Markers of translator gender: do they really matter?

M Shlesinger, M Koppel, N Ordan, B Malkiel

Research output: Contribution to journalArticlepeer-review

Abstract

Given the impressive record of machine learning in telling male- from female-authored texts in various genres, we asked whether the computer could also be “taught” to tell male- from female-translated texts. Our corpus, downloaded from the website of Words Without Borders, consisted of 273 samples of literary prose translated into English from a variety of languages. We found that despite its ability to isolate particular features of male- vs. female-translated texts, the computer could not be trained to accurately predict the gender of the translator. We see the difference between our results and those for original texts as highlighting the limitations of the classical social-science methodologies; i.e. notwithstanding the successful application of methods for isolating discrete features of male-translated vs. female-translated texts, these features were found to have little or no predictive value when tested in a cross-validation experiment. In other words, the same cross-validation approach that has been shown to be highly predictive in the case of author-gender attribution has proven unreliable for translator-gender attribution. We explore the implications of these results, both with regard to the competing methodologies and in terms of their implications for Translation Studies.
Original languageAmerican English
Pages (from-to)185-198
JournalInternet. Disponível em http://u. cs. biu. ac. il/~ koppel/Publications. ht ml (consultado em 31de março de 2011)
StatePublished - 2009

Fingerprint

Dive into the research topics of 'Markers of translator gender: do they really matter?'. Together they form a unique fingerprint.

Cite this