Towards incorporating language morphology into statistical machine translation systems

In this paper, a novel algorithm for incorporating morphological knowledge into statistical machine translation (SMT) systems is proposed. First, word stems are acquired automatically 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 statistical framework. The combined lexical and morphological SMT system is implemented using late integration and lattice re-scoring. The system is then evaluated on the Europarl corpus, using automatic evaluation methods for various training corpus sizes. It is shown, that both the BLEU and NIST scores of the lexical-morphological system improve by about 14% over the baseline English to Greek translation system when using a 1M word training corpus