Rich morpho-syntactic descriptors for factored machine translation with highly inflected languages as target

The baseline phrase-based translation approach has limited success on translating between languages with very different syntax and morphology, especially when the translation direction is from a language with fixed word structure to a highly inflected language. There are two main points to improve on: morphological translation equivalence and long range reordering. Translating the correct surface form realization of a word is dependent not only on the source word-form, but it also depends on additional morpho-syntactic information. In addition, the rich morphology of a highly inflected language permits a flexible word order, thus making difficult to model long range word order differences between languages.