8.4 Machine-aided Human Translation

Existing products now are those of Trados (MultiTerm), IBM (Translation Manager), and SITE-EuroLang (EuroLang Optimizer). They are available on PC/Windows, PS/OS2, or Unix-based workstations. The intended users are competent translators working in teams and linked through a local network. Each translator's workstation offers tools to: access a bilingual terminology. access a translation memory. submit parts ot the text to an MT server. These tools have to be completely integrated in the text processor. The software automatically analyzes the source text, and attaches keyboard shortcuts to the terms and sentences found in the terminogical data base and in the translation memory. One very important design decision is whether to offer a specific text processor, as in IBM's Translation Manager, or whether to use directly one or more text processors produced by third parties, as in EuroLang Optimizer. The server supports tools to: manage the common multilingual lexical data base (MLDB), often a multilingual terminological data base (MTDB), and the common translation memory, where previous translations are recorded. Here, concurrent access and strict validation procedures are crucial. manage the translation tasks (not always offered). Let us take the case of the most recent product, EuroLang Optimizer. One instance is available on Sun workstations under Unix. The server uses a standard DBMS (data base management system) (Oracle or Sybase) to support the terminological data base and the translation memory. The translator's workstations use Interleaf or Framemaker as text processors, while their data base functions are degraded versions of those of the servers, and are implemented directly in C++. In the other instance, the server runs on a PC under Windows NT, again with Oracle or Sybase, while the translator's workstations use Word 6 on PCs under Windows 3. Source languages currently include English, French, German, Italian and Spanish. There are 17 target languages (almost all languages written with the Latin character set). When a document has to be translated, it is preprocessed on the server, and sent to a translator's workstation with an associated kit, which contains the corresponding subsets of the dictionary and of the translation memory, as well as (optionally) translation proposals coming from a batch MT system. MAHTrelated functionalities are accessible through a supplementary menu (in the case of Word 6) and keyboard shortcuts dynamically associated with terms or full sentences. The translator may enrich the kit's lexicon. When translation is completed, the document is sent back to the server with its updated kit. On the server, the new translation pairs are added to the translation memory, and updates or additions to the dictionary are