Linguistics 550
4. X' theory 2

The special status of modals in English

English has a special class of elements called modals, which include can, could, may, might, shall, should, will and would. These elements resemble ordinary verbs, but differ from them in several important respects, both morphologically and syntactically.
Note: In what follows, be sure to distinguish modal can follows from the homonymous, but etymologically unrelated ordinary verb. The two differ phonologically: the ordinary verb must be pronounced [kæn], whereas the modal can be reduced to [kn].

At least one lexical item exists as a doublet: need. As a modal, it combines with a bare infinitive, but as an ordinary verb, it combines with a to-infinitive.

Modal: (9) a. i. ok S/he needn't write the letter.
ii. * S/he needn't to write the letter.
(10) a. i. ok Need s/he write the letter?
ii. * Need s/he to write the letter?
Ordinary verb: (11) a. i. ok S/he doesn't need to write the letter.
ii. * S/he doesn't need write the letter.
(12) a. i. ok Does s/he need to write the letter?
ii. * Does s/he need write the letter?

Because of their special properties, modals are assigned to a syntactic category of their own: I(nflection). The reason for the name is that the modals in English serve the same function that inflections serve in many other languages. A good example is the modal will, which marks future tense in English; another is can, which expresses possibility. In many other languages, both of these grammatical categories are expressed by verbal endings.

It is worth noting that certain lexical items–namely, auxiliary be, do and have– do not exhibit the morphological properties exemplified in (1)-(4), yet nevertheless behave like modals from a syntactic point of view.

Surprisingly, main verb be patterns together with this group as well. Main verb do and have, on the other hand, behave like ordinary verbs, as one would expect.

Note: Up until the 1950s, main verb have is attested with the syntactic behavior of a modal in written American English. Here's an example:
Seeing a murder on television can help work off one's antagonisms. And if you haven't any antagonisms, the commercials will give you some.–Alfred Hitchcock

(15) a. ok Is s/he reliable? S/he isn't reliable.
b. * Does s/he be reliable? S/he doesn't be reliable.
(16) a. * Does s/he the job? S/he doesn't the job.
b. ok Does s/he do the job? S/he doesn't do the job.
(17) a. * Has s/he a television set? S/he hasn't a television set.
b. ok Does s/he have a television set? S/he doesn't have a television set.

Notice also that main verb do has an infinitive, where auxiliary do does not.

(18) a. ok They do repairs.
b. ok They claim to do repairs.
(19) a. ok They do { not, so } like your brother.
b. * They claim to do { not, so } like your brother.

Extending the X' schema to I

So far, we have been treating sentences as maximal projections of V. However, while the sentence = VP approach is successful for sentences without modals like (20a), it breaks down for sentences containing them, like (20b), because there is no slot for the modal will within VP.

(20) a.   The child found the toy.
b. The child will find the toy.

The solution to this problem is to extend the idea that lexical items project an X' schema from lexical categories to functional categories like I. This allows the modal to project the structure in (21a). Substituting an NP specifier and a VP complement then yields (21b) as the full structure for the sentence.

(21) a.       b.  

Subject movement

Conceptual considerations

As usual in syntax (and life more generally), no sooner do we solve one problem than we find another one on our hands. Specifically, despite the minimal difference between the sentences in (20)–they differ only in tense–they now differ quite substantially in structure, since (20a) is a VP, whereas (20b) is an IP. Related to this is the fact that the subject occupies a different position (Spec(VP) versus Spec(IP)) in both cases, even though it is interpreted as an argument of 'find' in both cases. These categorial asymmetries, which are conceptually undesirable, are thrown into high relief by the emphatic past tense variant in (22), which is truth-conditionally equivalent to (20a), but must be assigned a different structure–one analogous to (21b)–due to the presence of the modal.

(22)     The child díd find the toy.

We will solve the problem posed by our current analysis of (20)-(22) in three steps.

A few remarks are in order about the process of movement just introduced. Movement is not intended as a psychologically real process. Rather, it is a metaphor to represent the fact that a single phrase may satisfy more than one function in a sentence. In the case at hand, the child satisfies two distinct functions. First, it is a semantic argument–the agent, to be precise–of the predicate find. Second, it is the grammatical subject of the entire sentence. (As we will see later on in the course, grammatical subjects are not always semantic arguments.) In order to clearly express the phrase's double function, we do not actually simply move the phrase from one position to another. Instead, movement leaves a so-called trace in the phrase's original position. Moreover, the two positions share an index.

Note: By convention, these indices are represented as subscripts just as in the case of binding, although it is not actually clear whether the binding indices and the movement indices are really the same animal. For clarity, we use the natural numbers as binding indices, and the lowercase letters i, j, k, and so on as movement indices.
A constituent together with any traces of movement (possibly none, in the absence of movement) are referred to as a chain. The elements of a chain are called its links. For instance, the tree in (26a) contains one chain with two links ([ the child ]i, ti) and another chain with a single link ([ the toy ]). Higher links in a chain are often referred to as the antecedents of lower ones. In (26a), for example, the phrase the child in Spec(IP) is the antecedent of the trace in Spec(VP). The motivation for the terminology is that traces of movement must be c-commanded by their antecedent, just as reflexive pronouns must be.

Evidence for subject movement from stranded quantifiers

If subject movement strikes you as a rather hokey move, perhaps you will find the following piece of evidence more convincing than the conceptual argument presented so far. Consider (27).

(27) a.   All the children should find their toys.
b. The children should all find their toys.

Notice that in both word order variants, the quantifier all must be construed with the children.

Note: Quantifiers are words that denote quantity (all, both, every, many, much, some, and so on) Not all of these words behave alike from a syntactic point of view, however. In what follows, we will focus on all ; both shows the same syntactic behavior.

This is not surprising in (27a) given the quantifier's adjacency to the noun phrase. But in (27b), one might have expected additional or other interpretations (for instance, a construal of all with their toys; cf. The children should find all their toys). However, such additional interpretations are not available. How can we explain this fact?

Notice that under the subject movement hypothesis just discussed (more commonly known as the VP-internal subject hypothesis), one of the intermediate steps in the derivation of (27a) is (28a). The surface word order of (27a) is then derived by moving the QuantNP from Spec(VP) to Spec(IP), as shown in (28b).

For the moment, we consider QuantNP (= quantified noun phrase) to be a subcategory of noun phrase. (See Notes 5 for a more sophisticated treatment of how QuantNP and NP are related.)

(28) a.  
b.

But now notice that the structure in (28a) also allows us to derive (27b) by moving just the NP the children, rather than the QuantNP all the children, to Spec(IP). In the resulting structure, shown in (29), the quantifier remains in Spec(VP), signalling the subject's original position.

(29)    

Note: When an element that is able to undergo movement together with some larger constituent fails to do so, it is said to be stranded. The alternative (when the strandable element moves along with the other constituent) is called pied-piping (the idea being that the strandable element is swept along by the irresistible lure of the larger constituent, just like the children of Hameln were swept along by the pied piper's music). (27a) and (27b) illustrate pied piping and stranding of the quantifier, respectively. Another category that allows pied piping or stranding in English is P, as illustrated in (30). We discuss the type of movement involved in deriving (30) in detail later on in the course.

(30) a. Pied piping: [PP Under [NP which table ] ]i were they hiding [PP ti ] ?
b. Stranding: [NP Which table ]i were they hiding [PP under [NP ti ] ] ?

Given the derivation just outlined, it is no longer surprising that all can (and indeed must) be construed with the subject in both (27a) and (27b). The reason is that there is a common stage in the derivation of both sentences at which the quantifier and the ordinary NP form a constituent. By contrast, there is no stage in the derivation of the sentence at which all and the object form a constituent. This is the reason that all cannot be construed with the object in (27b).

Complementizers

Another example of a functional category that projects syntactic structure is C(omplementizer), which corresponds closely (though not exactly) to the traditional category of subordinating conjunction. This category includes if and that (and, for some speakers, whether). The treelets that these lexical elements project are shown in (31).

(31) a.       b.       c.  

As is evident, complementizers take IP complements. Their maximal projections (that is, CP) can in turn function as the complements of verbs, nouns and adjectives. For some reason, hardly any prepositions in modern English take CP complements (an exception is in). This is by no means a universal, however, for Middle English and modern languages such as Spanish have prepositions that take CP complements.

Note: Although English prepositions take CP complements only exceptionally, there are several that take IP complements. In traditional grammar, the lexical items in question are assigned the separate syntactic category of subordinating conjunction, but generative grammar sees no reason to recategorize an element depending on the syntactic category of its complement. After all, a verb is a verb, regardless of whether its complement is a noun phrase (NP), a prepositional phrase (PP) or a clause (CP). This issue is taken up in Assignment 4, Exercise 2.

Determiners

The dialectics of cross-categorial parallelism

Recall that X' theory was originally motivated by the parallelism between sentences and noun phrases like (32) (= (1) of Notes 3).

(32) a.   The Romans destroyed the city.
b. the Romans' destruction of the city

Before extending X' theory to the functional categories, we were able to capture this parallelism by the structures in (33) (= (4) of Notes 3).

(33) a.       b.  

But now that we have introduced the category of I(nfl) and assume that sentences are projections of I rather than V, the structure that we assign to (32a) is (34), which is no longer parallel to (33b).

(34)    

The generalization of X' theory thus negates its original motivation, and we face the following contradiction:

Thesis:
In the original formulation of X' theory, sentences and noun phrases are structurally parallel.

Antithesis:
In generalized X' theory, sentences and noun phrases are no longer structurally parallel.

What we need is a synthesis that resolves the contradiction between the thesis and the antithesis without giving up any of the insights that we have gained so far.

The DP hypothesis of noun phrase structure

The desired synthesis comes from faithfully pursuing our path of generalizing X' theory to all syntactic categories. Note that we have not yet provided D(et) with its own projection. This category includes what are traditionally called articles (a(n), the) and demonstrative adjectives/pronouns (this, that). We will also assume that D is the syntactic category of the possessive morpheme in English (realized as ’s in the singular and as ’ in the plural). Although not everyone accepts this assumption, one argument in favor of it is that the possessive morpheme cannot cooccur with another determiner. If both share the same syntactic category, they compete for the syntactic slot, and the ungrammaticality of (35a) can be treated as parallel to that of (35b).

(35) a. * { John's the, John's this, John's that, John the's book, John this's book, John that's } book
b. * { the this, the that, this the, this that, that the, that this } book

Assuming that D projects to DP then allows us to represent (32b) as in (36).

(36)    

As is evident, the representation of the noun phrase as a DP restores the parallelism between sentences and noun phrases. In both cases, the entire structure under consideration (the sentence or noun phrase) consists of a core headed by a lexical category (V, N) with a 'shell' headed by a functional category (I, D).

Note: It is important to distinguish carefully between the pretheoretical term 'noun phrase' and the theoretical concepts NP and DP. The pretheoretical term (noun phrase) refers to something real out there in the world (actually, to be more precise, in there in your mind). This is the material of investigation whose properties we are trying to describe and/or explain. The theoretical concepts, on the other hand (DP, NP, and so on), are the tools that used in the investigation, not the material under investigation itself.

It is true that the label for the theoretical concept 'NP' was originally chosen to be mnemonically related to the pretheoretical term 'noun phrase'. However, there is no logically necessary identity relation between the entity 'noun phrase' and the label 'NP'. When we analyzed noun phrases as NPs, we assumed this identity. But if the assumption proves to be more trouble than it's worth, we are free to modify it, or even discard it completely.

There is no getting around the fact that it is often confusing and a bit painful to get used to new tools. It is also worth noting that the distinction between reality and representations of reality is not by any means peculiar to linguistics; for instance, the masters of Zen Buddhism are constantly exhorting us not to confuse the finger pointing at the moon with the moon itself.

More arguments in favor of the DP hypothesis

In addition to restoring cross-categorial parallelism between sentences and nouns, the DP hypothesis solves a further anomaly. Recall the general X' schema in (37) (= (5) in Notes 3).

(37)    

Note that the specifier in this schema is a maximal projection. However, if noun phrases are represented as NPs, Spec(NP) is problematic because in addition to being able to be filled by a maximal projection (a phrase), as in (38a), it can also be filled by a head, as in (38b).

(38) a.       b.  

But if the noun phrases are represented as DPs, as in (39), the anomaly vanishes, since there is now a separate position reserved for heads.

(39) a.       b.  

Another conceptual argument in favor of the DP hypothesis is that it allows us to recast the traditional distinction between articles and demonstrative adjectives/pronouns in terms of the independently motivated distinction between transitive and intransitive heads. Articles differ from demonstratives because they can never appear in isolation.

(40) a. I'll buy { a, the } book.
b. * I'll buy { a, the } .
(41) a. I'll buy { this, that } book.
b. I'll buy { this, that } .

Instead of being forced to distinguish three separate syntactic categories (articles, demonstrative adjectives, and demonstrative pronouns), the DP hypothesis allows us to make do with a single category D. Articles are then obligatorily transitive determiners, whereas demonstratives are determiners that are associated with both transitive and intransitive treelets, as shown in (42).

(42) a.       b.       c.  

From this cross-categorial perspective, articles correspond to verbs like devour and prepositions like of, whereas demonstratives correspond to verbs like eat and prepositions like in. The question of course immediately arises whether there are obligatorily intransitive determiners corresponding to verbs like wait. It has been suggested that the answer is 'yes', and that the elements in question are pronouns (or more precisely, a subset of them, as the contrast between the (b) examples in (43) and (44) shows).

(43) a. I, he, she, it, they
b. * I linguist, he fool, she idiot, it piece of junk, they traitors
(44) a. we, you
b. we Americans, you fool(s)

(45) shows the treelets corresponding to the pronouns in (43) and (44).

(45) a.       b.       c.  

Evidence for assigning pronouns the syntactic category D comes from the impossibility of combining them with articles, as shown in (46).

(46)   * the he, this she, that you, a they

This fact follows straightforwardly if pronouns and articles both occupy the same structural slot.