220 likes | 471 Views
Post-editing machine translation – a usability test for professional translation settings. Silke Gutermuth & Silvia Hansen-Schirra University of Mainz Germany. Post-editing?. “term used for the correction of machine translation output by human linguists/editors” (Veale & Way 1997 )
E N D
Post-editing machine translation – a usability test for professional translation settings Silke Gutermuth & Silvia Hansen-Schirra University ofMainz Germany
Post-editing? • “term used for the correction of machine translation output by human linguists/editors” (Veale & Way 1997) • “taking raw machine translated output and then editing it to produce a 'translation' which is suitable for the needs of the client” (one student explaining post-editing to another) • “is the process of improving a machine-generated translation with a minimum of manual labour” (TAUS Report 2010)
Degrees of Post-editing • lightorfastpost-editing • essential correctionsonly • time factor: quick • full post-editing • morecorrections=> higherquality • time factor: slow • (O‘Brien et al. 2009)
Background • Motivation: evaluationofmachinetranslation (MT), post-editingof MT, eye-enhanced CAT workbenches(e.g. O‘Brien 2011, Doherty et al. 2010, Carl & Jakobsen 2010, Hyrskykari 2006) • Project: in cooperationwithCopenhagen Business School (http://www.cbs.dk/Forskning/Institutter-centre/Institutter/CRITT/Menu/Forskningsprojekter) • Experiment: • English-German • translation vs. post-editing vs. editing • 6 sourcetexts (ST) with different complexitylevels(Hvelplund 2011) • 12 professional translators, 12 semi-professional translators • eye-tracking (Tobii TX 300), key-logging (Translog), retrospectiveinterviews, questionnaires
Translators‘ evaluationof MT quality Professional translators: conscious, subjective rating of machine translated output is extremely negative. Can eye-tracking tell a different story dealing with objective and measurable facts?
Processing time edited texts quite often suffer from a distortion of meaning => source text needed for good quality translation => post-editing
Processing metricsTranslation vs. Post-editing Translation: correlationbetweenincreasing ST complexityand TT processingmetrics Post-editing: no significant influence of ST complexity on TT processing metrics
Processing of ST Translationvs. Post-editing => post-editing more efficient WHY?
Fixation Duration ofclauses Average fixationduration (in milliseconds ) per clause
Fixation Duration ofclauses Average fixationduration (in milliseconds ) per clause Goodqualityof MT for non-finite clauses ST: to end thesuffering TT-P: um das Leiden zu beenden ST: Althoughemphasizingthat TT-P: Obwohl betont wird, dass ST: toprotestagainst TT-P: um gegen … zu protestieren ST: in thewakeoffightingflaring TT-P: im Zuge des Kampfes gegen ein erneutesupagain in Dafur Aufflammen in Darfur
PreliminaryConclusions • Efficient post-editing is possible under the following conditions: • good machine translation quality • post-editors who are language experts, i.e. they need • knowledge of the conventions of the source and target language • knowledge of the text type and register
What‘snext? • Analysis ofothercontrastivedifferencesandgaps • Analysis ofambiguitiesandprocessingproblems • Comparisonofcomplexitylevels • Analysis ofmonitoringprocessesduring TT production (withTranslog) • Comparisonof professionals vs. semi-professionals • Correlationsbetweenprocess data and quality of participants’ outputs • Comparisonwithothertranslationpairs
Bibliography • Carl, Michael andJakobsen, ArntLykke (2010): RelatingProduction Units andAlignment Units in Translation Activity Data, In Proceedingsof International Workshop on Natural Language Processing andCognitive Science (NLPCS), Madeira, Portugal. • Doherty, Stephen, O'Brien, Sharon and Carl, Michael (2010): Eye tracking as an MT evaluation technique. Machine Translation, 24, 1, pp1-13. • Hvelplund, Kristian Tangsgaard (2011): Allocation of cognitive resources in translation an eye-tracking and key-logging study. PhD thesis, Department of International Language Studies and Computational Linguistics, Copenhagen Business School. • Hyrskykari , Aulikki (2006): Eyes in Attentive Interfaces: Experiences from Creating iDict, a Gaze-Aware Reading Aid. Dissertation, Tampere University Press. • O'Brien, Sharon and Roturier,Johann and De Almeida, Giselle (2009): Post-Editing MT Output Views from the researcher, trainer, publisher and practitioner. http://www.mt-archive.info/MTS-2009-O’Brien-ppt.pdf • O'Brien, Sharon (2011): Towards Predicting Post-Editing Productivity. Machine Translation, 25, 3, pp197-215. • Postediting in Practice. A TAUS Report, March 2010 p.6 • Veale, T. and Way, A. (1997). Gaijin: A Bootstrapping Approach to Example-BasedMachine Translation. RecentAdvances in Natural Language International Conference, 239-244.
Silke Gutermuth & Silvia Hansen-Schirra gutermsi@uni-mainz.de & hansenss@uni-mainz.de http://www.staff.uni-mainz.de/hansenss/ Contact