Machine Translation: New Approaches to Parameter Optimization and Alignment

Abstract

The aim of this research is to improve the translation process. On the one hand, the standard loglinear model parameter optimization process, the MERT algorithm, presents high computational costs, so, as an alternative we propose a novel technique based on Support Vector Machines. This procedure calculate the coefficients of a log-linear combination that minimize a desired loss function. On the other hand, we present a new statistical machine translation alignment model that is an extension to IBM Model 1 to train word-to-word lexicon probabilities. This model takes into account a given fixed segmentation of the source and target sentences in the estimation of the statistical dictionary. Aditionally, we describe the process to create a translation system, from crawling the Internet to search for suitable texts to the training of the final translation system.