Error Correcting Parsing for Text-to-text Machine Translation using Finite State Models

This paper describes the approach followed to perform text-to-text Machine Translation (MT) in the first phase of the European project EUTRANS. This project aims at performing text and speech input MT in limited domain tasks. The EUTRANS system relies on Subsequential Transducers (SSTs), which are finite state translation models that can be automatically learned from training samples. Error Correcting Parsing is employed to increase the robustness of SSTs. After reviewing our approach, this paper presents results with the corpora defined within the EUTRANS project.