A Simplex Armijo Downhill Algorithm for Optimizing Statistical Machine Translation Decoding Parameters
نویسندگان
چکیده
We propose a variation of simplex-downhill algorithm specifically customized for optimizing parameters in statistical machine translation (SMT) decoder for better end-user automatic evaluation metric scores for translations, such as versions of BLEU, TER and mixtures of them. Traditional simplexdownhill has the advantage of derivative-free computations of objective functions, yet still gives satisfactory 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 efficiently 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 trajectories. We propose to embed the Armijo inexact line search within the simplexdownhill algorithm. We show, in our experiments, the proposed algorithm improves over the widelyapplied Minimum Error Rate training algorithm for optimizing machine translation parameters.
منابع مشابه
Tuning machine translation parameters with SPSA
Most of statistical machine translation systems are combinations of various models, and tuning of the scaling factors is an important step. However, this optimisation problem is hard because the objective function has many local minima and the available algorithms cannot achieve a global optimum. Consequently, optimisations starting from different initial settings can converge to fairly differe...
متن کاملAugmented Downhill Simplex a Modified Heuristic Optimization Method
Augmented Downhill Simplex Method (ADSM) is introduced here, that is a heuristic combination of Downhill Simplex Method (DSM) with Random Search algorithm. In fact, DSM is an interpretable nonlinear local optimization method. However, it is a local exploitation algorithm; so, it can be trapped in a local minimum. In contrast, random search is a global exploration, but less efficient. Here, rand...
متن کاملA Hybrid Machine Translation System Based on a Monotone Decoder
In this paper, a hybrid Machine Translation (MT) system is proposed by combining the result of a rule-based machine translation (RBMT) system with a statistical approach. The RBMT uses a set of linguistic rules for translation, which leads to better translation results in terms of word ordering and syntactic structure. On the other hand, SMT works better in lexical choice. Therefore, in our sys...
متن کاملFast Decoding and Optimal Decoding for Machine Translation
A good decoding algorithm is critical to the success of any statistical machine translation system. The decoder’s job is to find the translation that is most likely according to set of previously learned parameters (and a formula for combining them). Since the space of possible translations is extremely large, typical decoding algorithms are only able to examine a portion of it, thus risking to...
متن کاملTuning Methods in Statistical Machine Translation
In a Statistical Machine Translation system many models, called features, complement each other in producing natural language translations. In how far we should rely on a certain feature is governed by parameters, or weights. Learning these weights is the sub field of SMT, called parameter tuning, that is addressed in this thesis. Three existing methods for learning such parameters are compared...
متن کامل