Machine Translationese: Effects of Algorithmic Bias on Linguistic Complexity in Machine Translation

Recent studies in the field of Machine Translation (MT) and Natural Language\nProcessing (NLP) have shown that existing models amplify biases observed in the\ntraining data. The amplification of biases in language technology has mainly\nbeen examined with respect to specific phenomena, such as gender bias. In this\nwork, we go beyond the study of gender in MT and investigate how bias\namplification might affect language in a broader sense. We hypothesize that the\n'algorithmic bias', i.e. an exacerbation of frequently observed patterns in\ncombination with a loss of less frequent ones, not only exacerbates societal\nbiases present in current datasets but could also lead to an artificially\nimpoverished language: 'machine translationese'. We assess the linguistic\nrichness (on a lexical and morphological level) of translations created by\ndifferent data-driven MT paradigms - phrase-based statistical (PB-SMT) and\nneural MT (NMT). Our experiments show that there is a loss of lexical and\nmorphological richness in the translations produced by all investigated MT\nparadigms for two language pairs (EN<=>FR and EN<=>ES).\n