English to Urdu Hierarchical Phrase-based Statistical Machine Translation

摘要

This paper addresses the Hierarchical Phrase-based (HPB) models which are used in development of different Statistical Machine Translation (SMT) Systems for many modern languages. Any SMT System needs large parallel corpora for accurate performance. Therefore, availability of a large parallel corpus is a pre-requisite for designing a reliable, robust SMT system between any two languages. The HPB models have shown strong capability of generalization and reordering, which in turn gets improved results for the sparse resourced languages. This paper considers English as Source and Urdu as target language for experiments. For this study, Hierarchical phrase-based Baseline SMT system is used for English to Urdu translation. At the end automatic evaluation of system is performed by using BLEU and NIST as evaluation metrics. Average BLEU evaluation score the developed system got is 13% which is a good competitive score for any sparse resourced language.