The First Web-based Language Translator

The proof-of-concept Hobbes’ World Translator (Figure 1) integrated the Globalink machine translation software engine and supported four languages: French, German, Russian, Spanish. The web interface provided an HTML textarea field for entering the text to be translated, and select fields for choosing whether to translate from or to English and the dictionary (glossary) to be used. This was deployed internally to MITRE’s network on Hobbes’ Sun SPARCstation running NCSA HTTPd.

MITRE staff working with Sergey Chapkey of UNOJUST, the United Nations Online Justice Clearinghouse, found out about the web translator and approached Robert H’obbes’ Zakon to adapt his proof-of-concept into a prototype to show how clearinghouse documents could be automatically translated. The prototype was then demoed at the 9th United Nations Congress on the Prevention of Crime and the Treatment of Offenders in Cairo, Egypt, 29 April-8 May 1995.

Around 1993, Zakon developed an open sources coverage and capabilities database (CCDB) for the U.S. intelligence community. Initially using the Gopher networking sharing protocol, CCDB was converted for use on a World-Wide Web server and deployed on the Open Source Information System (OSIS), an information sharing service running on NIPRNet – the government’s unclassified network connecting military and intelligence locations worldwide. The National Air Intelligence Center (NAIC) was a key node on OSIS and one of the funders for the Systran machine translation system. With his experience on OSIS and web language translation, Zakon was tasked with converting his prototype to use the Systran engine for deployment as a service on OSIS. The initial capability (Figures 2 and 3) supported nine languages: Chinese, French, German, Italian, Japanese, Korean, Portuguese, Russian, Spanish. The different iterations developed went by various names, including WebMT, WebManTra, SystranMT, WebTrans, and The Web Translator. Zakon’s OSIS work received multiple honors including a MITRE Director's Distinguished Accomplishment Award and a Program Achievement Award.

The Systran-based web translation service also became available to other nations through OSIS-International (OSIS-I). This effort to expand OSIS to U.S. allies had been stalled for lack of a secure means of interconnecting information networks. Upon his return from one year as a Tech Ambassador visiting MITRE sites worldwide, Zakon was asked if he could make OSIS-I a reality. Using the Netscape Proxy Server in an innovative way, Zakon was able to demonstrate the secure sharing of Web services with untrusted sources by configuring reverse proxies at each peer point. This extended the web translation service to intelligence and military components of Australia, Canada, Denmark, Germany, Israel, Netherlands, and United Kingdom.

Another network that benefitted from the OSIS web translation service was Intelink, the U.S. government’s classified information sharing network, for which Zakon had been an architecture advisor during its conceptual phase. With the web translation service on three worldwide networks, there were now hundreds of thousands of potential users for the web translation service.

In 1996, Zakon hosted the first MITRE Foreign Language Technical Exchange Meeting (FLTEM). This brought together nearly fifty practioners in the areas of multilingual computing and language translation. Among the sixteen presenters were Susann Luperfoy on Voice-to-Voice Machine Translation, Margot Peet on Speech Processing and Foreign Languages, David Vogel on Retrieval of Foreign Language Documents, Marc Vilain on Multi-lingual Name Tagging for Name Finding, and Zakon on Web-Based Machine Translation.

When Altavista released their Babel Fish Web translator (Figure 4) in 1997, it was remarkably similar to the original OSIS translator prototype, even using the same translation engine (Systran), however with a separate field for translating URLs and only supporting five languages: French, German, Italian, Portuguese, Spanish. This was followed a few years later by Google Translate (Figure 4) in 2001, also using Systran until 2007 when Google switched to using its own statistical machine translation system.