Combining data-driven MT systems for improved sign language translation

Abstract

In this paper, we investigate the feasibility of combining two data-driven machine translation (MT) systems for the translation of sign languages (SLs). We take the MT systems of two prominent data-driven research groups, the MaTrEx system developed at DCU and the Statistical Machine Translation (SMT) system developed at RWTH Aachen University, and apply their respective approaches to the task of translating Irish Sign Language and German Sign Language into English and German. In a set of experiments supported by automatic evaluation results, we show that there is a definite value to the prospective merging of MaTrEx’s Example-Based MT chunks and distortion limit increase with RWTH’s constraint reordering.