Selection of Discriminative Features for Translation Texts

Beginning in the first century AD, Buddhist texts underwent a series of translations during a period of nearly 1300 years. The identification of the translator, textual apocrypha, and translation style in Buddhist texts are always important issues. This study proposes an approach to find the most discriminative features that characterize the different Buddhist translation texts or other translation texts. We studied five different kinds of features that can be extracted from translation texts and exploited the F-score and SVM classifier to find the most discriminative features. Not only did we use the translated Buddhist texts, Kalama Sutta, for our experiment, but we also chose The Canterbury Tales to perform the same experiment and compare the results. According to our experiment results, the newly considered fifth-type features are very effective to identify translators. The selected features will be very useful for further studies of translator characteristics.