The first generation of machine translation engines have been great travellers. Many of the original rule based systems – some of them now evolved into hybrid statistical/symbolic configurations – first saw the dawn decades ago. They represented a huge investment in person years of design, optimisation, maintenance and redesign. But no one seemed to make genuinely viable businesses out of them. Apart from Systran, the great survivor, more of which anon.
One classic example of an itinerant system is METAL, originally developed at a university in Texas in the 1970s, then acquired by Siemens in Germany in the late 1980s to drive its huge documentation localisation programme, and later sold off in parts (usually language pairs) to various smaller translation companies in Europe over the subsequent years. Today much mutated versions of METAL it are still hard at work in numerous incarnations from Spain to Germany, even though their software core has had to be largely rewritten.
Another fascinating story is the Logos system, originally developed as a very large (and beautifully-crafted) bespoke rule system in the US during the Vietnam war to translate weapons documentation into Vietnamese, and later extended to another strategic language on the weapons agenda – namely Farsi –just as regime change arrived in Iran in 1979. It continued commercially with a German-English pair into the 1990s, but since 2010 has gone open source at SourceForge (led by DFKI experts, offering a massive linguistic resource for developers but (so far) of very little commercial value.
So the outright commercial champion of the first half-century of automated translation software is Systran . Sourced in the early experiments in Georgetown (US) mostly for US intelligence end users, Systran was founded by Peter Toma in 1968 and the company has never disappeared from sight or been dissolved into a larger service supplier. Today it is the great brand name of the world of machine translation.
Like its successors, it too has travelled. The system was partially acquired by a French businessman in the late 1980s, while certain language pairs were owned and developed by translation services at the European Commission as part of a first effort to apply MT to solving the multilingual information barrier in the Europe Union.
Today, Systran is still innovating, and in 2013 won an LT-Innovate Summit Award for its SystranLinks solution, designed to optimize and accelerate the automation of website localisation. Prior to this it was almost certainly the first MT system to be linked up to the early Web, driving the free Babelfish translation service ever since the Internet paleolithic age of 1995. And can even claim to be the first online translation service ever, providing automated translations through the French Minitel videotext service back in the early 1990s.
The company has also weathered the statistical tsunami by extending its technology stack to include data-driven benefits on top of its fundamentally ‘symbolic’ architecture – i.e. one based on the properties of words and phrases rather than on the probabilities of strings of letters.
Systran also offers a useful benchmark for mapping the sector’s monetary value. It is one of the very few publicly-listed companies in its sector (SDL and other similar listed vendors operate in a galaxy of multilingual service markets, not just MT) so its financials are public. In a translation technology software market that LT-Innovate estimates to have grown by some 15.5% in 2012 to a volume of $739.2M, Systran has been posting sales in software and services of around €10M. Other software service suppliers do better.
But remember that Systran has remained independent of the much more valuable translation services segment and focused constantly on improving its core technology and value proposition as a “pure play” MT supplier. Where it has been particularly successful recently is in helping translate content for national intelligence agencies, especially in the US.
The company has run the gamut from offering a free online translation service to providing highly domain-tailored services to enterprises and industries. Systran’s forty-five years of loyal service to its clients in the MT segment constitutes a pretty rare track record. How will it innovate tomorrow?