Any translator can do a terminologist’s work. The best translators compile lists of terms, equivalents, maybe a piece of context or even a definition before or at least while they are translating. So, theoretically the above statement is correct. But let’s take another look at the focus of a translator and the focus of a terminologist with regard to terms.
Although a term can be at the same time a unit of translation and a term described and defined in a terminology database, translators and terminologists treat that unit differently. A translator works in context and arrives at a target solution that is correct for that particular text. Based on Saussure, Juan Sager calls terms in a translation text “instances of parole” or “language in use” (Routledge Encyclopedia of Translation Studies).
In Quasi dasselbe mit anderen Worten, Umberto Eco says “in light of [all the] meanings made available by a dictionary entry and its applicable encyclopedic information, the translator must choose the most probable, reasonable and relevant sense for the context in question and this possible world” (translation by BIK). That means that the translator cannot simply copy what he finds in a dictionary or terminology database; he actually has to be, as Robin Bonthrone put it years ago, “switched on.” If that wasn’t a condition, machine translation would have long since taken over.
That context then becomes part of the translated text, which in our scenario of technical translation, usually becomes part of a translation memory (TM). And it also becomes part of a product. As part of the product, the term is now part of history, as it were. As part of the TM, the term may be reused for the next version of the product, and it may also serve as reference material to others. But a translation memory does not equate to managed terminology. Strings in TMs contain terminology, but TMs are generally static and hardly ever managed.
In applied terminology, the starting point might be the term in the translation environment above. But a terminologist must research and understand the term not only in one particular context, but in as many as it takes to uniquely identify its meaning. Once that meaning has been identified, the terminologist creates a terminological entry. According to Sager, terminologists use the term, the “instance of parole”, to get to langue, i.e. the abstract system behind the linguistic sign. The entry is part of the terminological system in the database and can now be applied back in parole, in more than one situation or context, to more than one product or company. Therefore, it must be comprehensible to people other than the terminologist, and it must reflect the understanding and knowledge of the subject matter expert (see also Terminology by Maria Theresa Cabré).
While both translators and terminologists research terms, the product of their work is different. The translator is responsible for the delivery of a correct target language text with correct technical terms (parole or language in use). The terminologist is responsible for the creation of a 
correct and complete terminological entry in a database (langue or the abstract system underlying speech acts). That entry may over time be used for many different products and versions inside or outside the company; the entry may become obsolete or even incorrect and the terminologist may need to modify it or add a new entry to the database accordingly. Monetary compensation, as described in What do we do with terms? method and goal of translators and terminologists are different. Therefore, translators translate, terminologists research and document.
[This posting is based on an article published in the Journal of Internationalisation and Localisation, which can be downloaded for free.]
