Experimenting with reordering model of phrase-based machine translation system for english to Hindi

Abstract

Hindi is the national language of India. However, most of the government records, resolutions, news, etc. are documented in English which native urban users may not understand. This fact motivates to develop an automatic language translation system from English to Hindi. Grammatical structure of the Hindi language is very much complex than the English language. This structural difference makes it difficult to achieve good quality translation results. Translation, reordering and language model are the main working components of a translation system. The translation quality depends on how these individual components of the system are configured. Many times the values of these components are language-dependent. Hence, proper settings of these components are very much essential. This paper discusses various settings of the reordering model and through experimentation demonstrates the proper values of the parameters for getting a quality translation from English to Hindi. The freely available n-gram-based BLEU metric and TER metric is used for evaluating the results.