The problem with science focused exclusively on correct predictions is that it encodes implicit assumptions, and if the surrounding conditions change the previous evidence is no longer valid. It also has a tendency to lead people to think they are observing some final and universal Truth, when really they are just observing a system that is in a relatively stable equilibrium.
As for the second paragraph, I completely do not understand what it means.
Later on, something happens that forces him to wear clothes like everyone else is wearing. And how people react to him changes completely. All his testing proved that people react one certain way, but he didn't understand why so he didn't realize what his testing had showed was actually just a small subset of the actual behavior and was actually an extremely rare case.
Edit: oh well, what the heck, maybe mods can delete this if it's inappropriate. http://www.putlocker.com/file/718CD52BD479CE4A#, the joke is at minute 2:00!
The second paragraph was me trying to, very briefly, sum up the consequences of positivist epistomology. The alternative that Chomsky proposes is that observational knowledge can only describe the things that are true given the process that generates the interactive behaviors we observe. This applies to everything from humanities and linguistics to physical interactions (which is why people have such difficulty explaining quantum physics). Consider, for example, the idea that computers think in zeros and ones: it is usually plenty accurate, even though it is really an abstraction of high/low voltage signal. But if we encounter random bit flips it may be helpful to know that actually we are working with high/low signal and so a cosmic ray interacting with our RAM could produce a flip. Equally, we write programs in higher level languages without worrying that loops aren't an inherent property of the computer. "Loops" aren't a lie, but they also aren't a fact: they are useful knowledge.
Most knowledge falls into this area, of being the product of current, generative interactions. In this world it is possible to say that some statement isn't supported by evidence, so we can still have science based on falsifiablity. We can even fully describe the observed behavior of some system under a set of conditions. That knowledge is likely to be useful, even. But it is incomplete, and shouldn't be elevated to a place of infallible "fact" lest we miss the additional capabilities we might acquire by manipulating those assumptions.
Does Chomski at any point addresses these issues as well? http://www.scribd.com/doc/41399205/Alfred-Korzybski-General-... (jump to p86).
that version of the book is incomplete, but page 86 seems alright.
Edit: by the way, I'm a bit suspicious of your phrasing in "mathematical formalism in his linguistics". Mathematical verbalism does not qualify as sane. I know a case of a person who does speak as a mathematician (very convincing!), even uses mathematics in his expositions, yet he is completely insane and what he says makes no sense whatsoever. I can share a recording of one of his lectures if you wish, but unfortunately he speaks in Romanian.