Publication - A Review of the PRESEMT project.

A Review of the PRESEMT project.

Research Area:  
Other topics in Computer Science
In Proceedings


Year: 2013
Authors: George Tambouratzis; Marina Vassiliou; Sokratis Sofianopoulos
Editor: Sima'an, K., Forcada, M.L., Grasmick, D., Depraetere, H., Way, A.
Book title: Proceedings of the XIV Machine Translation Summit
Pages: 437
Address: Nice, France
Date: 2-6 September, 2013
PRESEMT (Pattern REcognition-based Statistically Enhanced MT) was a project funded by the EU under the FP7 topic "ICT-2009.2.2: Language-based Interaction". PRESEMT has led to a flexible and adaptable Machine Translation (MT) methodology, overcoming well-known problems of existing MT approaches, e.g. compilation of extensive bilingual corpora. In order for PRESEMT to be easily amenable to new language pairs, only relatively inexpensive, read-ily available language resources, in the form of monolingual corpora as well as bilingual lexica, are used. Since for the majority of language pairs the amount of available parallel cor-pora is very limited, PRESEMT extracts modelling information from monolingual resources. Only a very small bilingual corpus is used to provide information for structural modifications from SL to TL. Translation context is modelled on syntactic phrases, as they have been proven to improve the translation quality.