Can Crowds Build parallel corpora for Machine Translation Systems

Abstract

Corpus based approaches to machine transla-tion (MT) rely on the availability of parallel corpora. In this paper we explore the effec-tiveness of Mechanical Turk for creating par-allel corpora. We explore the task of sen-tence translation, both into and out of a lan-guage. We also perform preliminary experi-ments for the task of phrase translation, where ambiguous phrases are provided to the turker for translation in isolation and in the context of the sentence it originated from. 1