Efficient Beam Thresholding for Statistical Machine Translation
Deyi Xiong, Min Zhang, Ai Ti Aw, Haizhou Li · 2009
Beam thresholding is a widely-used pruning approach in decoding algorithms of statistical machine translation. In this paper, we pro-pose two variations on the conventional beam thresholding, both of which speed up the de-coding without degrading BLEU score. The first variation is the dynamic beam threshold-ing, in which the beam threshold varies with the length of source sequences covered by hy-potheses. The second one incorporates a lan-guage model look-ahead probability into the beam thresholding so that the interaction be-tween a hypothesis and the contexts outside the hypothesis can be captured. Both thresh-olding methods achieve significant speed im-provements when used separately. By com-bining them together, we obtain a further speedup, which is comparable to that of the cube pruning approach (Chiang, 2007). Ex-periments also display that the dynamic beam thresholding can further improve the cube pruning. 1