A Hybrid Approach for Word Alignment in English-Hindi Parallel Corpora with Scarce Resources

This paper presents an approach which improves the performance of the word alignment with scarce resources for English-Hindi language pair. We obtain an improvement in the performance of IBM Model 1-2 algorithm by applying part of speech (POS) tag prior to the computation of word alignment probability. This paper demonstrates the increase of precision, recall and F-measure by approximately 15%, 11%, 14% respectively and reduction in Alignment Error Rate (AER) by approximately 14% with IBM Model 1. Similarly it shows an increase of precision, recall and F-measure by approximately 6%, 6% and 6% respectively and reduction in Alignment Error Rate (AER) by approximately 6% with IBM Model 2. Experiments of this paper are based on TDIL corpus.