I''m afraid that the analogy might go like this:

the 1950s implementations of “Fully Automatic High-Quality Machine Translation" (FAHQT) are to Babelfish

as

the 1960s BASEBALL or LUNAR or maybe the 1990s START Natural Language Question Answering Systems will be to wolframalpha .

That is to say, the first 80-90% is easy, but the rest is reely, reely hard.

But I gotta admit that Wolfram is plenty smart and well-staffed.