On Explainable Features for Translatorship Attribution: Unveiling the Translator’s Style With Causality

Translatorship attribution deals with accurately attributing a translation to its translator. The task is challenging because several factors can confound the such as original author’s style, genre, and topic of text. identification translator’s style could contribute fields including studies forensic linguistics. In this paper, we pose translatorship multiclass classification problem employ machine learning algorithms. To address confounding, use corpora English translations same source material (parallel corpora) identify translators’ personal style. We propose two novel feature sets in task: i) list cohesive markers without their surrounding punctuation ii) syntactic n -grams capture real information. $\chi ^{2}$ selection and, using 10-fold cross-validation, assess accuracy classifiers trained our proposed features word, punctuation, POS, POS-punctuation -grams. results show that yield comparable even higher than reported literature on prove are an effective set for task. also recover most distinctive provide examples stylistic interpretations them each Finally, insights from causal inference, where confounding well-defined studied, explanation accepted need parallel contemporaneous different among types features.