Bent Jacobsen The Origin and Rationale of X-bar Syntax

The present paper is intended as a reasonably elementary introduction to the nature of Xbar syntax, an important module in the structure of a modern transformational-generative grammar. The examples have been taken from English; however, since X-bar syntax is an integral part of the overall structure of Universal Grammar, the analyses presented here extend to any language.1 1. The essence of X-bar syntax X-bar syntax can initially be defined in terms of two features: 1. multilayered structuring between lexical categories like N, V, A and P and maximal phrasal projections like NP, VP, AP and PP, some of the layers being recursively generated; 2. The replacement of category-specific by category-neutral rules to achieve cross-categorial generalization. 2. Constituent structure a simple example In American structural linguistics an important approach to syntax (in so far as syntax was ever reached) was the analysis of phrases and sentences into their I(mmediate) C(onstituent)s.2 Consider the following example: (1) The professor of Russian lives in this house IC-analysis would assign structure (1’) to the subject-NP of (1) (the dots are constituent-nodes and the integers are for ease of reference): 9 Hermes, Journal of Linguistics no. 10 – 1993 1 The paper is an adapted version of a chapter to appear in Jacobsen (forthcoming). I wish to thank my colleagues Kjær Jensen, Per Anker Jensen, Henning Nølke and Margrethe Petersen for helpful comments. I am grateful to Lis Rejnert Jensen for knocking the unwieldy manuscript into shape. 2 The classical statement of the IC-analysis of American structuralism is Wells (1947). IC-analysis was preeminently an operational procedure based on a set of analytical techniques applied to the physical data. Node-labelling was not automatic. See for ex-ample Gleason (1965: 151ff.) and Jacobsen (1978: 7ff.). In the Phrase Structure component of a TG-grammar, node-labelling is automatic. A more technical discussion may be found in Postal (1964).


1.
The essence of X-bar syntax X-bar syntax can initially be defined in terms of two features: 1. multilayered structuring between lexical categories like N, V, A and P and maximal phrasal projections like NP, VP, AP and PP, some of the layers being recursively generated; 2. The replacement of category-specific by category-neutral rules to achieve cross-categorial generalization.

Constituent structure -a simple example
In American structural linguistics an important approach to syntax (in so far as syntax was ever reached) was the analysis of phrases and sentences into their I(mmediate) C(onstituent)s. 2 Consider the following example: (1) The professor of Russian lives in this house (1´) Subsequent labelling of nodes would define 1 as an NP, 3 as a PP, 4 as a determiner, 6 as a preposition, 5 and 7 as nouns. In this context, the important node to note is 2. 2 defines professor of Russian as a constituent, to which we may provisionally attach the label nominal, and which represents a structural layer in between the full NP and the lexical category (i.e. word class) noun. There is a good deal of syntactic evidence that such an analysis is correct. First professor of Russian can be coordinated with another similar word sequence: (2) The professor of Russian and dean of the faculty Only constituents can be coordinated. Secondly, professor of Russian can be pronominalized: (3) The new professor of Russian is better than the old one Only constituents can be pronominalized. Thirdly, professor of Russian can undergo Right Node Raising: (4) Tom may be, and George certainly is, professor of Russian 3 Only constituents can undergo Right Node Raising. Fourthly, professor of Russian can occur independently in certain structures: (5) He is professor of Russian In (5) professor of Russian is clearly a constituent -a nominal. The question arises: is it also an NP? The answer is yes, but some kind of 10 3 In classical TG (4) was assumed to be derived from (i): (i) Tom may be professor of Russian, and George certainly is professor of Russian by the following two elementary operations: The first occurrence of professor of Russian is deleted under identity; the second occurrence of professor of Russian is raised away from its predicate position and attached as the right daughter of the highest S (see for example Jacobsen (1986: 253)).
"defective" NP. Thus (6), (7) and (8) show clearly that our constituent and full NPs have only overlapping distribution: (6) He is professor of Russian (7) He is the professor of Russian The professor of Russian was present

{ *Professor of Russian
In sum, there is ample syntactic evidence that professor of Russian should be layered as a constituent occurring between NP and N. 4 The Phrase Structure rules of a traditional TG-grammar would analyse the professor of Russian by a rule like (9): (9) NP → Det N PP which would assign structure (9'): (9') This is manifestly inadequate: professor of Russian is not defined as a constituent. Hence (9') could not account for the possibility of onepronominalization in (3). 5 11 3.

The origin of X-bar syntax within the TG-paradigm -Chomsky's "Remarks on Nominalization"
Consider (10) and (11): (10) Anne criticized her boyfriend (11) Anne's criticism of her boyfriend (11) is a Nominalization of (10). Until the late sixties it was assumed that (11) was derived from (10) by a transformational rule. Chomsky pointed to a number of fundamental difficulties involved in the transformational approach. Details need not concern us here.
Chomsky then went on to propose the Lexicalist Hypothesis. By this proposal items like criticize and criticism form a single neutral lexical entry specified as +N or +V (criticism would be formed by a low-level morphological rule). This neutral lexical item has a fixed set of selectional, subcategorization, and semantic features, such that a dependency relation may hold between one subset of the features (in particular semantic features) and +N, and between another subset of the features and +V (thus do (things) and deeds do not mean the same).
Chomsky notes that all major English categories, N, V, A and P may be preceded by specifiers and followed by complements and proposes a rule schema generalizing over the phrasal categories NP, VP, AP and PP. 6 In particular, he introduces the variable X to range over N, V, A and P and the symbol X (= X') 7 to denote the node immediately dominating X. Rule (12), then specifies that all major lexical categories can have one or more complements: where Comp(lement) is a functional term (on a par with 'subject' or 'object') with no independent status in the grammar: it merely serves as an abbreviation for some sequence of nodes, all of them maximal projections. (12), then, is an abbreviation for (12') 8 : 6 Chomsky is not very specific on PPs, but we will include them here. 7 The bar-notation and the prime-notation are equivalents. We shall use the prime-notation for typological reasons. We shall continue to talk about X-bar syntax. 8 I owe this explication of Chomsky's notation to Jackendoff (1977). The specifier is introduced by (13): (13) X'' → Spec x X' (13) is an abbreviation for (13'): 'Specifier', like Comp, is a functional term 9 The conjunction of (13ii) and (12ii) would generate a structure like (14): (14) If Comp N = the PP of Russian and N = professor, we have now defined professor of Russian as an N'-constituent.
Between them (12') and (13') represent what may be referred to as the Two-Level Hypothesis: all major phrasal categories consist of two structural layers. At this point, the canonical form of Phrase Structure rules can be stated as in (15)  In (15') X is the head of the entire construction and of X', and X' is the head of X''; ... indicates that heads may be preceded or followed by some category. Notice that (15) is an explicit formal statement of the endocentricity constraint on base structures. This was never formalized in traditional Phrase Structure rules (for discussion, see Lyons (1968: 330ff)). In section 6 we will give a more precise formulation of (15).
Armed with these notions, we now return to (10) and (11). They would be assigned structures like (10') and (11') (adapted from Chomsky (1970) and assuming of course the initial rule S → N'' V''): It is now obvious what multilayered X-bar syntax has achieved: 1. A structural similarity between nominal and sentential structures (as indicated by the boxed nodes) and the similarity between the encircled nodes V' and N'. 10 These generalizations are made possible by X-bar syntax in conjunction with the Lexicalist Hypothesis. 11 The spurious nominalization transformation can be eliminated.
Since 1970, X-bar syntax has come to play an integral part of transformational-generative grammar. In the next section we shall show in a little more detail the parallel analyses of the internal structures of the four major phrasal categories. After this we set up the required generalizations.

4.
The internal structure of NP, VP, AP and PP

NP
To generate NPs we will formulate the following set of rules (henceforth we will assume that Spec is optional): We will refer to (16i) as the specifier rule. Rule (16ii), which is recursive and optional, is the adjunct rule. Rule (16iii) is the complement rule; the asterisk indicates that there may be any number of complements. 12 Thus an N and its sister complement(s) expand into an N'; and an N' and its sister adjunct (there is always only one) expand into a new N'. Consider the following example: (17) The theory of grammar that I admire most By (16) this would be assigned structure (17'): (17') with the relative clause generated by the adjunct rule. 13 Relative clauses can stack up, as in (18): (18) The car that you bought that I admire most 16 12 At first sight this might appear to lead to massive overgeneration. However, this is prevented by the lexicon, as we shall see in section 8. 13 In section 9 we shall see that the sentence-schema can also be accommodated by Xbar syntax. In particular, the relative clause in (17') is an istantiation of XP.
The structure of (18) would contain one more N' than (17); and so on. In order to account for premodifying adjectives, we would need the "mirror-image" of (16ii). Something like (19)  That the adjectives are stacked is revealed by one-pronominalization facts. Consider the following sentences, in which N' stands for beautiful old yellow English cottage: In some cases, complements can precede their nominal heads. Consider the relationship between (28) and (29) (we ignore the ambiguity of (29)): (28) He is our professor of Russian (29) He is our Russian professor To account for (29), we need the "mirror image" of (16iii): (30) N ' → XP N Given (30), we can assign structure (29') to Russian professor: We will assume that it is only in the marked case that complements can occur to the left of their heads in English. 14

VP
Consider the bracketed VP in (31): (31) John will [study the letter in the office this afternoon] The Phrase Structure rules of a traditional TG-grammar would assign structure (31') to this VP: In this the complement and the two adjuncts are all daughters of VP and right sisters of V. Again, this is inadequate, as revealed by a simple do so pronominalization test: (32) (i) John will study the letter in the office this morning and Mary will do so too (ii) ... and Mary will do so this afternoon (iii) ... and Mary will do so in the kitchen this evening In (32i) do so replaces the entire VP; in (32ii) do so replaces study the letter in the office; in (32iii) do so replaces study the letter. These two strings are not defined as constituents in (31').
To account for these facts we need rules as those in (33) (parallels of the rules in (16)): 14 On this, see also section 6. For a full account of nominal premodifiers (and premodifiers in general) in the X-bar syntax approach, see Radford (1988). See also Haegemann (1991).

AP
Consider (35) Traditional Phrase Structure rules would assign structure (35') to the bracketed AP in (35): Intuitively, this is not satisfactory; thus it seems reasonable to assume that desirous of success is a head plus a complement (analogous with the VP desire success), but then desirous and of success should form one constituent; by the same token, it is arguable that very is a specifier. To capture this, we need rules like (36): Given (36), we assign structure (37) to the AP in (35): In (38) the adjunct rule has applied: (38)

PP
Arguably PPs also have specifiers. An example would be right in (39) If we now formulate the rules in (40) (analogous to (16), (33) and (36)) to account for PPs: (40) (i) P'' → (Spec) P' (ii) P' → P' XP (iii) P' → P XP* we can assign structure (41) to the bracketed PP in (39): One PP may occur inside the other as in (42) The rules would assign structure (42') to the bracketed PP in (42): Japanese would have the canonical structure (47): (47) Clearly (44) would not apply to a nonconfigurational language. In other words, (44) may be subject to parametric variation. We will leave it at that.

The Projection Principle
In connection with our discussion of rule (16iii), we noted that the grammar would massively overgenerate. How should this be avoided? The answer to this question is: by means of the lexicon. In particular, each N, V, A and P, that is each lexical head, will have associated with it a complementation specification. Thus, the complementation specification for a verb like give will look as in (48): NP] (48) states that give takes two NP-complements, no less, no more. In the typical instance heads take only one or two complements (if any at all). Therefore, although rule (16iii) (or rule (44iii)) allows any number of complements to be generated, this overgeneration is properly curbed by thelexicon. Only structures that can be projected from the complementation specifications of heads are proper syntactic structures. This is the essence of the Projection Principle, which is of crucial importance in modern trandformational-generative theory. It will be seen that X-bar syntax, in conjunction with the Projection Principle, has virtually eliminated traditional Phrase Structure rules from transformational-generative grammar. 16

X-bar syntax and the sentence-schema
In traditional transformational grammar, the structure of a sentence looked roughly as in (49): Later, in the seventies, especially under the influence of Bresnan's work (Bresnan (1974;1979)), the following sentence-schema evolved: There where good arguments for adopting (50) over (49). These will not be reviewed here. Notice that neither (49) nor (50) conforms to X-bar syntax in any obvious way. In recent years, however, a new sentence-schema has been developed and empirically motivated (see for example Chomsky (1986a;1986b)). It looks as follows: The three encircled constituents are heads (I = inflection, C = complementizer). V is head of V' (and V''); I is the head of I' (and I''); C is the head of C' (and C''). We will not be concerned here with the details of (51). It is sufficient for our purposes to note that (51) represents an extension of X-bar syntax to the entire sentence. 17 In effect, the sentence is defined as the largest phrase.