Phrase reordering is a challenge for statistical machine translation systems. Posing phrase movements as a prediction problem using contextual features modeled by maximum entropy-based classifier is superior to the commonly used lexicalized reordering model. However, Training this discriminative model using large-scale parallel corpus might be computationally expensive. In this paper, we explore recent advancements in solving large-scale classification problems. Using the dual problem to multinomial logistic regression, we managed to shrink the training data while iterating and produce significant saving in computation and memory while preserving the accuracy.

In state-of-the-art phrase-based statistical machine translation systems, modelling phrase reorderings is an important need to enhance naturalness of the translated outputs, particularly when the grammatical structures of the language pairs differ significantly. Posing phrase movements as a classification problem, we exploit recent developments in solving large-scale multiclass support vector machines. Using dual coordinate descent methods for learning, we provide a mechanism to shrink the amount of training data required for each iteration. Hence, we produce significant computational saving while preserving the accuracy of the models. Our approach is a couple of times faster than maximum entropy approach and more memory-efficient (50% reduction). Experiments were carried out on an Arabic-English corpus with more than a quarter of a billion words. We achieve BLEU score improvements on top of a strong baseline system with sparse reordering features.

Stochastic gradient descent has been advanced as a computationally efficient method for large-scale problems. In classification problems, many proposed linear support vector machines are very effective. However, they assume that the data is already in memory which might be not always the case. Recent work suggests a classical method that divides such a problem into smaller blocks then solves the sub-problems iteratively. We show that a simple modification of shrinking the dataset early will produce significant saving in computation and memory. We further find that on problems larger than previously considered, our approach is able to reach solutions on top-end desktop machines while competing methods cannot.

In phrase-based statistical machine translation systems, variation in grammatical structures between source and target languages can cause large movements of phrases. Modeling such movements is crucial in achieving translations of long sentences that appear natural in the target language. We explore generative learning approach to phrase reordering in Arabic to English. Formulating the reordering problem as a classification problem and using naive Bayes with feature selection, we achieve an improvement in the BLEU score over a lexicalized reordering model. The proposed model is compact, fast and scalable to a large corpus.