《crowdsourcing-translation》.pdf

Crowdsourcing Translation: Professional Quality from Non-Professionals Omar F. Zaidan and Chris Callison-Burch Dept. of Computer Science, Johns Hopkins University Baltimore, MD 21218, USA ozaidan,ccb@ Abstract Arabic–English, where there is ample data. How- ever, large bilingual parallel corpora exist for rela- Naively collecting translations by crowd- tively few languages pairs. sourcing the task to non-professional trans- There are various options for creating new train- lators yields disfluent, low-quality results if ing resources for new language pairs. These include no quality control is exercised. We demon- harvesting the web for translations or comparable strate a variety of mechanisms that increase corpora (Resnik and Smith, 2003; Munteanu and the translation quality to near professional lev- Marcu, 2005; Smith et al., 2010; Uszkoreit et al., els. Specifically, we solicit redundant transla- tions and edits to them, and automatically se- 2010), improving SMT models so that they are bet- lect the best output among them. We propose a ter suited to the low resource setting (Al-Onaizan set of features that model both the translations et al., 2002; Probst et al., 2002; Oard et al., 2003; and the translators, such as country of resi- Niessen and Ney, 2004), or designing models that