This paper describes the statistical machine translation system developed at ITC-irst for the evaluation campaign of the International Workshop on Spoken Language Translation 2005. The system exploits two search passes: the first pass is performed by a beam-search decoder which generates an n-best list of translations, the second by a simple re-scoring algorithm. The two passes apply log-linear phrase-based models with an increasing number of feature functions. Runs have been submitted under the supplieddata and manual-transcription conditions for three language pairs: Chinese-to-English, Japanese-to-English and Arabicto- English. Moreover, the Japanese-to-English system has been also employed under the ASR first-best condition. Significant improvements are reported by exploiting alternative word-alignments, and by using novel feature functions in the re-scoring step.