Évaluation de la qualité de la traduction automatique franco-polonaise - Typologie des erreurs

Abstract

This paper evaluates the performance of Translator jêzyka francuskiego, a French–Polish translator program, edited by Kompas in 2001. Articles from French papers were selected for translation by the program according to the domain they represented, i.e., computer science, biology and law. The quality of their translation was evaluated on the basis of predefined criteria of comprehensibility and faithfulness. This study establishes an error typology reflecting three levels of the analysis made by the program during the translation process of each lexical unit present in the French articles. In order to perform a correct translation, the program has to first recognize a unit as either simple or complex, and next assign to it a grammatical function and meaning. Since errors can appear at each level, our typology reflects all these stages and we can classify all the incorrectly translated lexical units according to their semantic value and the grammatical functions they assume. Our investigation has allowed us to distinguish the following groups of the incorrectly translated lexical units: complex lexical units recognised by the program as simple ones, simple polysemous lexical units, and idiomatic expressions. Each of these groups consists of a few subgroups that correspond to the grammatical categories such as pronouns, articles or prepositions.