Intuitions about Structure 2.

2

2

Consider first our intuitions about the way in which the words in (I) are grouped into successively larger constituents. For example. we'd all agree that incrcdihlv 'goes with' or 'modifies' stupid in (I). so that the sequence iincrediblv stupid] is a (phrasal) constituent of the sentence. Likewise. this modifies bov in (I). so that the sequence [this /lor] forms a single structural unit. a constituent or the sentence: in much the same way. we might argue that that modifies girl, so that the sequence [thilt ~irl] is also a constituent of the sentence. Furthermore. it is intuitively obvious that to goes with the phrase [that ~irl]. so that [to that girl] is also a constituent. And we might also claim that since the phrases [incredihly stupid] and [to that girl] both modify seem. then the whole sequence [seem incrediblv stupid to that girl] is also a constituent. What we need is some way of representing all this information about constituent structure in diagrammatic form. In fact, this can be done in terms of a tree diagram such as (2) below:

Structure

2.1 Overview

In the previous chapter, we argued that native speakers' competence in their native language(s) is reflected in their intuitions about sentence well-Iormedness (derived from their acceptability judgments) on the one hand, and their intuitions about sentence-structure on the other. We have argued that the ability to make judgments about well-Iormedness and structure holds at all four major linguistic levels Phonology, Morphology. Syntax. and Semantics. Hence. it follows that a native speaker's syntactic competence will be reflected in his intuitions about the grammaticalitv ( = syntactic well-forrnedness) of sentences on the one hand, and their S1'l1t1lCtic structure on the other. We have discussed in some detail the problems of deciding whether a given sentence is grammatical or not. In this chapter, we examine the question of what it means to say that a sentence has a syntactic structure; we discuss the evidence in support of that claim; and we look at ways of representing syntactic structure.

(1)

ThIS boy

must seem incredibly stupid

to

that girl

( I )

This boy must seem incredibly stupid to that girl

Each point in the tree is called a node: and each node represents a constituent (i.c. a syntactic unit). Thus. (2) tells us that not only are all thc individual words in (2) syntactic uruts (i.c, const ituentsi of the sentence, but so too are the phrases [this horj, [illcredihlr .llUllidj, [t!tilt gir/j. [to that gir/j, and [.1'1'1.'111 incrcdiblv stupid to that ~ir/l. In the diagram (2), we have indicated the nodes by large dots: however, since nodes are predictable (they occur at the top and bottom of the tree. and at points where two or more branches intersect) we shall suppress them in subsequent tree-diagrams.

But while a tree-diagram such as (2) provides a convenient pictorial representation ofour intuitions about the coust nucnt structurc of a sentence such as (I) (r.e. what the basic syntactic units in the sentence are), it does not provide any representation or our intuitions about which constituents are constituents otth« saute type. So if someone asked us whether we feel intuitively that hOI' and this In (I) arc constituents or the same type, we'd probably say 'No, hoy is the same kind of word as girl. man, II·O/l/lIn. etc., whereas this is the same kind "I word as that, a. the. etc.' The traditional way or describing the similarities

2.2 Intuitions about Structure

Part of the evidence for claiming that sentences have a syntactic structure in language comes from the native speaker's intuitions about the structure of sentences in his language. The structural intuitions which native speakers have about the Syntax of their languages are of two types, namely (i) intuitions about how sound-sequences in sentences are structured into successively larger structural units which we call constituents; and (ii) intuitions about whether particular sets of constituents (i.e, structural units) belong to the same category or not. We can illustrate the nature of these intuitions about constituents and categories in relation to a sentence such as:

50

<; I

Structure

Intuitions about Structure 2.2

and differences between constituents is to say that they belong to categories of various types. Thus, words like bov, girl, man, woman, etc. are traditionally said to belong to the category of Nouns, whereas words like a, the, this, and thai arc traditionally said to belong to the category of Determiners. Similarly, we'd probably recognise that seem is the same kind of word as appear, /£'1'1, become, etc. the kind of word that traditional grammarians call a ('main', 'full', or 'lexica!') Verb. By contrast, must is the same kind of constituent as mal', might, can, could, will, would, shall, and should - i.e. is a constituent of the type traditionally called a Modal (Auxiliary) (Verb). And we'd intuitively recognise that to is the same kind of constituent as with, from, bv. at, in, for, etc. namely a Preposition. In addition, we'd probably agree that incrcdiblv is a word like rcallv fundamentally, potentially, etc. ~ i.e. a word traditionally labelled an Advc·rh. In much the same way, we might feel that stupid is the same kind of word as clever, kind, randy, iliad, nice, etc. i.e. the kind of word traditionally called an Adjective.

Just as words belong to different categories, so too do phrases. For instance, we might capture the fact that [this hoy] and [that girl] seem to bc phrases of the same type, and that in both cases the head (i.e. key constituent) of the phrase is a Noun, by assigning these two constituents the categorial status of Noun Phrases. And in order to capture the fact that the phrase [incredihly stupid] is a constituent of the same type as [rathcr nice], [very handsome], [quite tren(h'], [somewhat rude] etc. -- i.e. a phrase whose head in each case is an Adjectivc (stupid, nice, handsome, trendy, rudes - we might call such constituents Adjectival Phrases. Similarly, in order to mark the fact that [to that girl] is a phrase of the same type as e.g. [with his mother], !.from the professor], (into the room], (jor Mary] etc., and that in each case the head of the phrase is a Preposition (to, with, [rom, into, /IJr), we might call such phrases Prepositional Phrases. And to capture our intuition that [seem incredibly stupid to that girl] is a phrase of the same type as [speak rather rudelv to his mother], l~ive a present to Mary], [run hack home], [close the door], [die tomorrow]- i.e. a constituent whose head is a Verb ~ we might classify such constituents as Verb Phrases. Finally, we might want to recognise the fact that the whole sequence [This hoy must seem incrediblv stupid to that girl] is a special type of constituent traditionally termed a Clause or Sentence.

Note that none of this vital categorial information is contained in the tree diagram (2) which we earlier gave to represent the constituent structure of (I). However, the relevant categorial information can be included in this type of diagram if we simply attach an appropriate category label to each of the nodes in the tree: the resultant structure is then known as a labelled tree diagram. Given this assumption, we can represent the categorial constituent structure of

sentence (I) in terms of the labelled tree diagram (3) below:

(3)

~~----------

N P M »>: V P ____________

/ '" I .i->: I ____________

D N must V AP PP

I I I r >; r >.

This boy seem ADV A P NP

I I I / ""

incredibly stupid to D N

I I

that girl

(Abbreviations: S = Clause/Sentence; M = Modal; D = Determiner; ADV = Adverb; P = Preposition, PP= Prepositional Phrase; N = Noun, NP = Noun Phrase; V = Verb, VP = Verb Phrase; A = Adjective; AP = Adjectival Phrase).

A diagram such as (3) provides a visual representation of the categorial constituent structure of sentence (I). Equivalently, we might say that since (3) shows us how sentence (I) is structured out of its constituent phrases, and how each of the phrases is structured out of its component words, (3) provides a visual representation of the Phrase Structure of sentence (1). Hence, the type of labelled tree diagram used in (3) is referred to as a Phrase-marker (Pvmarker), because it marks the hierarchical grouping of words into phrases, and phrases into sentences. The P-marker (3) thus provides a visual representation of the superficial syntactic structure (or Ssstructure, as Chomsky calls it) of the sentence (I).

The method of representing syntactic structure visually by the use of labelled tree diagrams (P-markers) such as that in (3) is in fact only one of many alternative systems which have been devised in order to provide a visual representation of structure. Another (logically equivalent) method of visual display frequently used in the linguistic literature is to make use of labelled bracketing rather than labelled tree-diagrams. Within this alternative system, wc could represent the categorial status of each of the words in (I) as in (4) below:

(4) [D this] [N boy] [M must] [V seem] [ADV incredibly] [A stupid] [p to] [D that] [N girl]

Likewise, we could use the system of labelled bracketing to represent the fact Ihallthis hor] and [that girl] are NPs (Noun Phrases), that [to that girl] is a PP (Prepositional Phrase), that [incredihly stupid] is an AP (Adjectival Phrase), that 1,11'1'11/ incredibly stupid to that girl] is a VP (Verb Phrase), and that [This

52

53

Structure

Intuitions about Structure J J

boy must seem incredibly stupid to that girl] is an S (Clause) as follows:

(5)

[S [NP [0 this] [N boy]] [M must] [VP [V seem] [AP [AOV incredibly] [A stupid]] [pp [p to] [NP [0 that] [N girl]]]]]

representation of structure (because they occupy less space on the printed page). But recall that the two systems of representation are logically equivalent (whatever structure can be represented in the one system can be represented in the other and vice-versa), so that the question of whether you adopt one or the other is entirely a matter of typographical convenience, not a matter of theoretical significance. Accordingly, we shall usc both systems freely in our exposition here.

We can summarise our discussion so Iar in the following terms. Sentences are not just unstructured sequences of sounds; rather, they havc a hierarchical constituent structure in which sounds arc grouped together into words, words into phrases, and phrases into sentences. Each constituent (word or phrase) in a sentence belongs to a specific svntactic catcgorv, Overall, then, we can say that sentences have a catcgorial constituent structure: or in other words, sentences are built up out of sets of constituents, each of which belongs to a specific category. This catcgorial constituent structure can be represented schematically in the form of a Phrase-marker (= labelled tree-diagram), or labelled bracketing.

A key assumption which we are making here is that all sentences have a catcgoria! constituent structure i.e. that all sentences are hierarchically structured out of words and phrases, and that each of the component words and phrases in a sentence belongs to a specific category. Thus far, we have implicitly assumed that the evidence in support of this claim comes from native speakers' intuitions about the syntactic structure of sentences within their language. But this is only a very small part of the available evidence; moreover, it is a type of evidence which ean be rather unreliable. For while experienced linguists over a period of years tend to acquire fairly strong intuitions about syntactic structure, untrained informants by contrast tend to have very weak, uncertain, and unreliable intuitions. For this reason, it is more satisfactory (and more accurate) to regard constituents and categories as having the status of theoretical constructs. That is to say, they are part of the grammatical apparatus which the linguist finds he needs in order to explain certain facts about language (just as 'molecules', 'atoms', and 'subatomic particles' are part of the apparatus which the physicist finds he needs in order to explain the nature of matter in the universe). Inevitably, then, much of the evidence for constituents and categories hangs on heavily theory-internal arguments - arguments of an essentially empirical character (i.e. arguments based on observed facts about language). In the remainder of this chapter, we shall present a substantial body of empirical evidence in defence of the assumption that sentences have a catcgorial constituent structure. We'll begin (in the next section) hy look iug at the evidence that there are a variety of different word-

(By convention, only one member of any pair of brackets is usually labelled with the relevant category label, generally the lefthand one.) The resultant diagram (5) is entirely equivalent to (3) - i.e. the two diagrams contain exactly the same information as each other. Generally speaking, tree-diagrams are easier to read (because the information they contain is less condensed), and this is the reason why they are frequently preferred by many linguists as a form of visual representation of syntactic structure.

Diagrams like (3) and (5) provide a virtually complete representation of the syntactic structure of a sentence like (I). Quite often, however, the linguist may prefer to give a less detailed account of the structure of a sentence, where he does not want to include absolutely every minute detail of the structure, but rather just wants to give a partial representation of the structure. It sometimes happens that the linguist will simply want to indicate the major phrases within a sentence, without worrying about the internal structure of those phrases. For this reason, it is quite common to find in the relevant literature partial tree-diagrams, or partial labelled braeketings. For example, suppose that, for the purposes of some point that I am making, it is essential to assume that a sentence like (I) contains three major constituents, the Noun Phrase [this hoy], a Modal Auxiliary Verb must, and a Verb Phrase [seem incrediblv stupid to that girl], but that it is irrelevant to my argument what the internal structure of the Noun Phrase and Verb Phrase may be. In such a case, in place of the full tree diagram (3), I might prefer the partial P-marker (6) below:

(6)

(In partial P-markers like (6), it is quite common to use a 'triangle' to represent constituents with a complex internal structure that you don't choose to represent.) Or, in place of the detailed labelled bracketing (5), we might prefer the partial bracketing (7) below:

(7)

[S [NP this boy] [M must] [VP seem incredibly stupid to that girl]]

Generally speaking, linguists tend to use labelled tree-diagrams (P-markers) for a full representation of the syntactic structure of a sentence (because at that level of detail they are easier to read), and labelled bracketings for partial

54

55

characteristic of natural languages. In this connection, it is important to distinguish between two very different types of ambiguity: (i) lexical ambiguity, and (ii) structural ambiguity. Lexical ambiguity is ambiguity attributable to the fact that some particular lexical item (= word) has more than one meaning. For example, the word hall is ambiguous as between one sense in which it means a 'round object used for playing games', and a second sense in which it means a 'dance'; the ambiguity here is purely lexical. But there is a second type of ambiguity characteristic of natural languages, illustrated by the sentence:

Structure

level categories: and in subsequent sections, we'll go on to look at the evidence in support of saying that words are grouped together to form phrase-level categories of various sorts.

You should now be able to tackle exercise I

2.3 Word-level categories

What evidence is there that words belong to various categories of different types - categories like Noun, Verb, Adjective, Adverb, Preposition, Modal, Determiner, and so forth? As we shall see, evidence in support of this assumption comes from a wide range of sources.

Part of the evidence is phonological in nature: that is to say, constituents and categories prove essential constructs for any adequate description of the phonology of natural language sentences. Consider, for example, the problem of formulating a phonological rule which will provide a principled account of word-stress in cases such as the following (where the primary-stressed syllable is capitalised):

Word-level categories 2.3

(II) Mistrust wounds

In a case such as (II), the ambiguity might be argued to be structural in nature: that is to say, how we interpret (II) depends on whether we take it to be a Noun + Verb sequence parallel to 'Suspicion wounds', or a Verb + Noun sequence parallel to 'Mistrust sores'.

In fact, traditional grammars typically go much further than this, and claim that categories can be defined (in part) in terms of their semantic properties (i.e. meaning), along the lines indicated in (12) below:

(8) (a) (b) (9)(a) (b) (10) (a) (b)

We need to inCREASE productivity We need an INcrease in productivity Why do you torMENT me?

Why do you leave me in TORment?

We might transFER him to another club He's asked for a TRANSfer

(12) (i) (ii) (iii) (iv)

In each of the (a) sentences, the stress falls on the second syllable (capitalised) of the relevant word; whereas in each of the (b) sentences, the stress falls on the first syllable. Why should this be? An obvious answer is that in each case the relevant word (increase, torment, transfer) has the categorial status of a Verb in the (a) sentences, but the categorial status of a Noun in the (b) sentences. That is, the relevant generalisation is that words like increase carry primary stress on the second syllable when used as Verbs, but on the first syllable when used as Nouns. However, any such generalisation presupposes that phonological rules have access to categorial information: i.e. that such rules have to 'know' what the category of a constituent is before they can apply to it. Overall, then, the claim that words belong to a variety of different categories seems to have strong empirical support from phonological facts.

A second type of evidence in support of positing that words belong to categories is semantic in nature. More particularly, such an assumption makes it possible to provide a principled account of certain types of ambiguity

(v) (vi)

Verbs denote actions (go, destroy, huv cat, ctc.) Nouns denote entities (car, cat, hill, Reagan, etc.) Adjectives denote states (ill, happy, rich, etc.)

Adverbs denote the manner in which something is done tbadlv, slowly, painfully, cynically, etc.)

Prepositions denote location (under, ovcr, outside, in, on, ctc.) Determiners serve to specify (e.g. in 'this book', this serves to specify which book)

Semantically-based criteria for identifying categories such as those in (12) above are generally referred to as ..!!.!!..!ig!!E!..~:.it~r.~a... They are, however. extremely unreliable: for example, assassination denotes an action, but is a Noun (not a Verb); illness denotes a state, but is a Noun (not an Adjective); uv fast [[Link] indicates the manner in which the food is prepared, but is an Adjective (not an Adverb); in tall women, tall serves to specify what kind of women we are talking about, but is an Adjective (not a Determiner); and Cambridge denotes a location, but is a Noun (not a Preposition). Not surprisingly, therefore, we shall not make much use of notional criteria here!

l-ur more reliable are morphological and syntactic criteria: and in fact, the bulk or the evidence in support of postulating that words belong to categories I' crt her morphological or syntactic in nature. Let's examine first the morpholuginil evidence, This concerns the fact that certain types of inflection (i.e.

56

57

Structure

Word-level categories 2.3

grammatical ending) attach only to specific categories: we can thus identify individual categories according to the range of inflections which they permit. For example, Verbs in English can be recognised by the fact that they have up to five distinct forms: they have an uninflected hase form, and may take as many as four different inflections (the present tense -.I, past tense -d, participle -n, and gerund -ing inflections), as illustrated in the Table of Verb Forms (13) below:

( 13) Base Part icipl« Past Present Gerund
hew hewn hewed hews hewing
mow mown mowed mows mowing
prove proven proved proves proving
sew sewn sewed sews sewing
shave shaven shaved shaves shaving
show shown showed shows showing
strew strewn strewed strews strewing (15) ean - could - *cans - *cannen *canning

shall - should - "shalls - *shallen - *shalling may - might - *mays - *mayen - *maying must - *musted _. *musts - *musten - *musting

So, in other words, we can justify our distinction between Verbs and Medals on morphological grounds (though, as we shall see later, this distinction is supported by syntactic evidence as well).

We can also use morphological criteria to identify Adjectives and Adverbs, since these are the only categories (a subset of) which have a comparative form In -er: cf.

(16) tall taller: fat fatter: fast faster: etc.

(The criterion is complicated by the fact that Adjectives and Adverbs longer than two syllables have no -er form, so that we don't find Adjective forms like "intclligenter.i And we can differentiatc the two morphologically in that Adverbs generally carry a distinctive -Ir inflection, as the paradigm in (17) below illustrates:

Like most morphological criteria, however, this one is complicated by the irregularity of English inflectional morphology: for example, many Verbs have irregular PAST or PARTICIPLE forms, and in some cases either or both of these forms may not in fact be distinct from the base form, so that a single form may serve two or three functions, as illustrated in the Table of Irregular Verbs (14) below:

( 14)

Participle gone spoken seen

Pasl went spoke saw

Base go speak see

- - - - - - - - come - - - - - - - - came

meet

- - - - - - - - met - - - - - - - -

- - - - - - - - - - - - - - cu t - - - - - - - - - - - - - -

Present Gerund
goes going
speaks speaks
sees seeing
comes coming
meets meeting
cuts cutting ( 17) Adjcctiv« Adverb
sad sad/r
lucky luckily
desperate despera tell'
arrogant arrogant/r
nonchalant nonchalantlr
aggressive aggressive/r (It should be noted, however, that there is a handful oftrrcgular" Adverbs like tast. hard, etc. in English which do not end in -lv. and may have the same form as the corresponding Adjectives.) Moreover, we can differentiate Nouns from (lor example) Adjectives morphologically by virtue of the fact that only Nouns not Adjectives take a plural inflection (usually -.I): hence the contrast in:

The picture becomes even more complicated if we take into account the Verb he, which has eight distinct forms! In fact, the most regular Verb inflection in English is the -ing inflection, which can be attached to the base form of almost any Verb in English to form a gerund (though cf. minor exceptions such as bewarei.

This morphological property of having five (potentially distinct) forms dilferentiates Verbs from so-called Modals (or Modal Auxiliaries), which have no PARTICIPLE (-n) or GERUND (-ing) forms (and may also lack other forms): cr.

5X

(I X) They are idiots (Noun)/*idiolies (Adjective)

I [Link]. we can define Prepositions (somewhat negatively') by saying that they .vcinvariablc forms which cannot take the Verb inflections -s, -d, -n, or -ing, or t lu: comparative inflection -cr, or the Adverb inflection -Iv, or the Noun plural

\ mtlccuon: hence. a Preposition like at always remains completely unIldkl'll'd' d.

11(/) at *~Its *attcd*atten!*atting/*atter!*atly

Structure

Word-level categories 2.3 Using the same type of distributional criterion, we could argue that only a Verb (in its base form) can occur in the position marked - in (23) below to complete the sentence:

And as for Determiners well, I was rather hoping you wouldn't ask' Morphologically speaking, they arc a heterogeneous bunch: most are invariable, though this has the plural form these, and that has the plural form those: but as an overall category, they have no single defining morphological characteristic which they all share. (Determiners are best defined in svntactic terms, as we shall see shortly.)

Having seen that there is ample morphol ogical evidence in support of positing that words belong to categories, let's now turn to examine the relevant syntactic evidence. This is distributional in nature: that is to say, it rests on the assumption that categories are distributional classes. The distribution of a word in this sense is the set of sentence-positions it can occur in. For example, if we want to complete the sentence in (20) below by inserting a single word at the beginning of the sentence, in the position marked

(23) They/it can-

And support from this claim comes from the contrasts in (24) below:

(24)(a) (b) (c) (d) (e)

They can staylleavelhidel dielstarvel cry [Verb] "They can gorgeouslcutelgrottyltrendy [Adjective] "They can upl [Link] lofflon [Preposition] *They can womanldoorlbiblelgoldlcamera [Noun]

Etc.

(20)- can be a pain in the neck

Conversely, the only type of word which could be used to begin a three-word sentence such as (25) below:

we can use a Noun (in either its singular form, or plural form, or both), but not a Verb, Preposition. Adjective. Adverb. or Determiner. as we see from (21) below:

is a Modal: cf.

(25) --I be frank?

(21) (a) (b)

Lingllistics/Jo/tll/Girls/Telnisioll [Noun] can be a pain in the neck * Went [Verb]," For [Prepositionj/*Older [Adjectivej;*Conscielltiousll' [Adverb];*T/tc [Determiner] can be a pain in the neck

(26) (a) (b) (c) (d)

Thus. using our distributional criterion. we might define the class or Nouns as the set of words which can occur in the position marked in (20). However. it's important to be clear about what we mean by this. What we do mean is that any word which can occur in the position marked is a Noun; what we emphatically dont mean is that the position in (20) is the only position which Nouns can occur in. Any such claim would clearly be preposterous, since a typical Noun such as Linguistics can occur in a wide variety of sentence positions, as we see from (22) below:

(22) (a) (b) (e) (d) (e)

C(fIl/CouldjSha///Should/Mustl be frank? [Modal] *Go/* Wanted/* Tried/* Am I be frank') [Verb]

* Slow]" Ready]" Keen]" Proud I be frank? [Adjective] Etc.

And the only category of word which can occur after very (in its sense of cxtrcmclv; is an Adjective or Adverb, as we see from (27) below:

(27) (a) He is very slow [very + Adjective]

(b) He walks very slowly [very + Adverb]

(c) "Very girls love to have fun [very + Noun]

(d) *He very adores her [very + Verb]

(e) *Jt happened very alia the party [very + Preposition]

Moreover. we can differentiate Adjectives from Adverbs in distributional terms. For example, only Adverbs can be used to complete a four-word sen-

tcncc such as 'He treats her ': ef.

Is Linguistics a spectator sport?

Some people consider Linguistics kinky

Has anyone ever been known to die from Linguistics') Do you hate Linguistics more than monetarism?

Etc.

L'X) (a) (11) (c) Id)

Hence, it's important to be clear about the logic of the distributional argumentation here: there's an obvious difference between 'Only X occurs in Y' and 'X occurs only in V': if you don't see tht; difference, try an elementary Logic textbook'

60

He treats her badlvl politelyiarrogantly [Adverb] *He treats her nice/proper/good/strange [Adjective] *lle treats herfriendl.f(iol/woman [Noun]

Etc.

.: \ lid smcc Adjectives (but not Adverbs) can serve as the complement of the \l',h bc, we can delimit the class or Adjectives uniquely by saying that only

61

Structure

Wor d-lcvel categories 2.3

Adjectives can be used to complete a four-word sentence or the Iorrn They are very - ': cf.

(33)

A word-level mtcf!,ory is a set or words which share a common set or linguistic (especially morphological and syntactic) properties

(29) (a) They are very talllprettvlkindlnice [Adjective)

(b) *They are very slowlvtinnocentlvinicelv [Adverb)

(c) *They are very gClltlcl1lcn!lailiesj/ilXi's [NoLIn)

(d) Etc.

One important point which we should clarify, however, is that individual words may belong to more than one category. For example, the word need can function as a Verb (in which case it takes the characteristic range of verb inflections such as -.1'), as a Modal (and so can be used in structures such as (25) above), and as a Noun (and hence can be premodified by a Determiner such as (1): the examples in (34) below illustrate these three uses of need:

As for the distributional properties of Prepositions. well they alone can be premodified (i.e. modified and preceded) by right in the sense ofcompletely':

(30) (a) (b) (c) (d) (e) (f) (g)

Go right up the ladder

He fell right down the stairs He arrived right 0/1 time It's right under the bed

I'm right out of matches He went right inside

He tell right of/the ledge

(34) (a) (b) (c)

He needs to see a doctor [Verb) Need he be there? [Modal)

I feel a need to explore my roots [Noun)

He wrote thel same'[ew'rnunyino other works [Determiner) *He wrote succcssj'ul/had/lonK other works [Adjective)

*He wrote hc/lwl'e/see/hear other works [Verb)

Etc.

Thus far, we have argued that we need to posit that words belong to a restricted set of categories (such as Noun, Adjective, Adverb, Verb, Modal, Preposition, Determiner) in order to attain observational adequacy - i.e. in order to arrive at an accurate description of the facts of English Phonology, Morphology. and Syntax. However, we can go much further than this and suggest that we cannot attain our ultimate goal of explanatory adequacy (accounting for language acquisition) unless we do recognise that words belong to categories. For, if there were no ca tegories in language, and if every word had its own utterly idiosyncratic set of linguistic properties, then the task of acquiring competence in a language would be an impossible one (within the constraints that the child operates under). By contrast, if words are grouped into a small finite set of categories, and if phonological, morphological, syntactic and semantic rules are all category-based, then the child's acquisition task is enormously simplified. For, if in fact all grammatical rules are category-based, then the task facing the child acquiring (for example) the Syntax of a language becomes the relatively simple one of identifying which words belong to which categories in the target language, and which categories can occur in which sentence positions.

To avoid making our discussion too abstract, let's take a concrete example.

Let's suppose that a child learns that in English the words a, the, this, that, m\" vour . his, etc. all belong to the category D (= Determiner). It follows that ifhe hears a sentence like:

By contrast. other categories cannot be intensified by righ! (in standard varieties of English): cf.

(31) (a) * He right despairer! [Verb)

(b) *Life right can be cruel [Modal)

(c) *She is right pret tv [Adjective]

(d) *She looked at him right strangclv [Adverb)

(e) *She chose right this one [Determiner)

(though in some dialects or English, (31) (c) and (d) are well-Formed). Finally, we might delimit the class or Determiners in distributional terms by saying that they are the only class or items which can occur in the position marked-below to complete a five-word sentence such as 'He wrote - other wor kis)': cf. (32) below:

(32) (a) (b) (c) (d)

Thus, all six of the categories we have encountered so far can be justified on distributional grounds.

Given the wide variety or different types or evidence in support of positing that words belong to categories, we might attempt to arrive at a composite definition of a word-level category in the following terms:

62

(35) I want ([ toy

then. grvcn the assumption that all syntactic rules are category-based, he can [Link] infer that not only a but also all other members or the same catl'gory (l )) Gill occur in the italicised position i.e., having heard (35), he will be

Struciure

Phrasal categories: nonsynt actic evidence 2.4

ahle to infer that the sentences in (36) below arc also gralllInallL",i1 in English:

(37) (a) (b) (c) (d) (e)

I want a hall

I want a banana

I want a blackboard I want a hike

Etc.

In the remaining sections of this chapter, we look at the evidence in support of the assumption tha t sen tenees are structured not only out of words belonging to various word-level categories, but also out of phrases belonging to the corresponding set of phrasal categories. I n this section, we shall briefly outline some non-syntactic evidence in support of this assumption; but the remaining sections of our chapter will be concerned with presenting a vast array of syntactic evidence in support of our claim that sentences are structured out of phrases of various different categorial types. The nonsynt actic evidence we shall present in this section is of three types: morphological. semantic. and phonological: let's begin by looking at the relevant morphological evidence.

Generally speaking, Morphology is <I property of words rather than phrases: for example, inflections (e.g. plural -s, past tense -d, comparative -er) attach to the end of words of a particular type, not to the end of phrases, But there is one inflectional morpheme in English which is indeed a phrasal inflection and that is the genitive (= possessive) 's inflection. Of course, this is difficult to tell if we take a simple Noun Phrase which ends in a Noun, as in:

(36) (a) I want the toy
(b) I want this toy
(c) I want that toy
(d) I want my toy
(e) I want your toy
(I') I want his toy
(g) Etc. Furthermore, given the same assumption that <III syntactic rules arc categorybased (or structure-dependent, to use the equivalent term suggested by Chomsky), he can also infer that in place of 101" in (36) he can substitute any other item helonging to the same category N ( = Noun), and thus generate, for example:

(38) This crown is th« king's

In short. the twin assumptions that syntactic rules are category-based, and that there are a highly restricted finite set of categories in any natural language (perhaps no more than a dozen major categories). together with the assumption that the child either knows (innately) or learns (by experience) that all rules arc structure-dependent (=c<ltegory-hased), provide a highly plausible model of language acquisition, in which languages hecome learnable in a relatively short. finite period of time (a few years).

For in a case like (3X), we cannot be sure whether the genitive '.\' inflection is attached to the end of the Noun king, or to the end of the whole Noun Phrase th« king. But if we choose <In example of a Noun Phrase in which the head Noun has a postrnodificr of some sort (i.e. an expression positioned after the NOlin which modifies the Noun), the situation becomes clearer: the genitive '.\' inflection must clearly attach to the end of the whole Noun Phrase, not to the end of the head Noun precisely as we see from (39) below:

(39) (a) (h)

This crown is [NP the [N king] o] Englandj's *This crown is [NP the [N killgj's of England]

You should now be able to tackle exercise /I

Here. the Noun king is followed by the postmodifying PP of England: and the genitive inflection's attaches to the end of the whole Noun Phrase [1171' king of I:"lIglalldj, not 10 the Noun king. So, it seems that genitive's is a Noun Phrase inflection, not a Noun inflection. Moreover, the genitive intlection in English can attach only to <I Noun Phrase, and not (for example) to an Adjectival Ph rase: cr.

2.4 Phrasal categories: nonsyntactic evidence

Thus far. we have presented evidence in support of the claim that words in natural language belong to a highly restricted finite set of word-level categories such <IS Noun, Verb, Modal, Adjective, Adverb, Preposition, Determiner. etc. But recall that in section 2.1 we posited that the major wordlevel categories can be expanded into the corresponding ph rasa I categories by the addition of other constituents. For example, Nouns can be expanded into Noun Phrases, Verbs into Verb Phrases, Adjectives into Adjectival Phrases, Adverbs into Adverhial Phrases, and Prepositions into Prepositional Phrases.

(40)

*This crown is [AP l'('IT handsomeis.

lhus, t hc postulation ora NOlI/1 Phrascconstituent is justified on morphologic<ll grounds, since it IS not obvious how we could describe the grammar of the )!L'IIIII\l' '.\ inflection in FIl)!lish without saying that it's a Noun Phrase inflecnon M ore )!l'nL'r<lII" st ill. morpholouica I facts force us to recognise tha t sen-

64

tences arc structured out of Phrases belonging to various phrasal categories such as (in our present case) Noun Phrase,

In addition to morphologica! evidence such as that discussed above, we can also adduce scnurnt ic evidence in support of our assumption that sentences are structured out of Phrases. Part of the evidence relates to the phenomenon of structural ambiguitv which we discussed earlier. In simple cases, the structural ambiguity may relate to the catcgorial status of a particular phrase, as shown in the following example (adapted from Huddleston (1976), p. II).

Structure

Phrasal categories: nonsvntactic evidence 2.4

scope over (i.e. to modify) the Verb Phrase [rat if), the treaty]. If these assumptions are correct, then (42) would have the simplified structure (45) (i) below on interpretation (43) (i), and the structure (45) (ii) on interpretation (43) (ii):

(45) (i)

(41) Mary looked I'('IT hard

(ii)

NP~r---------Vp

/~ I ~~

0he Preside~ could not ratify the treaty

the ambiguity concerns the categorial status of the italicised phrase [very hardi, On one interpretation, it is an Adjectival Phrase (AP) paraphraseable by 'very severe'; but on the second interpretation, it is an Adverbial Phrase (ADVP) pa raphrascablc by 'very intensely'.

A rather different kind or structural ambiguity occurs when a given word or phrase (with a given categorial status) can be taken as modifying anyone of two (or more) different constituents. For example, consider the following sentence:

(42)

The President could not ratify the treaty

For our present purposes, what is interesting about (45) is that the negative S('OP(, olllhiguity of (42) is represented in structural terms. The key premise of our analysis is that clauses like those above have three immediate constituents INP M VP], and that not may have scope either over M, or over VP. Since we are saying that not lies outside the VP in (45) (i) but inside the VP in (45) (ii), it is clear that the constituent VP is a key construct in our account of the scope ainbieuit v in (42). Or, to stand the argument on its head, negative scope facts provide strong support for the postulation or a VP constituent.

It is interesting to note that there seems to be some independent svntactic evidence in support or the structural ambiguity proposed in (45). For, an Adverb like simplv would be positioned after not on interpretation (43) (i ), but before not on interpretation (43) (ii ): cr.

If you think about it very carefully. you'll realise that (42) is ambiguous as between the two interpretations (4.1) (i ) and (ii) below:

(4.1) (i) (ii )

It would not be possible for the President to ratify the treaty It would be possible for the President not to ratify the treaty

The two different interpretations become clearer if we use (42) in the two different contexts in (44) below:

(44) (a)

(46) (i)

The President could not sintplv ratify the treaty (= (43) (i))

(b)

Even if he wanted to, the President could not ratify the treaty, because he doesn't have the aut horny to do so ( = (43) (i))

If he really wanted to make it clear that he isn't interested in disarmament, the President could not ratify the treaty. By not ratifying it, he would make his opposition to arms reduction patently obvious (= (43) (ii))

( ii ) The Presiden t could simplv not ratify the trea ty ( = (43) (ii))

Now. if we suppose that sintplv in such cases is positioned in front of the leftIllost constituent or the Verb Phrase. then the dual analysis ill (45) would provide a very straightforward account of the dual position or simply. For. in I'~~) ( i}, ratitv is the leftmost constit uent or Vf'. so that simply is positioned in 11(\11 I oj' ratii v in (46) (i ). But in (45) (ii), not is the first consti t uent of the Vf', so IILII simpl, is positioned in front of not in (46) (ii),

I 11II her support Ior the claim tha t 110/ modifies the Modal could on in terpre- 1.111(\11 (4.1) (i) but modifies the VP [ratif), the treaty] on interpretation (43) (ii) ,(l11lL', 110m the ract that the sentence (42) has two different pseudo-cleft ,"llllInpalls. (Psc udo-clcft sentences are sentences of the schematic form \\11;11 .X did \\;IS Y' ) Thus. the pseudo-cleft counterpart of sentence (42) is

The ambiguity of (42) is structural in nature. and relates to the scope of the negative particle not (i .e. relates to the question or which constituent is being modified by not v. for this reason. this type of ambiguity is known as scope ambiguitv. On the first (and most obvious) interpretation (43) (i), not might be argued to have within its scope (i.e. to modify) the Modal could; whereas on the second (less obvious) interpretation (43) (ii), not might be argued to have

66

67

Structure

Phrasal categories: distributional evidence 2.5

(47) (i) (ii)

What the President could not do is ratify the treaty (= (43) (i)) What the President could do is not ratify the treaty (= (43) (ii))

which it applies. More precisely, the rule applies to contract a negative which modifies a Modal, but not to a negative which modifies a YP. But any such account of NEGATIVE CONTRACTION presupposes that sentences have as their immediate constituents [NP M YP], and thus requires us to posit the existence of phrasal categories such as (in particular) YP. More generally still, we can say that the claim that sentences arc structured out of Phrases of various sorts receives independent support from phonological facts.

Overall, then, we can conclude this section by saying that the claim that sentences are structured out of Phrases is supported by morphological, semantic, and phonological evidence. But it should be remembered (as we pointed out at the beginning of this section) that the bulk of the evidence in fa vour of this assumption is svntactic in nature: and accordingly, the rest of our chapter is devoted to presenting various different kinds of syntactic evidence in support of this claim.

(47) (i) below on interpretation (43) (i), and (47) (ii) on interpretation (43) (ii):

And it seems clear that not goes with the Modal could in (47) (i ), but with the YP [rati(\" the treaty] in (47) (ii). Thus, although the structural difference between (4S) (i) and (4S) (ii) is not reflected in a word-order difference in our original sentence (42), there are variants of(42) in which this structural difference does indeed directly correlate with a word-order difference e.g. in sentence pairs such as (46) and (47) above.

Just as differences in syntactic structure between sentences correlate with semantic differences (i.e. if two sentences differ in syntactic structure, then they will also differ in semantic structure), so too they correlate with phonological differences (so that, for example, sentences with different superficial syntactic structure will have different stress patterns, etc.). By way of example, let's consider a phonological rule in English which I shall call NEGATIVE CONTRACTION. Modals in English generally have contracted negative forms, so that alongside the 'full forms' in the lefthand column in (48) below, we find the corresponding contracted forms in the right hand column:

You should now be able to tackle exercise 11/

(49)

NEGATIVE CONTRACTION is usually only possible where the negative modifies the Modal, and not where it modifies the following Yerb Phrase

2.5 Phrasal categories: distributional evidence

So far, we have seen that there is morphological, semantic, and phonological evidence in support of the assumption that sentences have a cutcgorial constituent structure. However, it should be repeated that the bulk .)1' the relevant empirical evidence is syntactic in nature. Many such syntactic arguments relate to facts about the distribution of various sequences of words I recall that many of the arguments we gave in support of positing word-level [Link] were of this kind): i.e. the argumentation takes the form: 'Unless we postulate that sentences are structured out of Phrases belonging to various [Link]. we cannot account for which sequences of words can occur in II 11Ieh positions in which types of sentence.' And the whole of this section will hl' devoted to considering distributional arguments in support of our claim 111;1 t sentences are structured out of Phrases belonging to categories of various k 1I1ds.

()I1l' such distributional argument concerns a phenomenon which we shall ulrr to simply as Proposing. Under appropriate stylistic conditions (e.g. to .u IIICH' a particular rhetorical effect), certain parts of a sentence may be pre\''''''d 'tllr emphasis' (especially where the preposed sequence is being con- 11.I\il'd WI t h some other sequence). For example, one way of emphasising the ,"<llll'lll'l' 1(1111' cl.k-r sister in a sentence like:

(48) Fullform Contractedform
would not wouldn't
should not shouldn't
could not couldn't
might not mightn't However, NEGATIVE CONTRACTION appears to be subject to a structural condition, which we can gloss informally along the lines 01'(49) below:

If this is so, then we should expect contraction to be possible in a structure like (4S) (i) above, but not in (4S) (ii). And this does indeed seem to be the case: for, a sentence such as:

(SO) The President couldn't ratify the treaty

can only have interpretation (43) (i) (= 'It would not be possihle for the President to ratify the treaty'), and not interpretation (43) (ii) (= 'It would he possible for the President not to ratify the treaty'). What this means is that our NEGATIVE CONTRACTiON rule is sensitive to the structure of the sentence t«

68

69

Structure

Phrasal categories: distributional evidence 2.5

(52)

is to prepose it i.e. position it at the front of the rest of the sentence, as in:

[Your elder sisler], I can't stand (though your brother's OK)

the up + NP sequence docs not form a phrasal constituent, it cannot therefore be preposed 'for emphasis'. So, it would seem that the relevant generalisation governing Preposing is that:

However, it turns out that while some parts of sentences can be preposed in this way, others cannot. For example, various types of phrasal constituent (such as those bracketed in (53) below) can freely undergo preposing: cf.

But only a whole phrase (and not just port of« phrase) can be preposed in this way: thus, in a sentence such as (51) above I can't stand your elder sister, only the whole NP (Noun Phrase) [your elder sisler] can be preposed for emphasis, not just part of the phrase: cf

(53)(a) (b)

(54) (a) (b) (c) (d) (e)

(57)

Only phrasal constituents (i.e. whole phrases) can undergo Preposing

(c)

[ThaI kind of behaviours; I simply will not tolerate (Noun Phrase)

I went to see the new James Bond film yesterday, and [very exciting] it was, too (Adjectival Phrase)

[Very shorlly], he's going to be leaving for Paris (Adverbial Phrase)

[Down the hill] John ran, as fast as he could (Prepositional Phrase) [Give in 10 hlackmai/], I never will! (Verb Phrase)

But note that this account of Proposing presupposes that sentences have a catcgorial constituent structure.

A parallel argument can be formulated in relation to the converse phenomenon of Postposing Under particular circumstances, it may be possible (or necessary) far a particular expression in a sentence to be postposed. This is particularly frequent with expressions which are (in some vague intuitive sense) felt to be especially 'heavy' or 'long'. For example, the bracketed expression in (58) (a) below can optionally be postposed - i.e. moved to the position it occupies in (58) (b):

(d) (e)

(58) (a)

He explained [alllhe terrible problems that he had l'nCmlfllered] to her

He explained to her [all the terrihlc problems that he had encounlered]

Your elder sister, I can't stand * Your elder, I can't stand sister * Elder sister, I can't stand your * Sisler, I can't stand your elder * Your, I can't stand elder sister

(b)

However, not just any random string of words can be postposed in this way; for example, in (58) (a) above, it would not have been possible to postpose the string terrible problems that he had encountered. as we see from the ungrarnmaticality of:

Likewise, a nonconstituent sequence (i.e. a string of words which do not together form a (phrasal) constituent, but which belong to two or more separate constituents) cannot be preposed: hence, in the case of sentences such as:

(55) (a) (b) (c)

(59) *He explained all the to her terrible problems that he had encoun-

t ered

John rang up his mother John stood up his date

John looked up her phone number

Why should it be that we can postpose the string all the terrible problems that he had encountered. but not the string terrible problems that he had encountered'! The obvious answer is that all the terrible problems that he had encountered is a complete Noun Phrase, whereas terrible problems that he had encountered is not (rather, it is a subpart of the larger containing Noun Phrase). And we might assume that just as only complete phrases can undergo Proposing, so too only complete phrasal constituents can undergo Postposing . In other words, we might revise our earlier principle (57) as:

we cannot prepose up together with the rest of the sentence, as we see from the ungrammaticalityof:

(56) (a) (b) (c)

* Up his mother, John rang * Up his dale, John stood

* Up her phone number, John looked

«(,0)

Only phrasal constituents can undergo Preposing or Postposing

The reason is that in each of the sentences in (55), the particle up forms a constituent together with the Verb (giving us the so-called 'Phrasal Verbs' ring up, stand up, and look up), not with the immediately following Noun Phrase: since

70

Since Prl'po,l"illg and PO,l"lpo,l"illg arc both types of Movement; we can refine (60) Iurt hLT as:

Structure

Phrasal categories: distributional evidence 2.5

reply the cinema, isn't he using just part of the Prepositional Phrase to the cinema, in apparent violation of the principle (64) that only complete phrases can be used as sentence-fragments? The objection is an interesting one (shows you're beginning to think like a syntactician at lastl), but entirely misconceived! To see why, consider the structure of the Prepositional Phrase [to the cinema], which we might represent as in (66) below:

While it is perfectly true that the cinema is part of the PP to the cinema (it is the object or complement of the Preposition to), it is equally true that the cinema is itself a Phrase namely, a Noun Phrase (NP). So in replying simply The cinema' in (65), speaker B is in fact using a complete Phrase (to be more precise, a complete Noun Phrase), and hence there is no violation of our earlier principle (64).

Let's now turn to look at one final distributional argument in support of the assumption that sentences have a categorial constituent structure. This particular argument relates to the distribution of adverbial expressions. From a syntactic point of view. there are two different classes of Adverbials in English: one including Adverbs such as cert ainlv ; and the other including Adverbs such as completely. If we consider a simple sentence such as:

(61) Only phrasal constituents can undergo Movement (from one position in a sentence to another)

But once again. note that this account of Movement presupposes that sentences have a hierarchical constituent structure.

A further distributional argument in support of constituent structure can be formulated in relation to facts relating to what are traditionally known as sentence-fragments. Consider the following dialogue:

(66)

(62) SPEAKER A: Where did he go?

SPEAKER B: Up the hill

Instead of replying with a full sentence like 'He went up the hill', speaker B chooses to give a 'short-form' reply using the sentence-fragment 'U p the hill'. Any adequate grammar of English will clearly have to specify what kinds of sequences can and cannot be used as sentence-fragments in this way, since some sequences cannot be so used: cr.

(63) SPEAKER A: Who were you ringing up?

SPEAKER B: *Up mv cider sister

Why the contrast between (62) and (63)'1 An obvious answer is to suggest that up the hill is a constituent of the sentence He went up the hill, whereas up my sister is not a constituent of a sentence such as I was ringing up my sister. We might then propose that only constituents can serve as sentence-fragments. Moreover, it seems likely that only complete phrases can function in this way, not parts of phrases; thus, for example, speaker B could have replied with the Noun Phrase my elder sister in (63), though not with the simple Noun sister. The relevant restriction would therefore appear to be that:

(67)

PP

P~ ~NP

t~ D~ ~N

I I

the cinema

The team can rely on my support

(64)

Only phrasal constituents (i.e. whole phrases) can serve as sentence-fragments (in an appropriate context)

we find that these two classes of Adverbs can occupy rather different positions in the sentence, as (6!l) below illustrates:

Thus, any account of the use of sentence-fragments also requires us to assume that sentences have a categorial constituent structure.

However, it might at first sight seem as if there are some obvious potential counter-examples to the claim in (64) that only complete phrases can be used as sentence-fragments. Consider, for example, the following dialogue:

(68) (a) (b) (c) (d) (e) (0 (g) (h)

(65) SPEAKER A: Where are you going to?

SPEAKER B: To the cinema/The cinema

Certainlvl=completelv, the team can rely on my support

*The certainly/completely team can rely on my support The team certainly/*completcly can rely on my support The team can certainlylcompletelv rely on my support The team can rely complet elyl+certainly on my support

*The team can rely on certainly/completely my support "The team can rely on my certainly/completely support The team can rely on my support completely/certainly

Here, speaker B can reply either with the complete Prepositional Phrase to the cinema, or simply with the cinema. But- we might object - in using the shorter

l l ow can we account for the different distribution of Adverbs such as certainly on the one hand. and complctct» on the other') We can provide a simple answer

72

Structure

Adverbs like certainly are S-adverbs, and hence can only be attached to an S-node

(ii) Adverbs like completely are VP-advcrbs, and so can only be

Phrasal categories: other syntactic evidence 2.6

J

i.e. presupposes that sentences are hierarchically structured out of word-level

and phrase-level constituents, and that such constituents belong to categories such as N . N P. V·· VP. P PP. A - AP. ADV ADVP. etc.

(69) (i)

to this question in constituent structure terms, along the lines of (69) below:

attached to a VP-node

You should now be able to tackle exercise IV

(70)

Let's assume that sentence (67) has the structure (70) below:

Now, since certainlv is an S-adverb. it can occur in positions where it is directly attached to S (viz. at the beginning or end of the overall sentence. between team and can. and between can and relv], but not in positions where it would be directly attached to VP (c.g. between rely and on), or to PP (e.g. between on and my). or to NP (e.g. between my and support). Possible positions for S-adverbs can be indicated schematically by the asterisks in (71) below:

Possible positions for S-adverbs like certainly

[S * The team * can * [VP rely on my support Vp] * S]

Conversely, since completely is a VP-adverb, it can occur only in positions where it is immediately attached to VP (viz. at the beginning of the whole VP as in (68) (d), at the end of the whole VP as in (68) (h), and in the middle of the VP between rely and on as in (68) (e)), and cannot occur in positions where it is directly attached to S (e.g. before the, or between team and can), or to PP (e.g. between on and my), or to NP (e.g. between my and support). Possible positions for VP-adverbs can be indicated schematically by the asterisks in (72) below:

(71 )

~S~

NP ~ VP

r >. I / ~

o N can V PP

I I I /~

The team rely P NP~

o~ 0/ N

I I

my support

2.6 Phrasal categories: other syntactic evidence

Thus far, we have argued that distributional evidence provides strong syntactic support for our claim that sentences arc structured out of Phrases belonging to various categories (NP, VP, PP, AP, ADVP. etc.). In this section, we arc going to look at three other types of syntactic evidence supporting this claim.

The first such evidence comes from a phenomenon known as Coordination.

English and other languages have a variety of Coordinating Conjunctions (CONJ) like and. or, hut, etc. which can be used to coordinate or conjoin words or phrases: cf.

(73) (a) (b) (c) (d)

He has a (at and a dog

I met your mother andjl/ther

Is she ill tlu: kitchen or ill tlu: bathroom" He speaks I'crr slowl v but I'err art iculatclv

In each 01" the sentences in (73), a Coordinating Conjunction has been used to conjoin the italicised pairs of words or phrases. Clearly, any observationally adequate grammar of English will have to provide a principled answer to the

I

questions:

(74) What kinds or elements can and cannot be coordinated in English')

Now. it turns out that we can't just coordinate any random set of elements in English. as we see from the ungrammaticality of:

(75) '" John rang up his mother and up his sister

Why can the italicised items be conjoined in (73), but not in (75)'! One possible answer could be that the italicised elements can be conjoined in (73) because they are constituents, whereas the coordinated sequences in (75) are not constitucnts (because up in (75) forms part of the Phrasal Verb ring up, so that the sequence up his mother is not a constituent). That is, we might suggest a principle along the lines 01":

(7(') Only constituents can be conjoined: nonconstituent sequences ca nnot be conjoined

Possible positions for VP-adverbs like completely

[S The team can [VP * rely » [pp on my support] '" Vp] S]

The preceding discussion about Adverbials serves to illustrate very clearly the fundamental principle that any statement about the distribution of particular types of expression requires reference to constituents and categories

(72)

74

7~

Structure

Phrasal categories: other syntactic evidence 2.6

But can a constituent of one type (or category) be coordinated with a constituent of a different type? Consider the following paradigm:

(77) shows us that we can conjoin two PPs or two NPs quite freely, but that any attempt to conjoin an NP with a PP, or a PP with an NP leads to an unnatural, unidiomatic, and 'forced' sentence (the type of sentence which, precisely because of its very unnaturalness, may sometimes be used for humorous effect). This suggests that we might revise our principle (76) as:

(78) Only identical categories can be conjoined, idiomatically

Let's now test the predictions made by the principle (78) for a wider range of categories; it predicts that the following sentences will all be grammatical, because in each case the italicised pairs of words belong to the same category:

(77) (a) (b) (c) (d)

(79) (a) (b) (c) (d) (e) (f)

John wrote to Mary and to Fred (= PP and PP) John wrote a letter and a postcard (= NP and NP) *John wrote a letter and to Fred (= NP and PP) "John wrote to Fred and a lelia (= PP and NP)

egory . This discussion serves to underline our remark at the beginning of this chapter that the real status of constituents and categories is that of theoretical constructs - i.e. abstract elements without which it is not obvious how we could provide a principled explanation of linguistic phenomena such as Coordination.

The type of Coordination illustrated in sentences like (79) and (80) above is what we might call Ordinary Coordination. However, there is a second (rather different) type of Coordination which we shall call Shared Constituent Coordination (though for irrelevant historical reasons it is more widely known in the literature as Right Node Raising). This second type of Coordination can be illustrated by sentences such as:

(81) (a) (b) (c)

John walked (and Mary ran) [up the hill]

John denied - but Fred admitted - [complicity in the crime] John will, and Mary may, [go to the party]

Good linguists and philosophers a re rare (N and N)

John is a very kind and considerate person (A and A) There arc arguments/iii' and against this claim (P and P) J.R. walks and talks like a true Texan (V and V)

You can bring these and those books (0 and D)

He opened the door quite slowly and deliberately (ADV and ADV)

In a sense which we shall not attempt to make precise here, the italicised sequence in each example is shared (so to speak) between the two conjuncts. Thus, for example, up the hill seems to go with both John walked and Mary ran in (81) (a). What is more important for our present purposes is that there are restrictions on this type of Coordination - as illustrated by the ungrammaticality of examples such as:

(82) * John rang (and Harry picked) up Mary's sister

In each case in (79) we have conjoined identical word-level categories; but the principle (78) also predicts that we should be able to conjoin identical phrasal categories as well. And this prediction appears to be borne out by sentences such as those in (80) below:

Why are sentences like (81) grammatical, but those like (82) ungrammatical? We might seek an answer to this question along the following lines. We might argue that a sentence like (81) (a) is (in an obvious sense) somehow an elliptical (i.e. 'short') form of:

(80) (a) (b) (c) (d) (e)

(83)

John walked up the hill and Mary ran up the hill

[NP The man next door] and [NP his wife] are very nice He is a lAP very shy] and [AP rather inarticulate] man He went [ppto London] and [pp to Paris]

He may [VP go to London] and [VP visit his mother]

John drives [ADVP very slowl)'] and [ADVP very carefully]

and that in both of the italicised conjuncts, the sequence up the hill is a constituent of the italicised clause containing it. By contrast, we might say that (82) is an elliptical form of:

(84) John rang up Mary's sister and Harry picked up Mary's sister

It thus seems that the principle (78) has considerable empirical support. But note that the principle makes crucial reference to constituent structure and category membership: in other words, it is not clear how we could provide any principled account of Coordination if we did not posit that sentences are structured into sets of constituents, each of which belongs to a particular cat-

76

and that in neither of the italicised conjuncts is up Mary's sister a constituent of the Clause containing it: for, in the first conjunct up 'goes with' the Verb rang to form the Phrasal Verb rang up, and in the second conjunct up 'goes with' the Verb picked, forming the Phrasal Verb picked up). So, it seems that we might try and answer the question: 'Why is Shared Constituent Coordina-

77

Structure

lion possible in cases like (X I) but not in (X2)'/' in terms of a principle such as the following:

Phrasal categories: other svnt actic evidence 2.6

boring hook Oil Transformational Grammar) by him, so that speaker A could equally have asked:

(X5)

Shared Constituent Coordination is only possible where the shared string is a possible constituent of each of the conjuncts

(88)

What do you think of him')

But note that the answer we have given to our question about when this type of Coordination is and is not possible makes crucial reference to the constituent structure of sentences: i.e. in order to provide a proper account of Shared Constituent Coordination, we have to posit that sentences have a categorial constituent structure.

Let's now turn to another linguistic phenomenon whose systematic description also seems to require the postulation of an abstract categorial constituent structure associated with sentences namely, Proninninalisation, It is a general property of natural languages that they possess devices for referring to entities mentioned elsewhere in the same sentence or discourse. Consider. by way of example, the following discourse:

Thus, the term Pronoun is doubly inappropriate, since it is not an accurate description either of the distribution of such items (i.e. they don't occur in sentence positions typically occupied by a Noun), or of the class of expressions which can serve as their antecedents (since Pronouns typically refer back to Noun Phrases, not to individual Nouns). Words like him, it, etc. are more accurately described as pro-NP (pro-Noun Phrase) constituents, since they occur in NP positions in a sentence, and generally have NPs as their antecedents (though we shall shortly qualify this remark).

In more general terms, since Pronouns 'replace' or 'refer' back to other constituents. we might refer to them as pro-constituent s, or proforms. Not all proconstituents are pro-NPs, however. One might argue, for example. that in a discourse such as:

(86)

SPIAKER A: What do you think of the guy who wrote that unbelievably boring book on Transformational Grammar') SPI:AKER H: I can't stand him

(X9)

SPEAKER A: Have you ever been to Paris') SPEAKER H: No. I have never been there

the word ther« is a pro-PP (pro-Prepositional Phrase). in that it occupies the same sentence-position as the Prepositional Phrase 10 Paris. In much the same way, we might argue that in sentences such as the following. the italicised words function as pro-VP constituents. and hence refer back to the bracketed VI':

In (X6) the italicised Pronoun provides a handy way of referring to the individual concerned without the need to repeat the full phrase [/h(' gill' who wrote 11/(/1 unbclicvablv boring hook 0/1 Transformational Grammar]. Using traditional terminology. we can say that [Ille gill' II'I/(} wrote that unbclicvablv boring hook 0/1 Transfonnational Gramtnari is the antecedent of the Pronoun him. However. the term PrOIlOUi/ (although a traditional one) is peculiarly inappropriate here, simply because the antecedent of a Pronoun like him is not in fact a Noun. but rather a whole Noun Phrase: that is. him refers to the whole Noun Phrase [/he gUI' wlto wrote that unbclie vablv boring hook on Transformational Grammar). and not to any individual Noun within the Noun Phrase. Likewise, the position occupied by so-called 'Personal Pronouns' like him in sentences is not that of a Noun, but rather that ofa Noun Phrase- hence the fact that we could not replace the Nouns gUl', hook, and Grammar in speaker A'S utterance in (86) by the corresponding Pronouns him and it, as we see from the ungrarnrnaticality of:

(90) (a) (b) (c)

John might [VP go home), and so might Bill John might [VP resign his post], as might Bill

If John can [VP speak French fluently) which we all know he can why is he shy with French girls')

And in (91) below, we might argue that so functions as a pro-AP (proAdjectival Phrase) constituent:

lhus, there's more to preforms than at first meets the eye! In fact, there seem l(l he almost as many different kinds of proform as there are Phrases (so, for n:lI11pk. hint is a pro- N P. III('/'(' can be a pro-PP, as can be a pro- VP, so can be ,I pro-A], or a pro-Vl' and so on). It is interesting to note that proforms ",'I\,'r:lll\ replace /,II/'{/\('-/('\'('/ const itucnt s, not word-level constituents.

(')1)

(87)

"Wha t do you think or the him who wrote that unbelievably boring il on Transformational i()

But precisely because pronouns like him occur in typical NP positions in sentences, we could replace the whole NP [/he guy who wrote that unbelievablv

78

Many people consider John [AP ex tremcl v rllde), but I've never found him so

79

Any adequate description of pro-constituents or proforms will have to be concerned with both the syntax and semantics of them. Any adequate syntax of proforms will have to specify which sentence-positions particular proforms can occur in; while any adequate account of the semantics of proforms will have to specify what kind of expressions can be construed as the antecedent of a given proform. But it turns out that any adequate description of either the syntax or the semantics of proforms will have to make crucial reference to constituents and categories.

We can make this discussion more concrete in relation to the proform it. An adequate syntax for it will have to answer the question: 'What range of sentence-positions can it occupy in English?' The answer is that it can occupy the same range of sentence-positions as a typical Noun Phrase like the book on the table, as we see from (92):

Structure

(92) (a)

Phrasal categories: other syntactic evidence 2.6

(94) (a)

SPEAKER A: Mary hasfinished her assignment SPEAKER B: [ don't believe it

Jean is pregnant again, l just know it [I'I don't turn up, it won't be fair on you SPEAKER A: Is there snow outside? SPEAKER B: Does it matter?

They say he's planning to resign, but it may not be true

(b) (c) (d)

(e)

(b)

{The book on the table}. . .

IS mterestmg

It

[ {the book on the table}

s you~

it

[ like t~e book on the table}

I' d b {the book on the table}

ve rea a out

it

Each of the italicised phrases in (94) can serve as the antecedent of it, and each appears to have the categorial status of S (= Clause). So it would appear that not only an NP, but also an S can be interpreted as the antecedent of the proform it.

This shows up an important discrepancy between the syntax and semantics of certain types of proform: syntactically, it occurs only in typical NP positions, not in typical S positions. Thus, for example, a verb like hope can take a following S complement, but not a following NP complement: cf.

(95) (a) [ hope [vou will come] (= hope S)

(b) *[ hope [the pen on the table] (= hope NP)

(c)

Significantly, it cannot be used as a direct complement of hope:

(d)

(%) *[ hope it

[M.I' ("{Ir] is OK when it works

[The back scat ofmy car] has got books on it The trouble with [the table] is that it's too small [I' [ find [the lid ofthe kettle], I'll give it to you

[ never use [that bone-handled kniti>], because it's blunt

[Link] (96) shows us that it cannot occur in a typical S position in a sentence; on the contrary. syntactically it occurs only in N P positions. But examples like (94) show us that semantically it can have an S as its antecedent. This reveals an obvious discrepancy between the syntax and semantics of it.

There are two conclusions to emerge from our discussion of PronominalisatlOI1. The first is that Syntax is autonomous of Semantics, so that in con-cquence. any adequate description of proforms must include both a syntactic .uul a semantic characterisation of their function (since these two functions ma v he distinct). It's not much use simply specifying that such-and-such a Iorrn is a pro-NP, pro-PP, pro-S, or whatever: is that meant to be a description O!"lts syntactic distribution, or of the class of expressions which can serve as its .uucccdcnt , or both') The second (and more general) conclusion to be drawn 1'0111 our discussion is that any adequate account of either the syntax or ,,'111;111 tics of pro-constituents will have to make crucial reference to structural I"' lpnt les of sentences. For it turns out that proforms have the same distribu- 11<>11 ;1S cnta I n phrasal ca tegories on the one hand, and that the class of expres',I<>I1S w hul: can serve as their antecedents is most simply characterised in ,.II<T<>'"'' rcrrn-, 011 till' other. Once again, then, we find a linguistic phenom-

Syntactically, then, it has the same distribution as a typical Noun Phrase: i.e. syntactically, it functions as a pro-NP.

But what about the semantics of it? What kinds of expression can be interpreted as the antecedent of it? All the italicised expressions in (93) below are possible antecedents for it:

(93) (a) (b) (c) (d) (e)

What have the italicised expressions in (93) got in common? The obvious answer is that they are all Noun Phrases, since they can all occur in the (bracketed) NP positions in (93). Thus, an obvious generalisation about the semantics of it is that it can be interpreted as having an (inanimate, singular) NP as its antecedent.

But is it only Noun Phrases which can serve as the antecedent of proforms [ike it? Data like (94) below cast doubt on this:

80

XI

Structure

Phrasal categories: other syntactic evidence 2.6

enon (i.e. Pronominalisation) which cannot he described adequately without appealing to notions olconstituent structure.

Now let's turn to further empirical evidence supporting the same conclusion. Under certain discourse conditions, it is possible in English (and other languages) for some part of a sentence to undergo Ellipsis (i.e. to he omitted), provided that the missing part of the sentence can be 'understood from the context'. For example, in a discourse such as (97) below, it is possihle to omit the slashed material. since it is identical to something said hy the previous speaker. and is thus in some sense 'redundant':

(101) John won't put the vodka into the drink, hut his brother will p~f

t~¢ y¢~~;-i irltf¢ 111¢ ~tI!'I~ ,

Why is it that the slashed material can he omitted in (10 I) but not in (lOa)? We can give a principled answer to thi~ question in terms of constituent structure. Let us assume that the structure of a sentence such as His brother will put the vodka into the drink is as follows:

( 102)

~s~

NP ~ VP

r >. I ~I~

D N will V NP PP

I. I I / -, / -.

HIS brother put D N P NP

I I I / ~

the vodka into D N

I I

the drink

(97) SI'EA"LR A: John won'tll·[Link] the dishes

SPI:AHR B: I bet he will W,H~ tl1¢ ¢i~~¢$ if you're nice to him

The same type of Ellipsis is also found in (9X):

(9X) John won't help me with the dishes, hut his hrother will I1¢Jp Jjl¢ with rhe di~l1es

Given the assumption of a constituent structure such as (102) above, we can suggest a tentative answer to the question we asked earlier ahout what wordsequences can and cannot undergo Ellipsis, namely:

In a fairly obvious sense, the slashed material in (97) and (9X) can he omitted because it is identical to the italicised material. and is thus in some sense 'recovcrablc. An obvious qucsuon which any observationally adequate grammar of English is going to have to answer is:

(IOJ) Only VPs (Verb Phrases) can undergo Ellipsis (under appropriate discourse conditions)

(99)

What kind 01" clements can undergo Ellipsis (i.c, he omitted) in this wav. and under what conditions')

Principle ( I OJ) would provide a ready account of why omission of the slashed rna terial was possihle in ( 101 ), hut not in ( 100); for in (10 I) the slashed material constitutes a VP constituent, whereas in (100) (a) it was an Nt oun), in (100) (h) an NP (Noun Phrase), in (100) (c) a PP (Prepositional Phrase), in (100) (d) a nonconstituent N PP sequence, and in (100) (e) it was a nonconstituent NP PP sequence. Further support for the principle (IOJ) comes from the fact that it correctly predicts that the slashed VPs in (104) helow can he omitted since they are identical to the italicised VPs (and hence are 'recoverable'. or 'red unda nt, in some sense):

In actual Iuct. we can't just omit 1II1l' random set of words in a sentence when they are identical to another set, as we see from the fact that the following sentences arc ungrammatical If the slashed material is omitted:

(100) (a) "John won't put the vodka into the drink, hut his brother will put the vodka into the ~til~~

(h) *John won't put the vodka into the drink, hut his brother will put the vodka into tnt ¢rin~

(c) *John won't put the vodka into the drink, hut his brother will put the vodka ivl(v) t~¢ ¢til~k

(d) "John won't put the vodka into the drink, hut his brother will put the v¢¢~~ i¢l¢ 111¢ ¢til'lk

(e) "John won't put the vodka into the drink, hut his hrother will put the v¢¢~~ i¢(¢ nn¢ ¢tI!'I~

(104) (a) He may come home earl,", hut then again he may not ¢¢Jjl¢ \II¢Jjl¢

¢<ttl'f

(h) Mary wants to close the shop; hut I don't want to ¢J¢¢¢ fl1¢ ~11¢P

(c) Felch 1111' 1/11 app!«. if you can r¢f¢11 Jjl¢ ~¢ ~pf>J¢

(d) sprAKER A: Could you have a look at the car') SI'IAHR B: OK, I willl1~v¢ ~ l¢¢~ ~f tl1¢ ¢#

Although we can't omit any of the slashed sequences in (100) above, we can omit that in (10 I) helow:

What's inll:resting about the above discussion is that it tells us that we cannot attain ;IIlV adequate description of the phenomenon of Ellipsis unless we ITl"<')!Ill\C that 'l'nknccs have a catcgorial constituent structure.

X2

Structure

Words used as Phrases 2.7

In our discussion in sections 2.3-2.6 in this chapter, we have presented a substantial body of empirical arguments in support of the key claim that sentences are hierarchically structured into word- and phrase-level constituents, each of which is assigned to a particular category. In general, these twin assumptions are uncontroversial -- i.e. virtually any linguist working on Syntax from any theoretical perspective will accept the need to recognise the existence of an abstract syntactic structure associated with sentences. Where linguists disagree with each other is over the question of just how many different categories there are in a particular language or universally - not over the question of whether there are ca tegories.

an NP (= Noun Phrase); and that useful, in addition to being an A (= Adjective) is also an AP (= Adjectival Phrase); thus, in place of (106) above, they would prefer the analysis (107) below:

(107)

~s~

NP ~ VP

I I /~

N can V AP

I I I

C~ ~ A

I useful

You should now be able to tackle exercises Vand VI

But -- we might object - surely it is absurd to say that cars is a Noun Phrase and useful an Adjectival Phrase when each of these constituents contains only one word: after all, isn't it in the very nature or things that a Phrase is a sequence of two or more words? The answer to this is simply: 'No, not the way that we use the term Phrase here.' Any theory defines the technical terms it uses in relation to the way in which it uses them: and within the kind of model we are discussing, the term Phrase is used to mean simply 'a set of elements which form a constituent', with no restriction on the number of elements that the set mayor must contain. So, for example 'Noun Phrase' can be taken as meaning something like: 'an expression containing a head Noun', without any implication that a Noun Phrase must necessarily contain anything other than the head Noun.

But of course it isn't enough merely to say that if we define the notion Phrase in the appropriate way, then we could call expressions consisting of a single Noun like cars Noun Phrases. The real question posed by such an analysis is: 'What is the empirical justification for doing so?' - i.e. what set of facts could we explain thereby which might otherwise go unexplained? One set of facts concerns the distribution of expressions like cars. Notice that in place of cars in a sentence like (105) above we could also have any of the italicised expressions in (108) below:

2.7 Words used as Phrases

So far, we have assumed that there are essentially two different levels of categories in language: on the one hand there are word-level categories like N, V, P. A, ADV, etc., and on the other hand there are phrase-level categories such as NP, VP, PP, AP, ADVP, ctc.; and the distinction between the two seems obvious enough: e.g. dog is a Noun, and [this dog] is a Noun Phrase. However. empirical evidence which we shall present in this section leads us to the conclusion that the relationship between word-level categories and their phrasal counterparts is hy no means as straightforward as one might Imagine. By way of illustration, consider a sentence such as:

( 105)

Cars can be useful

There doesn't seem to be anything controversial in saying that cars is an N (= Noun) in (105), can is an M (= Modal), he is a V (= Verb), and useful is an A (= Adjective); in addition, given the assumptions we have been making, we might argue that he useful is a VP (= Verb Phrase), and that the whole sequence Cars can he useful is an S (= Clause/Sentence). Hence it might seem reasonable to suppose that (105) has the S-structure (106) below:

( 106)

S

N~~~VP

I I /~

Cars can V A

I I

be useful

(I08)(a) [Fast cars] can be useful

(b) [Veryfast cars] can be useful

(c) [Those veryfast cars] can he useful

And the same is true of other sentence-positions: wherever we can have a full Noun Phrase like [l'ery./ast cars], we can also have the simple Noun cars, as the paradigm in (109) helow illustrates:

But in fact, this analysis would probably be rejected by most linguists; instead, they would want to say that cars, in addition to being an N ( = Noun) is also

( I ()I))

(a)

1· verv last cars}

Do . . turn you on?

cars

84

85

Structure

. {verYfast cars}

(b) I really enjoy

cars

. {very last cars}

(c) I'm Just crazy about .

cars

( 112)

In other words, a single unmodified Noun can have the same distribution as a Noun Phrase. Accordingly, we are justified on distributional grounds in saying that unmodified Nouns have the status of Noun Phrases (as well as that of Noun): and this is the reason why we posited that cars in (107) above is not only an N but also an NP. Consequently, we define 'Noun Phrase' as meaning 'Phrase containing a head Noun' (irrespective of whether or not it also contains nominal modifiers of one sort or another).

The same kind of reasoning can be extended to arguing that an Adjective like [useful] is actually also an Adjectival Phrase in sentences like (105) [Cars can he useful]. For, of course, here [useful] is occurring in exactly the position in which we could find an Adjectival Phrase like [very useful] (cf. Cars can he very uscful s. And more generally, wherever we can have a full Adjectival Phrase, we can have a simple unmodified Adjective on its own, as we see from the paradigm in (110) below:

Words used (/.1' Phrases 2.7

~~~

Nr ~ /Vp~

N must V ADVP

I I I

John speak ADV

I clearly

And the same distributional argument also extends to the analysis of simple Verbs that lack following complements, e.g. leave in a sentence such as:

( 113)

John may leave

Here we might be tempted to say that leave is simply a Verb: but note that in place of leave we could have a Verb Phrase like [/('al'(, home]: cf.

( 114) John may leave home

Wherever we can have [leave home] we can also have simply leave: hence, on distributional grounds, we might claim that leave in (113) has the same catcgorial status as [leave home] in (114) i.e. that of VP. Thus, the structure of (113) will be:

I" summarise. then: wherever a simple category (like N, V, P, A, ADV, etc.) in ;1 glVCIl sentence can be replaced by the corresponding phrasal category (e.g. N 1', VI', Pl', AP, ADVP, ctc.), then the simple category is to be regarded as an [Link] of the corresponding phrasal category. So, for example, an unmodilil'd Noun which can be replaced by a full Noun Phrase is to be taken as an 111\t;IIlCC of NP; an unaccompanied Adjective which is replaceable by a full Adlcctlval Phrase is to be regarded as an instance of AP ... and so on, and so l ort h What this in effect means is that we regard the key property ofa Noun 1'111;1\(' as being that of containing a head Noun (though not necessarily any- 111111).' else). the key property of an Adjectival Phrase as being that of containIllf' ;I head Adjective (though not necessarily anything else) ... and so on and ',,, I "II h.

III else you are feeling sceptical about this line of reasoning (I sensed you \\('11"1, let me I ry and present more evidence in support of this crucial point of "IIIlI·lplc. Ior example. why should I want to say that in a sentence such as:

(110) (a)

She is {very keen on sport}

keen

I, I I' d h {very keen on sport}

ve a ways roun er

keen

{VerVkeenon~port} .,

. though she IS, she s hopeless

Keen

( 115)

(b)

(c)

Thus, we might say that because a single unmodified Adjective has the same distribution as a full Adjectival Phrase, then it follows that an unmodified A should be assigned the status of AP as well as A -- precisely as with useful in (107) above. Accordingly, we might define an Adjectival Phrase as: 'a Phrase which contains a head Adjective' (with or without modifiers).

Of course, exactly the same reasoning also extends to other constituents: for example, we could argue on distributional grounds that in:

(111) John must speak clearly

the word clearly as well as being an ADV (= Adverb) is also an ADVP (= Adverbial Phrase), since clearly here occurs in exactly the kind of sentenceposition in which we could find an ADVP like [rather more clearly]. Thus, the structure of (111) would be:

86

S

NP~~~VP

I I I

N may V

I I

John leave

87

Structure

Testing the Structure 2.8

(116) Most people can't stand hypocrisy

(121 )

S

NP~~~VP

.>>: I / <.

D N can't V NP

I I I I

Most people stand N

I hypocrisy

the word hypocrisy is a Noun Phrase comprising simply the head Noun hypocrisy? Well, for one thing, the word hypocrisy can be preposed for empha-

SIS, as In:

(117) Hypocrisy, most people can't stand

As we saw earlier in relation to sentences such as (54) above, only full phrases can be preposed in this way: it therefore follows that hypocrisy in (116) must be a full phrase of some sort; and since the phrase concerned is headed by the Noun hypocrisy, it must be a Noun Phrase.

Moreover, hypocrisy in (116) can serve as a sentence fragment, as we see from the dialogue in (118) below:

The obvious conclusion is thus that unmodified Nouns have the status of Noun Phrases (as well as Nouns): and the principle extends from Nouns to all other categories (so that, for example, wherever an unmodified Adverb can be replaced by a full Adverbial Phrase, it is assigned the status of an ADVP containing a head ADV).

(118)

SPEAKER A: What can't most people stand? SPEAKER B: Hypocrisy

You should now be able to tackle exercise VII

And sinee we earlier suggested that only full phrases can serve as sentence fragments, this again suggests that hypocrisy in (116) must be a full NP.

A third argument leading to the same conclusion can be formulated in relation to Ordinary Coordination. For, note that hypocrisy can be coordinated with a full Noun Phrase such as [the kind of glib lies that politicians tell], as we see from (119) below:

2.8 Testing the Structure

What we have argued hitherto is that a whole variety of linguistic phenomena cannot be described adequately without assuming that every sentence in every natural language has a specific syntactic (constituent) structure. Hut an obvious question that arises is:

(119) Most people can't stand [hypocrisy] or [the kind of glib lies that politicians tell]

( 122) How do we determine what the constituent structure of a given sentence in a given language is?

(120)

{hYPOCriSY} Most people can't stand it

Sometimes we can rely on intuition as a first approximation: but while linguists tend to acquire fairly refined intuitions of this sort over a period of years, non-linguists may not have particularly firm intuitions about the struct life of sentences in their language. So, when our intuitions fail us, how can we determine what the constituent structure of a given sentence is?

Perhaps we can best answer this question by asking another question: 'Why do we posit the existence of constituents and categories?' The answer we have Vlven here is: 'In order to provide a principled account of, for example, Distrihution, Preposing, Postposing, Coordination, Adverbials, Pronominalisation, I llipxix, etc.' An obvious answer to the question: 'How do we determine II bet her a given sequence of words is a constituent of a given type?' is thus: .( )11 the basis of whether it behaves like a constituent of that type with respect I" l rixtribution, Preposing, Postposing, Coordination, Adverb position, ProI" .minulisa tion, Ellipsis, etc.' More precisely, we might propose the following ./1"1~1I().\li("s for determining whether a given set of words in a sentence is a [Link] or not, and ifso, of what type (=category):

Since we earlier posited in (78) that 'Only identical categories can be conjoined', and since it is clear that the second bracketed conjunct in (119) is a full NP, then it follows that the first conjunct must also be a full NP.

Support for this conclusion comes from Pronominalisation facts. We suggested earlier in relation to sentences such as (93) above that proforms such as it function as pro-NP constituents, and that they replace full Noun Phrases (not individual Nouns). But note that hypocrisy in (116) can be replaced by the pro-NP it: cf.

This provides still further evidence that hypocrisy in (116) must have the categorial status of a full Noun Phrase. In other words, sentence (116) must have the structure (121) below:

88

X9

Structure

Testing the Structure 2.8

(123) (a) Does it have the same distribution as (i.e. can it be replaced by) an appropriate phrase of a given type? If so, it is a phrase of the relevant type

(b) Can it undergo movement (i.e., preposing or postposing)? If so, it is a phrase of some sort

(c) Can it serve as a sentence-fragment'! If so, it is a phrasal constituent

(d) Does it permit positioning of S- or VP-adverbials internally? If so, it is an S or VP, and not, for example, an NP or PP

(e) Can it undergo Ordinary Coordination with another string? If so, it is a constituent of the same type as the one with which it is coordinated

(f) Can it serve as the 'shared constituent' in Shared Constituent Coordination') Ifso. it is a phrase

(g) Can it be replaced by, or serve as the antecedent of an appropriate preform" If so, it is a phrase of the same type as the proform

(h) Can it undergo Ellipsis under appropriate discourse conditions? If so, it is a VP

(b)

S

NP~ ~ --------------- VP

I I /~

N would V NP

I .r>. / -,

Drunks V P D N

I I I I

put off the customers

Now at first sight. the two sentences in (124) might appear to be parallel in structure. After all. both comprise a sequence of a Noun (drunks), a Modal (would), a Verb (get/put). a Preposition (off), a Determiner (the), and a Noun (bus/customers). And yet, what I'm going to suggest here is that the two sentences have very different constituent structures. More precisely, I'm going to try and convince you that (124) (a) has the structure (125) (a) below, whereas (124)(b) has the structure (125) (b):

The essential difference between the two structures is that in (125) (a) the Preposition off' 'goes with' the following Noun Phrase [the bus) to form the Prepositional Phrase [off the hus): whereas in (125) (b), the Preposition off'goes with' the Verb put to form the complex 'Phrasal Verb' [put offl Thus, in traditional terms, we might say that get in (125) (a) is a Prepositional Verb (because it is a Verb which takes a Prepositional Phrase after it), whereas put in (125) (b) is a Phrasal Verb because the sequence (put offl seems to form some kind of ' compound verb'.

However, before we examine the differences between the two structures in (125), let's look first of all at the obvious similarities between the two. For example, in both cases I have claimed that drunks is not only a Noun, but also a Noun Phrase: but what justification is there for such an analysis') The simple answer is that the NP analysis can be justified on the basis or the structural diagnostics which we listed in (123) above. For example, if we apply the distributton criterion (123) (a), we can argue that drunks must be an NP because it has the same distribution as (and hence can be replaced by) a full NP such as [people who get drunk): cf.

( 126) (a) People It'/IO get drunk would get off the bus

(b) People who get drunk would put off the customers

Having outlined our structural diagnostics, let's see how we might apply them to help us determine the constituent structure of the following pair of sentences:

(124) (a) Drunks would get off the bus

(b) Drunks would put off the customers

And if we apply the movement criterion (123) (b), we might argue that drunks must be a full phrase (i.e. an NP) because it can undergo movement - e.g. it can be ~ostposed in questions such as:

(127) (a) Would drunks get off the bus?

(b) Would drunks put off the customers?

(125)(a)

~~ -------------

NP M VP

I I / ~

N would V PP

I I /~

Drunks get P NP <,

I / <,

off D N

I I

the bus

Likewise, the sentence-fragment test (123) (c) leads us to the same conclusion, given that only full phrases can occur as sentence-fragments, and that drunks l';1 n function in this way: cr.

I 1 ~~) (a) SPEAKER A: Who would get off the bus?

SPEAKER H: Drunks

(h) SPFAKFR A: Who would put off the customers'?

SI'I'AKI'R II: Drunks

90

91

Structure

Testing the Structure 2.8

Since criterion (123) (d) is not directly applicable to the question of the categorial status of drunks. let's turn instead to consider the Ordinary Coordination test (123) (c). This again supports the NP analysis. since drunks can he coordinated with a full NP such as [other undesirable ele/llents]: cf.

(b) If drunks would [put off the customers] - which they obviously would - then we ought not to allow them in the restaurant in the first place

(129) (a) Drunks and other undesirable clements would get off the bus

(h) Drunks and other undesirable clements would put ofT the customers

And I leave it to you to verify for yourself that the various other 'diagnostics' which we have devised likewise support the VP analysis for both sentences.

Similarly. we can apply the same set of tests in (123) to support our claim that the three major constituents of the sentence in each case are [NP M VP]. For example. such an analysis would correctly predict that S-adverbs like certainly can be positioned between NP and M, or between M and VP, but not between any other pairs of constituents; and this prediction is entirely correct in both cases, as we see from (134) and (135) below:

The Shared Constituent Coordination test ( 123) (f) is not applicable here. since the test only works with strings which are on the rightmost hranch (i.e. at the righthand end) of their containing Phrases or Clauses; and we see from (125) above that drunks in each case is the leftmost constituent of the S which contains it. However, the Pronominalisation test (123) (g) is applicable, and again supports the N P analysis. since drunks can he replaced hy the pro-N P thev: cf.

(130) (a) (h)

They would get off the bus

They would put off the customers

(134) (a) Drunks certainly would get off the bus

(b) Drunks would certainly get off the bus

(c) *Drunks would get certainly ofT the bus (d) "Drunks would get off certainly the bus (e) "Drunks would get off the certainly bus

Finally, we note that the Ellipsis criterion is not applicable to drunks here. since Ellipsis is a test for VP-hood. not for NP-hood. Thus. all the tests in (123) which are applicable here provide strong empirical support for our claim tha t drunks is not only a Noun. but also a Noun Phrase.

We might use the same set of tests in (123) to check our claim that the sequences [get ofT the hus) in (125) (a) and [put offthe custolllers) in ( 125) (b) are Verb Phrase constituents. For example. note that the VP in either case can undergo 'preposing' for emphasis in an appropriate discourse setting: cf.

(135) (a) Drunks certainly would put off the customers

(b) Drunks would certainly put off the customers

(c) * Drunks would put certainly off the customers (d) "Drunks would put off certainly the customers (e) "Drunks would put ofT the certainly customers

By now. you should be excruciatingly familiar with the logic of this type of argument, so I won't subject you to any further mental torture! Suffice it to say that there is abundant evidence in support of the claim in (125) that both sentences comprise an [NP M VP] sequence.

Of course, where the two sentences differ is in the internal structure of the VP which they contain. As we noted earlier, in the case of the VP [get off the hus], the Preposition off 'goes with' the following Noun Phrase [the bus] to form the PP [off the bus]; whereas in the case of the VP [put off the customers], (he Preposition off 'goes with' the Verb put to form the Phrasal Verb [put oUl But at the moment, this claim is based on nothing more substantial than my own personal structural intuitions. However, you may well have no more faith III Ill)' structural intuitions than I have in yours, so let's see whether we can find vornc independent empirical evidence in support of the claimed structural difterence!

The distribution argument (123) (a) would seem to lend support to our anah.,ls. After all, if the sequence [oflthe bus] in (125) (a) is a Prepositional I'hrase. then we should expect that it can be replaced by other PPs with a

(131) (a) I f the driver told the drunks that they had to get off the bus, then [get off the hus] they would

(b) The restaurant manager thinks that drunks would put ofT the customers, and [put offthe customers) they undoubtedly would

Moreover, both VPs can function as sentence-fragments in an appropriate context: cf.

(132) SPEAKER A: What would drunks do?

SPEAKER B: {Get off the bus }

Put offthe customers

And both VPs can serve as the antecedent for a pro-VP constituent such as which: cf.

(133) (a) If, whenever they needed to heed the call of nature, drunks would [get off the bus]- which they obviously would - then we ought not to allow them on in the first place

92

93

Structure

Testing the Structure 2.8

related meaning: since on is the antonym (i.e. 'opposite') of 01f~ we might expect to be able to replace the off~phrase in (124) (a) by an on-phrase. And as (136) below indicates, this is indeed possible:

Verbs like get ojrand get on select the same range of complements (i.e. can be followed by the same range of expressions):

(139) (a)

Drunks would get off the busl the train/the tablellthe seailthe wind/

(136)

Drunks would {get off' the hUS} get on the bus

'kindness

(b) Drunks would get on the busl thc train/the t ablellthe seaj'the wind! 'kindness

The fact that the sequence [off the hus] can be replaced by the PP [on the hus] would seem to confirm its PP status. But what is interesting is that in the case of the second example, the sequence off' the customers has no on counterpart: cf.

By contrast, no such parallelism of Selection Restrictions holds for Phrasal Verb pairs such as put off' (in its sense of 'deter') and put on (in its sense of 'wear'): cf.

( 137)

Drunks would {put offthe customers} * put on the customers

(140) (a)

Drunks would put off the customers/the waiterslldirty clothes/ 'tattered trousers

(b) Drunks would put on dirty clothesltauered trousersllthe customers/

Why should this be? Well, the answer suggested by analysis (125) above is that the sequence off' the customers is not a PP (indeed not a constituent at all), and hence cannot be replaced by a PP like [on the customers]. Thus, distributional facts seem to support the claim that put offv: a Phrasal Verb, whereas get offi« a Prepositional Verb.

However, we are over-simplifying a hit here, by ignoring the fact that the sequence put on can indeed be used as a Phrasal Verb, e.g. in sentences such as:

'the waiters

( 138)

Mary would put on a new dress

This parallelism of Selection Restrictions in the one case but not in the other seems related to the fact that Prepositional Verb pairs such as get off and get on have related meanings, whereas Phrasal Verb pairs such as put offand put on do not.

Let's now turn to further syntactic evidence in support of claiming that Prepositional and Phrasal Verbs have different structures. Our criterion (123) (b) tells us that only full phrases can undergo movement. In this connection, it is interesting to note that in get off' structures like (125) (a), the whole sequence [off the hus] can be preposed for emphasis: cf.

(141) Every afternoon, the big red bus would stop in front of the village clock, and [off' the hus] would get a dear old lady carrying a shopping bag

But whereas there is a consistent parallelism between get off' and get on, no such parallelism seems to hold between put off and put on. What kind of parallelism holds in the first case but not the second? Well, for one thing, get seems to have a constant meaning (in both cases, it is roughly synonymous with 'climb') irrespective of whether it is combined with on or off. But this is not at all the case with put; thus put [Link] means more or less the same as 'deter', whereas put on (in its most familiar sense) is roughly synonymous with 'wear'. Generalising somewhat, we could say that Prepositional Verbs have a consistent, componential meaning (i.e. the meaning of the whole expression is a simple function of the meaning of its component parts), whereas Phrasal Verbs tend to have an idiosyncratic or idiomatic meaning. What I mean by saying that Phrasal Verbs have idiosyncratic meanings is that there's no way of telling what put [Link] or put on mean simply by knowing the meaning of put, and on, and off,

A second parallelism which we find between different but related uses of Prepositional Verbs is a parallelism of Selection Restrictions - i.e. a parallelism in the range of different expressions which can be used after a Prepositional Verb. We see from the examples in (139) below that related Prepositional

Since only full phrases can undergo movement, it follows that the italicised sequence in (141) must be a full phrase; and since it contains the head Preposition off; it clearly must be a Prepositional Phrase. By contrast, note that the sequence offthe customers can't be preposed in (125) (b): cf.

(142) "The manager suspects that drunks would put off the customers,

and off' the customers they certainly would put

Why should this be? The answer suggested here is that only full phrases can be preposed in this way, and - as we see from the tree diagram in (125) (b) above the sequence of] the customers isn't a phrase: in fact, it isn't even a constituent. So, our 1I100'1'1I11'lIt test (12:1) (h) provides further support for the analysis proposed here.

94

95

Structure

Not surprisingly, our sentence-fragment test (123) (c) yields the same results for, as we see from (143) below, the sequence [off the hus) can serve as a sentence-fragment in (12S) (a):

(143) SPEAKER A: Did he get off the train?

SPEAKER R: No, off the hus

Since only full phrases can serve as sentence-fragments, this confirms the PP status of the italicised sequence. By contrast, the string off the customers in (12S) (b) cannot function as a sentence-fragment, as (144) below illustrates:

(144)

SPEAKER A: Would drunks put off the waitresses? SPEAKER B: "No, off the customers

Why can't off the customers function as a sentence-fragment? The answer suggested by analysis (I 2S) (b) is that the italicised sequence in (144) is not even a constituent, let alone a Phrase (and recall that only full Phrases can function as sentence-fragments). So, our sentence-fragment test lends further support to analysis (I 2S).

Our fourth test (123) (d) - relates to the distribution of Adverbial Phrases.

Recall that in 2.S we drew a distinction between Ss Adverbials (which occur in positions where they are attached to an S node), and Vl'-Adverbials (which occur in positions where they are attached to a VP node). The class of VPAdverbials includes expressions such as quickly, slowlv, completely, etc, Now, since VP-Adverbials can occur internally within VPs, then we should expect that such an Adverbial could be positioned in between the Verb get and the Prepositional Phrase [off the hus] in (I 2S) (a); and (14S) below shows that this is indeed the case:

( 14S)

Drunks would get slowly off the bus

By contrast, it is not possible to position a VP-Adverbial between put and off in (l2S) (b), as we see from the ungrammaticality of( 146) below:

(146) *Drunks would put completely off the customers

Why should it be possible for a VP-Adverbial to intrude between get and offin (12S) (a), but not between put and (~ff in (l2S) (b)? We can see the answer immediately, if we look at the structures which would be assigned to each of the Verb Phrases in (l4S) and (146) above: these are as in (147) below:

(147) (a)

96

Testing the Structure 2,8

(b) *

.i-> VP ~

V --------- -...... N P

v<6~p ~usto~

I I I

put completely off

(For brevity here. we omit the fact that slowly and completely have the status of ADV as well as ADVP: this is a simplifying convention which we shall adopt for other phrasal categories subsequently.) It should now be immediately obvious why (147) (a) is OK. but (147) (b) is not. For. since Adverbials like slowly and completely are VP-Adverbials, they can only attach to a VP node, as in (147) (a), What's wrong with (147)(b) is that here we have a VPAdverbial attached to a V node, not to a VP node, In other words, structures like (I 2S) (a) and (b) make precisely the right predictions about the distribution of VP-Adverbials with Prepositional Verbs on the one hand, and Phrasal Verbs on the other.

Let's now turn to our fifth constituent structure test, the Coordination test in (123) (e). Given our assumption that [off the hus) is a PP constituent in ( 12S) (a), we should expect that it can be coordinated with another PP of the same type: and as (148) below shows, this is indeed the case:

( 14X)

Drunks would get [off the hus) and [on the train]

But given our assumption that the sequence offthe customers is not a constitu"Ill or any type in (I 2S) (b), we should expect that it cannot be coordinated wit h another similar sequence: and this is exactly the right prediction, as the ungrummaticality of(149) below illustrates:

II ~l)) "Drunks would put off the customers and otT the waitresses

lhus, Ordinary Coordination facts support the analysis we proposed in (I 2S),

BUI what about the Shared Constituent Coordination test in (123) (f)? Given 'III r assumption that the sequence [off the hus] in (I 2S) (a) is a PP constituent, \\l' <hould expect that it can function as the 'shared constituent' in sentences 1I1\t1lving Shared Constituent Coordination: and we see from (I SO) below that 11m IS Indeed the case:

I I "(1) Drunks would get and junkies would fall - 0[( the bus

11\ '''Illras\. the sequence ofT the customers cannot be used in the same way in I I "I (h)' cr.

I I ',11 • Drunks would pUI and junkies would also put off the customers

'", I ill' ,\/iarl'd tonst n ucnt Coordination test also supports analysis (I 2S).

97

Structure

Let's move on now to consider the Ellipsis test, (123) (h). So far we have looked at one specific type of Ellipsis, which we might call VP ELLIPSIS, for obvious reasons. However, there are other types of Ellipsis found in English. One such type is known as GAPPING, because it has the effect of leaving a 'gap' in the middle of some Phrase or Clause. For example, the second occurrence of the Verb bought can be gapped in this way in a sentence such as:

Testing the Structure 2.8

Noun'. Quite right! After all, simple Coordination facts tell us that they must be Noun Phrases, since they can be coordinated with other Noun Phrases: cf.

(157)(a) (b)

Drunks would get off [the hus] and [the train]

Drunks would put off [the customers] and [the waitresses]

(152) John bought an apple, and Mary I'>¢~t~t a pear

Now, if they are Noun Phrases, we should obviously expect that they can be replaced by an appropriate pro-NP constituent such as it or them. And yet, while this is true of the object of a Prepositional Verb like get off' cf.

When a Verb is gapped, any Modal preceding it can also be gapped along with the Verb, even if the two do not form a continuous sequence, as in (153) below:

The trouble with the bus was that drunks would want to get off it every few miles, to exercise their natural bodily functions

it is not true of Phrasal Verbs such as put off ef.

"What worries me about the customers is whether drunks would

put off them

So, it would seem that Prepositional Verbs can take pronominal Objects, but Phrasal Verbs require non-pronominal Objects. Why this should be is not entirely clear. However, what does seem clear is that if we draw a systematic structural distinction between Prepositional and Phrasal Verbs, we can attain the level of observational adequacy simply by stipulating that Phrasal Verbs don't permit (certain types of) pronominal Objects. Of course, we would ultimately hope to find a principled explanation for this rather curious restriction, and thereby attain a higher level of adequacy. But the essential point remains that unless we draw some distinction such as that in (125) between Phrasal and Prepositional Verbs, we shall be unable to attain even the level of observational adeq uacy.

In actual fact, our claim that Phrasal Verbs don't take pronominal objects is something of an oversimplification, as (160) below illustrates:

(160) (a) "Drunks would put off them

(b) Drunks would put them off

We see in (160) that a Phrasal Verb like put of{ can indeed take a pronominal object, but only when the Preposition is positioned at the end of the sentence. Moreover, it isn't just a pronominal object which can appear between put and off: as (161) below indicates, an ordinary nominal NP can also appear in this position:

(158)

(153) Could John close the window, and ¢¢~,~ Mary ¢,¢$¢ the door?

( 159)

The exact conditions determining what kind of constituent can and cannot undergo GAPPING in a given sentence are extremely complex, and need not be of concern to us here. What is of more immediate interest to us here is the fact that the Verb get can be gapped along with the Modal would in structures such as (125) (a), resulting in sentences such as:

(154) Drunks would get off the bus, and junkies V<¢~,~ t¢t off the train

However, what is even more interesting is that we cannot gap the Verb put along with the Modal would in structures like (125) (b), as illustrated by the ungrammaticaiity of:

(155)

"Drunks would put off the customers, and junkies v<¢M~ pM off the waitresses

By contrast, we can gap the whole expression put of{ along with would: cf.

( 156)

Drunks would put off the customers, and junkies v<¢M~ -PM ¢rr the waitresses

Thus, it would seem that as far as GAPPING is concerned, the whole expression put ofl is somehow treated as a single 'compound Verb'. But this is the very intuition which we encapsulated in analysis (125) (b), by giving the sequence [put offl the status of a single V constituent (i.e. by analysing it as a Phrasal Verb). Thus, facts about GAPPING lend yet further support to analysis (125).

Finally, let's turn to the Pronominalisation argument in (123) (g). And here, we find something of a surprise waiting for us. But let me prepare you for the surprise first of all. Just think about the status of the expressions [thf' bus] and [the customas] in (125). What are they? Well, by now it should come as a reflex reaction to say: They are Noun Phrases consisting of a Determiner and a

(161 )

Drunks would put [the customas] off

By contrast, a Prepositional Verb like get off does not permit the Preposition to he moved to the end of the VP in this way: cf.

( 1(2)

9X

*Drllnks would get the bus ()ff

99

Structure

Thus, whereas a Phrasal Verb allows its accompanying Preposition to be positioned either before or after a Noun Phrase Object (though when the Object is pronominal, the Preposition must be positioned after the Object), a Prepositional Verb only allows the Preposition to be positioned before the NP Object.

The obvious question to ask at this point is what is the structure of a sentence such as (161), in which the Preposition is positioned at the end of the VP. We shall argue here that (161) has the structure (163) below:

( 163)

S

NP~ ~ --------------VP ______

I I ~ I ______

N would V NP PP

I I ~~ I

Drunks put the customers P

I off

It seems clear that ofrmust be a constituent of the VP here, since it can be preposed along with the other constituents of VP for emphasis: cf.

(164) The manager said that drunks would put the customers off, and

[put the customers offl they certainly would

But what might seem more controversial here is the claim that off· in (163) is not just a P. but also a PP. After all, didn't we claim that in (125) (b) Drunks would put off the customers that oflwas a P, but not a PP? So, haven't we gone wrong somewhere?

No, not at all (you should know me better by now!). In fact, there is strong empirical evidence in support of saying that in a VP like [put off the customers], (Iff· is merely a P, whereas in [put the customers off], ojfhas the status of PP. We can defend this analysis on distributional grounds (i.e. using criterion (123) (a). After all, if offis a P when immediately adjacent to the Verb, but a PP when separated from the Verb, we should expect that only when it functions as a full Phrase (= PP) will it be able to take a Premodifier (i.e. some preceding expression modifying it), or Postmodifier (i.e. some following expression modifying it), or both. And sure enough, this is exactly what we find: for, when offi« separated from the Verb, it can take both pre- and postmodifiers such as those italicised below:

(165) (a) (b) (c)

Drunks would put the customers right OFF

Drunks would put the customers OFF theirfood Drunks would put the customers right OFF their food

But when off occurs immediately adjacent to the Verb, it cannot be premodified or postmodified in any way: cf.

100

Testing the Structure 2.8

(166) (a) "Drunks would put right OFF the customers

(b) "Drunks would put OFF theirfood the customers

(c) * Drunks would put right OFF theirfood the customers

This is exactly the pattern we should expect to find if off has the status of the simple word-level category P in structures like (125) (b), but the status of a phrase-level category PP in structures like (163).

So, our constituent structure tests (123) lead us to an interesting conclusion about Phrasal Verbs namely that the prepositional particle has the status of a simple Preposition when it immediately follows the Verb and precedes the Object, but the status of a full Prepositional Phrase when it follows the Object and is separated from the Verh. This shows the value of constituent structure tests like those in (123) in showing up structural differences which we would probably never have arrived at on the basis of intuition alone.

Overall, then, we see that we do not have to rely on (often rather fickle) intuitions to support the postulation of a categorial constituent structure associated with sentences. For, it is possible to confirm - or indeed refute- our intuitions using a variety of simple diagnostics. The status of the hypothetical constituent structures which we posit to be associated with sentences is thus not just intuitive, but also empirical (i.e. open to confirmation or disconfirrnation by using empirical procedures such as those outlined in (123) above).

Of course, no 'constituent structure test' can ever be expected to be foolproof; on occasions, a test may fail to yield the anticipated results, simply because some independent factor is 'interfering' with the test and making it inapplicable in some particular context. By way of example, consider the Adverb distribution test, in relation to a sentence such as:

(167) John will finish the assignment

Our standard constituent structure tests lead us to suppose that the sequence finish the assignment is a VP: for example, it can be preposed, proforrned, coordinated with another VP, and ellipsed: cf.

(168) (a) John promised that he will finish the assignment, and (finish the

assignment] he will

(b) John will (finish the assignment], and so will Mary

(c) John will [finish the assignment] and [hand it in]

(d) Paul won't lfinish the assignment], but John will rtM$Vt tVt¢ ~#jt'/llft¢r/tt

So it seems reasonahle to conclude that the sequence [finish the assignment] is a VP containing the Vfinish and an NP object Ithe assignment] (which can be

101

Structure

proformed by a pro-NP like it). And yet, Adverb distribution facts seem to run partly counter to this analysis, as we see from (169) below:

(169) (a) (b) (c)

John will [VP completely finish the assignment] John will [VP finish the assignment completely] >I< John will [VP finish completely the assignment]

Given the assumption that completely is a VP-adverb, then we should expect that it can occur not only at the beginning of the VP (as in (169) (aj), and at the end of the VP (as in (169) (bj), but also in the middle of the VP (as in (169) (c)). But this latter prediction is incorrect, as the ungrammaticality of (169) (c) shows. Does this mean that the whole Adverb distribution argument is flawed? Not at all! It simply means that sometimes perfectly good tests get fouled up by independent factors which interfere with them. What is the factor which is interfering with our Adverb test here? Keyser (1968, p. 37 I) notes that there is a general 'tendency to prevent anything from intervening between a Verb and the following Noun Phrase'. We might describe this restriction in terms of a principle such as (170) below:

(170)

STRICT ADJACENCY PRINCIPLE

An NP complement of a Verb must be positioned strictly adjacent to (i.e. right next to) its Verb

This restriction against separating a Verb from its NP object accounts for the contrast between (17 I) and (172) below:

(17 I) (a) John argued [about every point] with the tax-man

(b) John argued with the tax-man [about every point]

(172) (a) John argued [every point] with the tax-man

(b) "John argued with the tax-man [every point]

In (17 I), the bracketed complement of the Verb argued is a PP (= Prepositional Phrase), and can be positioned before or after the other PP [with the tax-man]; but in (172), the bracketed complement is an NP, and (in accordance with the STRICT ADJACENCY PRINCIPLE (170)) has to be positioned immediately after the verb argued. The same restriction also accounts for contrasts such as that between (J 73) and (174) below:

(173) (a) They loaded [hay] on the cart

(b) "They loaded on the cart [hay]

(174) (a) They loaded [the cart] with hay

(b) "They loaded with hay [the cart]

102

Testing the Structure 2.1'1

In both cases, the bracketed expression is an NP complement of the Verb loaded, and hence has to be positioned immediately after the Verb, given the restriction (170) that nothing can intervene between a V and its NP object.

In the light of the restriction that an NP object must immediately follow its V, we can now understand why our Adverb distribution test appeared to break down in the case of (169) (c); for by attempting to position the Adverb completely between the Verb finish and its NP object the assignment, we were violating the constraint that an NP complement must be positioned immediately after its governing V. This underlines the point made earlier that constituent structure tests sometimes appear to 'break down' because independent factors render them inoperative in special contexts. The simple moral of the story is that we have to make judicious use of constituent structure tests, and bear in mind that their validity may be affected (in particular contexts) by a whole range of syntactic, semantic, pragmatic, stylistic, etc. factors.

We can underline this crucial methodological point by looking briefly at two further cases where traditional constituency 'tests' appear to give the wrong results. Consider for example the following dialogue:

(175)

SPEAKER A: Who will clear up the mess? SPEAKER B: The caretaker will

Speaker B'S reply here seems at first sight to comprise an NP [the caretaker] and an M will. Given our postulate in (64) that only complete constituents can function as sentence fragments, then it might seem reasonable to conclude that Speaker B'S reply in (175) provides us with empirical evidence that the sequence [NP M] in a sentence forms a constituent of some sort. And yet, any such assumption would fly in the face of all the empirical arguments we presented earlier. So why does our sentencefragment test yield the wrong results here? The answer is that it doesn't! For, in applying the test, we failed to take into account the fact that Verb Phrases like [clear up the mess] can undergo Ellipsis in certain discourse contexts. Now, suppose we assume that when a VP undergoes Ellipsis, what happens is that it is stripped of all the lexical material (i.e. words) it contains, and survives only in the form of an empty category. If we use the symbol 'e' to designate an empty category, then we might suppose that Speaker B'S reply in (175) is an S which has the simplified form (176) below:

(176)

10.1

Structure

If this is so, then Speaker 8'S reply in (175) isn't in fact a sentence fragment at all, but rather afull sentence (S constituent) of the canonical form [NP M VP], in which the VP constituent was 'empty', as in (176) above. So, we have to be careful to differentiate between sentence fragments and elliptical sentences (= sentences containing empty categories).

Empty categories may also provide the answer to the question of why coordination tests sometimes appear to give the 'wrong' results. For example, consider the italicised sequence in (177) below:

( 177)

John will give his father a shirt and his mother a skirt

At first sight, (177) might seem to indicate that the sequences hisfather a shirt and his mother a skirt must be constituents, since they have been conjoined by ordinary coordination. But surely this cannot be right: after all, a sequence such as his/ather a shirt cannot be 'preposed for emphasis', or replaced by a single proforrn, as we see from the ungrammaticality of:

(178) (a) * His/ather 1.1 shirt John will give (b) * John will give it] them

So, why does our Ordinary Coordination test appear to give the wrong results in (177)? Well, we have overlooked the possibility that the italicised conjunct in (177) is in fact a Verb Phrase which contains an 'empty' Verb which has undergone the special kind of Ellipsis which we earlier called GAPPING, so that the VP has the schematic form (179) below:

(179)

VP

V~~P~NP

~ 4 fat~ ,fa shi~

If this is so, then (177) will have essentially the same structure as (180) below:

(180)

John will give his father a shirt, and give his mother a skirt

In both cases, the italicised constituent will be a Verb Phrase which has been conjoined with the VP [give his father a shirt]: the only difference between the two is that in (180) the Verb is overt, whereas in (177) it is covert (or empty). Given this assumption, then our Ordinary Coordination test does indeed work perfectly well. Once again, the moral of the story is that we have to take extreme care in applying constituent structure tests, and try to avoid jumping to hasty conclusions which may turn out to be based on a fundamentally flawed premise.

It is perhaps appropriate that our discussion in this chapter should end with

104

Summary 2.9

a brief mention of empty categories; for, these will come to play an increasingly important role as our discussion develops in later chapters. Indeed, a large part of Volume Two will be devoted to the syntax of empty categories.

You should now be able to tackle exercises VIII and IX

2.9 Summary

In this chapter, we have argued in detail that sentences are structured out of words and phrases assigned to various categories. In 2.2 we argued that the native speaker's intuitions about the structural relations between constituents within a sentence can be represented in the form of a tree diagram. However, we pointed out that the true status of constituents and categories is that of theoretical constructs which must therefore be justified on empirical grounds (i.e. we have to show that they are necessary in order to provide a principled account of certain linguistic phenomena). In 2.3 we put forward a variety of arguments (mainly morphological and syntactic in nature) in support of claiming that words belong to a variety of categories such as Noun, Verb, Modal, Adjective, Adverb, Determiner, etc. We then went on to show that similar evidence could be adduced in support of positing that sentences are structured out of Phrases assigned to categories such as Noun Phrase, Verb Phrase, Adjectival Phrase, Adverbial Phrase, Prepositional Phrase, etc. In 2.4 we examined evidence relating to morphological, semantic, and phonological facts; in 2.5 we put forward a series of distributional arguments; and in 2.6 we examined evidence relating to Coordination, Pronominalisation, and Ellipsis. In 2.7, we argued that individual words can function as Phrases, so that. for example. an unmodified Noun which can be substituted by a full Noun Phrase, or coordinated with a full NP, or pronominalised by a pro-NP, should be assigned the status of a full Noun Phrase. Finally, in 2.8 we suggested a variety of simple diagnostics designed to test whether a given sentence has a given constituent structure or not.

EXERCISES

Exercise I

In the bracketed structures below, the lefthand member of each pair of hr.u k ctx has been labelled. First, label the righthand member of each pair of brackets, .uul t hen convert the bracketed structure into a tree diagram, providing as much (or as hn h-j inforruation in your tree as is provided is the bracketed structure.

I II [S [N I' lIis mother] [M might] [VI' get [AI' very angry]]]

105

Structure

(2)

(3)

(4)

[S [NP [D The] [N President]] [M may] [VP [V reject] [NP [D their] [N proposal]]]]

[S [NP [D The] [N weather] ] [M will] [VP [V change] [ADVP [ADV extremely] [ADV suddenly]]]]

[S [NP [D The] [N boys]] [M should] [VP [V buy] [NP [D a] IN present]] [pp [p for] [NP [D their] [N teacherj j ] ] ]

Exercise II

Discuss the categorial status of each of the words in the following sentence, giving detailed reasons in support of your analysis.

(I)

That nice young boy in the corner probably will fall off his bed onto the cold hard floor early one morning

Now consider the status of the italicised nonsense words in the following sentences, giving arguments in support of your analysis.

(2) (a) (b) (c) (d) (e)

(3) (a) (h) (c)

(4) (a) (b)

(5) (a) (b)

John likes to glonk in the afternoons He never glonks on Sundays

He started glonking when he was fourteen He once glonk ed an out-of-work actress He's never glonk ed any of his classmates

John was feeling nurgv, hut happy He's nurgier than anyone I know

He's heen behaving very nurgily all week

John is a bong and so is Fred

In fact, they're hoth typical bongs

She put the car ung the garage

She made sure that it was right ung

Exercise III

The sentences below might be argued to be structurally ambiguous. Discuss the nature of the ambiguity in each case.

(1) (2) (3) (4)

(I) (a) (b)

106

Mary might seem very keen on the boat The police will shoot terrorists with rifles John will run down the new road

They may meet with scepticism

Exercise IV

Account for the contrasts in the following pairs or sentences:

Up a huge hill John ran "Up a huge bill John ran

(2) (a) (b)

(3) (a) (b)

(4) (a) (b)

(5)(a) (b)

Exercises

You can rely on my help entirely, naturally *You can rely on my help naturally, entirely

He obviously will appeal passionately for support "He passionately will appeal obviously for support

Pandas live entirely off bamboo shoots

*Poor weather kills entirely off hamboo shoots

He pledged to her all his worldly goods "He pledged all to her his worldly goods

Exercise V

Assuming that the following sentences are ill-formed (and hence can be used only for humorous effect, for example), say why they are ill-formed:

(I) (2)

(3) (4) (5)

*Could you turn off the fire and on the light?

*Your mother won't put (and your wire is red) up with your disgusting be-

haviour

*1 know the truth, and that you are innocent

* A nuclear explosion would wipe out plant lire and out animal life "He ran down the road and down the President

Exercise VI

Discuss the syntax and semantics or the italicised proforms as they are used in the sentences helow:

(I)

(2) (3)

(4) (5)

I don't know whether the President will retire next year. but I certainly hope .\'0

They say he's extremely intelligent, and .\'0 he may be

The junkies stashed the hash in the trash-can, but the fuzz found out about it

You should see Paris, if you've never been there

lf the Chairman is in Paris (which he is), how do we contact him there?

Exercise VII

Draw tree diagrams (and the corresponding labelled bracketings) to represent the structure or the following sentences, providing detailed empirical arguments in support or your analysis:

(I) (2)

Parents can treat young children badly John will probahly go inside

Exercise VIII

Decide whether the Verh in each or the following sentences functions as a Phrasal Verb, or as a Prepositional Verb, giving detailed empirical arguments in support or your analysis:

107

Structure

(I) (2)

He will pick up the ladder He will climb up the ladder

"Exercise IX

The two sentences below, although superficially similar in form, might be argued to have distinct constituent structures. More precisely, we might argue that in (I), the AP black is an entirely separate constituent from the NP [my coffees, whereas in (2) the AP present is part of the whole Noun Phrase [the people present]. What empirical evidence can you find in support of this claim?

(I) (2)

I don't like [NP my coffee] [AP black] I don't know [NP the people present]

108

3

Phrase-markers

3.1 Overview

In the previous chapter, we argued that sentences are not just unstructured sequences of sounds; rather, they have a hierarchical constituent structure in which sounds are grouped together into words, words into phrases, and phrases into sentences. Each constituent (word or phrase) in a sentence belongs to a specific syntactic category. The categorial constituent structure of sentences can be represented in the form of a Phrase-marker, in which the different nodes are labelled according to the category of the constituent they represent. But what are Phrase-markers? What kind of rules might we devise to generate (i.e. tell us how to form) Phrase-markers? And what exactly is the nature of the categories which are used to label the various constituents which Phrase-markers contain? These are the three essential questions which we address in this chapter.

3.2 The nature of Phrase-markers

In the previous chapter, we used Phrase-markers to represent the constituent structure of sentences. But what exactly is a Phrase-marker? It's useful to think about this problem for a while, and to devise some appropriate technical terminology for describing the internal structure of P-markers, since this will turn out to be of vital importance in subsequent chapters. But why do we need technical jargon? Well, any adequate description of any phenomenon In any field of enquiry (in our present case, Syntax) must be maximally cxplicit, and to be explicit, it must beformal- i.e. make use only of theoretical constructs which have definable formal properties. The use of formal apparatus (involving a certain amount of technical terminology) may seem confusing at first to the unfortunate reader, but as in any other serious field of enquiry (c.g. Molecular Biology) no real progress can be made unless we try to convtr uct formal models of the phenomena we are studying. It would clearly be II rational to accept the use offormalism in one field of enquiry (e.g. Molecular

109

167

Noun Phrases

4

Noun Phrases

There can surely be little doubt that the overall sequence in (2) is a Noun Phrase: for example, like other NPs, it can take the genitive's inflection, as in:

(3)

the king of England's crown

Likewise, there seems to be plenty of evidence that the sequence [of England] is a PP constituent in (2). After all, it can be coordinated with another similar PP (= or-phrase), as in:

(4) the king [pp of England) and [pp of the Empire)

4.1 Overview

Most of our discussion in the previous two chapters has been concerned with providing empirical substantiation for the claim that sentences are hierarchically structured out of constituents belonging to a restricted (perhaps universal) set of categories, and with considering the nature of categories. Implicitly, we postulated a two-level Theory of Categories: that is to say, we tacitly assumed that there are two levels of categories in natural language, namely

It can also function as the 'shared constituent' in cases of shared constituent coordination, as in:

(5)

He is the king, and she is the queen, [pp of England]

And it can function as a sentence fragment in an appropriate context: cf.

(1) (i) word-level categories, e.g.

N = Noun: V = Verb: A = Adjective; P = Preposition; ADV = Adverb; M = Modal; D = Determiner, etc. phrase-level categories, e.g.

NP = Noun Phrase; VP = Verb Phrase; AP = Adjectival Phrase; PP = Prepositional Phrase; ADVP = Adverbial Phrase, etc.

(6) SPEAKER A: Was he the king of France?

SPEAKER B: No, [pp of England)

Moreover, it can be preposed, e.g. in questions:

(ii)

(7) [pp Of which country] was he the king?

And it can be replaced (in a somewhat archaic style) by the pro-PP thereof cf.

In this chapter and the next, however, we are going to argue that our existing Theory of Categories should be extended to include a third type of category intermediate between word-level and phrase-level categories. That is to say, we are going to argue in favour of positing tha t there are nominal constituents larger than the Noun but smaller than a full Noun Phrase, verbal constituents larger than the Verb but smaller than a full Verb Phrase, adjectival constituents larger than the Adjective but smaller than a full Adjectival Phrase ... and so on. We'll begin our discussion in this chapter by looking at the internal structure of Noun Phrases: in the next chapter, we shall go on to look at the syntax of other types of Phrase.

(8) He dwelled in England, and was the king thereoJfor many a year

The obvious conclusion to draw is thus that there is overwhelming evidence that [oj England] is a PP constituent of the overall Noun Phrase [the king of England].

But what is the immediate constituent structure of the whole Noun Phrase?

In his influential 'Immediate Constituents' article, Rulon Wells argued (1947 [1957, p. 188)):

4.2 Small nominal phrases

Let's begin our story by introducing the hero - namely, the fictional character designated in (2) below:

that the ICs [= immediate constituents] of the king of England opened Parliament are the king of England and opened Parliament, that those of the former are the and king of England and those of the latter are opened and Parliament, and that king of England is divided into king and of England.

(2)

the king of England

What is of interest to us here is Wells' implicit claim that the phrase [the king

168

Small nominal phrases 4.2

170

Noun Phrases

(9)

of England] has the structure (9) below:

(13)

(12): it will generate the structure (13) below:

NP

D~ ~NP

But rule (12) can now re-apply to the structure (13), to expand the NP node at the bottom of the tree in (13) into another [0 NP] sequence, resulting in the structure (14) below:

have not attached any specific category label to the constituent [king of England] here, since Wells gives no label for it. But let's speculate on what our mystery constituent (designated by? in (9) above) might be.

Well, since it's a phrase containing the Noun king, an obvious suggestion is that the sequence king of England is just another Noun Phrase. In other words, we might assume that [the king of England] has the skeletal structure (10) below:

(10)

(14)

NP

D~ ~NP

D~~NP

And (14) also contains an NP node at the bottom of the tree which can likewise be expanded into a [0 NP] sequence by re-application of rule (12), thereby deriving the structure (15) below:

( 15)

But this seems to be wrong, for several reasons, For one thing, the 'small' nominal phrase [king of England] does not have the same distribution as a 'full' Noun Phrase such as [the king of England], as we see from (\1) below:

{The king of England} opened Parliament '" King of England

They crowned {the king of England} yesterday "king of England

Parliament grants little power to {the king of England}

. "king of England

Secondly, if [king of England] were a Noun Phrase, then it would mean that Determiners like the would be analysed as premodifying full Noun Phrases: for, a structure such as (10) above would need to be generated by a Phrase

(\\)(a)

And it should be obvious to you by now that we can go on recursively (= repeatedly) re-applying the same rule to produce an NP structure containing not just three Determiners (as in (15) above), but four, five, six ... in fact a potentially infinite number.

'Well, what's wrong with that?' you might ask. The problem is that multiple Determiner sequences are ill-formed in English, as we see from the impossibility ofNPs such as those in (16) below (each of which contains a sequence of just two Determiners):

(b)

(c)

(16) (a) (b) (c) (d) (e) (I) (g)

Structure Rule such as (12) below:

(12) NP -+ 0 NP

But rule (12) is recursive (in that the symbol 'NP' occurs on both thelefthand and the righthand side of the arrow), Now, this means that the rule will generate NPs containing multiple Determiners (in fact, NPs containing indefinitely many Determiners). To see this, consider what happens when we apply rule

*[0 the] [0 the] king of England *[0 the] [0 this] king of England *[0 a] [0 the] king of England *[0 our] [0 your] king of England *[0 the] [0 our] king of England *[0 an] [0 our] king of England *[0 that] [0 our] king of England

Of course, it may well be true that part of the reason why some of the Phrases in (16) are ill-formed is semantic in nature. And indeed, this might be argued to be the case in (16) (a) - (d); for example (16) (c) might be said to be odd

169

because a is indefinite and the is definite, so that we have a contradiction of some sort. And we might argue that if multiple Determiner sequences result in some form of semantic anomaly, then there's nothing syntacticallv wrong with them: i.e. we might argue that such sequences are grammatical, but semantically ill-formed. If this is so, then there's no reason to prevent syntactic rules like (12) from generating such sequences, since the task of syntactic rules is to generate svntactically well-formed structures.

But the question is whether it is plausible to claim that all multiple Determiner sequences in English can be ruled out as ill-formed on semantic grounds. As we have already suggested, this is plausible enough for examples such as (16) (a)- (d) above. But it could surely not be said that Phrases such as (16) (e) - (g) are semantically ill-formed. Why not?

Well. for one thing, ungrammatical Determiner + Possessive sequences such as those in (16) (e) - (g) have perfectly grammatical paraphrases in English, as we see from the following paradigm:

Now. if multiple Determiner sequences are simply meaningless, then we wouldn't expect to find an alternative grammatical way of expressing the same meaning in English: and yet examples such as (17) - (19) show that we can indeed find grammatical ways of expressing the relevant concept. The fact that the ill-formed Noun Phrases in the (a) examples in (17)-(19) above have perfectly grammatical synonymous counterparts in the (b) sentences suggests that the nature of the ill-formed ness in the (a) examples is syntactic rather than semantic.

A second argument for analysing the ill-formedness of multiple Determiner sequences in English as svntactic is that many such sequences which are illformed in English are well-formed in other languages. For example, Determiner + Possessive sequences have grammatical counterparts in languages such as r talian, Spanish. and Romanian - as the following examples show:

(17) (a) (b)

(18)(a) (b)

(19) (a) (b) (c)

(20) (a)

Small nominal phrases 4.2

172

Noun Phrases

esas ideas tuyas [Spanish]

those ideas yours ('those ideas of yours')

(c) cartea fa [Romanian]

book + the your ( = book the your = 'your book')

(b)

"a mv book

a book ofmine

Now, if multiple Determiner sequences were semantically anomalous, one would expect synonymous sequences to be equally anomalous in other languages: the fact that they are not suggests that the ill-formedness of multiple Determiner sequences in English is syntactic rather than semantic in nature (though in the case of NPs such as (16) (c) [a the king of England], the ill-formedness may be both syntactic and semantic). In other words, it seems likely that 'multiple Determiner' sequences are ruled out in English by some syntactic principle, not by semantic considerations alone. And the obvious principle to invoke is one to the effect that Determiners in English modify a type of nominal phrase which is smaller than a full Noun Phrase, though larger than a single Noun.

Overall, then, it would seem likely that our mystery constituent (indicated by'?') in (9) above is an 'intermediate' type of nominal phrase, larger than N, but smaller than NP. But what label can we attach to it? Well. I'm afraid that our existing inventory of categories summarised in (I) above simply doesn't provide us with enough category labels to go round. For, (I) recognises only two types of nominal constituent, namely Nand NP: it has no label for a constituent 'intermediate' between the two. So, we need a rather more sophisticated set of category labels. But where can we find them?

Fortunately, Zellig Harris' (1951) Structural Linguistics provides us with a simple answer. Harris (ibid., p. 266) suggests a system of what he calls 'raised numbers' to label successively larger phrasal expansions of a given head constituent. Adapting his numerical superscript system in minor ways (e.g, Harris starts counting at 'I', but we're going to start counting at 'O'!), we might then resolve the problem posed by the phrase [the king of England] in the following way. We might argue that the Noun king is an N°, that the 'small' nominal phrase [king of England] is a single phrasal expansion of king and hence an N 1, and that the full NP [the king of England] is a double phrasal expansion of the head Noun king, and hence an N2. Given this notation, (2) would have the structure (21) below:

"this your tie

this tie ofvours

"some your friends some friends of yours some ofyour friends

(21 )

un mio Iibro [Italian]

a my book ('a book of mine')

171

173

Noun Phrases

Small nominal phrases 4.2

(22) NUMBER NOT A TION BAR NOTATION PRIME NOTATION
N° (N-zero) N N
NI (N-one) N (N-bar) N' (N-prime)
N2 (N-two) N (N-double-bar) N" (N-double-prime) 4.3 Evidence for N-bar

Having managed to disentangle ourselves from the notauor... knots we were tied up in, we can now go on to ask ourselves what eviden, there is that Wells was right to assume that an NP such as [the king of Englan .. contains the 'small' nominal phrase (i.e, what we are calling an N-bar) lkillg , . England] as one or its immediate constituents. In this connection. it is intere-ting to consider the arguments which Wells himself put forward in support ," his own analysis (though it should he borne in mind that Wells was writing I~. a different era. and within a different theoretical framework). One such argument which he adduces (1957. p. 192) is a distributional one to the effect tha: [king ofEnglands must be a constituent because it can occur as an independen: unit in other types of sentence-structure. as in Wells' example:

In the numerical superscript notation, N° corresponds to the simple category N of our earlier system in (I), NZ corresponds to NP, and N I has no counterpart at all in our original system.

At this point, however, we should mention two rival (but entirely equivalent) notational alternatives to Harris' number notation. One is the bar notation introduced in Chomsky's (1970) 'Remarks on Norninalisation' paper; and the second is the prime notation used (for example) in an influential study of Phrase Structure by Jackendoff (1977a). The three systems are notational variants of each other (i.e. different ways of saying the same thing), and the correspondences between the three can be summarised as in (22) below:

(25) He became [king a/England]

Thus, the skeletal structure of [the king of England] could be represented in exactly equivalent fashion in each of the three systems as in (23) below:

A second argument which he puts forward (ibid .. p. 191) is that such an analysis will enable us to capture the structural parallelism between the two phrase' 111 (26) below:

(23) (a) (b) (c)

[N2 the [N 1 [N0 king] of England]] [N the [N [N king] of England]] [N" the [N' [N king] of England]]

(26) (a) (b)

the [English king]

the [king ofEnglands

Given that these three notational systems are entirely equivalent, it is not surprising to find that they are used interchangeably: for example, Jackendoffs (1977a) book uses the bar notation in its title (it is called X Syntax), but uses the prime notation throughout the rest of the book!

For typographical reasons (if you use a typewriter or a word-processor, you'll understand what they are!), we'll henceforth use the prime-system [N, N', N' '] in our tree diagrams, though (somewhat schizophrenically!) we'll refer to the relevant constituents as N, N-bar, and N-double-bar. Believe it or not, this is standard practice! Given these conventions, the constituent structure of our (in)famous phrase [the king of England] will now be represented in the manner outlined in (24) below:

Wells assumes (though does not argue) that the bracketed sequence [English king] is a constituent in (26) (a), and argues that a parallel analysis of (26) (b I along the lines of (24) above would be 'the best analysis of that phrase' because 'it harmonizes with other analyses' (i.e. with his analysis of phrases like [the English king]). He is implicitly invoking a principle of maximising structural symmetry between related constructions.

While there are potential pitfalls in Wells' argumentation, other independen t evidence can be adduced in support of his analysis. For example, a further argument in support of the key claim that the sequence [king of England] is a constituent concerns the fact that it can undergo Ordinary Coordination with another similar sequence, as in:

(24)

(27) Who would have dared defy the [king of England] and [ruler of the Empire]?

N"

D-------- ------------ N'

t~e N------- ------- pp

ki~g ~Engl~

And we'll say that king is an N, [king of England] is an N-bar, and [the king of England] is an N-double-bar (hence also an NP, since we earlier said that N" corresponds to the traditional category of NP).

Moreover, it can function as the 'shared constituent' in cases of Shared Constituent Coordination: cf.

(28)

He was the last (and some people say the best) [king of Englandi

Given our assumption that only a unitary constituent can undergo Simple Coordination, or can function as the 'shared constituent' in cases of Shared Con-

174

Complements and Adjuncts 4.4

176

Noun Phrases

However, any such hasty conclusion would ignore the traditional distinction between two different types of postnominal phrase - namely (i) those which function as Complements, and (ii) those which function as Adjuncts. We can illustrate the difference between these two types of postmodifier in terms of the contrast in (30) below:

In the case of (30) (a) [a student of Physics], the bracketed PP [of Physics] is (in an intuitively fairly obvious sense) the 'Complement' of student: the PP tells us what it is that the individual concerned studies. Hence the NP [a student of Physics] can be paraphrased by a clausal construction in which Physics functions as the Complement of the Verb study: cf.

stituent Coordination. the obvious conclusion to draw is that the sequence [king of England] must indeed be a constituent. And this provides empirical support for the analysis in (24) above.

An additional type of argument in support of the N-bar analysis can be formulated in relation to Pronominalisation facts. Recall from our discussion in Chapter 2 that only a unitary constituent can be replaced by a proform - and indeed only a phrasal constituent of some sort. In the light of this observation, consider the use of the proform one in the following examples:

(30) (a) (b)

(29) (a) (b)

The present [king of England] is more popular than the last one *The [king] of England defeated the one of Spain

How can we account for the contrast here? Well, if we posit that [king of England] is a 'small' nominal phrase of some sort (an N-bar, to be precise), then we could say that one in English is the kind of proform which can replace a 'small nominal phrase': in other words, we can say that one is a pro-N-bar. Thus. we could argue that [king of England] in (29) (a) can be replaced by one because it is an N-bar; whereas king in (29) (b) cannot be replaced by one because it is only an N and not an N-bar (and we already know that proforms replace phrasal constituents, not individual words). But any such analysis naturally presupposes that [king of England] is indeed a phrasal constituent of some sort, as in Wells' analysis (24).

So, both Coordination and Pronominalisation facts provide strong empirical support for the N-bar analysis. Accordingly, we shall henceforth assume that this analysis is correct, and that there is indeed an intermediate type of nominal constituent (namely N-bar) which is larger than N but smaller than NP. We thus posit that there are three types of nominal constituent in English, namely N, N' (= N-bar), and N" (= N-double-bar= NP). This means that we no longer recognise only two categorial/evels of nominal constituent (N and NP): on the contrary, we are now assuming that there are three categorial levels of nominal constituent, namely N, N-bar, and N-double-bar.

(31) (a) (b)

a student [of Physics] (= Complement) a student [with long hair] (= Adjunct)

He is [a student of Physics] He is [stua:ving Physics]

But this is not at ali the case in (30) (b), [a student with long hair]. In this case, the bracketed PP [with long hair] doesn't in any sense function as the Complement of student, so that we don't have any corresponding paraphrase in which [long hair] is used as the Complement of the Verb study: cf.

Thus, in (31) (a) [a student of Physics], the bracketed PP [of Physics] specifies what the student is studying: but in (32) (a) [a student with long hair] the bracketed PP doesn't tell us anything about what the student is studying; it merely serves to give us additional information about the student (i.e. that he happens to have long hair). In traditional terms, the kind of PP found in [student of Physics] (or indeed [king of England]) is said to be a Complement, whereas that found in [student with long hair] is said to be an Adjunct.

Of course, terms like Complement and Adjunct denote grammatical functions or relations, and thus have the same status as terms like 'Subject' and 'Object'. The obvious question to ask therefore is what is the structural correlate of the Complement-Adjunct distinction, and how do Complements and Adjuncts differ from the other class of nominal modifiers which we are already familiar with - namely Determiners. What we shall claim here is that the difference is essentially the following:

(32) (a) (b)

4.4 Complements and Adjuncts

What we have argued so far is that in an NP such as [the king of England], the postnominal PP [of England] expands the head Noun king into the N-bar [king ofEnglandi, while the Determiner the expands the N-bar [king of England] into the N-double-bar [the king of England]. Now, we might seek to generalise our conclusions about the function of the PP [of England] in this phrase by suggesting that all postnominal PPs (and indeed perhaps all postnominal phrases of any kind) have essentially the same constituent structure status, and thus serve to expand N into N-bar.

(33) (a) (b) (c)

175

He is [a student with long hair] =1= He is [studying long hair]

Determiners expand N-bar into N-double-bar Adjuncts expand N-bar into N-bar Complements expand N into N-bar

Complements and Adjuncts 4.4

178

Noun Phrases

Given the assumptions in (33), a Noun Phrase containing a Determiner, an Adjunct, and a Complement would have the schematic structure (34) below:

Given the 'no crossing of branches' restriction, it follows that the rules in (35) will generate Adjunct PPs to the right of Complement PPs as in (37) (a) below, not to the left as in (37) (b):

(37) (b) (which is the structure associated with the ungrammatical (36) (bj) is ruled out because it violates the 'no crossing of branches' restriction. But (37 (a) (which is the structure of the grammatical (36) (aj) contains no crossing branches, and thus is well-formed. So, it follows from (37) that Complements must occur closer to their head Nouns than Adjuncts. And this is precisely why the Complement phrase has to precede the Adjunct phrase in (36) - and why (more generally) postnominal Complements precede postnominal Adjuncts (as noted by lackendoff 1977a, p. 58).

But I bet you're wondering whether the rules proposed in (35) above really work! So let's see whether they do. If we apply the Determiner Rule (35) (i) [N" --> D N ']. we generate the substructure (38) below:

(34)

N"

.>: --------------- N'

Determiner ~ ~

N' Adjunct

/~

N Complement

(37) (a)

We can see from (34) that Determiners are sisters of N-bar and daughters of N-double-bar; Adjuncts are both sisters and daughters of N-bar; and Complements are sisters of N and daughters of N-bar. This means that Adjuncts resemble Complements in that both are daughters ofN-bar; but they differ from Complements in that Adjuncts are sisters of N-bar, whereas Complements are sisters of N. Likewise. it means that Adjuncts resemble Determiners in that both are sisters of N-bar, but they differ from Determiners in that Adjuncts are daughters of Nvbar, whereas Determiners are daughters ofN-double-bar.

Perhaps we can bring out the relevant distinctions rather more clearly in terms of the respective Phrase Structure Rules needed to generate Determiners, Adjuncts, and Complements. Given the claims made in (33) above, Determiners will be introduced by the rule (35) (i) below, Adjuncts by rule (35) (ii), and Complements by rule (35) (iii):

(35) (i) (ii) (iii)

N" --> D N' [Determiner Rule] N' --> N' PP [Adjunct Rule]

N' --> N PP [Complement Rule]

For ease of reference, we have called (35) (i) the Determiner Rule (since it introduces Determiners), (36) (ii) the Adjunct Rule, and (35) (iii) the Complement Rule.

Now, if you think about it, you'll realise that the rules in (35) make rather interesting predictions about the relative ordering of Adjuncts and Complements. More specifically, they predict that Complements will always be 'closer' to their head Noun than Adjuncts. In other words, our rules in (35) predict that if we modify student by an Adjunct PP such as [wi(h long hair], and a Complement PP such as [of Physics], then the Complement phrase must precede the Adjunct phrase. And, (as Hornstein and Lightfoot 1981 a, p. 22) note, this prediction is entirely correct - cr. their examples:

(38)

N'

N' .i>: ~

N --------------~ ~ pp pp

(Complement) (Adjunct)

(b)

N,~T

N~~PP

(Adjunct) (Complement)

N"

D~~N'

If we then apply the Adjunct Rule (35) (ii) [N' --> N' PP] to expand the N-bar in (38), we derive:

(39)

N"

D~~N' N'~~PP

(36) (a) the student [~( Physics] [wi(h long hair]

(b) *the student [1;'i(11 long hair] [a/Physics]

If we subsequently apply the Complement Rule (35) (iii) [N' --> N PP] to the N' at the bottom of the tree in (39), we derive the structure (40) below (we have

177

(40)

attached the relevant lexical items, to make the discussion less abstract):

Optional constituents of the Noun Phrase 4 .. '

180

Noun Phrases

However, it is not our purpose here to deal with the complex conditions under which Determiners can or cannot be omitted in English (or more generally): the reader interested in such questions should consult the relevant section on the use of Determiners in a detailed reference grammar such as Quirk et al. (1985). What concerns us here is simply the question: 'What is the structure of Noun Phrases which lack Determiners, and how will our existing set of rules (42) above have to be modified in order to cope with such NPsT

So, to return to a familiar example, what concerns us here is how we are to generate a simple Determiner-less nominal expression such as:

And this is precisely the constituent structure associated with a Noun Phrase such as [a student of Phvsics with long hair]. So, you see, the rules do actually work, after all!

(44)

Students of Physics with long hair

The first question we should ask about (44) is: 'What is the constituent status of the overall phrase?' In other words, is (44) an Nsbar, or an N-double-bar (i.e full Noun Phrase)') The answer is that (44) is indeed a full Noun Phrase, as can be shown by a variety of familiar constituent structure tests. For example, it can occur in isolation as a 'sentence-fragment', as in (45) below:

4,5 Optional constituents of the Noun Phrase

Thus far, we have heen looking at the internal structure of Noun Phrases of the schematic form (41) below:

(41)

Determiner+ Noun +Complement PP+ Adjunct PP

(45)

SPEAKER A: What kind of students do you hate teaching? SPEAKER B: [Students of Physics with long hair]

and we have argued that such NPs can be generated by a set of Phrase Structure Rules such as (35) above. repeated here for convenience as (42) below:

However. one fairly obvious point which we have overlooked in our rules in (42) is that Determiners, Adjuncts, and Complements are all optional constituents of Noun Phrases. Let's consider first the optional use of Determiners.

One thing which it is important to get clear at the very outset of our discussion is exactly what we do and do not mean by claiming that 'Determiners are optional constituents of the Noun Phrase'. All we mean is that 'Some Noun Phrases are used without Determiners': what we emphatically do not mean is that 'Any Noun Phrase of any kind can optionally be used with or without a Determiner'. Of course, there are complex conditions which determine when Determiners can or cannot be omitted from a Noun Phrase: for example, in general. Noncount Nouns and Plural Count Nouns can be used without an overt Determiner. but Singular Count Nouns cannot: cf.

(42) (i ) (ii) (iii)

(43) (a) (b) (c)

N" ~ D N' [Determiner Rule] N' ~ N' PP [Adjunct Rule]

N' ~ N PP [Complement Rule]

Given our assumption that only full Phrases can occur as sentence fragments, it follows that the bracketed Phrase [students of Physics with long hair] uttered by Speaker B in (45) must be a full Noun Phrase.

And indeed this analysis is independently confirmed by other sets of facts.

For example, such expressions can be coordinated with a full NP, and can also be pronominalised by a pro-NP such as them:

(46) (a)

[NP Students of Physics with long hair] and [NP their professors] often don't see eye to eye

[NP Students of Physics with long hair] sometimes think the world owes them a living

(b)

But given that expressions such as [students of Physics with long hair] can function as NPs, how can we account for the fact that they lack Determiners?

Sadly, the answer is disappointingly unspectacular! We say the obvious, and specify that Determiners are optional constituents of NP, Or, more precisely, we replace our earlier Determiner Rule (42) (i) [N" ~ D N'] by the revised rule (47) below:

(47) N" ~ (D) N' [new Determiner Rule]

Childhood can be traumatic ( = Noncount Noun) Children can be traumatic (= Plural Count Noun) *Child can be traumatic (= Singular Count Noun)

The parentheses around D in (47) indicate that the Determiner is an optional constituent of N" (recall that N" = NP), If we replace our earlier Determiner

179

Optional constituents of the Noun Phrase 4.5

182

Noun Phrases

optional, as we see from the paradigm in (52) below:

Rule (42) (i) by our new rule (47), our revised system of Phrase Structure Rules is now (4S) below:

(4S) (i) (ii) (iii)

N" ~ (D) N' [Determiner Rule] N' ~ N' PP [Adjunct Rule]

N' ~ N PP [Complement Rule]

(52) (a) (b) (c) (d)

Let's consider how our revised rule system works.

We start by applying the Determiner Rule (4S) (i), which tells us that we can expand an N" into an optional Determiner plus an N-bar: if we reject the option of having a Determiner, then our rule will generate the structure (49) below:

(49)

a student [o(Physics] [with long hair] (Complement and Adjunct) a student [with long hair] (Adjunct, no Complement)

a student [of Physics] (Complement, no Adjunct)

a student (no Adjunct, no Complement)

The obvious question to ask, therefore, is how we are to modify our existing Phrase Structure Rules (48) above so as to take into account the optionality of Complements and Adjuncts.

Let's consider first the question of how we generate NPs which contain an Adjunct but no Complement - i.e. NPs such as (52) (b) [a student with long hair]. The obvious suggestion is to deal with optional Complements in the same way that we dealt with optional Determiners - namely by specifying that a Complement PP is an optional constituent of N-bar. Thus, we might propose to replace our existing Complement Rule (48) (iii) [N' ~ N PP] b~ the revised rule (53) below:

N" I N'

We now apply the Adjunct Rule (48) (ii): this tells us that we can expand N-bar into another N-bar plus a PP Adjunct; thus, if we apply the rule to (49) above, we derive the structure (50) below:

(50)

(53)

N" I N'

N'/ ~PP

N' ~ N (PP) [new Complement Rule]

where the parentheses round PP in (53) indicate that a Prepositional Phrase Complement is an optional constituent of N-bar. If we now replace our earlier Complement Rule by our new one, our revised overall system of Phrase Structure Rules becomes (54) below:

Now let's move on to the Complement Rule (48) (iii), which tells us that we can expand N-bar into a head Noun followed by a PP Complement; applying this rule to (50) above will yield the structure (51) below (we have inserted appropriate lexical items for concreteness):

(51)

(54) (i) (ii) (iii)

N" ~ (D) N' [Determiner Rule] N' ~ N' PP [Adjunct Rule]

N' ~ N (PP) [Complement Rule]

Let's see how we can apply our revised system of rules to generate a Noun Phrase such as (52) (b) [a student with long hair].

First, we start by applying the Determiner Rule (54) (i): if we choose the optional Determiner, this generates the structure (55) below:

(55)

N"

D~~N'

We now go on to apply the Adjunct rule (54) (ii) to expand N-bar in (55) into the sequence [N' PP], as in (56) below:

Thus, our revised rules in (48) above can indeed generate NPs which contain no Determiner.

But hold on a minute' It isn't just Determiners which are optional constituents of Noun Phrases: for, as we noted above, Complements and Adjuncts are also optional constituents of NP. For example, in the case we are discussing, both the Complement PP [of Physics] and the Adjunct PP [with long hair] are

(56)

lSI

N"

D-~~N'

N'~~PP

Optional constituents ofthe NOlin Phrase -+.5

184

Noun Phrases

N"

D~ ~----N'

~ N"~ --------------------- pp

I ~"

]\I with long h~

I

rule (59) does no more than chase its own tail (well, if you want the relevant technical jargon, the rule is vacuously recursi~e!). Why? Because its output (= N') is the same as its input (= N'), so that the rule allows you to stack a potentially infinite number of non-branching N-bar constituents on top of each other. This means that we (quite wrongly) predict that any Noun Phrase should be potentially infinitely structurally ambiguous, according to how many non-branching N-bars it has stacked on top of other N-bars. For, we assume that differences in structure will generally correlate with perceived differences of meaning: so, rule (59) implies that a simple NP like [a boy] should be infinitely ambiguous according to whether it contains one, two, three, four, five, ... etc. non-branching N-bar constituents. But this is an absurd claim, and one which we shall treat with the contempt it deserves!

The bottom line of our argumentation here is that we cannot deal with the optionality of Adjuncts in terms of a rule such as (58) (ii) saying that an N-bar can be expanded into another N-bar plus an optional Adjunct PP. So, assuming that we don't opt for the 'tail-chasing' (sorry, I mean vacuous recursion') solution (59), how are we going to deal with the optionality of Adjuncts? Well, there's an even simpler solution than you might have expected: and that is to simply stipulate that our earlier Adjunct Rule (54) (ii) is an optional rule. That is to say, we can either choose to apply the rule, or not apply it, as we wish. In other words, we might revise our earlier rules (54) in the manner indicated in (60) below:

We now apply our new Complement Rule (54) (iii): this tells us that we can expand an N-bar into an N with or without an optional PP complement. Well. let's suppose that we decide not to choose the PP complement. In this case. N-bar will be expanded into N alone. as in (57) below (we have inserted relevant lexical items for the sake of making our discussion more concrete, and thus more intelligible):

Hence, we see that a very simple modification of our earlier Complement Rule will suffice to enable us to generate NPs in which the head N has no Complement.

So, now we've seen how to deal with optional Determiners on the one hand, and optional Complements on the other. But what about optional Adjuncts? In other words, how are we going to generate a structure such as (52) (c) [a student at' Phrsics]. which contains a Complement PP [of Physics], but no Adjunct PP" Well, the obvious suggestion is to deal with optional Adjuncts in exactly the same way that we dealt with optional Determiners and Complements. That is to say, we might simply stipulate that Adjuncts are an optional constituent of N-bar. The natural way of doing this might seem to be simply to put parentheses round the Adjunct PP introduced by our existing Adjunct Rule (54) (ii) above, so that our earlier rules (54) would be revised along the lines of (58) below (where (58) (ii) is the new Adjunct Rule):

(57)

(58) (i) (ii) (iii)

student

(60) (i) (ii) (iii)

N" ~ (D) N' [Determiner Rule]

N' ~ N' PP [Adjunct Rule: optional] N' ~ N (PP) [Complement Rule]

N" ~ (D) N' [Determiner Rule] N' ~ N' (PP) [new Adjunct Rule] N' ~ N (PP) [Complement Rule]

In the light of our claim that the Adjunct Rule (60) (ii) is optional (and hence can be 'skipped' if we wish to do so), let's go back to the question of how we generate a structure like (52) (c) [a student of Physics]. As before, we start by applying the Determiner Rule (60) (i) [N" ~ (D) N']: this tells us that we can form an NP (= N-double-bar) out of an optional Determiner and an N-bar. If we take the Determiner option, then we generate the structure (61) below:

But if you think about it. you'll realise that (58) (ii) can't be right at all. After all. one of the possibilities allowed for in our new Adjunct Rule (58) (ii) is that of omitting the PP Adjunct, in which case the Adjunct Rule would amount to:

(61 )

N"

D~ ------------- N'

(59) N' ~ N'

But rule (59) is vacuous, in a number of ways. For one thing, it's self-defining, and hence doesn't actually tell us anything about how to form an N-bar (it's a bit like a person who, when asked to define a troglodyte says 'Well, a troglodyte's a troglodyte, and that's all there is to say about it!'). More seriously,

Now we come to the Adjunct Rule, (60) (ii). Since this rule is optional, we can choose to either apply it, or not apply it. We'll take the latter course, and choose to 'skip' the rule altogether. This means that we go on to apply the Complement Rule (60) (iii), which expands N-bar into a head N plus an

183

Phvsics with long hair] will be assigned the structure (40) above: (52) (b) [a st~dent with long hair] will be assigned the structure (57); (52) (c) [a student of Phvsicsi will be assigned the structure (62): and (52) (d) [a student] will be assigned the structure (63). Now, if you look closely at these four structures (40). (57). (62) and (63), you'll see that the categorial status of the expression student changes from one example to another. More precisely, student has the status of a simple N in (40) and (62); whereas it has the status of an N-bar immediately dominating an N in (57) and (63). To underline this fact, we give the structure of the relevant examples in skeletal form in (64) and (65) below:

Optional constituents of the Noun Phrase 4.5

186

Since student cannot be replaced by the pro-N-bar one here, it therefore

Noun Phrases

optional PP Complement. Well let's suppose we take up the option of having a PP Complement: in this case, the result of applying the Complement Rule (60) (iii) to (61) will be the structure (62) below (as before, we have inserted appropriate lexical items, for the sake of clarity):

(62)

Since the PP [of Physics] in (62) is a Complement of the head of Noun student, we have now managed to achieve our goal of generating a Noun Phrase without an Adjunct.

So, we have managed to adapt our original rule system so as to cope with NPs which contain Adjuncts but not Complements, and conversely NPs which contain Complements but not Adjuncts. But what about NPs which contain neither Complements, nor Adjuncts - e.g. NPs such as (52) (d) [a student]. Can our revised system of Phrase Structure Rules (60) handle these? Well, let's see.

We'll begin by applying the Determiner Rule (60) (i) to generate the structure (61) above. We then 'skip' the optional Adjunct Rule (60) (ii): we go directly on to the Complement Rule (60) (iii), which tells us that we can expand an N-bar into a head N, plus an optional PP complement. But this time, let's take the option of not choosing to have a PP complement, so that we expand N-bar simply into an unmodified N: the result will be that by applying the Complement Rule (60) (iii) to (61) above we generate (63) below (as before, the relevant lexical items have been inserted):

(64) (a) (b)

(65) (a) (b)

[N" a [N' [N' [N student] of Physics] with long hair]] [N" a [N' [N student] of Physics]]

[N" a [N' [N' [N student]] with long hair]] [N" a [N' [N student]]]

The difference is that when the Noun student has an overt Complement like [ot' Physics] (as in (64) above), then it functions only as an N (because the corresponding N-bar is the whole Noun + Complement structure [student of Phrsics]). But when the Noun student has no overt Complement (as in (65) above), then it is not only an N, but also an N-bar.

Now, the assumption that student is an N in (64) but an N-bar in (65) has far-reaching consequences. We can see this if we look at the predictions the two structures make about the use of the proform one. Recall that we argued earlier that one in English can function as a pro-N-bar, but not as a pro-N (because proforms do not replace word-level categories). Now, if it is true (as our analysis claims) that student is an N-bar in (65), then since one is a pro-Nbar. we should expect that student can be replaced by one in structures like (65): and as (66) below illustrates, this prediction is entirely correct:

It therefore Iollows that student must have the status of N-bar in examples like (65) and (66). But by contrast. we find that student cannot be replaced by the proform one in examples such as (64) above. as (67) below illustrates:

(63)

N"

D------- ~N'

I I

a N

I student

(66) (a) (b)

Since the NP in (63) above contains neither a Complement PP nor an Adjunct PP, it seems clear that our existing system of rules in (60) is perfectly adequate to deal with NPs which lack both Complements and Adjuncts.

We have now seen that the very simple set of rules we devised in (60) will generate a full range of NP structures, with or without Determiners, with or without Adjuncts, and with or without Complements, and will assign appropriate structures to the NPs concerned: for example, (52) (a) [a student of

(67) (a)

185

The [student] with short hair is dating the one with long hair This [student] works harder than that one

Which [student] were you referring to? "The one of Physics with

long hair?

(b) *The [student] of chemistry was older than the one of Physics

(Lightfoot (1982). p. 54)

follows that student cannot have the status of N-bar in phrases such as (64), but rather must have the simple status of N.

We can use our one-pronominalisation test to provide further confirmation of the constituent structure analyses we have posited in (64) and (65) above. For. our proposed analysis also specifies that the sequences [student oj' Physics] in (64) (a). [student of Physics with long hair] in (64) (a), [student of Physics] in (64) (b), and [student with long hair] in (65) (a) are all N-bar constituents: hence we should expect that all four phrases can be proformed by the pro-N-bar one. And as (68) below indicates, this prediction is entirely correct:

Thus, one-pronominalisation facts provide quite remarkable independent corroboration of our analysis. In particular, they lend strong support to our claim that a Noun which has an overt Complement is simply an N, whereas a Noun which lacks a Complement has the status oj'N-bar (as well as 1'0.

(68) (a) (b) (c) (d)

More differences between Complements and Adjuncts 4.6

188

Noun Phrases

Adjuncts in (70) correlate in an obvious way with an associated difference in semantic structure. In relation to the pair of sentences in (71) below:

(71) (a) (b)

John is a [N' [N student] o(Physics]

John is a [N' [N' [N student]]lI'ith long hair]

Which [student of Physics]? The one with long hair? Which [student of Physics with long hair]? This one? Which [student of Physics]? That one?

Which [student with long hair]? This one?

they comment:

We also assume that syntactic constituent structure will playa role in determining the semantics of Noun Phrases, and specifically that each N-bar specifies a 'semantic property'. Therefore. to attribute (71) (a) to John is to attribute one property to him, that he studies Physics: to attribute (71) (b) to John is to attribute two properties, that he studies. and that he has long hair. Hence it follows that John is a student of Physics. meaning what it does (i.e. denoting only one property), cannot be assigned a structure like (7 I) (b): conversely. John is a student with long hair, meaning what it does (i.e. denoting two properties) cannot have a structure like (71 )(a).

(Hornstein and Lightfoot, Introduction to Explanation ill Linguistics (1981a), p. 21)

A related semantic argument can be formulated with regard to disambiguation. It should be obvious that the structural distinction we have drawn between Complement PPs (which modify N) and Adjunct PPs (which modify N-bar) will enable us to provide a principled account of the structural ambiguity of phrases such as:

(72) a student [o/high moral principles]

4,6 More differences between Complements and Adjuncts

Hitherto, we have argued that Determiners, Adjunct PPs and Complement PPs should be generated by the following set of Phrase Structure Rules (cr. (60) above):

(69) (i) (ii) (iii)

N" ---> (D) N' [Determiner Rule]

N' ---> N' PP [Adjunct Rule: optional] N' ---> N (PPJ [Complement Rule]

The NP in (72) is ambiguous as between the two interpretations:

As we have already seen, these rules specify (amongst other things) that Determiners, Adjuncts. and Complements differ from each other in the following ways:

(d) Complements are sisters ofN and daughters ofN'

In this section, we are going to look (rather more briefly) at a number of further arguments in support of the structural distinction between Complement PPs and Adjunct PPs drawn in (70) above.

One such argument is of a semantic nature. Hornstein and Lightfoot (1981 a, p 21) note that the structural differences between Complements and

(70) (a) (b)

(73) (i) (ii)

a person who studies high moral principles a student who has high moral principles

Determiners are sisters ofN' and daughters ofN" Adjuncts are sisters and daughters of N'

And we might characterise this ambiguity in structural terms by saying that on interpretation (73) (i) the bracketed PP (72) is a Complement (hence a sister) of the head Noun student; whereas on interpretation (73) (ii), the bracketed PP in (72) is an Adjunct, hence a sister of the N-bar headed by the Noun student. This would mean that on the first interpretation, (72) would have the skeletal structure (74) (a) below, whereas on the second interpretation it would have the structure (74) (b):

(74) (a) (b)

a [N' [N student] of high moral principles] (= 73(i))

a [N' [N' [N student]] of high moral principles] (= 73(ii))

187

Thus, our structural distinction between Complements and Adjuncts enables us to characterise some fairly interesting cases of structural ambiguity.

But let's return to syntactic arguments in favour of our claim that Complement PPs expand N into N-bar, whereas Adjunct PPs expand N-bar into N-bar. An important difference between the Adjunct Rule (69) (ii) above (which introduces Adjunct PPs), and the Complement Rule (69) (iii) (which introduces Complement PPs) is that the Adjunct Rule is recursive, whereas the Complement Rule is not: for convenience, we have repeated our earlier rules (69) above as (75) below:

(75) (i) (ii) (iii)

More differences between Complements and Adjuncts 4.6

190

Noun Phrases

A further syntactic argument in favour of the structural distinction between Complements and Adjuncts which we are assuming here can be formulated in relation to facts about Ordinary Coordination. Note that we can coordinate two PPs which are both Complements: cf.

(79) a student [of Physics) and [a/Chemistry]

And likewise we can coordinate two PPs which are both Adjuncts: cf.

(80) a student [with 10nK hair] and [with short arms)

But we cannot coordinate a Complement PP with an Adjunct PP: cf.

N" --> (D) N' [Determiner Rule]

N' --> N' PP [Adjunct Rule: optional] N' --> N (PP) [Complement Rule)

(81) (a) (b)

"a student [of Physics) and [with 10nK hair) "a student [frith long hair] and [olPhysics]

the student [with long hair] [with short arms] "the student [of Physics] [o(Chemistry]

Under the analysis proposed here, we can account for this in structural terms by assuming that Adjuncts and Complements are attached at different levels (Complements are sisters of N, and hence are attached at the N level; whereas Adjuncts are sisters of N-bar. and hence are attached at the N-bar level), and by positing that only constituents attached at the same level can be coordinated. It would then follow that the two Complement PPs in (79) can be coordinated (since both are attached at the N level), as can the two Adjunct PPs in (80) (since both are attached at the N-bar level): but it would also follow that an Adjunct cannot be coordinated with a Complement (as in (8 I l), since Complements are attached at the N level, whereas Adjuncts are attached at the N-bar level.

Incidentally, we might note in passing that our proposed analysis makes a number of further correct predictions about coordination. For example, since both Complements and Adjuncts are daughters of N-bar (i.e. both Noun + Complement and Noun + Adjunct sequences have the status of N-bar), our analysis correctly predicts that the Noun + Complement sequence [student of Physics) and the Noun + Adjunct sequence [student with long hair] have the same constituent status of N-bar, and so can be coordinated with themselves and each other in such a way that the whole conjoined sequence forms an Nvbar, and thus can be modified by a Determiner such as the: and the fact that examples such as (82) below are grammatical shows that this prediction is correct:

The Adjunct Rule (75) (ii) is recursive by virtue of the fact that the same symbol N' appears both on the left and on the right of the arrow, whereas the Complement Rule (75) (iii) is non-recursive. Since the rule generating Adjuncts is recursive, it predicts that indefinitely many Adjunct PPs can be 'stacked' on top of each other. But because the rule introducing Complements is not recursive, it does not allow PP Complements to be stacked in this way. And in fact, the prediction that PP Adjuncts can be 'stacked' but PP Complements cannot seems to be correct, as we see from the contrast in (76) below:

Moreover, our analysis predicts that PP Adjuncts can be stacked on top of each other in any order: and this again seems to be true, as examples such as (77) below (where both italicised phrases are Adjunct PPs) illustrate:

Furthermore, under our proposed analysis of (77) (a) the sequences [student], [student with long hair], and [student with long hair in the corner] would all be N-bar constituents, so that we correctly predict that all three bracketed strings can be replaced by the pro-N-bar one in an appropriate context. And as we see from (78) below, the bracketed sequences can indeed be proformed by one in each case:

(78) (a) Which [student]? The one with long hair in the corner?

(b) Which [student with long hair)? The one in the corner?

(76) (a) (b)

(77) (a) (b)

the [N' [N' [N' student] with long hair] in the corner] the [N' [N' [N' student] in the corner] with long hair)

(82) (a) (b) (c)

the [students of Chemistry and professors of Physics]

the [students with long hair and professors with short hair) the [students of Chemistry and professors with short hair)

(c)

Which [student with long hair in the corner]? That one?

Each of the italicised conjuncts in (82) is thus an N-bar, as indeed is each of the bracketed coordinate structures.

Once again, our analysis turns out to make just the right predictions.

189

an Adjunct, so that the ungrammaticality of (85) (b) suggests that an NP which is part of an Adjunct PP cannot be preposed. Thus, there is an obvious contrast insofar as the Object of a Complement Preposition can be preposed, but not the Object of an Adjunct Preposition.

Yet another syntactic argument in support of positing a structural distinction between Complement and Adjunct Phrases relates to Co-occurrence Restrictions. In the case of a PP Complement, there are severe restrictions on the choice of P heading the PP; particular Nouns require (or, in the terminology we shall introduce in Chapter 7, subcategorisei a PP introduced by a particular Preposition: for example, only some Nouns, not others permit an a/phrase Complement: cf.

(86) (a) a student of Physics

(b) *a boy of Physics

(c) *agirlofPhysics

(d) *a teenager of Physics

(e) "e punk of Physics

By contrast the type of PP which functions as an Adjunct can be used to modify any type of head Noun (subject to semantic and pragmatic restrictions), as we see from:

More differences between Complements and Adjuncts 4.6

192

Noun Phrases

An additional syntactic argument in favour of drawing a structural distinction between Complements and Adjuncts derives from Extraposition facts (we shall discuss this phenomenon more fully in Chapter 8). It appears that PP Adjuncts can be extraposed from their Heads (i.e. separated from their Heads and moved to the end of their Clause) more freely than PP Complements: cf.

(83) (a) (b)

a student came to see me yesterday [with long hair] *a student came to see me yesterday [of Physics]

It would seem that in some sense PP Complements are more 'inseparable' from their Heads than PP Adjuncts. Once again, our analysis provides us with a principled way of accounting for these differences in purely structural terms. Thus, we might posit that the more closely related a PP is to its Head, the less freely it can be extraposed. And (to extend the genealogical terminology introduced in Chapter 3), we might say that Complements are sisters to their Heads, whereas Adjuncts are aunts (an aunt being a sister of the mother of a given node). To clarify the term aunt, consider a structure such as (84) below:

(84)

We might say that in an abstract tree structure such as (84), E is the sister of D, whereas C is the aunt of D. In these terms. a Complement would be a sister of its Head Noun. and hence more closely related to the Head than an Adjunct (which would be an aunt of the Head Noun): and we might suppose that it is because there is a greater structural affinity between Heads and Complements than between Heads and Adjuncts. that Complements are more resistant to being extraposed.

Given that Extraposition involves Postposing ; the obvious question to ask is whether Complements and Adjuncts behave any differently with respect to Preposing . There is some evidence that this is indeed the case. It would seem that an NP which is the Object of a Preposition heading a Complement PP can be preposed more freely than an NP which is the Object of a Preposition heading an Adjunct PP: cf. the contrast below:

(87) (a) (b) (c) (d) (e)

a student with long hair a boy with long hair

a girl with long hair

a teenager with long hair a punk with long hair

Once again, it seems as if, in some informal sense, Complements are more closely linked to their head Nouns than Adjuncts. And we might argue that the N-bar analysis enables us to define 'closeness' in purely structural terms, in the manner outlined earlier: e.g. we might say that sisters are more closely linked to their Heads than aunts, and we could posit that subcategorisation restrictions hold only between a Head and its sisters, not between a Head and its more distant relatives (e.g. aunts). Thus, in the case ofa structure such as (40) above, repeated as (88) below:

(85) (a) (b)

[What branch of Physics] are you a student of? *[ What kind of hair] are you a student with?

(88)

Thus, in (85) (a). the preposed bracketed NP is the Object of the Preposition of, and afintroduces a Complement phrase, so that (85) (a) involves preposing an NP which is part of a Complement PP. But by contrast, the bracketed preposed NP in (85) (b) is the Object of the Preposition with, and with introduces

191

N"

D~ ----------- N'

I N'~ ~pp

a N/ ~pp 410n~

stu~ent ~PhY~

we find subcategorisation restrictions holding between the head Noun student and its sister Complement PP [of Physics], but not between student and its aunt Adjunct PP [It'ith long hair]. But, naturally, any such account of subcategorisation restrictions presupposes a structural distinction such as that in (70) between Complements and Adjuncts,

At this point, it might be useful to summarise our discussion in this section, We have argued that there are a vast array of facts (some semantic, and some syntactic) which lend strong empirical support to the claim that Complements are attached at the N-level, and Adjuncts at the N-bar level. More precisely, the key claim we have made is that:

(89) (i) (ii)

More differences between Complements and Adjuncts 4,6

194

So un Phrases

(92) (a) (b) (l)

the Sl'(jGESTION [Ihalll'e should abandon cruise missiles] the DEMAND L(in him /0 /,(,,\'Ign]

the <.)UESTION [II/zellz£'l' euthanasia IS ellzical]

By contrast. a much wider range of constituents can function as postnominal Adjuncts not just PPs. but also temporal NPs, APs and Clauses (more precisely. Restrictive Relative Clauses), For example. in (93) below. each of the italicised constituents is an Adjunct of the N-bar containing the capitalised constituents:

Complements expand N into N-bar

Adjuncts recursively expand N-bar into N-har

(93) (a) (b) (c)

Thus, Complements and Adjuncts are similar in that they are both daughters of N-bar; but they differ in that Complements are sisters of (i.e. modify) N, whereas Adjuncts are sisters of (i.e. modify) N-bar.

Given the arguments we have presented here, PPs such as those italicised in (90) below would be Complements:

(90) (a) (b) (c) (d) (e)

the [N' [N' ABOLITION OF TAXES] [NP nexlrear]]

those [N' [N' STUDENTS OF PHYSICS] [AP absen I from class]] the [N' [N' KING OF EJ"GLAJ"D] [5 who ahdicaled]]

But I guess you've had just about enough of postnominal Phrases by now. so I'll spare you the relevant argumentation" or rather, leave you to devise some of it for yourself in an appropriate exercise!

However. there's one additional complication which we'll touch upon briefly. So far. all the Noun Phrases we've looked at have been simple NPs comprising a head Noun with or without an optional Determiner, Complement. or Adjunct. What we have not considered is the structure of complex N Ps such as those in (94) below:

your reply [to my letter]

the attack [on the Prime Minister] the loss [iJfthe ship]

her disgust [al his behaviour]

his disillusionment [with Linguistics]

(94) (a)

whereas PPs such as those italicised in (91) below would be Adjuncts:

(91) (a) the book [0/1 the table]

(b) the advertisement [on the television]

(c) the fight [after the match]

(d) his resignation [because of the scandal]

(e) a cup [It'ith a broken handle]

an advocate of the abolition of indirect taxation

(b) a 1l'0/1l(1n with an umbrella with a red handle

(c) her dislike of men with big egos

(d) a girl with a dislike of macho men

For reasons which should now be familiar. the of~phrases in these examples are all Complement PPs (hence sisters of N), whereas the ~I'ilh-phrases in each case are all Adjunct PPs (hence sisters ofN-bar), But the crucial point that we want to make here is that although all the examples contain two PPs, the first PP in each case modifies the first italicised nominal. while the second PP modifies the second italicised nominal. To make matters clearer, let's briefly look in ra ther more detail at the internal structure of each of the N Ps in (94) (omitting the relevant argumentation for the sake of brevity),

In (94) (a), both o(~phrases are Complement PPs: but they are Complements of different head Nouns, More precisely, the first of phrase is a Complement of the Noun advocate. whereas the second one is a Complement of the Noun abolition. Given our earlier arguments that Complements are sisters of N. then (94) (a) will have the structure (95) below:

leave you to verify this for yourself. applying the various 'tests' we have devised in this section, and previous sections (l bet you won't bother!)

Of course, our discussion here has been limited to postnominal Prepositional Phrases, But the Complement-Adjunct distinction can be shown to be valid for other types of postnominal phrase as well (though we lack the space to do this here), Generally speaking, only Prepositional Phrases and Clauses can function as the Complements of Nouns, For example, the italicised constituents in (90) above are Complement PPs: and the italicised in (92) below are Complement Clauses- i.e. the italicised Clause in each case functions as the Complement of the capitalised Noun:

193

(95)

More' differences between Complements and Adiuncts 40

(lOa) (i) 196

Noun Phrases

N"

D/ ~N'

I /l~

an N PP

I /~

advocate P N"

I »<>;

of D 'N'

t~e N/ ~PP

abol1ition ~ect ~

In (94) (d) we have a rather different situation: the first PP (= the with-phrase) is an Adjunct (hence a sister of N-bar), whereas the second PP (= the or phrase) is a Complement (hence a sister of N). But whereas the with-phrase modifies girl, the of-phrase modifies dislike. so that (94) (d) has the structure (98) below:

(98)

Conversely, in (94) (b) both ~t'ith-phrases are Adjunct PPs. and hence (given our earlier arguments) both sisters of N-bar. But the crucial point is that they are Adjuncts of different N-bar constituents: the first ~1'ith-phrase modifies the N-bar woman whereas the second modifies the N-bar umbrella, so that (94) (b) has the structure (96) below:

N" D/~N'

I -r >;

a N' PP

I ,/ <.

N P N"

I I //'~

woman with D N'

I /~

an N' PP

~ ~red~

I

umbrella

In (94) (c). the o(phrase is a Complement PP whereas the with-phrase is an Adjunct PP (recall that Complements are sisters of N, and Adjuncts are sisters of N-bar). However. whereas the o(phrase modifies the N dislike. the with-

(96)

Thus, in more complex cases of postnominal modification we have to be concerned not only with the question of whether a given post modifier is a Complement or an Adjunct. but also with the question of which particular nominal it modifies. Not surprisingly, therefore, we find cases of structural ambiguity such as the following:

(99) a woman with three children with ginger hair

where the Adjunct PP [with ginger hair] might be taken to modify either the N-bar [woman with three children], or the N-bar children: in other words, it might be either the woman or the children who have ginger hair. And on that colourful note, we'll conclude our discussion of post nominal modifiers'

phrase modifies the N-bar men. so that (94) (e) has the structure (97) below:

N"

-: "~N

~ -: <.

her N' PP,

! .> <.

dislike P N'

1 I

of N'

-: <.

N" PP

~ 4big~~

I

(97)

You should now be able to tackle exercises I, /I, /II, IV, and V

4.7 Nominal premodifiers

So far, our discussion has been limited to the syntax of postnominal modifiers. But what about prenominal modifiers? We shall argue that there are three structurally distinct classes of nominal premodifier, namely (i) Determiners. (ii) Complements, and (iii) Attributes (this last term is borrowed from Bloomfield 1935, p. 195). We shall further argue that these three different classes of premodifier have the different structural properties described in (100) below:

men

195

Determiners expand Nvbar into N-double-bar

And it should (by now') be obvious to you that the PP [oj Physics] in (103) is a Complement (hence modifies N), whereas the PP [at Cambridges is an Adjunct (hence modifies N-bar). We can provide some empirical support for the analysis in (103) by standard constituent structure 'tests': for example, (103) specifies that student is an N, [studenr of Physics] is an N-bar, and [srudenr of Physics at Cambridges is also an N-bar. Therefore. the analysis predicts that the latter two constituents (but not the former) can be replaced by the pro-Nbar one: and as (104) below illustrates, this is indeed the case:

So, the analysis in (103) above seems to produce the right results.

Now, if we want to attain maximal structural symmetry between (101) [a Cambridge Physics studenti and (103) [a student of Physics at Cambridge], then the natural suggestion to make is that Physics in (10 I) should be analysed as a Complement, and Cambridge as an Attribute (recall that Attributes are the prenominal counterpart of Adjuncts). In other words, our structural symmetry principle would suggest that (101) should be assigned the structure (105) below:

(For didactic reasons, we use the label NP for Noun Phrases whose internal structure is not the focus of our interest, and which we choose not to represent, and N" for Noun Phrases whose structure we want the reader to concentrate on, and which we show in some detail: but recall that from a theoretical viewpoint, NP and N" are equivalent terms.) Given the structure (l05), then Physics would be a Complement because it is the sister of the N student, whereas Cambridge would be an Attribute, because it is the sister (and daughter) of an N-bar. A structure such as (105) could be generated by a set of Phrase Structure Rules such as (106) below:

Nominal premodifiers 4.-

198

Noun Phrases

(ii) Attributes recursively expand N-bar into N-bar

(iii) Complements expand N into N-bar

Since both Attributes and Adjuncts recursively expand N-bar into N-bar, it seems clear that the two have essentially the same function, so that Attributes are simply prenominal Adjuncts (though we shall continue to follow tradition and refer to attributive premodifiers as A ttributes rather than Adjuncts). For the time being, we shall concentrate on the distinction between Complements and Attributes: and more specifically, we shall concentrate on the distinction between Complement NPs and Attribute NPs.

In this connection, consider the following Noun Phrase:

(104) (a) (b) (c)

(101)

a [Cambridge] [Physics] student

Clearly, (10 I) is ambiguous, between the two interpretations which can be paraphrased as in (l02) below:

(102) (i) a student of Physics (who is) at Cambridge

(ii) a student of Cambridge Physics (i.e. the particular brand of Physics taught at Cambridge, as opposed to Oxford Physics)

In our discussion here, we'll concern ourselves solely with the first and most natural interpretation, namely (102) (i): this is purely for didactic purposes, to make our exposition as simple and concise as possible (our analysis can be extended straightforwardly to deal with the second interpretation (102) (ii), in ways that I'll leave you to work out for yourself).

In analysing (10 I) (on interpretation (l 02) (ij), we might like to bear in mind the principle of 'structural symmetry' which Rulon Wells invoked in his analysis of the king of England (recall that he wanted to treat this as structurally parallel to the English king). What this means is that we'd like to make our analysis of (101) a Cambridge Physics student as close to our analysis of (102) (i) [a student of Physics (who is) at Cambridge] as possible. So, let's start by looking at the internal structure of (102) (i), ignoring the material in parentheses. Given all the arguments we put forward in the previous section, (102) (i) will have the structure (103) below:

(105)

(103)

N"

D ---------- --------------- N' IN' .i-> --------------- PP

a .r>: /" ~

N PP Lat Cambridge

stu~ent 4ehY~

Which [student of Physics]? The one at Cambridge? Which [student of Physics at Cambridge]? This one? Which [student]" *The one of Physics at Cambridge?

N"

D ---------- ----~ N' _______

~ NP-------- ~N'

I ~ ______

Cambridge NP N

I 'I

Physics student

(106) (i) N" -> (D) N' [Determiner Rule: cf. (75) (i)]

(ii) N' -> NP N' [Attribute Rule: optional]

(iii) N' -> (NP) N [Complement Rule]

197

199

Noun Phrases

Nominal preniodificrs 4.7

We make the Attribute Rule (106) (ii) optional because not all Noun Phrases contain Attributes: and we make the NP Complement in (106) (iii) optional because not all head Nouns have NP Complements.

However. while it would obviously be extremely satisfying if we were to be able to establish structural symmetry between the prenominal and postnominal Phrases in (103) and (105) above. clearly we need to base our analysis in (105) on some firmer foundation than a mere desire to find structural symmetry. So, what evidence is there that (105) is the right analysis for (10 I) [a Cambridge Physics studen(l] Well, part of the evidence comes from wordorder facts. For, we have already argued that Complements must always come closer to their head Noun than Adjuncts (if we are to avoid 'crossing branches'). And if Attributes are the prenominal counterparts of Adjuncts, then we should expect that Complements must also come closer to their Head Noun than Attributes. And this does indeed turn out to be the case. For as we see from (107) below, the Complement NP Physics must come closer to the head Noun student than the Attribute NP Cambridge:

specifies that both [Physics student] and [Cambridge Physics student] are N-bar constituents, whereas student is not (it is simply an N). Given our assumption that one is a pro-N-bar, we therefore predict that the first two of these expressions can be proformed by one, but not the third. And as (109) below illustrates, this is indeed the case:

(109) (a) (b) (c)

Which [Physics student]? The Cambridge one? Which [Cambridge Physics student]? This one? Which [student]? "The Cambridge Physics one?

So, one-pronominalisation facts lend strong empirical support to our analysis.

A third argument in favour of (105) comes from facts about Simple Coordination. Given that both the sequences [Physics student] and [Cambridge Physics student] in (105) are assigned the status of N-bar, then we should expect that both can be coordinated with another N-bar such as [hockey player]: and as (110) below shows, this is indeed the case:

(107) (a) a [Cambridge] [Physics] student

(b) *a [Physics] [Cambridge] student

Now, if we posit that Complement NPs are generated to the left ofN (cf the Complement Rule (106) (iii) above) whereas Attribute NPs are generated to the left of N-bar (cf. the Attribute Rule (I06)(ii) above), then it should be obvious why (107) (b) is ungrammatical: for, the only way in which we can generate a structure like (107) (b) in which the Attribute NP is closer to the Head Noun than the Complement NP is to allow 'crossing branches' as in (108) below:

(1IO)(a) (b)

a Cambridge [hockey player and Physics student] a [hockey player and Cambridge Physics student]

Moreover, while we can coordinate two Complement NPs (as in (III) (a) below) or two Attribute NPs (as in (III) (b », we cannot coordinate a Complement NP with an Attribute NP (hence the ungrammaticality of (III) (c)), or an Attribute NP with a Complement NP (cf. "(III) (dj):

(108)

(III) (a) several [Physics] and [Chemistry] students

(b) several [Oxford] and [Cambridge] students

(c) *several [Physics] and [Cambridge] students (d) *several [Cambridge] and [Physics] students

N

D~~N' .r >:

NP N'

NP~~

I --- I I

a Physics Cambridge student

But a structure such as (108) is ill-formed because it violates our condition that branches should not be allowed to cross. By contrast, there is no violation of the 'crossing branches' condition in (105) above, so that we correctly predict that only the Attribute + Complement order found in (105) is possible,

So, it seems that Coordination facts provide strong support for our analysis.

A fourth argument can be based on the different properties of the rules introducing Complement NPs on the one hand, and Attribute NPs on the other. The rules which we gave earlier in (106) above are repeated in (112) below:

(112)(i) N" (0) N'[DeterminerRule]

(ii) N' NP N' [Attribute Rule: optional]

(iii) N' (NP) N [Complement Rule]

not the Complement + Attribute order found in (108).

A second argument in support of the analysis in (105) can be formulated in relation to one-pronominalisation facts. For note that our analysis in (105)

Note that the Attribute Rule (112) (ii) is recursive (since it has the symbol N-bar both in its input and in its output): thus, the rule predicts that indefinitely many Attribute NPs can be stacked on top of each other. For

200

Nominal premodifiers 4. -,

202

Noun Phrases

structure will be (117) below (assuming insertion of appropriate items):

example, if we apply the Determiner Rule (112) (i), we generate the structure (113) below:

(113)

(117) N

»>: --- -----------

D------ ~N'

~ NP~~N'

~ghqu~ NP~ ~N'

-L~ = >:

middle class NP N'

I I

Cambridge N

I student

N"

D~~N

If we now apply the optional Attribute Rule (112) (ii) to expand the N-bar in (113), we derive:

(114)

N"

D ----------------- ~ N' NP~----- ~N'

Each of the NPs in (117) is an Attribute: and we should expect that they can be freely stacked on top of each other in any order (subject to stylistic, etc. restrictions). And as (118) below shows, this is indeed the case:

But we can now re-apply the same Attribute Rule (112) (ii) to expand the N-bar at the bottom of the tree in (114), yielding:

( 115)

~N~

D---- ~N'

NP~~N' NP~~ ~N'

(\ 18) (a) (b) (c) (d) (e) (0

a [Cambridge] [high quality] [middle class) student a [Cambridge] [middle class] [high quality) student a [high quality) [Cambridge] [middle class] student a [high qualitv] (middle class) [Cambridge] student a [middle class] [high quality] [Cambridge] student a [middle class] [Cambridge] [high quality] student

So, our recursive Attribute Rule seems to make precisely the right predictions (I leave to you the task of working out in how many different ways the NPs in (118) a bove can be pronominalised by the pro-N-bar one!).

But now let's compare the Attribute Rule with the Complement Rule (112) (iii), repeated as (119) below:

(119) N' ---> (NP) N [Complement Rule)

And we can even re-apply the same Attribute Rule once more, again expanding the lowest N-bar, thereby deriving:

(116) N"

D~~N'

NP~~N' NP~~r-;' NP~~N'

We see from (119) that the Complement Rule is not recursive, so that we predict that Complement NPs cannot be recursively stacked: and (120) below suggests that this is indeed the case:

(120) *a [Physics) [Economics] [Agriculture] student

So, our analysis correctly predicts that Attributes can be recursively stacked (cf. (118) above), but not Complements (cf. (120) immediately above).

There seems little point in cataloguing more and more empirical evidence in support of our proposal that prenominal Complement NPs are sisters of N and daughters of Nvbar, whereas prenominal Attribute NPs are both sisters and daughters of N-bar. If I haven't convinced you by now, I'm obviously not

However, let's assume that at this point we get tired of playing the game of repeatedly re-applying the same recursive rule, and instead choose to skip the Attribute Rule (112) (ii) this time and pass on to the Complement Rule (112) (iii). Now, if we choose to omit the optional NP Complement and instead choose to expand the lowest N-bar in (116) only as N, the resulting

201

Nominal premodifiers 4.-

204

Noun Phrases

going to! Instead, let's comment briefly on some aspects of the rules in (112) which we have used to generate Complement and Attribute NPs: these rules are repeated in (121) below:

(125) (a) Which [Cambridge student]? This onet

(b) Which [student]? The Cambridge one?

So, our analysis seems to have strong empirical support.

But now let's use our rules in (12 I) to generate a rather different type of structure. As before. we first apply the Determiner Rule (121) (i) to generate the substructure (122) above. But this time, we 'skip' the Attributive Rule (121) (ii), and go straight on to the Complement Rule (121) (iii): if we select the optional NP Complement. then from (122) by application of the Complement Rule we will derive the structure (126) below (assuming insertion of the relevant lexical items):

(121) (i) N" -> (D) N' [Determiner Rule]

(ii) N' -> NP N' [Attribute Rule: optional]

(iii) N' -> (NP) N [Complement Rule]

The rules in (121) specify that Determiners, Attributes, and Complements are all optional constituents of a Noun Phrase: i.e. they tell us that the only constituent which an NP must contain is a head Noun. Let's see how we can use our rules in (121) to generate first an NP containing an Attribute but no Complement, and secondly an NP containing a Complement but no Attribute; and let's examine the different structures assigned by the rules in these two cases.

If we apply the Determiner Rule (121) (i) - selecting the optional Determiner - we generate the substructure (122) below:

(122)

N"

D~~N'

N"

~~

o N '--.____

~ NP ~N

I I

Physics student

Note that the sequence [Physics student] in (126) has the status of N-bar, but not student (which is only an N). Hence. we correctly predict that only the former and not the latter can be proformed by the pro-NP one: cr.

( 126)

(123)

If we now apply the optional Attribute Rule (121) (ii) to (122), we derive:

(127) (a) Which [Physics student]? This one'?

(b) Which [student]? "The Physics one?

Thus, our analyses in (124) and (126) show that the NPs [a Cambridge student] and [a Physics student] are similar insofar as both the sequences [Cambridge student] and [Physics student] have the status of N-bar; but they differ in that student in the case of [a Physics student] has the status of N, whereas student in the case of [a Cambridge student] has the status of N-bar. This follows from the observation we made earlier that an N which has an overt Complement has the status of N, whereas an N which lacks a Complement has the status of an N-bar dominating an N. And we have already seen that our analyses in (124) and (126) make precisely correct predictions about one-pronominalisation facts.

In this section, we have argued that just as some postnominal PPs function as Complements to Nouns, and others as Adjuncts, so too some prenominal NPs function as Complements to Nouns, and others as Attributes. In fact, the parallel between postnominal PPs andprenominal NPs is a very close one. For example, all the postnominal PP Complements in the (a) examples below have prenominal NP Complement counterparts in the corresponding (b) example:

And if we subsequently apply the Complement Rule (121) (iii) - omitting the optional complement NP - we derive the structure (124) below (we have inserted appropriate lexical items for the sake of clarity):

Given our earlier definition of an A ttribute as a sister and daughter of N-bar, then it is clear that Cambridge in (124) functions as an Attribute. Note that both Cambridge student and student in (124) are assigned the status of N-bar, so that we correctly predict that both expressions can be proformed by the pro-N-bar one, as in:

(124)

N"

D~~N'

I .i->: ~

a NP N'

I I

Cambridge N

I student

(128) (a) ( b)

the ban [on pornography] the [pornography] ban

203

Nominal premodifiers s:

Noun Phrases

(129)(a) recruitment [of personnel]
(b) [personnen recruitment
(130) (a) the appeal [for charity]
(b) the [charity] appeal
(131) (a) relief [from famine]
(b) [famine] relief
(132) (a) damage [to the brain]
(b) [brain] damage
(133) (a) the investigations [into fraud]
(b) the [fraud] investigations
(134) (a) a fan [of Debbie Harry]
(b) a [Debbie Harry] fan
(135)(a) the allegations [of treachery]
(b) the [treachery] allegations It thus seems clear that prenominal NPs are the natural counterpart of postnominal PPs.

The two are not completely equivalent however. There are obvious syntactic differences between the two: premodifiers have the status of NP and precede the N-bar they modify, whereas postmodifiers have the status of PP and follow the N-bar they modify. Moreover, this syntactic difference is reflected in parallel semantic differences. To be more precise, the semantic relation between a prenominal NP and the N-bar it modifies is much more vague (and has to be inferred from pragmatic clues) than in the case of a postnominal PP. By way of example, consider a phrase such as the following:

(144) the [proportional representation] campaign

The bracketed prenominal NP in (144) has two very different postnominal PP counterparts: cf.

(145) (a) the campaign [jor proportional representation]

(b) the campaign [against proportional representation]

And in much the same way, the postnominal PP Adjuncts in the (a) examples below all have prenominal NP Attribute counterparts in the corresponding (b)

The difference between (144) and (145) is that the postnominal modifier in (145) contains a Preposition whose semantics specifies the relation between the Head Noun and the NP [proportional representation]: but in (144) there is no Preposition, and hence no additional semantic information, so that the exact relationship between [proportional representation] and campaign has to be inferred on the basis of pragmatic clues (i.e. knowledge of the way the world is). Thus, the two bracketed NPs in (146) below are likely to be interpreted very differently:

example:
(136) (a) the shop [on the corner]
(b) the [corner] shop
(137) (a) the strike [in the shipyard]
(b) the [shipyard] strike
(138)(a) the lady [of iron]
(b) the [iron] lady
(139) (a) the bridge [over the river]
(b) the [river] bridge
(140) (a) a keyboard [for a typewriter]
(b) a [typewriter] keyboard
(141) (a) a sauce [with cream]
(b) a [cream] sauce
(142) (a) tea [from China]
(b) [China] tea
(143) (a) the weather [in winter]
(b) the [winter] weather
205 (146) (a) Nancy Reagan's [drugs] campaign

(b) Ronald Reagan's [re-election] campaign

Our knowledge of the views of the indi vid uaJs concerned helps us interpret the N-bar [drugs campaign] as paraphraseable by 'campaign against drugs' in (146) (a), but 'campaign/or re-election' (e.g. of his own party) in (146) (b). And your knowledge of Syntax should (by now) enable you to work out for yourself whether the bracketed Phrases in (144--6) above are Complements, Attributes, or Adjuncts!

There are also other important differences between nominal premodifiers and postmodifiers. For example, generally speaking it seems to be that an NP which is part of a postnominal PP can alternatively be positioned in front of the nominal which it modifies. Thus, the italicised NP contained within the bracketed PP Complement in (147) (a) below can alternatively be positioned prenorninally, as in (147) (b):

206

207

Noun Phrases

Nominal premodifiers 4.7

(147) (a) (b)

a lover [pp of [NP classical music]] a [NP classical music] lover

You should now be able to tackle exercise VI

But let's see what happens if we try to do the same in the case of the italicised NP in (148) below:

(148) a lover [pp of [NP the opera]]

4.8 Adjectival premodifiers

Thus far, all the examples of Attributes which we have considered have involved attributive NPs. But other categories can be used in an attributive function as well. The commonest class of Attributes are APs (Adjectival Phrases): for example, the bracketed expressions in (153) below are attributive APs:

What we'd expect to get is (149) below:

(149) *a [NP the opera] lover

But, as you can see, we don't get this: instead, we ha ve:

( I 53)(a) (b) (c) (d)

a [really excellent] film

a (most entertaining] evening

a (de/igh(tully mysterious] stranger a (patently obviolls]lie

(150) an (NP opera ] lover

Why? What's going on here? Well, there seems to be some restriction to the effect that prenominal NP Complements cannot contain a Determiner. Moreover, examples such as (151) below suggest that the same restriction operates in the case of Adjuncts and Attributes:

In many cases, attributive APs alternate with attributive NPs: for example, each of the (a) examples below involves an NP attribute which has an AP attribute counterpart in the corresponding (b) example:

(152) (a) the (All India] radio station

(b) an [aI/points] bulletin

(c) a (half frame] camera

(d) the president's (no compromise] policy

(e) an (each way] bet

(f) an [any topic] discussion

(g) an [every weekend] girl

(154) (a) the (England] football players
(b) the (English] football players
(155)(a) a [Paris] nightclub
(b) a [Parisian] nightclub
(156) (a) a [metal] finish
(b) a [metalliC] finish
(157)(a) a [prestige] project
(b) a [prestigious] project
(158)(a) the [II'inter] weather
(b) the [1I'intry] weather (151) (a) a/the/this strike [pp in [NP the shipyard]]

(b) *a/the/this [N P the shipyard] strike

(c) a/the/this [NP shipyard] strike

The exact nature of the restriction is anything but clear. It seems that only some kinds of Determiner are barred from occurring in Attribute NPs, and that others can indeed be used in this function. Thus, in the following examples:

Moreover, the following example (taken from confidential University papers!) shows that an attributive AP can be conjoined with an attributive NP:

the italicised constituents might be argued to be Determiners, so discounting the possibility of a 'blanket restriction' against the use of Determiners in prenominal NPs. It would seem that Articles (a, the) and Demonstratives (this/ that/these/those) are barred from occurring in attribute NPs, whereas Quantifiers like every/each/all/both/haff/any!some!no, etc. are not. Quite why this should be is not a question which need concern us here (OK ... I'll admit that I always say that when I don't know the answer to my own question!)

(159) Any change is bound to have numerous [AP academic] and (NP cost] implications

So, there does seem to be an apparent parallelism between attributive NPs and attributive APs.

Now, if attributive NPs are generated by rule (121) (ii) above, repeated as (160) below:

(160) N' -+ NP N' [Attribute Rule: optional]

208

and if attributive Adjectival Phrases seem to be structurally parallel to attributive Noun Phrases, then we might propose to generate Attributive APs by a parallel rule such as (161) below:

(161) N' ---+ AP N' [Attribute Rule: optionali

Moreover, it may well be that we can conftate our two Attribute Rules (160) and (161) above into a single rule. How? Well, recall that in Chapter 3 we argued that categories are analysable into matrices (= sets) of syntactic features, so that (e.g.):

Thus, Noun and Adjective might be argued to form a supercategory of [+ N] elements. And in the same way, we might say that NP and AP form a corresponding phrasal supercategory which we might designate as [+ NP] (i.e. a phrasal constituent with a [ + N] head). Given these assumptions, then our two Attribute Rules (160) and (161) could be conftated as (163) below:

(163) N' ---+ [+ NP] N' [Attribute Rule: optional]

If we incorporate our generalised Attribute Rule (163) into our earlier system of rules in (121) above, our revised set of rules becomes ( 164) below:

(164) (i) N" ---+ (D) N' [Determiner Rule]

(ii) N' ---+ [+ NP] N' [Attribute Rule: optional]

(iii) N' ---+ (NP) N [Complement Rule]

Now, since our revised Attribute Rule (164) (ii) is recursive, it predicts (amongst other things) that Noun Phrases can contain indefinitely many stacked attributive APs: and this does indeed seem to be the case, as (165) below illustrates (where each of the bracketed constituents is an attributive AP):

( 162)

(165) (a) (b) (c) (d) (e)

Adjectival premodifiers 4./>.

210

Noun Phrases

Applying the Attribute Rule (164) (ii) to (166), and selecting AP as our [+ NP] category. will yield (167) below:

( 1(7)

Reapplying the same rule in the same way to the lower N' in (167) will then derive ( 168) below:

Noun=[+N, -V]

Adjective = [ + N, + V)

( 1(8)

Applying the same Attribute Rule in the same fashion once more to expand the lowest N-bar in (168) will give us (169) below:

Finally. if we apply the Complement Rule (164) (iii) to expand the lowest N' in (169) into N. the result will be (170) below (we have inserted the relevant words for illustrative purposes; of course. AP will have further internal structure. but this need not concern us for the time being):

a [handsome) stranger

a [dark] [handsume] stranger

a [tall] [dark] [handsome] stranger

an [intelligent] [rail] [dark] [handsome] stranger etc.

By way of illustration, let's see how our rules (164) would generate a Noun Phrase such as (165) (c) [a tall dark handsome stranger]. Applying the Determiner Rule (164) (i) and selecting the Determiner option would generate the structure (166) below:

(166)

N" D~"~----,~·

209

Adjectival premodifiers 4./'1

212

Noun Phrases

Now, the structure (170) predicts that the sequences stranger, [handsome stranger], [dark handsome stranger], and [tall dark handsome stranger] are all N-bar constituents. Among the predictions we therefore make is that each of these can be proformed by the pro-N-bar one: and (171) below shows us that this prediction is exactly correct:

(175) (a) (b)

(176) (a) (b)

the best [available] person the best person [available]

the only [suitable] actor the only actor [suitable]

But there are extremely complex restrictions on when APs can or cannot be used prenominally, and when they can or cannot be used postnominally: the examples below illustrate some of the restrictions concerned:

(171) (a) Which [stranger]? The tall dark handsome one?

(b) Which [handsome stranger]? The tall dark one?

(c) Which [dark handsome stranger]? The tall one:

(d) Which [tall dark handsome stranger)? This one?

(177) (a) He has a [similar] car

(b) *He has a car [similar]

(178) (a) He has a [similar though subtly different] car

(b) He has a car [similar though subtly different]

So, it would appear that the structural parallels between attributive NPs and attributive APs are quite striking: both can be used to recursively expand N-bar into N-bar.

Now, if we are correct in positing that both NPs and APs can be used as Attributes (hence can recursively expand N-bar into another Nvbar), then it follows that we should expect that two different kind of Attributes can be recursively stacked on top of each other in any order. We can demonstrate this by the free relative ordering of the NP and AP Attributes in (172) below:

(172) (a) (b) (c) (d) (e) (f)

a [AP Japanese] INP toy] INP plastic] duck a [AP Japanese] [NP plastic] [NP toy] duck a [NP toy] [AP Japanese] INP plastic] duck a [NP lOY] [NP plastic] [AP Japanese] duck a [NP plastic] [AP Japanese] [NP toy] duck a [NP plastic] [NP toy] [AP Japanese] duck

(179)(a) (b)

(180) (a) (b)

(181) (a) (b)

(182)(a) (b) ( c)

He has a [similar enough] car He has a car [similar enough]

He has a [fairly similar] car He has a car (fairly similar]

*He has a [very similar indeed] car He has a car [very similar indeed]

* He has a [similar to mine] car He has a car [similar to mine] He has a [similar] car [to mine]

We are not here going to attempt to unravel the complex syntactic and stylistic factors which determine the position of APs used to modify N-bars: the interested reader should consult a good reference grammar such as Quirk et 01. ( 1985).

Well, I know you'd want me to end the chapter on a high, so I'll conclude with an argument of your favourite type - relating to structural ambiguity. No, don't groan -- this one is quite straightforward. Consider the Noun Phrase in ( 183) below:

(183) an English teacher

Thus, word order facts provide striking support for our analysis: they are exactly as predicted. I leave you to verify for yourself how many different ways [a Japanese toy plastic duck] can be pronominalised by one: think about it, next time you play with your toy duck in the bath (if you don't have a toy duck, a toy boat etc. will do just as well).

We have argued in this section that APs can function as prenominal Attributes: but recall that we argued at the end of section 3.6 above that APs can also be used as postnominal Adjuncts. Indeed. in certain types of construction, APs of a given class can be positioned either pre- or post-nominally: cf. the following examples from Quirk et 01. (1985, pp. 418-29):

(173) (a) (b)

(174)(a) (b)

This is ambiguous in a fairly obvious way between the two interpretations represented by the paraphrases in (178) below:

(184) (i) someone who teaches English

(ii) someone who teaches, and who is English

Now, it seems plausible to assume that the source of the ambiguity of (183) is structural in nature, and that part of the ambiguity relates to the fact that

the best [possible] use the best use [possible]

the greatest [imaginable] insult the greatest insult [imaginable]

211

Adjectival premodifiers 4.8 Eng/ish has two different categorial functions in (183). On the first interpretation (= someone who teaches English), Eng/ish functions as a Noun, and hence can be modified by an Adjective, as in (185) below:

On the second interpretation (= a teacher who is English), English functions as an Adjective, and hence can be modified by an Adverb, as in (186) below:

So, part of the ambiguity of [an Eng/ish teacher] lies in the categorial status of Eng/ish, which can either be a prenominal NP, or a prenominal AP: cf.

But this is only part of the story. I suppose you'd already guessed that the answer couldn't be that simple! But why not?

Well, what we're going to argue is that when [an English teacher] means 'someone who teaches English', then the prenominal NP English is a Complement; but when the NP means 'a teacher who is English', the prenominal AP Eng/ish is an Attribute. Now, if I'm right (am I ever wrong"), then it follows that the NP [an Eng/ish teacher] could have either of the two structures in (188) below:

(188) (a) [= a person who teaches English]

N"

0----------- ~ N'

aln NP---------- ~N

I I

English teacher

(b) [ = a person who teaches, and who is English]

(185)(a) (b) (c)

(186) (a) (b) (c)

(187) (a) (b)

an [NP O/d Eng/ish] teacher

a [NP Midd/e English] teacher a [NP New English] teacher

214

Noun Phrases

a [AP typically Eng/ish] teacher a [AP very English] teacher

a [AP disappointing/y English] teacher

(I leave you to work out for yourself which of our rules in (164) apply to generate each of the structures in (188).) Now, the analysis in (188) has a certain amount of semantic plausibility. After all, if we accept the claim (cited earlier) by Hornstein and Lightfoot (l98Ia, p. 21) that 'each N-bar specifies a semantic property', then from the fact that (188) (a) contains only the single N-bar [English teacher], it follows that (188) (a) will attribute only one semantic property to the person concerned, namely that (sjhe teaches English. But by the same token, the fact that (188) (b) contains two N-bar constituents, namely [teacher] and [English teacher], entails that (188) (b) will attribute two properties to the individual concerned, namely (i) that (s)he teaches, and (ii) that (s)he is English.

However, the proposed analysis also has independent syntactic support.

Well, I won't bore you with a whole battery of syntactic arguments in support of our analysis. Let's just look at one such argument, relating to word-order facts. Consider a sentence such as the following:

an [N P English] teacher (= someone who teaches English) an [A P English] teacher (= a teacher who is English)

(189)

I think it would be crazy to employ [a French English teacher]

Just think about what (189) means: does it mean (190) (i) below, or (190) (ii)?

(190) (i) I think it would be crazy to employ a French person to teach English

(ii) I think it would be crazy to employ an English person to teach French

Well, I think it's pretty clear that the NP [a French English teacher] can only be interpreted along the lines of (190) (i) as 'a French person who teaches English', and not along the lines of (190) (ii) as 'an English person who teaches French'. But if you think about it, that's exactly what our analysis in (188) predicts. For, as we have noted many times, our rules specify that Complements always come closer to their Heads than Attributes/Adjuncts. This means that when we have a prenominal Attribute and a prenominal Complement associated with the same Head Noun, they will occur in the order:

(191) Attribute + Complement + Noun

N"

0----------- ~ N'

aln AP---------- ~N'

I I

English N

I teacher

Thus, in a sequence such as [French English teacher], since English is closer to the Head Noun teacher, it must be a Complement; and since French is further away from teacher, it must be an Attribute. Hence, we correctly predict that the only possible interpretation for [a French English teacher] is 'a person who teaches English who is French'. So our analysis not only has semantic plausibility; but in addition it has independent syntactic support.

But what is even more exciting is that there is also phonological evidence in

213

215

Noun Phrases

Adjectival premodifiers 4.8

support of the structural ambiguity represented in (188) above. For, just as differences of syntactic structure correlate with differences of semantic structure (i.e. meaning), so too we might expect them to correlate with phonological differences. And this is precisely what we find. For if you think about it, you'll realise that the phrase [an English teacher] has two different stress patterns, namely those represented in (192) below (where the syllables receiving primary stress are CAPITALISED):

4.9 Summary

The general aim of this chapter has been to provide empirical support for the claim that there is a type of nominal constituent which is larger than the Noun but smaller than the Noun Phrase. In 4.2 we examined Rulon Wells' claim in 1947 that the NP [the king of England] has as one of its immediate constituents just such a 'small nominal phrase', [king of England]; and we proposed to adopt Chomsky'S bar-notation for such cases, whereby king is an N, [king of England] is an N-bar, and (the king of England] is an N-double-bar. In 4.3 we produced empirical evidence in support of the proposed analysis, based on Coordination and one-pronominalisation facts. In 4.4, we argued that N-bar has a crucial role to play in differentiating two different classes of postnominal modifier - Complement PPs (which are sisters of Nand daughters of N-bar), and Adjunct PPs (which are both sisters and daughters of N-bar). In 4.5 we looked at how to deal with the optionality of Determiners, Adjunct PPs, and Complement PPs in Noun Phrases. In 4.6 we presented further evidence in support of the structural distinction drawn between postnominal PP Complements and PP Adjuncts. In 4.7 we argued in favour of positing a parallel structural distinction for premodifiers between Complement NPs (which expand N into N-bar), and Attribute NPs (which recursively expand N-bar into N-bar). In 4.8 we argued that not only NPs, but also APs can function as Attributes, and accordingly we generalised our Attribute Rule to allow [ + NP] phrases (i.e. NPs and APs) to function as Attributes.

We have summarised the various rules we have posited in this chapter in (195) below. for your convenience:

(192) (i) an ENglish teacher

(ii) an ENglish TEAcher

And if you think even harder, you'll realise that the two different stress patterns correspond to the two different interpretations in (184) above: that is, (192) (i) means 'someone who teaches English', whereas (192) (ii) means 'a teacher who is English'.

Now, why should the two different stress patterns in (192) be associated with two different meanings? Well. let's assume that just as each N-bar in a sentence is a semantic unit (recall Hornstein and Lightfoot's (198Ia, p. 21) observation that 'each N-bar specifies a semantic property'), so too each N-bar is a phonological unit. More specifically, let's assume that the rule for primary stress assignment in English is along the lines given very informally in (193) below:

(193) Assign a separate primary stress to each separate N-bar (i.e. to an appropriate syllable of an appropriate word in each N-bar)

If we look at the two structures assigned to [an English teacher] in (188) above, we see that the N-bar constituents which each contains are as in (194) below:

(195) (i) N" ~ (D) N' [Determiner Rule]

(ii) (a) N' ~ N' PP [Adjunct Rule: optional) (b)N' ~ [+ NP] N' [Attribute Rule: optional) (iii) (a)N' ~ N (PP) [PP Complement Rule] (b)N' ~ (NP) N [NP Complement Rule]

(194) (a) an [N' English teacher] (= Complement = 'someone who teaches English')

(b) an [N' English [N' teacher]] (= Attribute = 'someone who teaches

who is English')

And our Stress Rule (193) above will accordingly assign primary stress only to the (first syllable of the) word English in (194) (a), but to (the first syllable of) both the words English and teacher in (194) (b). Thus, it seems clear that phonological facts provide strong independent empirical support for our claim that [an English teacher] exhibits the structural ambiguity characterised in (188) above. So, it's nice to find that an analysis which has syntactic and semantic plausibility turns out to have independent phonological motivation.

As we have already seen. the Determiner Rule (195) (i) has to apply, if we want to generate a Noun Phrase; the Adjunct Rule (195) (ii) (a) and the Attribute Rule (195) (ii) (b) are optional; and one of the two Complement Rules (195) (iii) (a) or (b) must also apply if we are to generate a properly terminated subtree (i.e. an NP which terminates in an N-node).

You should now be able to tackle exercises VI/- XVI

216

Exercises

(b) 218

Noun Phrases

EXERCISES

"Exercise III

In the text, we considered only one analysis of [rhe king of England], Evaluate this analysis against the two alternative analyses given in (I) and (2) below:

Exercise I

It might be argued that the bracketed Prepositional Phrase in (1) below is a Complement, whereas that in (2) below is an Adjunct:

(I) the discussion [of the riots]

(2) the discussion [in the bar]

(I)

Given this assumption, what structure would be assigned to each of the Noun Phrases in (l) and (2), and what rules would be necessary to generate the relevant structures (show each individual step in the derivation)?

Show how the assumption that the bracketed PP is a Complement in (1) but an Adjunct in (2) would account for the following contrasts (assuming the grammaticality judgments given):

(3) (a) (b)

(4) (a) (b) (c)

(5) (a) (b)

(6) (a) (b)

(7) (a) (b)

"Exercise IV

The discussion in the text ignored clausal postmodifiers, We might suppose that just as postnominal Prepositional Phrases can function either as Complements or as Adjuncts. so too can postnominal Clauses, For example, we might argue that the bracketed postmodifying Clause in (I) (a) below is a Complement of the head Noun claim, whereas that in (I) (b) is an Adjunct:

(2)

·The discussion of the match was more animated than the one of the riots The discussion at the match was more animated than the one in the bar

The discussion of the riots and of their implications was full and frank The discussion at the match and in the bar was full and frank

"The discussion of the riots and in the bar was full and frank

The discussion of the riots in the bar was full and frank "The discussion in the bar of the riots was full and frank

(l)\a) (b)

The discussion was rather misleading in the document "The discussion was rather misleading of the document

NP

NP~ ~pp

~eki~ ~ng~

the claim [that you made a mistake] the claim [that you made]

Which document did they ban the discussion of?

"Which document did they ban the discussion in?

This correlates with the traditional distinction drawn between Noun Complement Clauses (NCC) such as that bracketed in (I) (a), and Restrictive Relative Clauses (RRC) such as that bracketed in (I) (b), The two different types of Clause have differing lexical properties: for example, the particle rhat can be replaced by an appropriate wh-pronoun such as which or omitted altogether in Relative Clauses like (I )(b), but not in Noun Complement Clauses such as (I) (a): cf.

• Exercise II

Discuss the syntax of the bracketed Noun Phrases in the following sentences, presenting empirical arguments to support your analysis:

II)

(2) (a)

I met [a specialist in fibreop tics from Paris]

12) [The girl on the stage in jeans] is a friend of mine

13) [The journey from Paris to Rome on Sunday] was tiring

(4) [The ban on belts with studs in the school] has caused a lot of resentment

15) [The girl at the disco lust week] rang me up yesterday

(b)

the claim [that/·which/·{i} you made a mistake] (Noun Complement Clause)

the claim [that/which/{i} you made] (Restrictive Relative Clause)

More importantly, we might argue that the two types of Clause have different structural properties, in that NCCs are Complements, whereas RRCs are Adjuncts. Show how this difference might be represented in structural terms within the N-bar framework (assume that the bracketed Clauses in (1) have the status of S, but ignore their internal structure); and present empirical arguments in support of your analysis.

Why would examples such as the following prove problematic for such an analysis?

"ow discuss possible differences in structure between the bracketed NPs in (6) and (7) below:

16)

She's [another friend of Mary] She's [another friend of Mary's]

(3) (a)

I For the purposes of this exercise, simply assume that of Mary and of Mary's are PPs, and don't concern yourself with the internal structure of these PPs.)

217

The claim which Reagan made that no arms had been exchanged for hostages was greeted with scepticism

The claim has been reiterated that no arms were exchanged for hostages

Exercises

220

Noun Phrases

How would these problems be overcome if we assumed a general CLAUSAL EXTRAPOSInON rule along the lines of:

(4)

NP such as:

Any postnominal Clause (i.e. NCC or RRC) can be extraposed to the end of the minimal Phrase or Clause containing it

(I)

corresponding to the two (very subtly different I) interpretations in (2) below:

the tall girl in jeans

(2) (i) (ii)

• Exercise V

For the purposes of the exercise. assume that the Determiner any (in its existential function. where it is similar in meaning to 'some') is subject to the following restriction:

(I)

(2) (a) (b) (e)

(3) (a) (b) (c)

Exercise VII

Show how the rules in the text would assign two different structures to an

the girl in Jeans who is tall

the tall girl who is wearing jeans

Discuss which rules would be needed to generate the relevant structures. and how they would apply. In addition, show how your analysis accounts for the one-pronominalisation facts in (3) below:

Any must be preceded and c-commanded by a negative

(3) (a) (b)

(Recall that in Exercise IV of Chapter 3, we suggested a similar restriction to the effect that ever must be preceded and c-commanded by a negative.) Show how a restriction such as (I) above would account for the {un)grammaticality of existential any in the bracketed NPs in the following examples. concentrate only on the structure of the bracketed NPs. ignonng the rest of the sentence for the purposes of the exercise. Assume that no and lack are 'negative' constituents in the requisite sense, and assume the grammaticality judgments given.

(I) (2)

[No king of any country] abdicated [No king ofanv importance] abdicated

[No king of any country oj any importance] abdicated

Were you talking to the tall [girl in jeans], or the short one'?

Were you talking to the [tall girl] in jeans, or the one in a miniskirt?

* Exercise VIII

Discuss the ambiguity of the following

an old French student

the nuclear test ban treaty

How might it be represented in structural terms? What evidence is there is favour of your proposed analysis? And what rules would need to apply (how?) to generate the relevant structures?

[The lack of any discipline in some schools] worried them "[The lack of discipline in any schools] worried them

[The lack of teachers with any qualifications] worried them

(I)

··Exercise IX

A Noun Phrase such as

the English king

might be argued to be open to the two interpretations paraphrased in (2) below:

the king who is English

(2) (i)

(ii) the king of England

Compare and contrast the two following accounts (A and B below) of this 'dual interpretation':

* Exercise VI

Discuss the ambiguity of the following NPs, and how it might be represented in structural terms, giving evidence in support of your analysis. What rules would be required to generate the relevant structures?

Assume that no, an)" and some all have the categorial status of Determiners (more precisely, they belong to the subset of Determiners called Quantifiers).

Show how facts such as these could be used as the basis for an argument supporting the structural distinction we have drawn in the text between Complements and Adjuncts.

(I) (2) (3) (4)

the house in the wood near the park a toy factory

a brass button holder

the king of England's people

219

Analysis A: a structural ambiguity analysis

On interpretation (2) (i) English is an Attribute, whereas on interpretation (2) (ii) English is a Complement

Analysis B: a pragmatic analysis

On both interpretations, English is an Attribute: there is no structural ambiguity. The relationship between English and king is left vague, so that English king means more or less 'king connected in some unspecified way

Exercises

222

Noun Phrases

with England' (so that the phrase could mean all sorts of things, including 'king born in England', 'king who rules over England', 'king who behaves in a typically English fashion' (as in 'King Hussein of Jordan is the most English king I've ever met'), etc.).

Which interpretation of which sentence proves problematic for the assumptions we are making about one, and why? Can you think of any wayis) in which we might defend the assumption that one is a pro-N-bar in the face of such apparent counterevidence?

HINT: You might like to explore pragmatic rather than purely syntactic solutions to the problem.

Discuss the structures assigned by each of these two accounts to (I).

What predictions would each of the two analyses (A and B) make about possible and impossible interpretations of English in the italicised phrases below, and why? Which set of predictions are correct, and what is the implication of this for each of the analyses proposed?

(3) (a)

(4)(a) (b) (c) (d)

(4)

Now consider the following two dialogues:

Which reply by Speaker B proves problematic for the analysis of one given in the text, and why?

(b)

Why do philosophers always use examples involving a bald French king, rather than a bald English one?

There's not much to choose between the present English and French kings, except that the French king is Jess bald that the English one

Henry VIII is the best known English Protestant king

We've had relatively few English septuagenarian kings

Boedicea was the most famous English pagan queen

Henry IV was the last English French king

(5)

SPEAKER A: Pass me that picture over there

SPEAKER B: Which one? The one of a girl in a bikini? SPEAKER A: Pass me that picture over there

SPEAKER B: Which one? The one with a gold frame?

** Exercise XI

In the text, we argued that multiple attributive APs are hierarchically, not linearly stacked. But consider the alternative possibility that APs can be stacked in either way onto an Nvbar. giving rise to possible structural ambiguities such as the following:

··Exercise X

Much of the justification for the analysis of Noun Phrases presented in the text rests on the assumption that (i) the antecedent of one is always a unitary constituent, and (ii) it is an N-bar. Given these (and other related) assumptions, a phrase like a big hlack dog will have the structure (1) below:

(I)

(1)

N"

~~N'

9 .>: ~N'

a AlP .i-::" ~ ,

big AP N

I I

black N

I dog

(2)

For speakers such as myself (and the late David Kilby, who first pointed out the problem to me), sentence (2) (a) below can have either of the interpretations (2) (b) or (2) (c), and sentence (3) (a) can have either of the interpretations (3) (b) or (3) (c):

(2) (a) (b) (c) (3) (a) (b) (c)

Jane has a big black dog, and Jean has a small one ... a small black dog

... a small dog

Jane has a big black dog, and Jean has a brown one ... a big brown dog

... a brown dog

N"

D »>: ~N'

I <->: ~

my AlP N ~

fi rs tAP ---------- N '

I I

disastrous N

I marriage

marnage

This structural difference might correlate with a semantic difference: thus, we might suppose that disastrous has a restrictive interpretation in (I), and an appositive interpretation in (2). On the restrictive interpretation (I), the phrase would be taken to imply that I had other disastrous marriages; whereas on the appositive interpretation (2), the phrase would be taken as implying that I had other marriages, though they were not

221

Exercises

224

Noun Phrases

English. Show what problems appear to be posed for this assumption by the bracketed NPs in examples such as those below:

c.c..:essanly disastrous. Note that English spelling marks the difference between the two .[Link]: on the restrictive interpretation, there are no punctuation marks of any "'rl separating the two APsfirst and disastrous; but on the appositive interpretation, :he tw 0 APs are separated by a comma, a hyphen, or a bracket: cf.

I.' i ta) Ib) (c) (d)

(I) %[ Your last week's leller] was very interesting

(2) [My John Smith's best biller] has gone flat

(3) It's no more than [a silly old wives' tale]

(4) Look at [those three quaint old Cornish fisherman's two-bedroomed terraced cOllages]

my first disastrous marriage = (I) = restrictive

my first, disastrous, marriage = (2) = appositive my first (disastrous) marriage = (2) = appositive my first - disastrous - marriage = (2) = appositive

(5)

It was [a typical old men's drearyfisherman's tale]

How might these various problems be overcome? How will your analysis deal with the ambiguity of phrases like:

Of course, the punctuation differences here reflect intonational differences. What kind of additional evidence might be adduced in support of (or against) such an analysis? And how would our PS rules have to be revised so as to allow for structures such as (2) as well as (I)?

(6)

an old man's bicycle

•• Exercise Xli

Leonard Bloomfield in his classic work Language (1935, pp. 203-6) recognises the following classes of Determiners:

•• Exercise X I V

Discuss the syntax of the following pairs of examples, pointing out the similarities and differences between them. What complications (if any) do they pose for the analysis in the text, and what modifications to that analysis (if any) do they entail?

(J)(i) (ii) (iii) (iv) (v)

Definite Article the and Indefinite Article a Demonstratives this/that/these/those

Quantifiers such as e~ery/each/all/any/some/no/many/most!few, etc. Interrogatives such as which/what

Genitive (= possessive) NPs like the man next door's, and the corresponding Possessive Pronouns myiyourlhislher, etc.

(I) (a) (b)

(2) (a) (b)

(3) (a) (b)

(4)(a) (b)

(5) (a) (b)

In the text, we suggested that Determiners are non-recursive premodifiers which expand N-bar into N-double bar. Discuss the apparent problems posed for this claim by the bracketed Phrases in the examples below. To what extent can these problems be handled within our existing grammar, or do they require modifications to the grammar, and ifso of what sortts)?

(I) (2) (3) (4) (5) (6) (7) (8) (9)

[His many faults] can't be overlooked

Don't expect me to indulge [your every whim] [Myfew real friendsi would never desert me [All the African competitors] have withdrawn [What a silly fool] he is'

There's [many a slip] twixt cup and lip You are [such an insufferable bore] The hole has [too narrow an aperture] I've never seen [so pretty a picture]

Salvador Dali's latest portrait the latest Salvador Dali portrait

my family's income my family income

the film about aparthied last week last week's apartheid film

the presentation of the medals to the athletes the medal presentation to the athletes

my Linguistics lecture to the F acuIty my Faculty Linguistics lecture

"Exercise XV

Try and work out the nature of the restrictions illustrated in the examples below (inventing further examples of your own as necessary to expand the paradigm given below):

(I) (a) (b)

··Exercise Xl/I

It is generally assumed (cf. e.g. Bloomfield's classification cited in Exercise XII) that Genitive (= possessive) NPs function solely as (non-recursive) Determiners in

(2) (a) (b)

(3) (a) (b)

223

this/that green one this/that one

the tall one in the corner the one in the corner

a yellow one *a one

225

Exercises

my/your gold one %my/yourone

(b)

many/all old ones "many/all ones

I Dialectally, 'He's a one' can mean 'He's a strange person')

··Exercise XVI

In examples (I) and (2) below, the bracketed NPs might be argued to be -tructurally ambiguous in respect of which nominal is modified by the italicised Adjunct PP. Draw Phrase-markers to show how this ambiguity could be represented I limiting yourself to the structure of the bracketed N Ps, and ignoring the rest of the sentences). Which interpretation of each sentence proves problematic for the assumption made about Adjunct PPs in the text (i.e. that they expand N-bar into N-bar), and why?

11) 12)

I like [that picture of him in his pushchairi

[The workers and the managers in thefactorvi just don't get on

Suppose we were to redefine a nominal Adjunct as an expression which expands a nommal of a given type into another nominal of the same type (hence, expands N into N, or :\-bar into N-bar, or NP into NP). Would this help us deal with the problems posed by the italicised Adjunct phrases in (I) and (2) above?