Experimenting the use of catenae in Phrase-Based SMT

Following recent trends on hybridization of machine translation architectures, this paper presents an experiment on the integration of a phrase-based system with syntactically-motivated bilingual pairs, namely the so-called catenae, extracted from a dependency-based parallel treebank. The experiment consisted in combining in different ways a phrasebased translation model, as typically conceived in Phrase-Based Statistical Machine Translation, with a small set of bilingual pairs of such catenae. The main goal is to study, though still in a preliminary fashion, how such units can be of any use in improving automatic translation quality.