An anecdote should help illustrate this. I attended a talk by Chomsky when he last visited Australia (January 1995). It was a linguistics talk, which meant he got all obfuscatory: methodological dualism was introduced as equivalent to mental materialism, and vice versa, and if you missed the first two minutes of the talk, you'd have no idea what was going on. At one point, Chomsky said something like "if you're a physicist, you don't stick a video camera out the window and start filming. You conduct controlled experiments. Similarly, if you're a linguist, you don't just turn on a tape-recorded and start taping people. You construct sentences."
In one word: Bollocks.
In a few more words: constructing sentences does give us some insights into language we wouldn't otherwise get --- although the enterprise is fraught with difficulties and fine print. But to say that turning on a tape-recorder and analysing speech as it actually happens is not linguistics, and to say that abstruse artificialities like every sailor loves any sailor are a legitimate target of linguistic research --- and "um" isn't, is lunacy. And to say that in Australia, where linguists are fieldworkers and therefore stick tape-recorders in peoples' faces as a matter of course, will not win you favour...
opoudjis [AT] optusnet . com . au