An English-lingala Rule-based Machine Translation On Apterium Platform

Globalization being the real tendency,there is the need of sharing information worldwide in a short time.Unfortunately,there are many different languages in the world.The need to retrieve the information in all those different languages has motivated the field of machine translation.Machine translation is the mean by which one can automatically translate from a given natural language to another one,retrieving the meaning intended by the source language in the translated verse.The present work describes the development of a shallow-transfer rule-based machine translation translating sentences in English into sentences in the Lingala language.Apterium is the platform used to implement the system.The platform consists of eight modules responsible for the analysis,the transfer,and the generation of the English sentence into the Lingala language.The morphological analyzer performs the analysis process,and the lookup is done in the English dictionary.The structural transfer deals with the word order in the sentence of the target language and uses the transfer file.The lexical transfer looks up the bilingual dictionary for the corresponding translation.The morphological generator generates the correct form of the Lingala sentence by looking up the Lingala Dictionary.The definition of lexical selection rules is done in the same file as the definition of the structural transfer rules.The platform was intended to deal only with romance language of the Liberian Peninsula.These languages have their word inflection happening as a suffix making all the modules used to this structure when dealing with paradigms.In the Lingala language,it is not so;the inflection occurs as a prefix.A de-normalization of the lexical unit is done to allow the correct generation of words in the Lingala language.We evaluated the obtained system using the BLEU metric proposed by IBM with a selected corpus of 300 English sentences.These sentences were first given to a human translator to obtain the reference translation and then fed into the system for translation;we obtained the hypothesis translation.The BLEU metric was run in the hypothesis and references;an accuracy of 90.63% is obtained.