The Error Analysis Approach for the Assessment of Automatic Translation

Abstract

The Error Analysis Approach has been widely used to improve the learning of foreign languages and to evaluate human translation, but Error Analysis can also be applied to automatic translation, as shown by different scholars (see M. Koponen 2010). This article reports on a European project, called Organic.Lingua, which tries to demonstrate the potential of a multilingual web portal for Sustainable Agricultural & Environmental Education by using a machine translator in order to make available all materials in different languages. The selected texts are taken from the corpus compiled for this project. The source texts are translated into various target languages using computer translation tools. We have chosen some of these texts in English and their translations into Spanish, and have applied a classification of human errors to check whether the errors generated by machines could be avoided if we can ‘teach’ the machines or to what extent they are specific to computer tools or human beings. Some metric systems, such as BLEU (Bilingual Evaluation Understudy), are being employed to evaluate the quality of the target translation. However, our role as philologers is to concentrate on more specific linguistic features. By defining the different types of errors and by trying to establish a scale of gravity, we intend to determine the quality of the automatic translations. After analysing the data, we will propose linguistic measures for improvement that would be implemented by computer experts working on the project.