Performance Analysis of Different Smoothing Methods on n-grams for Statistical Machine Translation

Smoothing techniques adjust the maximum likelihood estimate of probabilities to produce more accurate probabilities. This is one of the most important tasks while building a language model with a limited number of training data. Our main contribution of this paper is to analyze the performance of different smoothing techniques on n-grams. Here we considered three most widely-used smoothing algorithms for language modeling: Witten-Bell smoothing, Kneser-Ney smoothing, and Modified Kneser-Ney smoothing. For the evaluation we use BLEU (Bilingual Evaluation Understudy) and NIST (National Institute of Standards and Technology) scoring techniques. A detailed evaluation of these models is performed by comparing the automatically produced word alignment. We use Moses Statistical Machine Translation System for our work (i.e.Moses decoder, GIZA++, mkcls, SRILM, IRSTLM, Pharaoh, BLEU Scoring Tool). Here machine translation approach has been tested on German to English and English to German task. Our obtain results are significantly better than those obtained with alternative approaches to machine translation. This paper addresses several aspects of Statistical Machine Translation (SMT). The emphasis is put on the architecture and modeling of an SMT system.