Source-side reordering using a Probalistic Inverse Transduction Grammar (PITG).
Source-side reordering is a Statistical Machine Translation (SMT) approach to account for differences in source and target language word order by imposing a reordering model on the source data as a pre-translation step.
To construct a reordering model various grammars can be used, in this paper we focus on the use of an Inversion Transduction Grammar (ITG). To constrain the model in a linguistically motivated we we impose source-side syntax constraints on the ITG rules that can be derived.
The results of our model given a reordered gold standard on two well-known sentence distance metrics can be examined and analyzed using BitPar.