Translationese Traits in Romanian Newspapers: A Machine Learning Approach.
Iustina Ilisei, Diana Zaiu Inkpen · 2011
Abstract. This paper presents a machine learning approach to the investigation of the translationese effect on Romanian newspapers texts. The aim is to train a learning system to distinguish between translated and non-translated texts. The classifiers achieve an accuracy well above the chance level, the results confirming the existence of translationese manifestation. Also, the experiments investigate whether there are any traits of the simplification universal within translated text. The learning system is enhanced with features previously proposed to stand for this universal, and their impact on the learning model is assessed.