Syntactic Reordering as Pre-processing Step in Statistical Machine Translation of English to Sesotho sa Leboa and Afrikaans

The output quality of statistical machine translation (SMT) depends to a large extent on the quantity and quality of the parallel corpora on which it is trained. In the case of resource-scarce languages where sufficiently large parallel corpora are not always available, alternative ways of improving the output quality of SMT systems must be sought. In this article, one such a method for improving the quality of SMT output is described – introducing a pre-processing step via syntactic reordering of the source language data. This preprocessing involves exploiting certain systematic differences in the syntax of the source and target languages. Apart from describing the method and language-specific rules, we also evaluate the resulting machine translation systems for translating from English to Afrikaans and Sesotho sa Leboa (Sepedi).