"Generative" serves them right

[Karel Oliva]

Hi,

being almost born as a dissident (beware, youngsters: this is *not* a kind
of a toothpaste !), I have to disagree with the general opinion, as usual

:-)


Liz, you have nothing to apologize for.

To me, "generative" is an *extremely felicitous* term for the kind of
grammar as GB/P&P/MP and the like. I guess that even Homer, had he
mentioned GB in Illias & Odyssey, would have sticked to this epitheton
constans, as his did with the "light-footed Achilleus" or "thunder-reigning
Zeus" (sorry for the miserable translations, I indeed did not read these
works in English).

The point about GB/MP and the like is that (in the above epic sense) they
indeed very much deserve the attribute "generative" because they are -  at
least as long as they rely on transformations, which is the rule -
GENERATIVE-ONLY. It is their distinguishing property. In the sense that one
cannot really PARSE with them, definitely not if one takes this task
seriously (yes, I do know there were attempts by Barton, Weinberg, Millies
and others, but I also do know they all were not really successful to put
it mildly - correct me if I am mistaken, at best by pointing out a working
GB/MP wide-coverage parser). That is, something like generating one by one
all possible strings and  checking each time whether the string created is
the one to be parsed, is not a serious approach to parsing to me.

However, the story has *more* to it than that it is difficult at best (and
I would say imposible) to work out a reasonable GB/MP parser.
The true point is that human beings not only generate language, but,
astonishing as this fact seems to be for  many GB proponents, human beings
also tend to understand it. And while such theories as HPSG, GPG, LFG, ...
describe the language in such a way that the distinction between language
production and language comprehension plays no (major) role, GB/P&P/MP
mainstream is oriented on production (generation) only, while it is
difficult even to think about modelling understanding. For those who do not
know the old-established fact: the point is not that one cannot invert a
transformation creating  from an "input" structure" an "output" one -
indeed this can be done, provided one has the "output" structure. The
trouble is we do not have any structure when we parse, we have just the
string.

In this sense, GB/P&P/MP cannot be adequate theories of human language -  a
GB/P&P/MP description of a language is only a "generative" grammar, but it
isn't and cannot be a real grammar.
In this sense, the attribute "generative" describes GB/P&P/MP-based
grammars absolutely correctly, distinguishing them from grammars which can
do more.



k