Statistics-based language translation supports growing international communications needs

Manufacturers doing business internationally often rely on third-party language service providers—or LSPs—to translate owner's manuals for them. With the advent of advanced computer translating capabilities, the process is changing, and some companies are even bringing the technology in-house.

Don DePalma, chief research officer for Lowell, Mass.-based Common Sense Advisory, says, “While automated translation is a more accurate term—because the computer automates some aspects of translation—machine translation is still a more common term,” explains DePalma. There actually are various types of machine translation, including rules-based, context-based, and statistical machine translation (SMT). This third technology, which uses mathematics and algorithms to compare bodies of documents and find commonalities, is gaining the most traction.

Rules-based translation, which attempts to correlate the rules and words of one language with those of another, suffers from there being so many exceptions and syntax problems.

“Statistical machine translation is the preferred market approach for anyone introducing technology,” says DePalma. “Companies that use rules-based systems are adding statistical capabilities to their engines.”

SMT was created in 2002 by Language Weaver. Using already-translated text as a reference, its solution makes alignments and matches, takes account of context, and builds a probability table that leads to the translation.

Prior to introduction of SMT, it could take an LSP about 14 days to translate 3,000 pages of documents and technical manuals. With Language Weaver, the LSP translates the same number of pages in seven to 10 days.

“Manufacturers should consider an LSP that uses statistical machine translation because it will offer time savings, which is obviously a cost savings,” explains DePalma.

Because of what SMT can offer—and the competitive nature of international markets—Language Weaver is seeing considerable interest in its technology from manufacturers themselves.

According to Kirti Vashee, a company VP for Language Weaver, “When companies only translate what is absolutely necessary to support international business activities, they normally use high-quality but relatively expensive human translation. But by not translating more, they leave opportunities on the table that may lead to higher productivity and revenues, as well as better communication.”

Language Weaver sees opportunities for manufacturers in four areas:

• Productivity systems (advertising & marketing literature, product documentation, and software user interface);
• E-solutions (technical bulletins, service information, training materials, and customer support);
• Communication systems (email and instant messaging); and
• Enterprise databases (Human Resources and ERP).

“LSPs are focused on the specific projects like producing documentation,” says Vashee. “Manufacturers see a much larger picture, and realize that language is an issue that affects international business success as a whole. Reducing language as a barrier facilitates international business.”

For example, when Language Weaver began working with a large PC hardware manufacturer, it started with the company's localization department—i.e., the people who produce manuals. When the localization department saw the potential for the technology, they realized it would be useful in translating a lot of other documents. “We can help companies with the rate at which information of all kinds becomes available in the global market,” says Vashee.

In 2007, the Society of Automotive Engineers approved J2450-Quality Metric for Language Translation of Service Information. Says Liz Bergeron, chair of the J2450 task force, “Companies are interested in processes that translate more content faster, while continuing to meet standards of quality,” states Bergeron. “Because the J2450 is process-independent, it can be uniformly applied to the output of any process used to generate technical translation, including statistical translation.”