed by MT or HT Feedback responses showed 85 1% satisfaction with MT compared with 78 1% for HT, while reservations and complaints represented 2. 3% of the total volume done by MT and 25% of that by HT