Extending Feature Decay Algorithms Using Alignment Entropy

摘要

In machine-learning applications, data selection is of crucial importance if good runtime performance is to be achieved. Feature Decay Algorithms (FDA) have demonstrated excellent performance in a number of tasks. While the decay function is at the heart of the success of FDA, its parameters are initialised with the same weights. In this paper, we investigate the effect on Machine Translation of assigning more appropriate weights to words using word-alignment entropy. In experiments on German to English, we show the effect of calculating these weights using two popular alignment methods, GIZA++ and FastAlign, using both automatic and human evaluations. We demonstrate that our novel FDA model is a promising research direction.