Towards Crowdsourcing translation tasks in library cataloguing, a pilot study

摘要

Although automated translation systems are increasingly impressive they are still far from perfect, and even casual use demonstrates that they can't produce robust, error free, transcriptions of arbitrary text. However it is becoming increasingly apparent that Crowdsourcing of translation tasks is not only viable but, in many cases, provides results equal to more expensive and slower alternatives. This paper provides a brief survey of academic and commercial systems that harness collective intelligence in the translation of text, and presents the results of a pilot conducted using MTurk for the translation of non-Roman scripts of unknown providence. It concludes by proposing a workflow for integration of Crowdsourcing in to the cataloguing of foreign texts.