Symmetric Probabilistic Alignment

摘要

We recently decided to develop a newalignment algorithm for the purposeof improving our Example-Based MachineTranslation (EBMT) system’s performance,since subsentential alignment iscritical in locating the correct translationfor a matched fragment of the input. Unlikemost algorithms in the literature, thisnew Symmetric Probabilistic Alignment(SPA) algorithm treats the source and targetlanguages in a symmetric fashion.In this short paper, we outline our basicalgorithm and some extensions for usingcontext and positional information, andcompare its alignment accuracy on theRomanian-English data for the shared taskwith IBM Model 4 and the reported resultsfrom the prior workshop.