Statistical machine translation into a morphologically complex language

In this paper, we present the results of our investigation into phrase-based statistical machine translation from English into Turkish - an agglutinative language with very productive inflectional and derivational word-formation processes. We investigate different representational granularities for morphological structure and find that (i) representing both Turkish and English at the morpheme-level but with some selective morpheme- grouping on the Turkish side of the training data, (ii) augmenting the training data with "sentences" comprising only the content words of the original training data to bias root word alignment, and with highly-reliable phrase-pairs from an earlier corpus-alignment (iii) re-ranking the n-best morpheme- sequence outputs of the decoder with a word-based language model, and (iv) "repairing" translated words with incorrect morphological structure and words which are out-of-vocabulary relative to the training and the language model corpus, provide an non-trivial improvement over a word-based baseline despite our very limited training data. We improve from 19.77 BLEU points for our word-based baseline model to 26.87 BLEU points for an improvement of 7.10 points or about 36% relative. We briefly discuss the applicability of BLEU to morphologically complex languages like Turkish and present a simple extension to compare tokens not in a all-or-none fashion but taking lexical-semantic and morpho-semantic similarities into account, implemented in our BLEU+ tool.