Affiliation: Associate Professor at Institute for Logic, Language and Computation at the University of Amsterdam.
The Hierarchical Structure of Translation Data
I will provide an overview of the research conducted within my group on statistical machine translation (SMT) and concentrate on the problem of how to exploit monolingual syntax to improve SMT models. Contemplating our findings over the past years, I will discuss the principles of how monolingual syntax can be made to fit with the hidden, bilingual structure of translation in parallel corpora, followed by a brief review of three successful syntax-driven models that we developed in the past 8 years. Subsequently I will briefly discuss major challenges in our ongoing work, particularly how to model the role of meaning in machine translation. The talk will cater for an audience with general background in language processing and linguistics, and basic knowledge of probabilistic modelling.