Here is a more detailed summary of the section. It is section 6.5.3 in G&S, with references to other sections.
The section is concerned with clauses where a wh-word functions as a subject:
(1) Who visits Merle?
First, we (the G&S readers) should be sure that the Who has a non-empty WH-set (our QUE list). This is because (2) is possible:
(2) Who the hell would eat that?
Now I am turning to section 6.5.1, page 229.
Recall that wh-words can be WH-specified or not, meaning their WH-set can be empty or nonemtpy. In G&S’s analysis, the leftmost wh-constituent will always be WH-specified, and it will be the only WH-specified constituent in the clause. This distinction plays an important role in G&S’s account of the fact that the hell can modify fronted wh-expressions but not the in-situ ones:
(3) *Who visited who the hell?
Back to section 6.5.3.
Assuming Who in (1) is WH-nonemtpy, we observe that (1) cannot be licensed by the HSR. This is because of the WHSP which is the WH Subject Prohibition.
Here’s what section 5.2.2 has to say about WHSP.
Yhe WHSP prevents subjects from being WH-specified. This has the effect that all wh-phrases that appear to be WH-specified subjects are really extracted phrases licensed by the head filler rule.
Back to 6.5.3:
G&S suggest a type for subject-wh-interrogative-clause. This type should, according to their analysis, inherit some key constraints. Here I will just use screenshots:
“Thus, in all subject wh-questions, the subject NP is in fact extracted and the verb heading such construction is slashed”.
I think I am not approaching this reading correctly, like I am trying to reason in the wrong direction or something? What is the premise here, or the motivation, and what is the consequence? Is the premise that wh-subjects must be extracted (and now we are modeling that and concluding that it indeed can be modeled)?