NEURONAL APPROACH FOR TEXT AUTOMATIC TRANSLATION ARAB TO FRENCH

للكاتبين :

REZEG K. 1, LASKRI M. T. 2

1Department of computer science Mohamed KHIDER University Biskra BP 145, 07000 Biskra,

2Department of computer science BADJI Mokhtar University Annaba BP 12, 23000 Annaba,

ABSTRACT

The automatic translation of the human origin texts is a complex strong implementation called

to apprehend the open universe textual without any constraints towards nature or to their

diversity. To resolve this problematic, several trials have been made every time with the goal

of obtaining the best quality of translation , but in front of the different ambiguities of the

natural language, this problem of translation is too far to be solved. In fact, in absence of the

language mastery of context, most of phrases are ambiguous. By this reason, the current

researches in this domain were oriented at first to the mastery of the meaning vehiculed by

phrases before proceeding to the translation process itself. In fact, translating without

understanding leads directly to failure. In order to elevate the translation quality, we suggest

a neuronal approach for the generation of the different semantic cases related to the different

parts of the phrase to wicle the sens first and to generate the translation in the aimed language

later. It is what allowed us to obtain satisfying results in comparison to the same works using

other technics. Effectively, the neuronal technics have good capacities of training and

generalization, hardiness, mainly tolerance breakdowns of it, and parallel treatment

possibility. In our approach we use a supervised training by a simple recurrent neurons

network to learn translation technics with a performance similar to the human. Our system

contains two phases, the first one is the training phase, the second is the test part and

generalization to measure the network generalization degree of the network according to what

was learnt during the training phase. The uses neurons network hold account the deepened

representation using the semantic cases and the surface representation using elements which

indicate the shape of phrase.

اترك تعليقاً

لن يتم نشر عنوان بريدك الإلكتروني.