Abstract
We introduce a method for learning to predict text completion given a source text and partial translation. In our approach, predictions are offered aimed at alleviating users' burden on lexical and grammar choices, and improving productivity. The method involves learning syntax-based phraseology and translation equivalents. At run-time, the source and its translation prefix are sliced into ngrams to generate and rank completion candidates, which are then displayed to users. We present a prototype writing assistant, TransAhead, that applies the method to computer-assisted translation and language learning. The preliminary results show that the method has great potentials in CAT and CALL with significant improvement in translation quality across users. © 2012 Association for Computational Linguistics.