We’ll conclude this section with a little extra terminology that goes along with the notion of headedness and projection. Consider the NP in (29). () NP D the NЈ NЈ PP N bag PP with the plastic handle of grocers Each of the N’s and the NP in (28) are called the projections of the N. The NP is the maximal projection and the N’s are intermediate pro- jections. ModiWcation relations are no longer expressed in terms of sisterhood to the head; instead, modiWers of a head are sisters to any projection of that head. 7.2.3 Binarity The constituency tests we’ve seen in this chapter and the rules given in (26) reveal another important property that X-bar theory might ac- count for. It appears as if the layers of structure operate in a binary manner.10 That is, as we add layers of structure, new material is added one element at a time to the existing structure, thus creating a binary branching tree structure. The three X-bar statements mentioned above can be modiWed to capture this: (29) (a) XP ! (YP) X’ (b) X’ ! (ZP) X’ (c) X’ ! X’ (ZP) (d) X’ ! X (WP) Instead of ellipses ( .),weusethevariable categories YP, WP, and ZP to stand in for the modiWers. These are listed as optional as the phrase can consist of a head without any modiWers at all. They are also all 10 See Chametzky (2000) for a discussion of how this insight is retained in the minimalist non-X-bar-theoretic Bare Phrase Structure system. 120 phrase structure grammars and x-bar listed as phrasal, including the one in (29a), which is typically occupied by a bare determiner. We will return to this contradiction below. Binarity is a common—but by no means universal—part of X-bar theory. 7.2.4 Distinctions among modifier types Given our three types of rule, which introduce three distinct layers of structure, we predict that we should have at least three distinct types of head-modiWers. This appears to be true. We Wnd good evidence that we need to distinguish among speciWers (the YP in the XP ! (YP) X’ rule), adjuncts (the ZP in the X’ ! (ZP) X’ and X’ ! X’ (ZP) rules) and complements (the WP in the X’ ! X (WP) rule).11 Consider the two prepositional phrases that are subconstituents of the following NP: (30) the bag [ PP of groceries] [ PP with the plastic handle] Using the X-bar schema, we can generate the following tree for this NP: () NP D the NЈ NЈ PP 2 sister to a bar level N PP 1 with the plastic handle bag sister to a hea d of groceries You will note that the two PPs in this tree are at diVerent levels in the tree. The lower PP 1 is a sister to the head N (bag), whereas the higher PP 2 is a sister to the N’ dominating the head N and PP 1 . Notice also that these two PPs were introduced by diVerent rule types. PP 1 is 11 Napoli (1989) argues against this distinction referring to the reverse ordering of adjuncts and complements in Italian NPs such as: (i) la distruzione brutale de Troia the distruction brutal of Troy spec head adjunct complement ‘‘the brutal destruction of Troy’’ See Longobardi (1994) for an alternative analysis that involves movement of the head around the adjunct to its surface position. x-bar theory 121 introduced by the rule meeting the X’ ! X (WP) schema and PP 2 is introduced by the higher-level rule type (X’ ! X’ (ZP)). An XP that is a sister to a head is a complement.PP 1 is a comple- ment. XPs that are sisters to single-bar levels and are daughters of another bar level are adjuncts.PP 2 in (31) is an adjunct. The third type of modiWer is a speciWer. These are sisters to the bar level and daughter of a maximal category. The determiner in (31) is a speciWer. If we abstract away from speciWc categories we can distinguish among modiWers as seen in (32). ()XP YP XЈ specifier adjunct XЈ ZP complemen t XWP We predict these diVerent kinds of modiWer to exhibit diVerent behav- iors. We’ll concentrate Wrst on the distinction between complements and adjuncts then turn to speciWers. Take NPs as a prototypical example. Consider the diVerence in meaning between the two NPs below: (33) (a) the bag of groceries (b) the bag with a plastic handle Although both these examples seem to have, on the surface, parallel structures (a determiner followed by a noun followed by a prepos- itional phrase), in reality they have quite diVerent structures. The PP in (33a) is a complement and has the following tree: ()NP D the NЈ N bag PP of groceries You will note that the circled PP is a sister to N, so it is a complement. By contrast, the structure of (33b) is: 122 phrase structure grammars and x-bar N ba g ()NP D NЈ NЈ PP with a plastic handle Here the PP with a plastic handle is a sister to N’, so it is an adjunct. Observe that rules that introduce complements also introduce the head (X). This means that the complement will be both adjacent to the head and more importantly closer to the head than an adjunct. (36) the bag [of groceries] [with a plastic handle] head complement adjunct (37) ??the bag [with a plastic handle] [of groceries] head adjunct complement Since the adjunct rules take an X’ level category rewrite it as an X’ category, adjuncts will always be higher in the tree than the output of the complement rule (which takes an X’ and rewrites an X). Since lines can not cross, this means that complements will always be lower in the tree than adjuncts, and will always be closer to the head than adjuncts. The adjunct rules are iterative. This means that the rule can generate inWnite strings of X’ nodes, since the rule can apply over and over again to its own output: ()XЈ XЈ YP XЈ YP XЈ etc. YP … Complement rules do not have this property. On the left side of such rules there is an X’, but on the right there is only X. So the rule cannot apply iteratively—that is, it can only apply once within an XP. What this means for complements and adjuncts is that you can x-bar theory 123 have any number of adjuncts (39), but you can only ever have one complement (40): (39) the book [of poems] [with a red cover] [from Oxford] head complement adjunct adjunct [by Robert Burns] adjunct (40) *the book [of poems] [of Wction] [with a red cover] head complement complement adjunct The tree for (39) is given below; note that since there is only one N, there can only be one complement, but since there are multiple N’s, there can be as many adjuncts as desired. () D NP NЈ NЈ PP NЈ PP by Robert Burns NЈ PP from Oxford N book PP with a red cover of poems It also follows from the iterative nature of adjunct rules that adjuncts can be reordered with respect to one another, but one can never reorder a complement with the adjuncts: (42) (a) the book of poems with a red cover from Oxford by Robert Burns (b) the book of poems from Oxford with a red cover by Robert Burns (c) the book of poems from Oxford by Robert Burns with a red cover (d) the book of poems by Robert Burns from Oxford with a red cover (e) the book of poems by Robert Burns with a red cover from Oxford 124 phrase structure grammars and x-bar (f) the book of poems with a red cover by Robert Burns from Oxford (g) *the book with a red cover of poems from Oxford by Robert Burns (h) *the book with a red cover from Oxford of poems by Robert Burns (i) *the book with a red cover from Oxford by Robert Burns of poems etc. Conjunction also distinguishes these types of modiWer. Conjunction is typically restricted to constituents of the same general kind and result in a complex constituent of the same type as its conjuncts. Imagine one were to conjoin a complement with an adjunct, resulting in a contra- dictory situation: something can not be both a sister to X’ and X at the same time. Adjuncts can conjoin with other adjuncts (other sisters to X’), and complements can conjoin with other complements (other sisters to X), but complements cannot conjoin with adjuncts: (43) (a) the book of poems with a red cover and with a blue spine12 (b) the book of poems and of Wction from Oxford (c) *the book of poems and from Oxford Finally, recall the test of one-replacement. This operation replaces an N’ node with the word one. Look at the tree in (44): ()NP DNЈ can be replaced by on e NЈ PP N book PP with a red cover of poems cannot be replaced b y one Two possibilities for one-replacement exist. It can target the highest N’, and produce (45): (45) the one It can target the lower N’ and produce (46): 12 If this NP sounds odd to you, try putting emphasis on and. x-bar theory 125 (46) the one with a red cover For many speakers—but not all—the N head may not be targeted. This means that one followed by a complement is ill-formed: (47) *the one of poems with a red cover13 Since complements are sisters to X and not X’, they cannot stand next to the word one. Adjuncts, by deWnition, can. The distinction between complements and adjuncts is not limited to NPs; we Wnd it holds in all the major syntactic categories. The best example is seen in VPs: the direct object of a verb is a complement of the verb, while prepositional and adverbial modiWers are adjuncts: (48) I loved [the policeman] [intensely] [with all my heart]. complement adjunct adjunct ()VP VЈ VЈ PP VЈ AdvP with all my heart V loved NP intensely the policeman Direct objects must be adjacent to the verb, and there can only be one of them. (50) (a) *I loved intensely the policeman with all my heart. (b) *I loved the policeman the baker intensely with all my heart. Did-so (did-too) replacement targets V’. Like one-replacement, this means that it can only apply before an adjunct and not before a complement: (51) Mika loved the policemen intensely and (a) Susan did so half-heartedly. (b) *Susan did so the baker. 13 Not everyone Wnds this NP ill-formed. There is at least one major US dialect where sentence (47) is entirely acceptable. 126 phrase structure grammars and x-bar The evidence for the adjunct–complement distinction in adjective phrases and prepositional phrases is considerably weaker than that of nouns and verbs. Adverbs that modify adjectives have an adjunct Xair—they can be stacked and reordered. Other than this, however, the evidence for the distinction between PPs and AdjPs comes mainly as a parallel to the NPs and VPs. This may be less than satisfying, but is balanced by the formal simplicity of having the same system apply to all categories. SpeciWers are the third type of modiWer. Thus far we’ve only seen one, the determiner in the NP: (52) [the] [book] [of poems] [with a red cover] speciWer head complement adjunct ()NP D the NЈ specifier NЈ PP adjunc t N book PP with a red cover head of poems complement The speciWer is deWned as the daughter of XP and sister to X’: (54) SpeciWer : An element that is a sister to an X’ level, and a daughter of an XP. SpeciWers are diVerent from adjuncts and complements. Since the speciWer rule is not recursive, there can only be one speciWer:14 (55) *the these red books The speciWer rule has to apply at the top of the structure, which means that the speciWer will always be the left-most element (in English): (56) *boring the book This example also shows that speciWers cannot be reordered with respect to other adjuncts or complements. As the Wnal diVerence between speciWers and other types of modiWer, speciWers can only be conjoined with other speciWers: 14 It is not hard to Wnd exceptions to this claim, as in all the books. If determiners are heads (Abney 1987), then this problem disappears. x-bar theory 127 (57) (a) two or three books (b) *two or boring books In the late 1980s, with the advent of X-bar theoretic functional categories (IP, TP, etc.) instead of S (see ch. 11), determiners and other categories that had been assumed to be speciWers shifted in their phrase structure position. Abney (1987) proposed that deter- miners headed their own phrase, which dominated the NP. What then became of the speciWer? Around the same time, it was suggested that speciWers have a special role, serving as the identiWers of subjects of various kinds of phrase (Stowell 1981; for a critique see Borsley 1996). For example, Koopman and Sportiche (1991) propose that subject arguments start as the speciWer of the VP (the VP internal subject hypothesis), and then move to the speciWer of some higher functional projection (e.g. IP). See also Fukui and Speas (1986) and Fukui (1995). This kind of approach is widely adopted in Chomskyan P&P theory and in the Minimalist Program, but not elsewhere. 7.2.5 Cross-linguistic variation Consider the position of direct objects (complements) in Turkish. In Turkish, the complement precedes the head: (58) Hasan kitab-i oku-du. Hasan-subj book-obj read-past ‘‘Hasan read the book.’’ Interestingly, X-bar theory provides an avenue for exploring the diVer- ences and similarities among languages. Travis (1989) proposed that a certain amount of cross-linguistic variation in word order could be explained by allowing languages to parameterize the direction of head- edness in the X-bar schema. Take, for example, the complement rule. In English, complements of verbs follow the verbal head. In Turkish, they precede it. There are two options in the rule: (59) (a) X’! X (WP) (b) X’ ! (WP) X The child learning English will adopt option (a), the child learning Turkish will adopt option (b). I am obscuring some of the details here, but this provides a relatively elegant account of cross-linguistic variation. X-bar theory allows individual languages to select among a Wnite set of phrase structure options: 128 phrase structure grammars and x-bar (60) (a) XP ! (YP) X’ or XP ! X’ (YP) (b) X’ ! X’ (ZP) or X’ ! (ZP) X’ (c) X’ ! X (WP) or X’ ! (WP) X For a contrasting analysis in the GPSG framework, see Fodor and Crain (1990). 7.2.6 Summary In this section, we have surveyed some of the motivations for X-bar theory as well as one particular formulation of the X-bar schema itself. Phrase structure rules can be overly powerful, thus motivating an endocentricity requirement. Evidence for intermediate structure (X’), takes us to the point where there are similarities across categories in terms of the kinds of phrase structure rule are allowed. Cross-categor- ial generalizations about structure tie up the knot, showing that a variable-based notation is motivated. Next, we saw that the distinct modiWer types of complement, speciWer and adjunct, predicted by the X-bar schema, seem to be well motivated. Finally, we saw how the X-bar schema at least partly allows for a straightforward and constrained theory of cross-linguistic variation. In the next section, we survey the history of X-bar theory, including an ontological discussion of what the X-bar formalism represents. 7.3 A short history of X-bar theory15 7.3.1 The origins: Harris (1946) and Chomsky (1970) The bar notation (actually the numerical N1,N2, etc. equivalents) is Wrst found in a work on the substitution task for identifying constitu- ency within the word in Harris (1946). This work emphasizes the diVerences among types of hierarchical modiWer, but it does not focus on the syntax, nor does it use the variable notation prevalent in modern X-bar theory. The point of Harris’s notation was to limit overgeneration of recursive structures within words. Chomsky (1970) adapted this part of the notation but developed it into the beginnings of X-bar theory. At the time, one of the driving issues in the theory was to explain the relations between constructions that appeared to have similar semantics but diVered in the categorial 15 For excellent histories of the X-bar theory, see Stuurman (1985), LeVel and Bouchard (1991), and Fukui (2001). x-bar theory 129 . insight is retained in the minimalist non-X-bar-theoretic Bare Phrase Structure system. 120 phrase structure grammars and x-bar listed as phrasal, including. kitab-i oku-du. Hasan-subj book-obj read-past ‘‘Hasan read the book.’’ Interestingly, X-bar theory provides an avenue for exploring the diVer- ences and