Predicting difficulty in translation : a pilot study

In this pilot study, we investigate a number of features that have been suggested as indicators of translation difficulty, namely error count, word translation entropy and syntactic equivalence. We correlate these product features with translation process features such as duration, editing, and gaze information. The dataset that we use contains both translations of professional translators and students. The product data contains manual error analyses of the translations as well as automatically calculated entropy values and syntactic reordering metrics whereas the process data consists of keystroke and eye tracking data together with general duration and pause information. In addition to correlations between product and process data, we also compared the datasets of the professional translators and students