Modelling human translator behaviour with user-activity data

Modelling Human Translator Behaviour with User-Activity Data Michael Carl, Arnt Lykke Jakobsen, and Kristian T.H. Jensen Copenhagen Business School, Languages & Computational Linguistics, Frederiksberg, Denmark carl@iai.uni-sb.de,alj.isv@cbs.dk,kthj.isv@cbs.dk Abstract. Nagao [8] suggests to model machine translation as an imitation of a process that can be observed in human translation behavior. However, not much work has been done in an MT context that looks into human translation processes. This paper seeks to introduce a research method for the investigation of human translation behavior based on User-Activity Data (UAD) which consists of the translator’s recorded keystroke and eye-movement behavior and which can be re- played and analysed off-line. The paper gives the background of this technique and an example on a English-to-Danish translation. Our goal is to elaborate and investigate cognitively grounded basic translation concepts which are material- ized and traceable in the UAD. 1 Introduction In his seminal paper, Nagao [8] suggests a three step procedure in which humans would produce translations and which should be adopted and emulated by the mecanical de- vice: 1) decomposing an input sentence into phrases 2) translating these phrases into other language phrases and 3) composing these phrase translations into one long sen- tence. To confine example-based MT from the predominant rule-based methods, Nagao claims that these three steps are mainly memory-based processes, relying on the experi- ence of the translator, rather than on the analytical knowledge of the languages involved. While today almost all data-driven MT implement some version of this 3-step schema, making use of ’previous’ translation knowledge (i.e. sets of bitexts), it is largely un- clear what and how human translators actually proceed, what chunks they decompose and how they are re-composed into a target sentence. However, a deeper understanding of the human process might also be helpful to advance the state of the art in machine translation. In this paper we first outline some traditional research methods in translation stud- ies and introduce our new method for investigating human translation behavior based on User-Activity Data (UAD), i.e. eye-movement and keystroke data. Section 3 gives an example of an experimental setting which investigates a particular translation diver- gence for English-Danish making use of UAD. 2 Techniques in Translation Research Over the past 25 years, human translation research has focused increasingly on investi- gating translation processes, either from a cognitive perspective or from the perspective 12th EAMT conference, 22-23 September 2008, Hamburg, Germany 21 of managing the process in interaction with new technologies and colleagues in increas- ingly globalised organisations. Using think-aloud as their preferred method of eliciting verbal data and viewing translation as fundamentally a decision-making process (for which the flow-chart was a suggestive analogy), the pioneers of process-oriented research Gerloff [3], Krings [6], and others succeeded in establishing a complex inventory of meaning operations or strategies performed by translators. Krings [7] found that think-aloud delayed transla- tion by about 25% in his experiments. The experiments reported by Jakobsen [1] indi- cate that the think-aloud constraint had a degenerative effect on segmentation. There- fore, …

Content truncated to 500 words