Errors in Machine Translated and Crowdsourced Post-Edited Texts

The initial objective of the present study was to identify the most and the least frequent error types in Google Translate (GT) raw outputs and the crowd(sourced) post-edited versions according to Vilar et al.’s (2006) typology. The second objective was to compare the results of error analysis between both outputs in order to address the significance of the decrease in the number of errors in post-edited texts. To this end, four English sports news texts were uploaded on Google Translator Toolkit (GTT), which is an online collaborative environment for post-editing the automatic translations rendered by GT. Subsequently, eleven M.A. students of translation studies which were categorized as unprofessional translators were invited to the online environment via email to modify the machine translations. Results of the error analysis revealed that the two categories of Incorrect Words and Unknown Words were respectively the most and the least frequent error types in both outputs. The study also showed less than fifty percent decrease in the number of errors in post-edited texts. However, some effective factors for improving the quality of crowd(sourced) post-edited outputs and the applicability of GTT were investigated based on the collected literature, an online interview with participants and the researchers’ own observations.