Paper: A Simplex Armijo Downhill Algorithm for Optimizing Statistical Machine Translation Decoding Parameters

ACL ID N09-2006
Title A Simplex Armijo Downhill Algorithm for Optimizing Statistical Machine Translation Decoding Parameters
Venue Human Language Technologies
Session Short Paper
Year 2009
Authors

We propose a variation of simplex-downhill algo- rithm specifically customized for optimizing param- eters in statistical machine translation (SMT) de- coder for better end-user automatic evaluation met- ric scores for translations, such as versions of BLEU, TER and mixtures of them. Traditional simplex- downhill has the advantage of derivative-free com- putations of objective functions, yet still gives satis- factory searching directions in most scenarios. This is suitable for optimizing translation metrics as they are not differentiable in nature. On the other hand, Armijo algorithm usually performs line search ef- ficiently given a searching direction. It is a deep hidden fact that an efficient line search method will change the iterations of simplex, and hence the searching trajectorie...