Towards incorporating language morphology into statistical machine translation systems

In this paper, a novel algorithm for incorporating morpho-logical knowledge into statistical machine translation (SMT) systems is proposed. First, word stems are acquired au-tomatically for the source and target languages using an unsupervised morphological acquisition algorithm. Then a word-stem based SMT system is built and combined with a phrase-based word level SMT system using a general sta-tistical framework. The combined lexical and morphologi-cal SMT system is implemented using late integration and lattice re-scoring. The system is then evaluated on the Eu-roparl corpus, using automatic evaluation methods for var-ious training corpus sizes. It is shown, that both the BLEU and NIST scores of the lexical-morphological system im-prove by about 14 % over the baseline English to Greek tran-lation system when using a 1M word training corpus. 1