@DW
> One of the dangers of contrasting (for example) statistics with logic is that this is really about comparing empiricist and rationalist methods.
I agree that one should not confuse mathematical methods with any kind of dogma — empiricist, rationalist, or theological.
Mathematical methods, including logic and statistics, are neutral with respect to any kind of application. For example, one could apply statistics to bridge either by a priori calculation of the probabilities or by gathering data about how people play the game.
@DW
> … every new science must find a fertile balance between these scientific methods, and the recent swing from “linguistics should be rationalist” to “linguistics should be empiricist” takes us to another glass ceiling.
As a science, linguistics is as old as Aristotle, and the pendulum is always swinging. In the 1950s, information theory and grammar discovery procedures were dominant, and Charles Fries did some very interesting work with the tiny corpora available.
In the late ’50s, Chomsky began his campaign against statistics, information theory, grammar discovery procedures, and finite-state machines. Instead, he promoted “the native speaker’s intuition” (i.e., his own intuition) as the ultimate standard.
Computational linguists have always been more empirical. Even when they used their own intuition to write grammar rules, they tested them by running their systems on actual data. That’s just as empirical as a physicist’s using intuition to write a theory and then testing its predictions against the data.
The ultimate criterion for science is the ability to make predictions about future observations. It’s irrelevant whether the methodology began with intuition, statistical analysis, or some combination of both.
John Sowa
Proofs are availabe.
