There are things that computers do so badly that its almost, but not quite, funny. Language translation may have pioneered that category 30-plus years ago, when my "How and Why Wonder Book of Robots and Electronic Brains" included the same example still used today: It was claimed, in that book as in many anecdotes since then, that an early English-to-Russian translator turned "The spirit is willing, but the flesh is weak" into "The vodka is good, but the meat has spoiled."
The "spirit is willing" story is probably just a persistent myth--ironically, one with its roots in a comment about incompetent translations in Parisian newspapers of the late 1930s. But despite decades of interest, and the growing commercial value of real-time language translation in international markets, language translation is still expensive to do well and easy to do quite badly.
The most convincing translation work that Ive seen is by Roger Schank, who demonstrated useful performance in translating newspaper stories by understanding the structure of the story first--and only then mapping the tokens in that structure from one language to another. Many silly mistakes of literal, word-by-word translation dont happen when the translator works with a model of roles and relationships, and not just a string of nouns and verbs.
But when we leave the simple, Who-What-When-Where model of a newspaper story to translate a live conversation between two people, Id worry about whether those people even had those primitive tokens in common with each other. Last week, for example, I mentioned my familys occasional use of the fictional Klingon language: Conversational Klingon has no real equivalent of the English "Hello." If you observed Klingons encountering each other, you might think that you had identified their word of greeting, but the expression that they most often use translates better as "What do you want?" If you encountered a Klingon and offered a friendly "Whats happening?" what hed hear from your Pocket PC running IBMs ViaVoice (with the as-yet unoffered Klingon option) might be something more like a starship captains "Report!"
Its tempting to assume that the hardships of translation, combined with the prevalence of English speakers in designing information systems, will lead to a convergence on what some have called "World English" as the worldwide language of business. Thats already the case, many would say, in international science, and English fluency is commonly assumed in air transportation--with unfortunate results when that assumption proves incorrect.
Living as I do in the Los Angeles area, I have a different view: Even in the tiny neighborhood within two blocks of my home, Id be on much more friendly terms with my neighbors if I could speak Chinese, Korean, Japanese and Farsi as well as English and high-school Spanish. (Oh, and a few words of Klingon--which I rarely get to use.)
Enterprises with international ambitions must consider the opportunities, as well as the costs, of language translation technologies--especially when Asian markets are the worlds largest growth opportunity. Someday, translation wont just be a stupid computer trick.