Abstract: This thesis on Statistical Machine Translation (SMT) has shown it is possible to combine succesfully hierarchical decoding with finite-state technology. On the other side, refinements on hierarchical grammars lead to faster translations with similar performance. The combined strategy has worked very well on several language pairs, including Spanish-English translation tasks, leading in many cases to systems that are search-error free, with state-of-the-art performance.
Index Terms: statistical machine translation, hierarchical de- coding, finite-state transducers.