Roger and Ray wrote:

>> But I confess, I'm talking about things I never knew
>> well, and have forgotten much, and haven't kept up
>> with... Perhaps Elliott can find the time to update me
>> and all of us as to how things stand in current
>> TG-derived grammars (without having to write a
>> dissertation before he's ready :-)))) )

>I have a suspicion that Elliot will update us sometime.  I shall certainly be interested to see his observations.

Sorry for the delay. Here is my 'generative analysis'.

Okay, so, I've already proposed the 'generative hypothesis concerning the sentence:

Ultima Cumaei venit iam carminis aetas.

The hypothesis is that there is a movement operation originally due to topicalization or some discourse related meaning. Furthermore, I elaborated the hypothesis by saying that given the constraints of the meter in this poem, this topicalization movement was reused in order to fit the meter by the poet (Virgil). 

Now, let me give the beginnings of analysis in a version current 'Minimalist' generative syntax.  ('Minimalist' in quotes because sometimes one can lose track of why exactly it's called Minimalist.)

Basically, by the time of Minimalism (which is sometimes said to have been started in Pollack's 1989 paper on the positioning of the French verb and English, although the work always quoted is Chomsky 1995), transformational operations have been reduced to one generalized transformation:  Move.  This was due to work in the 70s and 80s that basically saw that the Standard Theory which developed due Syntactic Structures (Chomsky 1957) and Aspects of the theory of syntax (Chomsky 1965) was fairly over complicated with all its construction specific transformations. 

So, we just have Move. What is Move driven by? Well, In Mimimalism there is a strong emphasis lain on the features of lexical items. Lexical items include grammatical/functional morphemes (which may be phonologically null) and contentful morphemes (verbs, nouns, etc).  Features include category features (V,N,A, etc), gender/number/person features (so-called 'phi' features for some reason), tense feature, case features, topic features, among many (usually undefined) others. Additionally, there is a major feature in Minimalism called variously 'EPP' (for historical reasons) or 'movement diacritic'. This is a second order feature that can be associated with any of the more contentful features (phi, tense, case). this feature is the direct cause of movement. 

The second operation in Minimalism is 'Merge'. This takes two lexical items and creates a syntactic object from them and then projects the category label of the head. This takes the place of phrase structure rules.

Finally, in minimalism there is a relation called 'Agree'. This is basically a relationship between two items in a tree, one higher and one lower. The higher one is called 'Probe' and the lower is called 'Goal'. The relationship is set up when the Probe has a feature whose specification is unvalued (or 'uninterpretable') and the Goal has the same feature whose specification is valued. This relationship values the Probe's feature. 

With those preliminaries, this is how I would analyze the sentence:

let us say we have the content lexical items:
ven- (V), iam (Adv), carmen- (N), ultim- (A), aetat- (N), Cumae- (A)

And the functional: C(Feature: TOP+EPP/movement), T(Feature: Past+EPP/movement, PHI:?).

Lets start by creating the subject phrase.
Carmen is merged with the noun Cumae- and 'N' is projected since N is the head
of this phrase:  [NP [N  Cumaei  N carmen-]]

The genitive case is usually said to be a case licensed under selection by a nominal head, so in this case, the genitive 'Cumaei' is licensed by virtue of being selected by another nominal. Note that 'carmen-' itself is still not case marked at this point, i.e. it has an unvalued case feature. This phrase is complete at this point so the label of the phrase is 'NP'. Note that the complement is to the left of the selecting head. This is because Latin is (like Ray said) mostly an SOV language, at least in prose.

Then the separate noun 'aetat-' is merged in, selecting the already formed NP as its complement:

[N [NP [N Cumaei N carminis]] N aetat-]

The genitive case of 'carmen' is added under selection by a nominal head. The case feature of 'aetat-' is still unvalued. 

Then the adjective is added. The phrase is projected to NP, since the phrase is complete: 
[NP [A ultim-] [N' [NP [N Cumaei N carminis]] N aetat-]]

The Agree relation holds between A and N(aetat-) and they agree (abstractly) for case, number and gender. There is a strain in Minimalism that allows us not to express this abstract agreement until after the derivation is done, that is, when the phrase leaves syntax and is interpreted by the phonological system. For this reason I will leave the endings out. This helps us because of the portmanteau nature of Latin endings - we can't actually fill in any ending at this point since we still don't know the case feature of the N (and by extension the A). 

So, our subject phrase is complete. We merge this with the verb:

[V [NP ....aetat-] V ven-]

To which we can add the adverb: iam:

[VP iam [V [NP .... aetat-] V ven-]]

Then we merge this with the functional item 'T', which has the feature 'Past+movement' as well as unvalued 'phi' features (number, person).  The label of T is projected.

[T T(PAST, PHI) [VP iam [V [NP ..... aetat-] V ven-]]]]

So, two Agree relations are set up: between the tense feature of T and the unvalued tense of the verb. Because of the movement diacritic, this Agree relationship forces the verb to 'move' to T. This represents the fact that the verb is associated with two functional layers: argument structure and event structure. It moves from the argument structure domain (the VP) to the event structure domain (the TP).  The second Agree relation is between the PHI of T and the number/person features of the subject noun.  Since this is 3S, the verb (which is now in T) gets the ending -it.  And also due to this relationship, the case feature of the subject is valued: +NOMINATIVE, and hence we can fill in the endings of the N and A now. The label of T is projected as TP to indicate the end of a phrase.

[TP T-V venit [VP iam [V [NP [A ultima] [N' [NP Cumaei carminis] N aetas]] <V>]]]]

(Note, that I indicate the movement of V by placing it in <> in the original position. 

After this, the null-complementizer C is merged (remember with features that indicate it is concerned with topicalization and that topicalization must be fulfilled by movement, i.e. C(Feature: TOP+EPP/movement).

[C [TP T-V venit [VP iam [V [NP [A ultima] [N' [NP Cumaei carminis] N aetas]] <V>]]]]]

The topic feature of C targets NP (although it is unclear from the current theorizing what determines what thing is targeted in this case). The NP is moved to adjoin to C and the label CP is projected.

[CP [NP [A ultima] [N' [NP Cumaei carminis] N aetas]] C [TP T-V venit [VP iam [V [NP ... ]]]]]

Now, comes a somewhat worrying feature of minimalism: the copy theory of movement. This theory is basically that movement doesn't move the original unit, but rather a copy of the unit. The evidence that people typically cite for this kind of theory are dialects of English (and other languages, Norwegian comes to mind as having been mentioned) where wh-words can be doubled as "Who do you think who is coming? (which sounds okay to me)...

So, with that in hand, I then have to mention the so-called 'scattered deletion' idea. This idea arises from the following question: "given that we have a theory of movement that says that copies are moved, how do we determine which copies are pronounced? (remember that all of this syntax stuff is happening, in Minimalism, before phonology and pronunciation: in the mind). So, scattered deletion says (in very unclear terms - there's a whole book about this thing) that units can be pronounced both their base and in their moved position or both, or parts of both can.

Given this, a possible analysis is to say that, 'ultima' and 'Cumaei' are retained in the CP domain but deleted in the VP domain while 'aetas' and 'carminis' are retained in the VP and deleted in the CP. 

I told you it was it is very scary.  I think there is much to be said for the copy theory and even deletion of this sort - however, in the Latin case, I can't really see how it would be supported (arguments for scattered deletion tend to involve prosodically weak elements that need to be supported by a strong element). 
There is a much more plausible story which Ray hinted at: the adjunction story.  In this analysis, 'carminis' and 'aetas' are part of an NP that is adjoined to the CP and 'ultima Cumaei' are part of an NP that is the actual subject:

[CP C [TP [NP ultima Cumaei] [T T-V venit [VP iam [V [NP <ultima Cumaei>] <V>]]]] [NP carminis aetas]]

This tree changes three things:
(1) the subject NP moves from the argument structure domain (VP) to the inflectional domain (TP), after its nominative case was licenced by agreement between T and the subject.  (This type of analysis comes directly from Chomsky's original analysis of English and is often applied to other languages. In this analysis, languages like Latin are said to have an optional subject movement and even optional subject, while English obligatorily has a subject and subject movement).  The lower copy (in the VP) is deleted.
(2) The C no longer is associated with topicalization - it is just there as a marker of declarative force (as opposed to interrogative or something like that)
(3) The final NP is not moved at all to attach to its high position. It is rather generated in that position as an adjunct. The link betweem 'ultima Cumaei' and 'carminis aetas' is therefore purely semantic.

I think that's my story. :-) 
(note that I am not a Latin expert - and thank goodness for that! The hoops that Latin generativists must jump through.  It makes Old Irish look easy. And that's saying something).