GPL covers the linguistic data in *spell dictionaries?

There are good annotated vocabulary in {a,i,my}spell dictionaries, giving the lexical class et all.

I'm writing a program that needs a large vocabulary, but with a reduced scope, so I don't need all the information on the *spell dictionaries. Then, I'm reading this dictionary data, processing it and rewriting in my own format. The dictionary I'm using is GPLv2 and I want to release my program in BSD.

As the data itself [the idiom] is public domain, but the representation of data [*spell .dic files] is not, maybe I must follow GPL.

But as I'm not using the GPLd code itself, but a derived trough machine processing from it, maybe I must not.