A new machine translation company, LinguaSys, announced its existence today. Responding to market requirements for MT software, CEO Brian Garr emphasized “extremely high quality” and the ability to “create whole new languages in a matter of months for an extremely low price.” We spoke to him about these oft-heard claims for his company’s product, Carabao (a type of Asian water buffalo).

What intrigued us about Carabao was the “creation myth” of the company — and the promise that it holds for users. CEO Garr most recently had been IBM’s Program Director for the WebSphere Translation Server (not the technology licensed to Lionbridge) and, prior to that, had been CTO of Globalink, a 1990′s era MT company acquired by Lernout & Hauspie. Carabao’s developer and LinguaSys CTO, Vadim Berman, created the MT technology based on deep sentence linguistic analysis he developed for a search engine tool currently sold by Digital Sonata in Australia.

When Garr left IBM, he ran into Berman and realized that they had half an MT engine in their hands. The two scoped out a more complete engine, had Berman add transformation and synthesis, and then hooked up with Can Unal in Germany who had a hosted data center offering Sugar CRM. The next step was TransGen MT Suite, a web services-based user interface for using Carabao in chat, document translation, web page translation, and other collaborative tools (from a naming perspective, TransGen is a bit of an inter-cultural eyebrow raiser).

As they developed their hybrid (statistical and rules-based) engine over the last year, the team decided not to rush into the starting-to-get-crowded MT market for French, Italian, German, and Spanish (FIGS) but instead to focus on Chinese, Japanese, and what Garr said were “some languages we cannot discuss.” Because LinguaSys has teamed up with the MAPA Group to sell to the U.S. government and NGOs, the list of other languages shouldn’t be too hard to guess. Garr told us that Carabao’s runtime executable is language-agnostic in that it stores no linguistic data, thus promising rapid market-driven rollouts of new language pairs. He claims that new domain-specific language pairs can be created in less than 90 days.

Garr has brought in some other ex-IBMers to sell the product. He has also enlisted some non-traditional kick-off partners, including the Dray Group which builds iPhone and iPad applications for the enterprise and Managing Worldwide, a Paris-based consultancy.

One thing that’s clear from this announcement is that the MT market is far from a done deal. While some people wonder whether they should just walk away from machine translation in light of Google and Microsoft’s investment and open-source Moses, companies such as AppTek, Asia Online, Language Weaver, PROMT, and SYSTRAN continue to innovate (see “Developers Work to Broaden Use of Machine Translation,” Mar10).