Chapter prepared in 1994 and ultimately published as a re-worked version in Wright, S.E., and G. Budin [eds.] Handbook for Terminology Management: vol. 2: Application-Oriented Terminology Management, Amsterdam/Philadelphia: John Benjamins, 2001, pp. 697-723) TERMINOLOGY AND MACHINE TRANSLATION Muriel Vasconcellos, Brian Avey, Claudia Gdaniec, Laurie Gerber, Marjorie León, and Teruko Mitamura 1. Introduction Machine translation will be defined here as the technology whereby computers attempt to model the human process of translating between natural languages. The computer, rather than a person, generates the "output" though it is only a rough draft, not yet fit for most types of consumption. The draft is usually polished into final form by a translator or a bilingual editor, though in some cases it may be used directly by a technical expert who is gathering data for ongoing research. If the right terminology has been supplied to a machine translation (MT) system, the target- language equivalents are retrieved not only automatically but also in their correct place in the output document. This is one of the advantages of MT: it dispenses with the need to look up terms, whether in hard-copy dictionaries or on-line. With large projects and multi-translator teams, fully utilized MT has proven to be very effective for keeping terminology uniform (e.g., Brace 1993). Indeed, in a group of recent reports from 36 MT users (Vasconcellos 1993), this was the advantage most often cited. Alternatively, even if a translator prefers to create translation in the traditional way, an MT printout can be a valuable guide, assuming that the system's database is well stocked with the required vocabulary. It's worth the spelling assistance alone. To cite a few well-known bugbears in medical terminology, it makes the right choice between gluco- and glyco-; it saves looking up erysipelas and paracoccidioidomycosis; and it never gives *military tuberculosis for tuberculosis miliar – a common translator error. The question is not so much whether MT is useful for banking and retrieving terminology, but rather what is involved in building up the terminological reserve so that it can produce needed terms on demand. The "opposite" of MT, so to speak, is MAT, or machine-aided (also called machine-assisted) translation, in which people do the translation and computers stand at the ready to look up isolated terms on command. The MT-MAT distinction is not water-tight, however, and recent developments are blurring it even more. Innovative approaches are making MAT more MT-like and vice versa. In the case of MAT, the computer is taking on an increasingly active role: systems have been developed that monitor human translation input and identify identical or similar strings of text in a database of previously stored documents, which are then flashed on the screen as candidates to be pasted into the ongoing translation. Meanwhile, in the MT world, interactive systems are finding ways in which to capture needed human reactions in midstream in order to produce output of better quality. For present purposes, however, the distinction between MAT and MT is quite useful. It defines the difference between terminology management software, which is covered elsewhere in this volume, and full-text MT, dealt with here, in which surrounding context interacts with terminology and, as a result, terms cease to be isolated entities.