Parallel texts alignment strategies: the semantic aspects

The paper deals with the problems of design and development of the linguistically motivated mechanisms for parallel texts alignment and grammatical (functional semantic) matches extraction for design of statistical language portraits to be further incorporated into the hybrid models of machine translation. By hybrid models are meant those employing both statistical and rule-based mechanisms within one framework for natural language processing. The presented approach consists in the use of the starting (“seed”) grammar to be further enriched by the matches discovered in parallel texts. The seed grammar is used featuring the cognitive and functional characteristics of phrase structures. The grammar formalism employed is the Cognitive Transfer Grammar based on the transfemes (bilingual phrase structures). The interpreters' transformations existing in parallel texts are of primary importance to the research efforts.