Lorena Guerra wrote: >> I was wondering if someone knows of any specific citations in >> publications or any article(s) that discourage MT for High Quality >> translation efforts, and/or those that specifically promote MT systems >> only for dissemination purposes using technical manuals.
Harold Somers replied: >Well almost any (respectable) article on MT will say something >along those lines. With apologies for shameless self-promotion, the >point is made repeatedly by me and other contributors to H. >Somers (ed.) >Computers and Translation: A Translator�s Guide (Benjamins >Translation Library 35), Amsterdam (2003): John Benjamins, I've actually started saying the opposite in recent articles (ie, encouraging the use of MT for high-quality translation) ONLY for the scenario that an MT system with a reasonable amount of MT postediting features is available to the posteditor. Definitely NOT using online web portals that are aimed for content gisting MT usage. See: ALLEN, Jeffrey. 2001. Postediting: an integrated part of a translation software program. In Language International magazine, 13:2, April 2001, pp. 26- 29. ALLEN, Jeffrey. 2002. Review of "Repairing Texts: Empirical Investigations of Machine Translation Post-Editing Processes". (KRINGS Hans, edited by Geoffrey KOBY. 2001. Ohio: Kent State University Press. (abstract and ordering information at http://bookmasters.com/ksu-press/ksu071.htm ) In Multilingual Computing and Technology magazine. Number 46. March 2002. Pp. 27-29. Available at: http://www.multilingual.com/allen46.htm ALLEN, Jeffrey. 2003. Panel talk on Controlled Language and Machine Translation Implementation and Use. Presented at the European Association for Machine Translation and Controlled Language Applications Workshop (EAMT/CLAW 2003). As mentioned during that panel talk, and indicated in the powerpoint presentation, the following article is living proof of MT for high-quality publication. English orginal containing just under 6,000 words was minimally to mid-level-postedited (in between minimal and full postediting) in French in 6 hours (ie 1000 words per hour on average). ALLEN, Jeff. 2002. The Bible as a Resource for Translation Software: A proposal for MT development using an untapped language resource database. In Multilingual Computing and Technology magazine. Number 51, Vol. 13, Issue 7. October/November 2002. Pp. 40-45. English version: http://www.multilingual.com/allen51.htm French translation: La Bible comme Ressource pour les Logiciels de Traduction: Une proposition de d�veloppement des syst�mes de traduction automatique (TA) en utilisant une ressource linguistique inexploit�e http://www.editionscle.com/bol/presse/article1/allen-mltc51-fr.htm Steps conducted: minimal to mid-level postediting, then translation review (by another person) and then final review by the original posteditor. Electronic copies of the French version were made at each hour of the postediting stage, as well as at the end of the postediting stage, the first review stage, and the final review stage. At this rate, a full work day could produce 8000-9000 postedited words, when the posteditor knows the subject very well. Note: My current survey on translation speed amoung many organizations and independent translators indicates an average translation speed of 2000-3000 words per work day. Note: An average typist (30 words per minute) doing full sprint typing for 8 hours (if the person could survive such a duration) could theoretically reproduce from softcopy paper to electronic format a total 14,400 words. However, most typists cannot keep such a typing rate for 8 solid hours. So, postediting can be equal to or better than the typing speed of an average typist. However, we must remember that translation includes cognitive effort beyond simply reproducing what is already written down. In this study, the initial postediting within an MT postediting + review cycle can be performed at nearly 3 times the average translation speed based on the specific conditions below: * the posteditor knows the subject very well. * posteditor is near-native (but not native) speaker of the target language * English to French translation pair * the posteditor is an expert user of the MT tool in question * no dictionary building was conducted for this specific experiment * first step of minimal to mid-level postediting followed then by 2 additional translation review steps to produce a high-quality target language document. A couple of upcoming articles will describe and build upon these results in more detail. Best, Jeff -- For MT-List info, see http://www.eamt.org/mt-list.html
