The KIT Translation system for IWSLT 2010

This paper presents the KIT systems participating in the French to English BTEC and in the English to French TALK Translation tasks in the framework of the IWSLT 2010 machine translation evaluation. Starting with a state-of-the art phrase-based translation system we tested different modifications and extensions to improve the translation quality of the system. First, we improved the word reordering by learning POS-based reordering rules from an automatically word-aligned parallel corpus. Furthermore, different experiments to adapt the machine translation system towards the target domain were carried out. In addition, for the BTEC task we tried to avoid data-sparseness problems by using word stems instead of the full word forms.