Translation of human language has been a goal of computer science from the very beginning of the field. In 1966, the infamous ALPAC (Automated Language Processing Advisory Committee) report, prepared by the National Academy of Sciences, concluded that automated translation was unattainable in the near future. The report recommended that government funding be redirected to basic cognitive science research, a suggestion that squelched U.S. research in automated translation for more than a decade. During that time, however, research continued in Europe and Japan.
In 1989, a 10-year study, supported by the Japanese Ministry of International Trade and Industry (MITI), found that virtually all of the circumstances underlying the ALPAC conclusions had reversed. This study concluded that machine translation is both practical and necessary for specific information search needs (information acquisition) and cost-effective production of certain classes of documents (information dissemination).
Early translation systems performed a word-for-word replacement of target language for source language. Such systems ignored the fact that sentence structure varies widely and that words play more than one role (e.g., noun or verb) and have more than one meaning. The lowest-cost translation systems available today on personal computers still suffer from these limitations. If you decide to play around with one, make sure there is a money-back guarantee.
To keep track of Russian technology during the Cold War, the U.S. Air Force funded development of a translation system by IBM called the Mark II, which performed word-for-word replacement. In 1970, the Mark II was replaced by a program called Systran, which had become operational in 1964. Systran, now owned by Systran Translation Systems, consists of low-level primitives to manipulate human language. Instead of performing word-for- word replacement, the Systran system translates through analysis of the sentence's syntax.
Systran is considering porting its translation program to workstations, but currently you can access it (for a price) via modem. Since 1981, Systran has been developing Japanese-to-English and English-to-Japanese modules. The company offers more than 20 language pairs.
The Pan American Health Organization (PAHO) has developed two translation systems, Spanam and Engspan. Today, these systems are used on a production basis, and they collectively have translated millions of words. PAHO has organized a consortium of public-sector supporters to fund porting of the PAHO system to workstations and to develop additional language pairs.
Logos was originally developed to translate U.S. Air Force equipment manuals into Vietnamese. Today, Logos runs on various classes of computers, including several IBM models and Wang departmental computers, and offers close to a dozen language pairs. Logos uses an internal representation that includes both syntax and semantics in the same data structure. Written in FORTRAN, Logos can be ported to workstations.
Originally developed on DEC VAXes, the Smart line of translation systems now is offered for Sun SPARC workstations and SCO Xenix on 386 and 486 workstations. Smart smoothly interfaces with a number of publishing systems, including Interleaf, FrameMaker, PageMaker, Microsoft Word, and WordPerfect.
In the early 1960s, the University of Texas started development of Metal, a linguistically sophisticated system for German-to-English translation. The result was a huge FORTRAN program that was tremendously resource intensive. In the late 1970s, Metal was rewritten in Lisp, and, in 1980, Siemens acquired the software. In 1989, Siemens introduced Metal as part of an integrated multilingual publishing system that was composed of a Unix workstation and a specialized Lisp-based translation server. Metal preserves markup codes and provides an expert system to help the user update its multilingual dictionary.
Tovna Machines was incorporated in Israel in 1985 to commercialize technology emerging from a research project begun in 1977. Tovna is commercially available for about half a dozen language pairs. Developed on Sun workstations, Tovna is written in C. The Tovna architecture uses a variety of AI methods to both translate and learn new rules by examining how the post-editor polishes the translation. Reportedly, this software has the capacity to create general rules from specific examples, and it gives an expert user the ability to refine the rules over time.
Alpnet offers its Automated Language Processing System software for use on personal computers. Users of ALPS have massive translation contracts with Alpnet and need in-house automation for some portion of the translation task. Alpnet's system provides machine-assisted translation. It works interactively with the translator to provide automated dictionary lookup and sentence-by-sentence translation.
Executive Communication Systems offers a series of toolkits for processing language, at costs ranging from $50,000 to $150,000. Reportedly, users can develop their own customized translation systems and create the necessary lexicons. At the other end of the price range is GTS, by Globalink, possibly the lowest-priced sentence-level translation system available.
BSO Language Translation in the Netherlands has been working since 1982 on prototypes of a Distributed Language Translation system. The company expects commercial versions to be available in 1993. It has ambitious objectives--to be able to translate general business correspondence and technical literature. BSO expects to achieve this goal by equipping DLT with an immense knowledge base and the ability to query the user about the copy. In essence, the system will ask the user to pre-edit copy when it contains ambiguities. Developed on Sun-3 workstations, DLT's software modules are written in C and Quintus Prolog.
In Japan, Fujitsu offers one commercially available translation system, called Atlas G, which uses a syntactic approach. The firm is seeking partners for a second system (still in a multiyear development stage) called the Atlas II. This system will incorporate a massive knowledge base of commonsense information.
Per comentaris i observacions sobre el servidor, poseu-vos en contacte amb l'Administrador WWW