Deep-syntax tectomt for english-spanish MT

Deep-syntax approaches to machine translation have emerged as an alternative to phrase-based statistical systems, which seem to lack the capacity to address essential linguistic phenomena for translation. As an alternative, TectoMT is an open source framework for transfer-based MT which works at the deep tectogrammatical level and combines linguistic knowledge and statistical techniques. This work describes the development of machine translation systems for English-Spanish in both directions, leveraging on the modules for the English-Czech TectoMT system. We show that it is feasible to develop basic systems with relatively low effort in 9 months. Our evaluation shows that despite not yet being able to beat a phrase-based statistical system, the TectoMT architecture offers flexible customization options, which considerably increase the BLEU scores.