Araucaria asks for ‘references’, so I’ve dutifully read everything that seemed to be relevant in CGEL. Dismayingly little of it was helpful; I quote a few passages primarily as jumping-off points for disagreement. The answer as a whole really came out of this article:
Ellen Dodge and Abby Wright, “Herds of Wildebeest, Flasks of Vodka, Heaps of Trouble: An Embodied Construction Grammar Approach to English Measure Phrases”, J. Larson and M. Paster (Eds.) Proceedings of the 28th Annual Meeting of the Berkeley Linguistics Society: 75-86.
It's not directly on topic, but it introduces an approach that inspired (or at least provoked) the following answer. The answer would probably be a lot better if I took six months off to get a handle on how Construction Grammar works; but the current need which Araucaria signals will presumably have expired by then.
"How long was spent on the job?"
Question #1: Would that sentence be acceptable as standard English to you?
Sure. The passive is sort of odd—it suggests a balding bureaucrat in round wire-rims wearing a standup collar and celluloid cuff-protectors, filling in a ledger with a fountain pen—but like Ben Kovitz, I can see it arising echoically.
Question #2: And how would you parse its subject "How long"? And would you consider the subject to be:
- a noun phrase (NP)
- an adjective phrase (AdjP)
- an adverb phrase (AdvP)
Let’s start by clearing the landing-strip. There’s a tendency to categorize every Subject as an NP because of the old S → NP VP thing; but in fact practically any sort of syntactic entity can act as a Subject: an NP, a VP, an AdjP, an AdvP, a Clause, a PP. (Observe, while we’re at it, that all those categories describe the phrase’s internal structure; they are not suitable designations for its external syntactic function. I’ll take up the matter of internal structure in my answer in my answer to Question 3.)
In any case, is how long the Subject? That seems to be the position of CGEL; the discussion of ‘open interrogatives’ gives examples [4i] Who broke the window? and [4ii] Which one did he choose?, and comments
In [4i] the interrogative phrase is subject and occupies the same position as the subject of a declarative (*Kim broke the window). In [4ii] the interrogative phrase is a non-subject in prenuclear position: we say that the interrogative phrase has been fronted. (p.856)
But earlier, when introducing the notion of prenucleus, CGEL takes a more stringent attitude. There they compare the ‘canonical’ sentence [4a] Liz bought the watch with its ‘non-canonical’ (interrogative and subordinate) use in [4b] I wonder what Liz bought and offer these diagrams:
In [b] what precedes the subject in what we call the prenucleus position: it is followed by the nucleus, which is realised by a clause with the familiar subject–predicate sructure. Withinn this nuclear clause there is no overt object present. But the prenuclear what is understood as object, and this excludes the possibility of inserting a (direct) object after bought: ∗I wonder what Liz bought a watch. We represent this by having the object realized by a gap, an abstract element that is co-indexed with what (i.e. annotated with the same subscript index, here ‘i’): this device indicates that while what is in prenuclear position, it also functions in a secondary or derivative sense as object of bought.
Note that it would not be satisfactory to replace the ‘prenucleus’ label by ‘object’, and then simply dispense with the object element on the right of bought. Functions, we have said, are relational concepts and ‘object’ is a relation between an NP and a VP construction. Directly labelling what as object would not show that it is the object of the VP headed by bought.
That strikes me as —I won’t call it weaseling, but certainly waffling springs to mind. It is the object, except it isn’t, really, except in some sort of ‘secondary or derivative sense’.
And I have to wonder why a displaced object leaves a ‘gap’, but a displaced subject doesn’t.
I don’t think CGEL goes far enough. I’m going to adapt (not adopt) their terminology and pretty pictures; but I’m going to suggest that a) the ‘prenuclear’ element is not the object of the verb but a cataphor which takes the ‘gap’ as its postcedent, and b the ‘gap’ is not an ‘abstract’ element but something which is actually missing, videlicet the answer to the question.
How long has two functions, closely related but not entirely inseparable:
Externally (to the clause, not to the phrase itself), how long marks what follows as a specific sort of illocutionary act: a question. (In fact I will argue presently that it is the Head of an interrogative clause.) I’ve never encountered a term for this syntactic role, so I’m going to invent one of my own: interrogator. Since How long is a phrase we should identify it as such, as an “interrogator phrase”; but IP is already in use for “inflectional phrase”, so we need a distinct abbreviation ... I’ll call this an I?P.
(If this approach catches on we’ll be able to call the corresponding function in imperatives an I!P. )
Internally (again, to the clause), how long is a ‘pro-form’, a cataphor which a) alerts the hearer to the gap and b) constrains the sort of response the hearer can employ to fill that gap. Here’s how I diagram the clause, following CGEL’s conventions:
How long is not the Subject, but a pointer to the Subject of a subordinate clause; in itself, as interrogator, it is the head of the Interrogation Clause. (Not “interrogative clause”—according to CGEL’s admirable paradigm I should reserve the term interrogative for a lexical category.)
I offer these points as informal evidence for this analysis:
But what kind of form is how long a pro-form of? ... Well, what sort of forms can we stick in the gap? The ‘canonical’ complement of spend is an NP: two months, 157 hours. That would make how long a pronoun—except that there’s a bunch of other things that can go there. Ben Kovitz puts what certainly looks like an AdjP there, “too long”. Perhaps you find that frivolous, but how about “three weeks long”? —a little odd, maybe, in the passive, but not in the active. And PPs are certainly acceptable: from April to June, since March. —But of course I’ve already pointed out that we should be looking at function, not internal structure: do these various constituent types serve a common and commonly recognized syntactical function? CGEL provides a couple of candidates; neither is formally defined or accorded a formal status, but the terms occur often enough to latch on to.
One is measure phrase: a phrase which provides a measurement. The ‘core’ type for measure phrases is probably the plain NP (six inches, two pounds, a bucketful), but genitive NPs (one day’s, a pound’s worth), PPs (from Ghent to Aix, between forty and fifty), AdjPs (*six feet tall, ) and comparatives are also possible (bigger than life, as long as your arm).
The other is locative. This category consists primarily of PPs†, but includes NPs (home, next door) and Adverbs (north, right), too. Traditional grammar usually classifies these expressions as ‘adverbials’, but they are just as likely to act as adjuncts and predicate complements to nouns, or as complements to verbs. CGEL devotes some 30 pages of its chapter on Adjuncts to “Adjuncts and complements expressing location and change of location” in both space and time (and metaphorical extensions of these categories). The rubric “location and change of location” also embraces notions of direction, orientation and extension. That last is what I think how long is pointing to and demanding: an expression signifying temporal extension. And CGEL also acknowledges locatives which involve scalar position and change, such as She increased her philosophy mark from 70% to 85%, so we can pull in the notion of measure phrase as well.
My answer, then, to Question #2: How long is a pro-locative of scalar temporal extension acting as an interrogator. I think that ranges how long neatly beside her sister pro-locative interrogators when, where, whence and whither on one side and her sister scalar interrogators how much, how many, how big and many many more on the other.
† In fact, I believe that this is the core function of PPs; and I’m coming to suspect that PP and locative are so closely related that perhaps ‘preposition’ should be renamed ‘locator’. But I’m not ready to argue that yet.
Question #3: And what word categories would the words "how" and "long" in that subject belong to?
I’d just as soon skip this question. Since nowadays a) we’re pretty much agreed that ‘word classes’ or ‘parts of speech’ or ‘lexical categories’ are defined by morphology on the one hand and syntactical role on the other, and b) just about any English word can be recategorized at will, I don’t really see what the categories are good for outside of a particular syntactic analysis. But here goes.
Given the syntactic analysis above, I don’t see how how can be placed in any category but interrogative—a term I intend to have the same relationship to ‘interrogator’ as determinative has in the CGEL scheme to determiner. The questions that raises—whether how it is still an ‘interrogator’ in fused relatives, and whether what we ordinarily call ‘relative pronouns’ should also be categorized as interrogatives—I will leave for another occasion.
As for long ... CGEL says the long in how long is an adverb:
Adverbial long, as in It won't last long, has a temporal meaning: "a long time". Its distribution is quite exceptional for an adverb, in that it can head phrases functioning as internal complement to a few verbs such as take, have, need, spend, give, and be:
ii a. How long can you give me?
The [bolded] phrases are functionally comparable to NPs: compare [...] How much time can you give me?; I won't be *more than ten minutes**. It is nevertheless clear from the dependents of long [...] how) that it is an adverb, not a noun. Notice, moreover, that such AdvPs cannot replace temporal NPs in subject function: A long time / Long had passed since their last meeting (except, somewhat marginally, in passives – How long was spent on the job?). (p. 569)
My problem with CGEL’s analysis is that it takes how to be an adverbial degree modifier of the phrase head:
[How old] is your father? [modifier of adjective]
[How many] children have they got? [modifier of degree determinative]
How modifies adjectives, degree determinatives, adverbs, and verbs to question degree, extent, quantity.
I cannot agree. I compare how long not with two feet long or two hours long but with how big, how much, how far, &c; I take how to be the head of the I?P, and long to be a modifier which defines the category in which how demands that the answer be cast.
ADDED to meet Araucaria's request for evidence:
Well, I did a crash course in X-bar theory (a form of generative analysis in which strict binary branching plays much the same role as strict circular motion does in Ptolemaic astronomy), and it seems that in its terms I got it about 75% right. I leave my original analysis in place as evidence of my hybris.
According to that school of analysis, How long does, as I suggested, lie outside the clause headed by was, and it does leave a 'trace’ (which is what CGEL calls a ‘gap’). It lies, specifically, in an entity called a CP; etymologically this derives from ‘complement phrase’, and it is posited so the grammar may represent all sorts of non-canonical finite clauses—not only wh- questions but also relative constructions and that and for complement clauses—within the same framework.
However, How long is not the head of the CP but its specifier, which is a hypernym for entities as diverse as complementizers, subjects, determiners and auxiliary verbs. And the clause headed by was is not a sister of the specifier but an entity three nodes down. Here's a diagram, as near as I can figure it, but with a bunch of uninformative epicycles omitted:
[t1], coindexed with How long, is the gap. The unrealized [C] is there to accommodate the auxiliary if subject/auxiliary inversion is called for. [IP] (<‘inflection phrase') is where the subject (or one of them) and [I'] (tense) reside; tense will eventually move down or up to the auxiliary or lexical verb.
Note that [C'], the head which How long ‘specifies’, is not a “real" entity, a ‘lexical head’, but an “abstract" entity, a ‘functional head’— I’m counting my head that as a moral victory.
My main source for this stuff is Thinking Syntactically by Liliane Haegeman (Blackwell, 2006), a syntax textbook written from a Minimalist point of view.
As for ‘how long’ as a pro-form ... ‘Interrogative pro-form’ is a common designation for the wh- terms in questions; see for instance the SIL Glossary of Linguistic Terms s.v. “What is a wh-question?” and “What is an interrogative pro-form?”. Granted, Trotta, Wh-clauses in English, rejects this characterization on the grounds that “it is not substituting for any item (40)”; but he also concedes that “even wh- interrogative words can be understood as pro-forms in the (loose) sense proposed by Quirk et al (1985: 77) in having ‘...a meaning something like “It has not been known before what this item refers to, and so it needs to be stated in full”’ (102).” I’m perfectly happy with this ‘loose’ sense; I see no practical difference between looking for a present referent (an antecedent or postcedent) assumed to be present and an absent referent (a gap) assumed to be absent.
In fact, most of this is beside the point. My original analysis rested on the parsing problem: hearers/readers don't have a diagram in front of them, and when they encounter an interrogator they set out looking for a gap to plug it into. If there’s no gap, they’ll never be able to fit it anywhere.