Translation Memory vs. Example-based MT – What’s the difference?
HAROLD L. SOMERS, María Gabriela Fernández Díaz · LA Referencia (Red Federada de Repositorios Institucionales de Publicaciones Científicas) · 2004
Throughout the 1970s and 1980s, Machine Translation (MT) research focused on the development of so-called “second generation” systems, which aimed to translate text by a process of rule-driven linguistic processing, usually in three stages: syntacticsemantic analysis of the source text, bilingual transfer at a more or less abstract level of representation, and target-text generation from syntactic representation. At the same time, from a practical point of view, there was much discussion on how systems built with this architecture could be used to deliver a reasonable standard of translation usable by real users. Most popular were ideas of restricting the input (the sublanguage and controlled language approaches), or involving the user in preand post-editing. Interactive MT, where the user and the computer would cooperate in resolving ambiguities and making choices, was also championed.