Recognition and characterization of translator attributes using sequences of fixations and keystrokes
Pascual Martínez-Gómez, Dan Han, Michaël Carl, Akiko Aizawa · Benjamins translation library · 2018
Abstract As the demand for high quality translation continues to increase, there is a growing interest in developing effective computer-assisted tools that support translators in their routines. However, the development of such tools requires a deeper understanding of the translation process that allows us to identify the motoric, perceptual and cognitive bottlenecks of the task. With the ubiquitous presence of computers and logging software, researchers could record the keystrokes of translators, which allowed us to analyze the motor activities of text production. However, the translation process is much more complex and keystroke logging can only capture a marginal fraction of the perceptual and cognitive activities of translators, which are often the source of most translation effort. In recent years, eye-tracking devices became more affordable and they were integrated in our logging interfaces, thus giving us access to both eye-movement and keystroke events during a translation session. Here we present a framework where we use keystrokes and eye movements as low-level measurements of translation behavior. These measurements are then interpreted as sequences of higher-level activities that we use to create interpretable models of translation. The parameters of these models are estimated with the objective to maximize the recognition accuracy of several translator personal attributes given measurements of their translation behavior. These classifier models can then be queried with the purpose to retrieve the most characterizing features of translators that distinguish a certain target personal attribute. In this chapter we formalize this framework, show its effectiveness in a real task and point the reader towards possible extensions of this work.