English Word Classes and Phrases - Aarts and Haegeman

English Word Classes and Phrases 115 Part II THOC06 Syntax 115 1/10/06, 1:48 PM 116 Bas Aarts and Liliane Haegeman THOC06 116 1/10/06, 1:48 PM English Word Classes and Phrases 117 6 English Word Classes and Phrases BAS AARTS AND LILIANE HAEGEMAN 1 Introduction: Aims and Scope In this chapter we introduce two concepts which are essential for the description of the grammar of a language: word classes and phrases. In the first part of the chapter (section 2), we examine the classification of words into categories and we highlight some of the many problems that may arise. Among other things, we will outline some of the solutions proposed for dealing with words that seem to have properties of different categories. In the second part of the chapter we turn to the grouping of words into phrases, and we examine in particular the constituency of what is referred to as the ‘verb phrase’ (sections 3–4). We will integrate our conclusions into a representation of the structure of clauses (section 5). Section 6 is a brief summary of the chapter. 2 Word Classes 2.1 Definitions Word classes (also known as parts of speech) are essential for any grammatical description, even though we can never really be entirely sure what their nature is. The reason for this uncertainty is that word classes are not tangible three-dimensional entities, but mental concepts, i.e. they ‘exist’ only in our minds. Word classes can be viewed as abstractions over sets of words displaying some common property or properties. In this section we will be looking at a number of approaches to word classes, asking in particular how we can define them, and whether they have sharp boundaries. For English, most linguists agree on the need to recognize at least the following word classes: noun, verb, adjective, preposition, adverb, determinative and conjunction. Each of these word classes is illustrated in the sentence below: THOC06 117 1/10/06, 1:48 PM 118 Bas Aarts and Liliane Haegeman (1) [determinative The] [noun chairman] [preposition of] [determinative the] [noun committee] [conjunction and] [determinative the] [adjective loquacious] [noun politician] [verb clashed] [conjunction when] [determinative the] [noun meeting] [verb started] Each member of the word classes can be the head of an associated phrasal projection, e.g. a noun can be the head of a noun phrase, an adjective can be the head of an adjective phrase, verbs head verb phrases, prepositions head prepositional phrases, etc.1 Phrases will be discussed in greater detail in sections 2 and 3. Sections 4 and 5 consider the way phrases are combined to form clauses. The question arises how to define word classes. The oldest way to go about this is by appealing to so-called notional definitions, an approach familiar from school grammars. In this tradition, a noun is defined as ‘a word that denotes a person place or thing,’ and a verb is an ‘action word.’ While perhaps useful in certain pedagogical settings, notional definitions are not adequate. For nouns, the definition clearly fails, for example, in the case of abstract words like freedom, intelligence and rudeness. As far as verbs are concerned, there are many words that do not refer to actions, but which we would nevertheless want to call verbs, e.g. sleep, think, concentrate, seem, please, etc. Moreover, in spite of their denotation, the words action and activity are nouns and not verbs. A variant of this semantic approach to defining word classes is to argue that word classes should be defined in terms of more abstract semantic criteria. Thus for Langacker (1987: 189) word classes are ‘symbolic units’ whose semantics determines the category the elements belong to. For example, a noun is a symbolic unit that semantically instantiates a schema referred to as [THING]. Verbs designate processes, while adjectives and adverbs designate atemporal relations (Langacker 1987: 189; see also Taylor 2002: 341ff). Other linguists stress that the definitions of word classes should make reference to the discourse roles of words. For Hopper and Thompson “the basic categories N and V are to be viewed as universal lexicalizations of the prototypical discourse functions of ‘discourse-manipulable participant’ and ‘reported event’” (1984: 703). To supplement these meaning-based definitions (or even to replace them), we can try to define word classes in terms of their morphosyntactic properties, i.e. by using inflectional and distributional properties. Under this view, nouns are words that can typically be associated with plural and genitive morphology,2 and which can occur in the position of X in the frame ‘determinative-adjectiveX.’ Following this line of thinking, the word cat is a noun because it has a plural form cats and a genitive form cat’s, and because it occurs in a sequence such as a beautiful cat. The word cheerfully is not a noun because it lacks a genitive or a plural, and because the string *the beautiful cheerfully is illicit. Verbs are words that can take tense inflections and that can occur to the immediate right of a modal auxiliary; thus arrive is a verb because it has a past tense form arrived and because it can occur in a string such as he will arrive tonight.3 In English, there are many words that can be assigned to different categories depending on their different syntactic environments. An often cited example is round, which can be a noun (this is your round, John), an adjective (a THOC06 118 1/10/06, 1:48 PM English Word Classes and Phrases 119 round surface), a verb (they rounded the corner), or a preposition (round the clock). (See also note 8.) In the vast majority of cases we can assign words to word classes without much difficulty, but there are words about which linguists disagree as to what might be the best way to classify them. We will discuss a few such problematic cases in the next section. In section 2.3 we deal with the issue of words whose properties would justify simultaneously assigning them to distinct classes. 2.2 Some problematic cases: determinatives Consider first the noun phrases in (2): (2) a politician/the politician The question arises to which word class we should assign words like a and the. At first sight, there are at least three possibilities: • • • a and the are adjectives a and the are articles a and the are determinatives The representation in (1) reveals which analysis we prefer, but we may ask ourselves whether there is any supporting evidence for this preference. Consider first the alternative possibility that the is an adjective. If this were indeed the case, it would be difficult to explain the contrasts shown in (3) and (4). (3) a. loquacious, boring politicians b. politicians are loquacious c. very loquacious politicians (4) a. *the a politician b. *politician is a/the c. *very a/the politician The data in (3) and (4) show that words like a and the are more restricted in their distribution than adjectives: While we can combine (‘stack’) adjectives to the left of the noun, as in (3), we cannot combine a and the, as (4) shows. Also, while adjectives can be positioned to the right of a verb like be (cf. (3)), this is not possible for a and the (cf. (4)). Finally, while adjectives can be preceded by intensifying words like very (cf. (3)), words like a and the cannot (cf. (4)).4 Notice also that while words like loquacious and boring have clear descriptive (or ‘lexical’) meaning, words like a and the do not have such lexical meaning. All they contribute, meaningwise, to the phrases in which they occur is ‘indefiniteness’ or ‘definiteness’ (hence the more specific labels indefinite article and definite article).5 Clearly, then, we have some arguments to assign a and the to a THOC06 119 1/10/06, 1:48 PM 120 Bas Aarts and Liliane Haegeman word class distinct from that of adjectives because their distributional and semantic properties are sufficiently different from those of adjectives. One way to separate a and the from adjectives would be to classify a and the as ‘articles.’ The traditional class of articles is usually taken to comprise just these two words, and no others. This is problematic, however, because there are a number of other words which behave very much like a and the. For instance, this, that, these and those (traditionally called demonstrative pronouns) are distributionally similar to the articles in that they can also immediately precede nouns. Like the, the demonstratives encode that the noun phrases they introduce are definite. (5) a. this/that politician b. these/those politicians Demonstratives differ from the definite article in that they also signal that the referent of the associated noun phrase is proximal (‘nearby’) or distal (‘far away’)6 and, unlike the definite article, the demonstratives have number inflection: this/that are singular in number, while these/those are plural. What is of interest to us, though, is the parallel distribution of a/the and the demonstratives. Given the above considerations, it seems that to accommodate all these words we need a class that is wider than the two-member class of articles. In recent discussions this more comprehensive class has been labeled the class of determinatives (Huddleston 1984; Huddleston and Pullum et al. 2002; see also n. 7). The case of words like a and the is relatively straightforward, and most present-day grammarians would agree that calling such elements adjectives is misguided. However, there are a number of other words, some with quite distinct properties, which have more controversially been claimed to belong to this class of determinatives. Quirk et al. (1985: 253ff) in fact distinguish three sub-classes of determinatives: predeterminatives, central determinatives and postdeterminatives.7 Here are some examples from each of these classes: Predeterminatives: all, both, half, double, such, etc. Central determinatives: a, the, this, that, these, those, my, his, etc. Postdeterminatives: two, three, second, third, last, next, few, many, etc. The three labels aim to reflect the distributional properties of the words belonging to the class. Quirk et al. claim that if there is more than one determinative only the order predeterminative – central determinative – postdeterminative is allowed. What is more, in any one noun phrase there can only be one item from the class of central determinatives and one item from the class of predeterminatives. Multiple postdeterminatives are possible. Thus, for example all the many questions, with one item from each of the determinative classes shown above, is fine, but *all both books with two predeterminatives is not permitted, and neither is *my this book. On the other hand, the last two days, which contains a central determinative and two postdeterminatives, is licit. THOC06 120 1/10/06, 1:48 PM English Word Classes and Phrases 121 While the classification above offers a neat descriptive taxonomy of the determinatives and captures some of their distributional (linear precedence) properties, there are problems with it. For example, why is it that we cannot select more than one element from the predeterminative and central determinative classes, while there is no such restriction in the case of postdeterminatives? And what about problematic examples such as the following: (6) a. many a good book b. these many good books In (6) the word many seems to be a predeterminative as it precedes a central determinative, while in (6) many follows a central determinative, and hence is arguably best classified as a postdeterminative. How do we solve this problem? Let’s consider some more data. Consider (7) and (8) below: (7) very many books (8) many books, more books, most books The fact that an intensifying element can precede many and that many itself has comparative and superlative forms suggests that perhaps many ought to be regarded as an adjective, not as a determinative, because adjectives generally allow intensification and the occurrence of comparative and superlative forms. But then, what about (6)? Surely this example shows that many cannot possibly be an adjective? This objection to classifying many as an adjective would be valid only if adjectives could never occupy the position occupied by many in (6), but this is not the case, as the following example shows: (9) Seldom have I seen so magnificent a palace! On the other hand, the word many is not quite like other adjectives either: in (10a) many is followed by a PP of the books; a similar pattern is not possible with the adjective nice (10b): (10) a. many of the books b. *nice of the students Given its contradictory properties, many has received different analyzes in the literature. Taking (10) as core evidence, Huddleston and Pullum et al. (2002: 539ff) analyze many as a determinative, but obviously this leaves questions as to how to account for its adjectival properties. In a generative framework, Kayne (2002) takes the view that many is adjectival and accounts for its determinative properties by assuming that it moves to a determinative position. Consider next the behavior of such, a similarly contentious word that is regarded by some grammarians as a determinative (cf. Quirk et al. 1985: 257), while others regard it as an adjective, cf. Huddleston et al. (2002: 435) and Spinillo (2003). THOC06 121 1/10/06, 1:48 PM 122 Bas Aarts and Liliane Haegeman (11) such a nice day (12) no such thing (13) the next such event In Quirk et al.’s approach, a word like such would again have to be regarded as anomalous because it can occur in the position occupied by predeterminatives, as well as in the position occupied by postdeterminatives. If we regard such as an adjective this problem does not arise: in (12) and in (13) the word would have the position typical of adjectives, in (11) it could have been fronted to a position to the left of the determinative. A third alternative is proposed in Biber et al. (1999: 280ff), who analyze such as a semi-determiner to reflect its intermediate status between determiners and adjectives. For a recent transformational analysis of such, which appeals to movement to account for its distribution, see Wood (2002). The discussion above does not pretend to be exhaustive and many other similar problems could be raised for the classification above. The discussion only serves to show that it is not always obvious how to classify specific words. 2.3 Word class boundaries and gradience The problematic cases discussed in the previous section raise the more general question whether the boundaries between the word classes can really be sharply delimited. Readers will have noticed that in assigning our problem words to word classes we systematically made an either-or choice. That is to say, we assumed that words like many and such belonged either to the class of determinatives or to the class of adjectives. Although we did conceive of the possibility that in one use a word may belong to one category, and in another use it may belong to another category, crucially, we did not envisage a situation in which in a particular use one word would simultaneously belong to more than one category. We also did not envisage that a word could partially belong to one category and partially to the other. Such a procedure is very much in keeping with a very dominant line of thinking in linguistic categorization that goes back to Aristotle. Aristotle held that as far as membership of categories is concerned a particular element A either belongs to a category α or to a category β, but not to both categories at the same time. In addition, he held that all members of a category are equal members, so that it is not possible to be a member of a category to a certain degree. The main attraction of the Aristotelian approach to categorization is that a grammar that has neatly delimited categories is less ‘messy’ than a grammar that doesn’t, and arguably it is necessary to impose such an abstraction (an ‘idealization’) onto the facts of language in order to be able to even begin to make sense of the often complex and intricate facts of natural languages. THOC06 122 1/10/06, 1:48 PM English Word Classes and Phrases 123 Formal approaches to linguistics (e.g. Noam Chomsky’s theory of language) have adopted a fairly strictly Aristotelian approach to categorization.8 This view was countered by other schools of linguistics whose thinking was influenced by the philosopher Ludwig Wittgenstein. In thinking about the notion ‘game,’ Wittgenstein had noticed that the concept is difficult to define: there are many activities which we would call games, but which are nevertheless quite different. For instance, skipping is a game and it is something you can do by yourself, while football is a game played by two teams. Wittgenstein’s solution to this classificatory problem was to say that all games bear a family resemblance to each other, in the same way that members of a family do. Wittgenstein influenced work in psychology by Eleanor Rosch and her collaborators who did experiments which involved showing subjects a large number of pictures of animals and objects, e.g. birds and chairs (cf. Rosch 1978). The subjects were then asked if a particular picture showed a good or bad example of the animal or object in question. The results revealed that subjects perceived particular instances of animals or objects as more prototypical than others. For example, a sparrow was perceived as a more typical example of a bird than a penguin. This type of work in Prototype Theory influenced cognitive linguists who refused to accept what we might call the categorial straitjacket, and strove to build the concept of prototypes into their theories. In such frameworks, there have been proposals to conceptualize grammatical categories in terms of prototypes. How would this work? One way to do this is to examine the syntactic behavior of a particular word, say a verb, in a given context, to compare it to the behavior of another such word, and to decide on the basis of that comparison which is the more typical verb. For example, if we compare the distributional potential of the word must with that of eat we find that the former cannot occur on its own, and always has to precede a verb (e.g. I must go to London but not *I must to London). Furthermore, must lacks a third person singular ending (*musts) and a past tense form (*musted). The word eat is not constrained in the same way: it can occur without an accompanying verb (e.g. I eat bagels every day), it has a third person ending (e.g. He eats bagels every day), and it has a past tense form (e.g. He ate bagels every day). On the basis of such data we might wish to introduce gradience: we could say that both must and eat are verbs but that eat is a more prototypical verb that must (see also notes 24 and 29). This approach leads to postulating what is called Subsective Gradience: grammatical categories involve a categorial core (the prototypes) as well as a periphery which consists of a number of less prototypical members. Note that if no gradience is allowed, there are two options. One might say that modals such as must are auxiliary verbs which are obligatorily tensed (and hence have a restricted distribution) or, alternatively, one could say that given their particular morphological and distributional properties modals are not verbs at all. Another dimension of gradience, which we will call Intersective Gradience (IG) involves categories resembling each other to varying degrees. The so-called gerund in English is a good example. Consider the examples below: THOC06 123 1/10/06, 1:48 PM 124 Bas Aarts and Liliane Haegeman (14) I’m so tired of [this builder incompetently plastering the walls] (15) [The builder’s incompetent plastering of the walls] was a frustratingly slow process. Both examples contain the word plastering, and in both cases this word has verb-like properties, as well as noun-like properties. In (14) the verbal properties are that plastering ends in -ing, a typical verbal inflection. In addition, this word appears to take a noun phrase as its subject (this builder) and as its complement (the walls), and is modified by a manner adverb (incompetently). In (15) plastering is preceded by a genitival noun phrase (i.e. the builder’s, cf. the builder’s van) and by an adjective phrase (incompetent), and is followed by a prepositional phrase (i.e. of the walls, cf. the color of the walls). These are all properties of nouns. Conversely, in (15) plastering cannot be preceded by an adverb (*incompetently plastering of the walls). In conclusion, it seems that plastering in (14) is more verb-like than plastering in (15). We can now approach these examples in at least three ways. Firstly, we could say that verbs and nouns are on a cline or gradient, such that these word classes shade into each other gradually.9 Another possibility is to say that plastering in these two examples is a hybrid element and belongs to the classes of verb and noun at the same time. This strategy is adopted in cognitive approaches to grammar. It is also proposed in Hudson (2003). Notice that both these strategies would mean abandoning the strict Aristotelian separation of the categories. A third possible strategy would be to retain the sharp boundaries between the verb and noun classes, and say that although plastering in (14) has verbal as well as nominal properties, the verbal ones (for instance taking an NP object and having an adverbial modifier) outweigh the nominal ones, and for that reason plastering is a verb. In (15) the converse situation obtains: here the nominal features (e.g. being modified by a genitival NP and by an adjective phrase) are more numerous than the verbal features, and we therefore conclude that plastering is a noun. We will say that the classes of verbs and nouns converge upon each other, and that this is manifested by the possibility of elements displaying verbal and nominal features at the same time in different proportions.10 3 From Word to Phrase 3.1 Grouping words Having discussed words as units of grammar, we now turn to phrases, which we regard as ‘expansions’ or ‘projections’ of words. Consider sentence (16), which consists of eight words. It is uncontroversial that these words are grouped into strings that form units, both in terms of form and in terms of meaning. For instance, in (16) it is generally agreed that the determinative the and the THOC06 124 1/10/06, 1:48 PM English Word Classes and Phrases 125 noun students form a unit. It is also agreed that the core of this constituent is the noun students, hence the string the students is referred to as a noun phrase.11 In the same vein, the string just recently is labeled an adverb phrase. (16) a. [det The] [N students] [V have] [V completed] [det the] [N assignments] [Adv just] [Adv recently] b. [NP [det The] [N students]] [V have] [V completed] [NP [det the] [N assignments]] [AdvP [Adv just] [Adv recently]]12 The structural grouping of the words in a sentence is represented either by a so-called labeled bracketing or by means of tree diagrams, a format that has been popular since the emergence of generative grammar in the 1960s and which we will turn to presently. Informally, one might define a noun phrase as a unit or a constituent whose most important element is a noun. This definition implies that NPs in fact need not contain more than just a noun: (17) [NP [N Children]] bring [NP [N happiness]] The definition will obviously have to be adapted to include NPs without a nominal head. To mention a few examples, consider the phrases the rich and the poor in (18) and the bracketed constituents in (18) which contain a determinative element but lack a head noun. (18) a. [NP The rich] do not understand [NP the miseries of [NP the poor]] b. The students have chosen their texts. [NP These three] have been selected by [NP many]13 Typically, noun phrases can be replaced by pronouns: in (19), from Quirk et al. (1985: 76), the pronoun he replaces the man and the pronoun her replaces the little Swedish girl. (19) The man invited the little Swedish girl because he liked her. An NP functioning as a predicate may be replaced by so: (20) Mary is [NP an excellent teacher] and so is her sister There also seems to be agreement that the italicized strings in the following sentences are NPs: (21) a. The discovery of the wreck caused consternation b. What we need is a careful examination of all the details c. We need a quick reappraisal of the situation THOC06 125 1/10/06, 1:48 PM 126 Bas Aarts and Liliane Haegeman By analogy with the definition of NPs above, we can say that an adjective phrase (AP) is a constituent whose core element is an adjective. The italicized strings in (22) are APs. (22) a. John is very envious of his sister. b. Mary is afraid of the consequences of this decision. It is possible to substitute the AP by means of so: (23) a. John is [AP very envious of his sister] and so is Bill. b. Mary is [AP worried about the consequences of this decision] and so am I. Prepositional phrases are constituents with a preposition as their core, as illustrated by the bracketed strings in (24): (24) a. Mary is [PP in London]. b. Mary arrived [PP on Tuesday]. And once again, these strings can be replaced, this time by pro-forms like there or then: (25) a. John is there too. b. John arrived then too. In (24a) the PP can also be replaced by so: (26) Mary is [PP in London] and so is John. 3.2 The verb phrase Identifying noun phrases, adjective phrases and prepositional phrases is usually fairly straightforward. We turn now to verb phrases, which require more extensive discussion. Analysing the grouping of words around verbs has led to many sharply different analyzes, two of which we will compare in this section. We will provide arguments for one of these analyzes and against the other.14 3.2.1 The problem: two approaches to the verb phrase In the representation in (16) above, repeated here for the reader’s convenience as (27), the affiliation (if any) of the verbal elements have and completed is left open. In fact, in the literature there is an interesting split in how such units are treated and in how the overall structure of clauses is elaborated. In one line of thinking have and completed are taken to form a constituent (labeled VP, for instance); in another, the string completed their assignments just recently would THOC06 126 1/10/06, 1:48 PM English Word Classes and Phrases 127 be one constituent (VP) of the clause, and the auxiliary is represened as a separate constituent of the clause. The first approach is represented by (27) based on Quirk et al. (1985: 39); the other is represented by (27). In (27b) the label auxiliary is used to signal that the node dominates an element belonging to the class of auxiliaries. In (27c) the label ‘Aux’ is provisionally introduced to signal a specific structural position in the clause which is occupied in our example by the finite auxiliary.15 (27) a. [NP [det The] [N students]] [V have] [V completed] [NP [Det their]] [N assignments]] [AdvP [Adv just] [Adv recently]] b. clause NP VP NP AdvP auxiliary main verb Det N adverb adverb Det N completed their assignments just recently The students have clause c. Aux NP VP V Det N The students have completed NP AdvP Det N adverb adverb their assignments just recently These two analyzes of the verb phrase have consequences for the overall structural relations in clauses. In (27) the subject NP, the VP complement and the VP adjunct are on the same hierarchical level; they are all immediate constituents of the clause. In (27) the subject NP is a privileged constituent of the clause: it is hierarchically more ‘prominent’ in that it is an immediate constituent of the clause, while the complement of the verb, the NP their assignments, is an immediate constituent of VP, itself an immediate constituent of the clause.16 In the next sections we show that structure (27) is preferable to structure (27). We will see that a closer look at some data reveals that postulating a VP along the lines of (27) is in conflict with the assumptions about structure elaborated in section 3.1. Representation (27) is similar to those adopted in earlier transformational approaches (Chomsky 1957, 1955/1975), while representations along the lines of (27) have been adopted in more recent versions of generative syntax. THOC06 127 1/10/06, 1:48 PM 128 Bas Aarts and Liliane Haegeman Interestingly, the two major comprehensive grammars of English also differ to some extent in terms of the structure they adopt, with Quirk et al. endorsing an approach along the lines of (27) and Huddleston and Pullum et al. adopting a variant of (27). For a more general discussion of the different status of the two representations see also Leech (2004). In fact, Quirk et al. (1985: 79) seem to assume something like the structure in (27) – in addition to (27) – when they introduce the category of ‘predicate,’ and provide a structure as in (28): Sentence (28) Independent clause Subject Predicate He Auxiliary + Operator17 Predication had given the girl an apple. Quirk et al. (1985: 90) discuss the co-existence of the two representations. They say: There are occasions, however, when such alternative analyzes seem to be needed, on the grounds that some of the generalizations that have to be made require one analysis, and some require another. It is for this reason that we have presented, in this chapter, two ways of analysing a clause: one analysis in terms of the elements S,V,O,C, and A,18 and the other in terms of subject and predicate, the predicate being subdivided into operator and predication. Given that their grammar remains relatively informal, these authors do not spell out in detail how the two analyzes are formally related, or which of the two is more basic. But see also Leech (2004). See also section 5 in which we actually integrate (27c) into a representation like that in (28). 3.2.2 The relation of the complement to the verb Consider again the examples of the uncontroversial phrases discussed in section 3.1. The italicized strings in (29), (29b), (29c) illustrate NPs, those in (29), (29e) APs, and those in (29), (29g) PPs: (29) a. b. c. d. THOC06 The discovery of the wreck caused consternation. What we need is a careful examination of all the details. We need a quick reappraisal of the situation. John is very envious of his sister. 128 1/10/06, 1:48 PM English Word Classes and Phrases 129 e. Mary is afraid of the consequences of this decision. f. This chapter is about categories and structure. g. Mary arrived on Tuesday. In each of these examples the complement of the head of the construction is taken to be part of the phrase. Thus, for instance, the complement of the N discovery is the string of the wreck, which is taken to be part of the NP. This is corroborated by the fact that the string the discovery of the wreck can be replaced by the pronoun it. Similarly, of her sister, the complement of the adjective envious in (29), is taken to be part of the AP, etc. With respect to NP and AP, the discussion in Quirk et al. (1985: 62ff) is fully compatible with such an analysis. When we turn to VPs, though, things are different. According to the approach in (27) the complement of the verb, whether it is a predicate, a direct object NP or a subcategorized PP, is not part of the VP. Rather, the verb and the auxiliaries form a constituent separate from the verb’s complement and from its adjunct. Quirk et al. (1985: 39) and many others use the label VP for this sequence of one or more auxiliaries and the lexical verb taken together; others use a different label but the implications for the structure are similar.19 3.2.3 Medial adjuncts One consequence of assuming that the verb phrase consists of just auxiliaries and the main verb is that very often the VP will have to be taken to be discontinuous. In the attested examples in (30) we find non-verbal material intervening between the auxiliaries and the verb. This has very much repeatedly been the story of staphyloccocus aureus. (The Guardian, 12.7.02, p. 6, col. 7) b. The result is a hobbled place, where working for public services can only with difficulty make you proud. (The Guardian, 1.29.03, p. 8, col. 6) c. The former Treasury minister, Geoffrey Robinson, was last night publicly upbraided for “self-indulgence” and playing “personality politics” . . . (The Guardian, 10.16.00, p. 2, col. 1) (30) a. We either have to say that the VP in such examples contains verbal elements, as well as any intervening (non-verbal) adverb phrase(s), PP(s) and NP(s), or else we have to say that the VP is discontinuous and that the italicized segments are somehow ‘outside’ the VP. Observe that the assumption implicit in the traditional literature is that constituents such as NP, PP etc. are not routinely discontinuous. Discontinuous NPs, for instance, are usually accounted for in terms of extraposition. Suppose we assume that VPs conceived of as in (27) are not normally discontinuous and that therefore the italicized adjuncts in (30)–(30c) are part of the VP. If this is true then, according to (27b), the manner adjunct very carefully will be part of the VP in (31) but not in (31), which is surprising, to say the least. THOC06 129 1/10/06, 1:48 PM 130 Bas Aarts and Liliane Haegeman (31) a. Jack will examine the evidence very carefully b. Jack will very carefully examine the evidence 3.2.4 Substitution In the literature there seems to be a consensus that proforms typically replace constituents, even though this assumption is not always made fully explicit. Thus, for instance NPs can be replaced by pronouns (see Quirk et al. 1985: 76), while predicative NPs, APs, and PPs can be replaced by so, as we have seen. Let us explicitly adopt the assumption that a proform replaces a constituent, which may be a word or a phrase. This assumption will lead us to the conclusion that the verb and its complement 20 must be a constituent. This is shown by the examples in (32): (32) a. b. c. d. John has left the office, and so has Mary. The evenings have turned very cold, and so have the mornings. John has left for another job, and so has Mary. John has passed the new information to the police, and so has Bill. In each of the above examples so substitutes for the verb and its complement. If substitution is structure-dependent, then the substitution data above are clearly much more compatible with the structure in (27) than with that in (27). These data also suggest that the inflected auxiliary is not included in the VP. The same conclusion also seems to follow from the following observation in Quirk et al. (1985: 76): “But so has a more important function in modern usage, namely to substitute – along with the ‘pro-verb’ do – for a main verb and whatever follows it in the clause” (our italics). The following attested examples illustrate how a verb + its complements (italicized here) can be replaced by do (see Miller 2002). [Linley] said: Why do you keep the cellar door locked? Have you always done’? (Elizabeth George, Missing Joseph, Bantam Books, 1993, p. 272) b. If I had wanted to hurt someone, believe me, I would have done. (Elizabeth George, Missing Joseph, Bantam Books, 1993, p. 172) c. If Sir Alex wants to sign somebody he can do. (The Guardian, 12.31.02, p. 14, col. 1) d. There was page upon page of tribute to “The Man who saved the mirror,” some of it from people who should have known better, and indeed had done a few years earlier. (The Guardian, G2, 11.5.01, p. 2, col. 3) (33) a. Again, if substitution is structure-dependent, then these data conflict with (27), which treats the VP as a string of auxiliaries + a lexical verb. Once again, the tensed auxiliary is not affected by the substitution process. THOC06 130 1/10/06, 1:48 PM English Word Classes and Phrases 131 3.2.5 Movement It is generally assumed that constituents have a canonical position in the clause, and that they may be moved from that position for particular communicative effects. For instance, in (34), (34b), (34c) an NP is fronted, in (34) an AP is fronted: Everything that doesn’t sell we give to Goodwill. (The Guardian, 1.3.03, p. 5, col. 1) b. The news, when it comes, he seems to take well enough. (The Guardian, G2, 7.26.02, p. 2, col. 1) c. A lot of the elements that surround you in the job, you sometimes think are just a vast conspiracy to divorce you from ordinary life. (The Guardian, 4.26.02, G2, p. 6 col. 4) d. Our dustmen arrive too early for me to check, but our fishmonger and his staff in Petersfield all wear ties (Letters, October 22) and very smart they look too. (Letters to the Editor, The Guardian, 10.23.02, p. 9, col. 5) (34) a. We assume that fronting a constituent is structure-dependent. (35) shows that the verb is fronted with its complement, whereas simply fronting a verb without its complement is not possible. Again this is unexpected under (27) but it follows naturally from (27). “But I couldn’t rewind time, I just had to get over it.” And get over it, she did. (The Guardian, 9.6.01, p. 15, col. 8) b. *And get, she did over it. (35) a. Consider also the following sentences from which we can draw the same conclusion: Pete says he will call his bank manager, and call his bank manager he will –. b. *Pete says he will call his bank manager, and will call his bank manager he –. (36) a. (37) a. Clear their debts though they must –, this isn’t going to be easy for them. b. *Must clear their debts though they –, this isn’t going to be easy for them. In (35), (36), and (37) verb + complement combinations are fronted, while the dummy auxiliary do in (35) and the modal auxiliaries will and must in (36) and (37) stay behind. See Aarts (2001) for discussion. Patterns referred to as ‘predicate inversion’ and illustrated by the attested example in (38) also offer support for (27). Here again, the lexical verb is fronted with its complement, leading to inversion of be around the subject. It is THOC06 131 1/10/06, 1:48 PM 132 Bas Aarts and Liliane Haegeman not clear how such patterns could be derived by movement on the basis of the structure in (27). (38) Competing with him are Jack Nicholson, who would set a record of four Oscars if he won for his portrayal of a retired widower in About Schmidt, Daniel Day-Lewis, who plays a ferocious, knife-wielding butcher in Gangs of New York; Nicolas Cage in Adaptation, and Adrien Brody, of The Pianist, the only one of the five not nominated previously. (The Guardian, 2.12.03. p. 5, col. 2) 3.2.6 Coordination Once constituents are formed they may be coordinated. We reproduce the following extract from Quirk et al. (1985: 46): [T]wo or more units of the same status on the grammatical hierarchy may constitute a single unit of the same kind. This type of construction is termed coordination, and, like subordination, is typically signalled by a link-word termed a conjunction: in this case a coordinating conjunction. The most common coordinating conjunctions are and, or and but. Coordination of clauses a. [[S It was Christmas Day] and [S the snow lay thick on the ground]] Coordination of prepositional phrases b. You can go [[PP by air] or [PP by rail]] Coordination of nouns c. His [[N son] and [N daughter]] live in Buenos Aires When we turn to coordinations involving verbs, it becomes clear that the coordinated segments containing a verb correspond more to the VP as represented in structure (27) than to the VP as represented in (27). Quirk et al. (1985: 949) give (39), while (39) is attested. Observe that in both these examples the complements of the verbs (and some adjuncts in (39b)) participate in the coordination. (39) a. You must take the course and pass the examination. b. Word spreads rapidly through a telephone tree, she said, which has galvanized activists in the West Yorkshire valley and already filled six Calderdale buses for next Saturday’s London demonstration. (The Guardian, 2.8.03, p. 4, col. 4) If coordination implies the linking of two constituents, then the data in (39) again tend to favor representation (27).21 From the discussion above we tentatively conclude that a structure like that in (27b), in which a VP does not include the complement(s) of the verb, is not THOC06 132 1/10/06, 1:48 PM English Word Classes and Phrases 133 compatible with a conception in which constituents are units of structure and units of sense. We therefore adopt the structure in (27). In the next section we elaborate the structure of clauses in terms of such a view of the VP. 4 Clause Structure Before we can integrate the type of VP we postulate here (cf. (27)) into the representation of the complete clause, we need to address two points. What happens when there is more than one auxiliary in a clause? What happens when there is no auxiliary at all? 4.1 Stacked auxiliaries Consider the following example:22 (40) This student might have been writing a letter On the basis of so-substitution in (41) and coordination in (42) we conclude that the string writing a letter is a constituent, a VP, as shown in (43): (41) Mary thinks this student might have been [writing a letter], and so he might have been (42) This student might have been [writing a letter] or [watching TV] (43) This student might have been [VP writing a letter] The question arises how to deal with the sequence of auxiliaries might have been. Morphologically and distributionally, the aspectual auxiliaries have and be share properties of verbs: they can be finite or nonfinite, and when finite they may show agreement morphology: (44) a. He has/had been writing a letter b. Having been writing letters all day . . . (45) a. He is/was writing a letter b. To be writing letters all day would be terrible If have and be are verbs, then they should be able to head verb phrases. We will say that unlike lexical verbs the aspectual auxiliaries select a VP as their complement. The examples in (46) provide evidence that the string been writing a letter in (40) is a constituent: in (46) so substitutes for been writing a letter and in (46) the string been writing a letter is coordinated with the string been watching TV.23 In THOC06 133 1/10/06, 1:48 PM 134 Bas Aarts and Liliane Haegeman the attested (46), (46d), the second of the coordinated VPs includes a non-finite auxiliary. In (47), so substitution and coordination show that the string have been writing a letter is also a constituent. Mary thinks the student might have been writing a letter, and so he might have b. The student might have been writing a letter or been watching TV c. He had claimed asylum in 1998 and been refused in 2001. (The Guardian, 1.16.03, p. 1, col. 4) d. Determining precisely how much money has made it to New York and actually been distributed is difficult. (The New York Times, 12.30.02, p. B4, col. 1) (46) a. (47) a. Mary thinks the student might have been writing a letter and so he might b. The student might have been writing a letter or have been watching TV Data such as those in (46)–(47) show that while the verb, its complement(s) and adjuncts form a constituent, the finite auxiliary can remain outside the VP. (But see also section 5.) Observe that modals remain in situ when verb+ complement combinations are displaced. The modal auxiliaries are inflected for tense; they are formally always either present or past. On the basis of these observations, we propose the provisional structure in (48): clause (48) NP Aux VP V VP V The student might have been VP V writing NP a letter The core VP is writing a letter, which expresses the kind of event denoted by the clause. The merger of the core VP with the auxiliary been creates another VP and adds progressive aspect to the event; the merger of have with the resulting VP adds perfectivity. The stacked structure in (48) manages both to express constituency relations and to encode the scopal relations of the auxiliaries. For clauses with one or more aspectual auxiliaries, but without a modal auxiliary, such as (49), we maintain the structure in (48) as a starting point, but in addition we assume that in such cases the finite aspectual auxiliary, which, as shown above, originates as the head of a VP, moves into the auxiliary slot THOC06 134 1/10/06, 1:48 PM English Word Classes and Phrases 135 (see Emonds 1970, 1976, 1987; Pollock 1989, 1997; Haegeman and Guéron 1999; Aarts 2001). We will account for this movement in section 4.2.24 (49) a. The student has been writing a letter. b. clause NP Aux VP V VP V The student – has been VP V writing NP a letter Infinitival clauses such as the bracketed constituent in (50) can be analyzed with to occupying the position ‘Aux,’ as in (50): (50) a. I expect [my students to have been writing numerous protest letters] clause b. NP Aux VP V VP V My student to have VP V NP been writing numerous protest letters In the representations above, clauses systematically contain three basic constituents: a subject, an Aux position (containing an auxiliary or the infinitive marker to) and a VP. This constituency has an intuitive semantic appeal to it: a clause could be seen as the application of a particular event/state of affairs to a referent, and the element occupying the Aux position serves to qualify the linking in terms of time, probability, etc.25 The representation singles out the subject as the most prominent NP in the clause: the subject is an immediate constituent of the clause. This is a positive result since we know that all finite THOC06 135 1/10/06, 1:48 PM 136 Bas Aarts and Liliane Haegeman clauses have subjects, even when the subject lacks semantic content,26 in which case impersonal it and there are inserted. In addition, the structural prominence of the subject can be related to a number of properties which single it out, for instance the fact that the subject is the most accessible to syntactic processes such as relativization (cf. Keenan and Comrie 1977), and the observation that subjects are often privileged antecedents for reflexives and anaphoric pronouns (Halmari 1994). 4.2 Clauses without auxiliaries The question arises what happens if a clause does not contain any auxiliaries. One might propose that in the absence of auxiliaries a clause such as (51) consists simply of a subject NP and a VP, as represented in (51b). (51) a. The student wrote a letter b. clause NP The student VP wrote a letter With respect to the informal semantics outlined above this is unattractive. In structure (48) there are three major components: (i) a predicate (the VP) as applied to the (ii) subject (NP), and (iii) the linking element in the position labeled ‘Aux.’ The element in the Aux position qualifies the subject-VP link in terms of modality or time. In (51b) there is no longer a linking position available. If we consider how (51a) behaves with respect to the various diagnostics for structure applied in the preceding sections, it is also not clear that the tense morpheme of the verb should be an integral part of the VP. Observe, for instance, that if we replace the VP by so, then the tense morpheme is stranded and realized on the auxiliary do: (52) The student wrote a letter and so did the professor. Similarly, if we front the VP, then we do not actually move the tense morpheme of the verb along, as shown by example (35) a repeated here as (53): (53) “But I couldn’t rewind time, I just had to get over it.” And get over it, she did. (The Guardian, 9.6.01, p. 15, col. 8) Furthermore, in negative clauses without aspectual or modal auxiliaries the tense of a lexical verb is not realized on the verb itself but it is realized separately on do: (54) The students did not write any letters THOC06 136 1/10/06, 1:48 PM English Word Classes and Phrases 137 These data suggest that the tense morpheme should have some independence with respect to the VP. When there is no aspectual or modal auxiliary in the clause, tense serves to link the subject and the predicate and locates that link in time. In clauses without auxiliaries, we will separate the tense structurally from the VP and locate it in the position previously labeled Aux. By adopting this analysis, we can generalize the ternary structure elaborated above and assume that all clauses consist of a predicate as applied to a subject, and that the link is encoded in a particular position, and that it can be qualified by a separate unit, realized by an auxiliary, by to or by the tense morpheme. The auxiliaries that were shown to occupy the linking position (originally labeled Aux) are inflected for tense. We can postulate that the crucial feature of this linking position is its inflectional nature, and we will relabel the position Aux as ‘I’ for ‘inflection.’ ‘I’ is an abstract functional head, which carries inflectional and agreement features, and hosts (modal) auxiliaries in finite clauses, as well as the element to in non-finite clauses.27 We represent (51) as in (55): clause (55) NP I VP 1 ±tense, ±agreement1 2 to 2 The student 3 modal 3 write a letter As discussed in section 4.1, a finite aspectual auxiliary is inserted as the head of a VP and moves up to the position ‘I,’ previously labeled ‘Aux’ (see (49b)). We can make sense of this movement now: the aspectual auxiliary moves up to ‘I’ in order to pick up its finite inflection in ‘I.’ In (56), a more accurate representation of (49), have moves to ‘I,’ and picks up the third person singular inflection, resulting in has: (56) clause NP I VP V VP V The student THOC06 137 have+3sg = has — been VP V writing 1/10/06, 1:48 PM NP a letter 138 Bas Aarts and Liliane Haegeman In case there is only a tense morpheme in the clause, this is either affixed to the verb,28 or it is spelt out by means of the auxiliary do. The latter arises in negative or interrogative clauses.29 One context in which the tense morpheme in the ‘I’-node in (55) is not affixed to V concerns clauses with so called emphatic do, when the actual validation of the link between subject and predicate is focused on: (57) contains some such examples: (57) a. The student did write the letter. b. I’m probably more benevolent towards Mr Livingstone than a lot of people and I actually do think he’s very brave in trying congestion charging. (The Guardian, 1.3.03, p. 3, col. 4) c. People close to Senate leader Tom Daschle say he should be considered a possible candidate, but many Democrats say they would be surprised if he does run. (Atlanta Journal Constitution, 12.1.02, p. A6, col. 5) 5 Rethinking the Structure of the Clause Structure (27c) displays ternary branching. We have consistently used such ternary branching structures in this chapter. However, there is an intuition that sentences are essentially organized on a binary scheme in that a subject combines with a predicate. This intuition is reflected in representation (28), which we reproduced from Quirk et al.(1985: 79). The two proposals can be combined into one structure. A potential counterexample to the ternary branching structure in (27c), and evidence for a binary branching structure along the lines of (28) is the following type of example:30 (58) The Smiths will have arrived and should have read their mail. We could address this point in two ways. One option would be to posit an ellipted subject (coreferential with the Smiths) before should: (59) [coordination [clause The Smiths will have arrived] and [clause Ø should have read their mail]] This isn’t entirely satisfactory, however, as we might then also posit ellipsis in cases like (39). Moreover, an ellipsis analysis becomes harder to maintain in view of data such as (60). The ellipted constituent in representation (60) could not be said to be ‘coreferential’ with no one, since no one does not refer to a particular entity. THOC06 138 1/10/06, 1:48 PM English Word Classes and Phrases 139 (60) a. No one could understand it or would take the trouble to read it. b. [coordination [clause No one could understand it ] or [clause Ø would take the trouble to read it.]]31 Alternatively, pursuing developments in generative grammar (see Haegeman 1997), we could adapt our structure (27) in the spirit of the binary branching format of (28), using a particular formalism in generative grammar. (61) IP NP I′ I VP V VP completed the assignments just recently The students have – According to (61), a clause is a projection of I, or an ‘Inflection Phrase’ (IP). I′ (‘I-bar’) is a constituent consisting of the inflection node I and the VP. I′ corresponds to Quirk et al’s ‘Predicate’ in (28). The subject NP combines with I′ to form IP. Under this hypothesis, (58) is derived by coordinating two constituents of the type I, each consisting of the modal in I and the VP: (62) [IP The Smiths [coordination [I′ will have arrived] and [I′ should have read their mail]] For more details on the implementation of this type of structure, the interested reader is referred to the literature, see e.g. Kayne (1984) and Haegeman and Guéron (1999) for an application to English. 6 Conclusion In this chapter we discussed the issue of how to classify words into categories (‘word classes’), and how words are combined into larger units (phrases). We discussed a number of problems that arise with respect to classifying words into categories. In particular, we have raised the possibility of gradience in categorizing words. THOC06 139 1/10/06, 1:48 PM 140 Bas Aarts and Liliane Haegeman In the discussion of phrases, one phrase type, the VP, was singled out. We argued for a conception of the VP as containing a lexical verb together with any complement(s) and adjunct(s). This account is shown to be preferable to one in which the VP contains merely auxiliaries (if present) and the main verb. We also propose that each clause contains a specific position, labeled ‘I,’ which hosts inflectional properties. In the final section of the chapter we show how the proposed structure can accommodate the traditional conception of sentences in terms of a combination of a subject and a predicate. NOTES Our thanks are due to Peter Collins and Rodney Huddleston for reading an earlier draft of this chapter. 1 Following the generative tradition, a phrase headed by a subordinating conjunction could be argued to be a clause (see Haegeman and Guéron 1999: ch. 10). 2 Obviously plural endings are restricted to countable nouns. 3 In general terms, morphosyntactic definitions are valid crosslinguistically, but the specific inflectional or distributional properties will be determined by the language in question. 4 Observe that not all adjectives have all the properties listed here: some cannot precede nouns (*an afraid cat), others cannot function as predicates (*the point is main; cf. the main point). Non-gradable adjectives cannot be modified by degree words (*a very nuclear war), but while adjectives will have at least a subset of the properties, the articles do not have any of them. 5 For some discussion of the semantics of the articles see, among others, Hawkins (1978) and Lyons (1999). 6 See also Cornish (2001) and the references in n. 5 above. 7 In fact Quirk et al. (1985) use the labels predeterminer, central determiner THOC06 140 and postdeterminer. They use the label determiner as a form label and determinative as a function label. In this chapter we follow Huddleston and Pullum (2002) in using determinative as a form label, and determiner as a function label. Quirk et al.’s labels have been adjusted in accordance with this practice. 8 It should be noted that in recent years there have been various attempt to elaborate a more refined conception of categorization. One approach tries to deal with what seem to be intermediate categories. See for instance Biber et al. (1999), Corver and Van Riemsdijk (2001), and Aarts (2003, 2004) for discussion. The approach referred to as Distributed Morphology proposes that categories such as nouns or verbs are not specified in the lexicon. Rather, categorially underspecified roots such as round are inserted in different positions in the structure and these positions will determine a particular nominal or verbal behavior. We will not elaborate on this approach here (for discussion see Halle and Marantz 1993). 9 On clines and gradients see Bolinger (1961), Halliday (1961) and Quirk et al. (1985). 1/10/06, 1:48 PM English Word Classes and Phrases 141 10 For further details of this approach, see Aarts (2003, 2004, forthcoming). For a selection of papers on linguistic indeterminacy, see Aarts et al. (2004). For a generative approach to nominalization see Fu, Roeper and Borer (2001) and the literature cited there. See also n. 14. 11 Since Abney (1987) it has been assumed in generative approaches that the head of the noun phrase is in fact the determinative (the DP-hypothesis). (i) [DP [ det The] [NP youngest children]] bought [DP [det a] [NP book of fairy tales]] We refer to the literature for that discussion. For an introduction see Aarts (2001) and Haegeman and Guéron (1999). 12 The bracketing in (16b) is incomplete as we have not indicated any VP. We return to this point in section 3.2. 13 As Peter Collins (p.c.) points out, the examples in (18) are subtly different, in that in (18a) the rich and the poor do not require an understood head to be available in the context. A question arises whether the NPs in (17) have a zero determinative: (i) [NP [ det Ø] [N Children]] bring [NP [det Ø] [N happiness]] Similarly, one might think of postulating a zero noun in (18): [NP The rich Ø] do not understand [NP the miseries of [NP the poor Ø]]. b. [NP These three Ø] have been selected by [NP many Ø] (ii) a. We won’t pursue these issues here, as it would lead us too far astray. THOC06 14 141 Gerunds pose an additional problem for the labeling of phrases. To accommodate the nominal and verbal properties of phrases whose head is a gerund, it has been proposed that in such cases the head of a phrase may, as a marked option, be of a different category from that of the phrase itself. Pullum (1991), for instance, argues (against Abney 1987) that the head of a gerund in English may be of the category V, while the containing phrase may be nominal. This ‘hybrid’ status of the projection would account for the fact that the internal structure of the gerund in (i) is clausal, with a verb taking a nominal complement ( pieces of paper) and being associated with adverbial modifiers such as often, while its external distribution is like that of an NP. (i) [John often throwing pieces of paper during class] bothered the teacher. 15 As will become clearer later, the position labeled Aux hosts finite auxiliaries, the finite form of the copula be, the infinitive marker to, and the finite inflection of the verb. See section 4.2. 16 Observe that an alternative representation could be one in which auxiliary and verb form a constituent which is the head of the predicate, which also contains complement(s) and adjunct(s). (i) [S The students [predicate [have completed] their exams just recently]] As will become clear below, the arguments in favor of (27c) suggest that the finite auxiliary should be separated from VP. This is not compatible with (i). 1/10/06, 1:48 PM 142 Bas Aarts and Liliane Haegeman 17 18 19 20 21 The class of operators includes all the auxiliary verbs, but in e.g. Is John here? and Have you any idea how old he is? both is and have are also operators. This would correspond to (27b). Bache and Davidsen-Nielsen (1997: 38) use the term ‘predicator,’ for instance. And indeed at least some adjuncts. A stacked structure internal to the VP will allow the distinction between complements and adjuncts to be made. For reasons of space we cannot go into this here. Rodney Huddleston notes (p.c.) that data like (i) could be argued to favor the analysis in (27): (i) I [have read] and [may recommend] Kim’s new textbook. However, this example involves a process that is called Right Node Raising, such that the verb read shares its (right-raised) direct object with recommend: (ii) I [have read – i ] and [may recommend – i ] [Kim’s new textbook]i 22 23 kind of left-peripheral ellipsis in the clause. They do not make the structural basis for this claim explicit, but note that by simply assuming coordination of VPs we can generate the patterns in (i) and (ii) without an additional appeal to ellipsis. See also section 5 on ellipsis and coordination. 24 It is not clear whether we should propose that like aspectual auxiliaries, English modals are inserted under a node V and move to Aux. The rationale for the analysis of aspectual auxiliaries in (49) is that these auxiliaries may also appear in nonfinite forms, in which case they follow a modal or another auxiliary. But modals themselves are always tensed, and they lack nonfinite forms. See also n. 29. 25 Interestingly, despite their conception of the verb phrase as in (27)b, Quirk et al. (1985: 121) also suggest an analysis of sentences containing sequences of auxiliaries which is very similar to (50), one in which each auxiliary selects a predication consisting of the next auxiliary combined with another predication. Thus the sentence He might have been being questioned by the police is analyzed as in (i): See Huddleston and Pullum et al. (2002: 1343ff) who call this phenomenon Delayed Right Constituent Coordination. The data are based on Radford (1988: 162–4) Quirk et al. (1985: 949) provide the examples in (i) and (ii): Most people will have read the book or have seen the film (ii) Most people will have read the book or seen the film (i) (i) They seem to suggest that these coordinations result from some THOC06 142 26 [S [Subject He] [Predicate might [Pred.1 have [Pred.2 been [Pred.3 being [Pred.4 questioned by the police]]]]] Using the binary branching format discussed in section 5, (i) can straightforwardly be made compatible with the hypothesis concerning VP structure that we endorse. “A subject is obligatory in finite clauses except in imperative clauses, where it is normally absent but implied” (Quirk et al. 1985: 725). 1/10/06, 1:48 PM English Word Classes and Phrases 143 27 For reasons of space we cannot elaborate the proposed structure in more detail. See Haegeman and Guéron (1999) and Aarts (2001) for further discussion. For more technical discussion in terms of the generative framework see Pollock (1989) and (1997). In the generative literature it has been proposed that the ternary structure be reinterpreted in terms of binary branching (Kayne 1984). For an introduction see Haegeman and Guéron (1999). See also section 5. 28 In the earlier generative literature this process was referred to as ‘Affix hopping.’ See Haegeman and Guéron (1999) for further discussion of why the inflection moves onto lexical verbs. 29 With respect to the status of modals discussed in n. 24 above, we might say that they are verbs/auxiliaries, and that they differ from aspctual auxiliaries in that they are necessarily finite. 30 As pointed out by Peter Collins(p.c.) 31 Thanks to Rodney Huddleston (p.c.) for the example. FURTHER READING Aarts, Bas and Meyer, Charles F. (1995) The verb in contemporary English: theory and description. Cambridge: Cambridge University Press. Haegeman, Liliane (1994) Introduction to government and binding theory, 2nd edn. Oxford: Blackwell Publishers. Haegeman, Liliane (1997) Elements of grammar. Dordrecht: Kluwer. Lakoff, George (1987a) Women, fire and dangerous things: what categories reveal about the mind. Chicago and London: University of Chicago Press. (Chs. 1 and 2 are reprinted in Aarts et al. 2004.) Langacker, Ronald W. (1987) Foundations of cognitive grammar, vol. 1: Theoretical prerequisites. Stanford: Stanford University Press. Langacker, Ronald W. (1991) Foundations of cognitive grammar, vol. 2: Descriptive application. Stanford: Stanford University Press. Newmeyer, Frederick J. (1998) Language form and language function. Cambridge, MA: MIT Press. Taylor, John (2004) Linguistic categorization, 3rd edn. Oxford: Oxford University Press. REFERENCES Aarts, Bas (2001) English syntax and argumentation, 2nd edn. Basingstoke: Palgrave Macmillan. Aarts, Bas (2003) Conceptions of gradience in the history of linguistics. Language Sciences 26, 343–89. Aarts, Bas (2004) Modelling linguistic gradience. Studies in Language 28 (1), 1– 49. THOC06 143 Aarts, Bas (forthcoming) Syntactic gradience. Aarts, Bas, Denison, David, Keizer, Evelien, and Popova, Gergana (eds.) (2004) Fuzzy grammar: a reader. Oxford: Oxford University Press. Abney, Steven (1987) The English noun phrase in its sentential aspect. Ph.D. dissertation MIT. 1/10/06, 1:48 PM 144 Bas Aarts and Liliane Haegeman Bache, Carl and Nielsen, Niels Davidsen (1997) Mastering English. Berlin and New York: Mouton de Gruyter. Biber, Douglas, Johansson, Stig, Leech, Geoffrey, Conrad, Susan, and Finegan, Edward (1999) Longman grammar of spoken and written English. Harlow: Longman. Bolinger, Dwight L. (1961) Generality, gradience and the all-or-none. The Hague: Mouton. (Chs. 1 and 2 are reprinted in Aarts et al. 2004.) Chomsky, Noam (1955/1975). The logical structure of linguistic theory. New York: Plenum. Chomsky, Noam (1957) Syntactic structures. The Hague: Mouton. Cornish, F. (2001) Modal that as determiner and pronoun: the primacy of the cognitive-interactive dimension. English Language and Linguistics 5, 297–317. Corver, Norbert and Henk van Riemsdijk (eds.) (2001) Semi-lexical categories: the function of content words and the content of function words. Berlin: Mouton de Gruyter. Emonds, Joseph (1970) Root and structure preserving transformations. Cambridge, MA: MIT dissertation, distributed by Indiana University Linguistics Club. Emonds, Joseph (1976) A transformational approach to English syntax. New York: Academic Press. Emonds, Joseph (1978) The verbal complex V′-V in French, Linguistic Inquiry 9, 151–75. Fu, Jingqi, Thomas Roeper, and Hagit Borer (2001) The VP within process nominals: evidence from adverbs and the VP anaphor do so. Natural Language and Linguistic Theory 19, 549– 82. Haegeman, Liliane and Jacqueline Guéron (1999) English grammar: a generative perspective. Oxford: Blackwell. Halmari, Helena (1994) On accessibility and coreference. Nordic Journal of Linguistics 17, 35–60. THOC06 144 Hawkins, John A. (1978) Definiteness and indefiniteness: a study in reference and grammaticality prediction. London: Croom Helm. Halle, Morris and Marantz, Alec (1993) Distributed morphology and the pieces of inflection. In: Jay Keyser and Ken Hale (eds.), The view from building 20. Cambridge, MA: MIT Press. Halliday, M. A. K. (1961) Categories of the theory of grammar. Word 17, 241–92. (Reprinted in Halliday 2002: 37–94.) Halliday, M. A. K. (2002) On grammar. Volume 1 of the collected works of M. A. K. Halliday, ed. Jonathan Webster. London and New York: Continuum. Hopper, Paul J. and Thompson, Sandra A. (1984) The discourse basis for lexical categories in Universal Grammar, Language 60, 703–52. (Reprinted in Aarts et al. 2004.) Huddleston, Rodney (1984) An introduction to the grammar of English. Cambridge: Cambridge University Press. Huddleston, Rodney and Pullum, Geoffrey K. et al. (2002) The Cambridge Grammar of the English Language. Cambridge: Cambridge University Press. Hudson, Richard (2003) Gerunds without phrase structure. Natural Language and Linguistic Theory 21 (3), 579–615. Kayne, Richard (1984) Connectedness and binary branching. Dordrecht: Foris. Kayne, Richard (2002) On the syntax of quantity in English. Ms New York University Keenan, Ed and Comrie, Bernard (1977) Noun phrase accessibility and universal grammar. Linguistic Inquiry, 8, 63–99. Langacker, Ronald W. (1987) Foundations of cognitive grammar, vol. 1: Theoretical prerequisites. Stanford: Stanford University Press. 1/10/06, 1:48 PM English Word Classes and Phrases 145 Leech, Geoffrey. (2004) A new Gray’s Anatomy of English grammar. English Language and Linguistics 8, 121–47. Lyons, Christopher (1999) Definiteness. Cambridge: Cambridge University Press. Miller, Philip (2002) Les emplois non finis de do auxiliaire. In Claude Delmas (ed.), Construire et reconstruire en linguistique anglaise: syntaxe et sémantique. CIEREC: Travaux 107. Publications de l’Université de SaintEtienne, 185–98. Pollock, Jean-Yves (1989) Verb movement, UG and the structure of IP. Linguistic Inquiry 20, 365–425. Pollock, Jean-Yves (1997) Notes on clause structure. In Liliane Haegeman (ed.), Elements of grammar. Dordrecht: Kluwer, 237– 80. Pullum, Geoffrey K. (1991) English nominal gerund phrases as noun THOC06 145 phrases with verb phrase heads. Linguistics 29, 763–99. Quirk, Randolph, Greenbaum, Sydney, Leech, Geoffrey, and Svartvik, Jan (1985) A comprehensive grammar of the English language. London: Longman. Radford, Andrew (1988) Transformational grammar. Cambridge: Cambridge University Press. Rosch Eleanor (1978) Principles of categorization. In Eleanor Rosch and Barbara B. Lloyd (eds.), Cognition and categorization. Hillsdale, NJ: Lawrence Erlbaum, 27–48. (Reprinted in Aarts et al. 2004.) Spinillo, Mariangela (2003) On such. English Language and Linguistics 7 (2), 195–210. Taylor, John R. (2002) Cognitive grammar. Oxford: Oxford University Press. Wood, J. L. (2002) Much about such. Studia Linguistica 56, 91–116. 1/10/06, 1:48 PM