POLIBITS, Vol. 60, pp. 5-12, 2019.
Abstract: Translation quality, crowdsourcing, translation framework
Keywords: When performed correctly, crowdsourcing can produce near-expert quality translations both quickly and inexpensively; however, the quality obtained from crowdworkers is rarely consistent. We propose a framework to obtain high-quality work from non-expert crowdworkers by incorporating intermediate mechanisms such as ranking and editing in addition to translation. We conduct three empirical experiments in which we explore the impact of these framework mechanisms on translation quality, time, and cost. We also demonstrate that our proposed framework is robust against spammers, verifiable at different steps, and consistent with little variance in quality. Our framework achieved a higher BLEU score than professional translators at a fourteenth of the cost but required about 50 percent more time to complete. Therefore, it is most appropriate when a task requester wishes to maximize translation quality and minimize cost.
PDF: A Framework to Build Quality into Non-Expert Translations
PDF: A Framework to Build Quality into Non-Expert Translations
https://doi.org/10.17562/PB-60-1
Table of contents of POLIBITS 60