Pattern REcognition-based Statistically Enhanced Machine Translation
The PRESEMT (Pattern REcognition-based Statistically Enhanced MT) project has been funded under "ICT-2009.2.2: Language-based Interaction". It is intended to lead to a flexible and adaptable MT system, based on a language-independent method, whose principles ensure easy portability to new language pairs. This method attempts to overcome well-known problems of other MT approaches, e.g. compilation of extensive bilingual corpora or creation of new rules per language pair. PRESEMT will address the issue of effectively managing multilingual content and is expected to suggest a language-independent machine-learning-based methodology.
Key innovation
The PRESEMT project proposes a novel approach to the problem of Machine Translation by introducing cross-disciplinary techniques, mainly borrowed from the machine learning and computational intelligence domains, in the MT paradigm.
To this end, a flexible MT system will be developed, which will be enhanced with (a) pattern recognition approaches (such as extended clustering or neural networks) towards the development of a language-independent analysis and (b) evolutionary computation (such as Genetic Algorithms or Swarm Intelligence) for system optimisation.
The PRESEMT project proposes a novel approach to the problem of Machine Translation by introducing cross-disciplinary techniques, mainly borrowed from the machine learning and computational intelligence domains, in the MT paradigm.
To this end, a flexible MT system will be developed, which will be enhanced with (a) pattern recognition approaches (such as extended clustering or neural networks) towards the development of a language-independent analysis and (b) evolutionary computation (such as Genetic Algorithms or Swarm Intelligence) for system optimisation.
Features
- Development of a novel method based on generalised clustering techniques, for creating a language-independent phrase aligner also adaptable to phrasing principles designated by the end users.
- Use of pattern recognition approaches for defining syntactic structure.
- Employment of techniques inspired by functional biological systems for disambiguating translations.
- Extensive use of automated optimisation techniques to define a process for methodically optimising system parameters.
- Application of machine learning methods for allowing system adaptation.
- Use of parallel computing architectures as well as mainstream multi-core architectures for PCs for substantial advances in translation speed.
'
Status
Completed
Start Date
End Date
Responsible
George Tambouratzis