Analysing Post-Editing Performance: Correlations with Years of Translation Experience

Analysing Post-Editing Performance: Correlations with Years of Translation Experience Giselle de Almeida Dublin City University Dublin 9, Ireland giselle.dealmeida2@mail.dcu.ie Sharon O'Brien Dublin City University Dublin 9, Ireland sharon.obrien@dcu.ie With the growing use of machine translation (MT), the market for post-editing (PE) is also expanding. However, there is currently little training available for post-editors, guidelines tend to vary from company to company, and there are no internationally adopted quality measurement standards yet. All of these factors Abstract Post-editing (PE) is still a new activity for many translators. The lack of training, clear and consistent guidelines and inter- national standards may cause difficulties in the transition from translation to PE. Aiming to gain a better understanding of these difficulties and using data gathered from a pilot project, this paper explores possible correlations between PE per- formance and previous translation ex- perience. We test a combination of the LISA QA Model and the GALE post- editing guidelines as a typology for clas- sifying post-editing changes implemented by six post-editors for French and Span- ish (three for each target language). This enables a comparison of the types of changes made for the two target lan- guages. We also measure speed and key- board/mouse activity and link those to translator experience. The insight gath- ered may be useful for devising future PE guidelines and training programmes. 1 Introduction © 2010 European Association for Machine Translation. This research is funded jointly by the Irish Research Council for Science, Engineering and Technology and VistaTEC Ltd. through the Enterprise Partnership Programme. contribute to the difficulties encountered by translators when post-editing. The over-riding objective of our research is to model post-editing behaviour for two languages which belong to the same language family in or- der to design specifications for computer-aided support of post-editing, as well as to design a training programme for new post-editors. As part of this research, we conducted a pilot project in June 2009. This was the initial step for a larger scale project, to be carried out in 2010. By ana- lysing PE performance we hope to gather data that may help improve future PE guidelines and training. This paper is structured as follows: Sec- tion 2 provides an overview of the pilot project and its objectives and constraints. Section 3 dis- cusses the profile of a good post-editor and in- troduces the typology we used for classifying the post-editing changes made during the project. Section 4 provides an analysis of the data, in- cluding number and types of changes made, pro- ductivity and keyboard/mouse data. Section 5 presents conclusions and recommendations for future work. 2 Pilot project One aim of the pilot project was to try to gather more insights about the influence that previous translation experience may have on post-editing performance. This stems from the frequently- asked question: do highly experienced (and pre- sumably highly efficient) translators also make for good post-editors? It also aimed at testing our methodology (which involved remote recording of post-editing activity) as well as the typology we customised …

Content truncated to 500 words