Machine translation and human translation often encounter the same difficulties at the level of the word and sentence. This chapter describes a methodology which encourages students to select examples of such problems from corpora, evaluate the performance of on-line machine translation engines with them, and compare the results with those of human translation. This has proved a way of providing training in linguistic analysis and the use of corpora as language resources in an interesting way. The methodology has been made possible by the project Linguateca, and the computational tools it develops also offer opportunities for research projects in the areas of human and machine translation, terminology extraction, corpus linguistics, contrastive linguistics and related areas.
It is clearly essential for future translators to learn to use the available translation technology but, given the many linguistic skills translators also need to acquire, the process needs to focus not just on the ability to use the technology, but also on encouraging a good understanding of the objectives, possibilities and limitations of the technology itself. Skilful use of the technology will come later after practice in the professional contexts in which it is needed. An understanding of the problems posed by integrating technology into both the curriculumand teaching practice led us to develop the Corpógrafo at the PoloCLUP of Linguateca1 at the University of Porto. It allows for the building and analysis of parallel and comparable corpora, extraction and management of terminology, as well as the collection and analysis of lexical items, particularly multi-word expressions that are relevant to text, genre or discourse analysis. It is a research environment for autonomous study, but it also offers various possibilities for education in translation, text analysis and terminology management, and has the advantage over commercial software of being freely available online.