PMEG says:

"Li pentras ŝin nudan. = Li pentras ŝin, kiam ŝi estas nuda. Se oni dirus nuda sen N, la signifo estus: Li pentras bildon, en kiu ŝi estas nuda. Sen N la signifo eĉ povas esti: Li pentras ŝin, kiam li mem estas nuda. Sed por tia signifo estas pli klare diri: Li nuda pentras ŝin."

I think it closes the case.

Miland:

T0dd:
Mi aĉetis botelon da vino. Ĝi ruĝa havas dolĉan guston.
.. well-formed or not?

I believe that should be La ruĝa, ktp.

T0dd:I can't think of any logical reason to forbid it

(The logic is: if it were good Esperanto, you (or I) would probably have come across something similar by now.)

The trouble with this approach is that it doesn't make much of a distinction between good Esperanto and merely common Esperanto.

Using the generative grammar idea, the generation of noun phrases from nouns in Esperanto is pretty well established.

Birdo flugas. That's a well-formed sentence.

I think it's fair to say that if NOUN VERB is well-formed, then either ADJECTIVE NOUN VERB or NOUN ADJECTIVE VERB is well-formed in Esperanto. So we can generate either,

Birdo blua flugas or Blua birdo flugas by applying that rule.

Pronouns are a subset of nouns, of course. So the question is, can we use the rule stated above to make well-formed sentences in which the NOUN is a PRONOUN? If we can't, then that's an interesting discovery, since the difference between nouns in general and pronouns is semantic, not syntactic--at least, I think that's the case. If that's right, then there are semantic barriers to the generative grammar rules, which would break the GG, I should think.

But maybe there's some purely syntactic way of restricting things. This takes me beyond my superficial understanding of GG.

T0dd:.. there are semantic barriers to the generative grammar rules, which would break the GG, I should think.

But maybe there's some purely syntactic way of restricting things. This takes me beyond my superficial understanding of GG.

I'm not sure that a GG can be "broken" by restrictions on the way words are used, because the GG, as I see it, is an abstraction from a reality. It's like describing a tree in terms of branches ('Hey, that branch has only 2 and not 3 sub-branches, it must be diseased..').

And if it turns out that some sentences compliant with your abstract formula are consistently rejected by speakers and other are acceptable (grammatical even if nonsensical?) then it undermines the whole formulaic approach as a universal account.

A possible example is Mi trovas ŝin mortan is OK and Mi trovis ĝin bonan is not.

You have to squeeze the meaning of well-formed to make the second sentence well-formed.

And it is not obvious how you can rescue the situation by imagining underlying deep structures so that one sentence is an exemplar of one pattern and the second another pattern.

sudanglo:And if it turns out that some sentences compliant with your abstract formula are consistently rejected by speakers and other are acceptable (grammatical even if nonsensical?) then it undermines the whole formulaic approach as a universal account.

Yes, false positives are counterexamples to the proposed GG, as I pointed out many pages back. They don't necessarily undermine the whole approach, however. They may show that the proposed details of GG are wrong.

A possible example is Mi trovas ŝin mortan is OK and Mi trovis ĝin bonan is not.

You have to squeeze the meaning of well-formed to make the second sentence well-formed.

Either that, or I haven't found the right rule.

And it is not obvious how you can rescue the situation by imagining underlying deep structures so that one sentence is an exemplar of one pattern and the second another pattern.

That's right, it's not obvious. It's a hypothesis that there are underlying deep structures that constrain possible GGs, but that hypothesis isn't terribly relevant here.

The point about GG is to try to give an account of what people know when they know a language. Part of what they know is syntax, since language is impossible without it. But just saying "They know syntax" is a woefully inadequate answer, unless you can actually say what the syntax is that they know. And this turns out to be a very difficult thing. Classical grammar fails miserably, so GG is an attempt to do a better job of it. But, to repeat the point yet again, no fully adequate GG of any natural language has yet been worked out. It's just a very hard job of work to do this.

As with any other scientific research program, it's an ongoing thing, and the details need to be empirically checked.

@Miland -- If it turns out to be impossible to define "well-formed" without introducing semantic constraints, then GG fails. This would show that semantics and syntax are entangled "all the way down", so to speak. It's a much disputed point.

T0dd:If it turns out to be impossible to define "well-formed" without introducing semantic constraints, then GG fails...

This reminds me of a tale about Nasrudin, a middle Eastern character used to illustrate human foibles. Nasrudin saw a tired falcon resting on a wall. 'Ho kompatindaĵo!' he cried. He took some scissors and trimmed its beak and wings. 'There, now you look much more like a real bird!'