What Prompts Translators to Modify Draft Translations? An Analysis of Basic Modification Patterns for Use in the Automatic Notification of Awkwardly Translated Text

What Prompts Translators to Modify Draft Translations? An Analysis of Basic Modification Patterns for Use in the Automatic Notification of Awkwardly Translated Text Takeshi Abekawa and Kyo Kageura Library and Information Science Course Graduate School of Education, University of Tokyo {abekawa,kyo}@p.u-tokyo.ac.jp Abstract In human translation, translators first make draft translations and then modify them. This paper analyses these modifications, in order to identify the features that trigger modification. Our goal is to construct a sys- tem that notifies (English-to-Japanese) vol- unteer translators of awkward translations. After manually classifying the basic modifi- cation patterns, we analysed the factors that trigger a change in verb voice from passive to active using SVM. An experimental re- sult shows good prospects for the automatic identification of candidates for modification. 1 Introduction We are currently developing an English-to-Japanese translation aid system aimed at volunteer transla- tors mainly working online (Abekawa and Kageura, 2007), As part of this project, we are developing a module that notifies (inexperienced) translators of awkwardly translated expressions that may need re- finement or editing. In most cases, translators first make draft trans- lations, and then examine and edit them later, often repeatedly. Thus there are normally at least two ver- sions of a given translation, i.e. a draft and the final translation. In commercial translation environments, it is sometimes the case that texts are first translated by inexperienced translators and then edited by ex- perienced translators. However, this does not ap- ply to voluntary translation. In addition, volunteer translators tend to be less experienced than commer- cial translators, and devote less time to editing. It would therefore be of great help to these translators if the CAT system automatically pointed out awk- ward translations for possible modification. In order to realise such a system, it is necessary to first clarify (i) the basic types of modification made by transla- tors to draft translations, and (ii) what triggers these modifications. In section 2 we introduce the data used in this study. In section 3, we clarify the nature of modifica- tion in the translation process. In section 4, we iden- tify the actual modification patterns in the data. In section 5, focusing on “the change from the passive to the active voice” pattern, we analyse and clarify the triggers that may lead to modification. Section 6 is devoted to an experiment in which machine learn- ing methods are used to detect modification candi- dates. The importance of the various triggers is ex- amined, and the performance of the system is evalu- ated. 2 The data The data used in the present study is the Japanese translation of an English book about the problem of peak oil (Leggett, 2005). The book is aimed at a popular audience and is relevant to the sort of texts we have in mind, because the majority of texts vol- unteer translators translate deal with current affairs, social issues, politics, culture and sports, and/or eco- nomic issues for a popular audience1. The data con- sists of the English …

Content truncated to 500 words