Improving the Performance of GIZA++ Using Variational Bayes
نویسندگان
چکیده
Bayesian approaches have been shown to reduce the amount of overfitting that occurs when running the EM algorithm, by placing prior probabilities on the model parameters. We apply one such Bayesian technique, variational Bayes, to GIZA++, a widely-used piece of software that computes word alignments for statistical machine translation. We show that using variational Bayes improves the performance of GIZA++, as well as improving the overall performance of the Moses machine translation system in terms of BLEU score. This work was supported by NSF grants IIS-0546554 and IIS-0910611.
منابع مشابه
Improving the IBM Alignment Models Using Variational Bayes
Bayesian approaches have been shown to reduce the amount of overfitting that occurs when running the EM algorithm, by placing prior probabilities on the model parameters. We apply one such Bayesian technique, variational Bayes, to the IBM models of word alignment for statistical machine translation. We show that using variational Bayes improves the performance of the widely used GIZA++ software...
متن کاملThe TÜBİTAK statistical machine translation system for IWSLT 2012
We describe the TÜBİTAK submission to the IWSLT 2012 Evaluation Campaign. Our system development focused on utilizing Bayesian alignment methods such as variational Bayes and Gibbs sampling in addition to the standard GIZA++ alignments. The submitted tracks are the ArabicEnglish and Turkish-English TED Talks translation tasks.
متن کاملImproving the Performance of Bayesian Estimation Methods in Estimations of Shift Point and Comparison with MLE Approach
A Bayesian analysis is used to detect a change-point in a sequence of independent random variables from exponential distributions. In This paper, we try to estimate change point which occurs in any sequence of independent exponential observations. The Bayes estimators are derived for change point, the rate of exponential distribution before shift and the rate of exponential distribution after s...
متن کاملOn Variational Bayes Algorithms for Exponential Family Mixtures
In this paper, we empirically analyze the behaviors of the Variational Bayes algorithm for the mixture model. While the Variational Bayesian learning has provided computational tractability and good generalization performance in many applications, little has been done to investigate its properties. Recently, the stochastic complexity of mixture models in the Variational Bayesian learning was cl...
متن کاملFinding hypergraph communities: a Bayesian approach and variational solution
Data clustering, including problems such as finding network communities, can be put into a systematic framework by means of a Bayesian approach. Here we address the Bayesian formulation of the problem of finding hypergraph communities. We start by introducing a hypergraph generative model with a built-in group structure. Using a variational calculation we derive a variational Bayes algorithm, a...
متن کامل