Improving Statistical Machine Translation through co-joining parts of verbal constructs in English-Hindi translation

Verb plays a crucial role of specifying the action or function performed in a sentence. In translating English to morphologically richer language like Hindi, the organization and the order of verbal constructs contributes to the fluency of the language. Mere statistical methods of machine translation are not sufficient enough to consider this aspect. Identification of verb parts in a sentence is essential for its understanding and they constitute as if they are a single entity. Considering them as a single entity improves the translation of the verbal construct and thus the overall quality of the translation. The paper describes a strategy for pre-processing and for identification of verb parts in source and target language corpora. The steps taken towards reducing sparsity further helped in improving the translation results.