About a month ago, I brought up structures like
- a further two months passed
- a surprising 50 students failed
- a mere two books were lost
- *a two months ago
- *a 50 students
- *a two books
In the meantime, I had asked Rodney Huddleston about it. If I follow his explanation, the gist of his argument is that the adjective somehow demotes the noun phrase (NP) to an undetermined nominal by changing the numeral from a determiner to a modifier. (Huddleston doesn't say it this way, and I may be misrepresenting which constituent selects which.)
So, if we start with the example 50 students failed, we get the following tree for 50 students:
This means that we have 50, a determinative functioning as a determiner in a noun phrase (NP). The head of the NP is always a nominal. The CGEL has nominals as the head noun, plus any dependents except the determiner. Here, the nominal consists of only the head noun, since there are no other dependents.
Now, if we add the adjective before the determiner, we get the following structure:
Notice that 50 students is no longer an NP (determiner + nominal). Instead, it is a nominal (modifier + head noun). This nominal is nested in another nominal and modified by the adjective surprising. Huddleston says that the numeral "plays no part in marking the definiteness in these cases." To get a full NP, we need a determiner:
On p. 353, the CGEL addresses such constructions. It says:
"(c) Dependents (or sequences) that select a singular or quantified plural headI now feel that I have a better understanding of the constituent structure, but I still don't know why the nominal requires a determiner. Plural nominals don't, typically, you know. We can say exuberant readers rejoiced over the explanation. But we can't say *exuberant 50 readers...
 another | an additional | a further | a good
 i [Another body/*bodies] had been discovered.
ii [Another three bodies] have been discovered.
iii [a further few/*many volunteers] were needed.
iv He ate [a good three hefty steaks] before leaving the table.
A plural head is permitted only if it is quantified by a numeral or by few."
I must admit, I have trouble understanding why the numeral plays no part in marking the definiteness.
[After reading this post, please see the response by Rodney Huddleston]