The Projection of DP (and DegP)

Transcription

1 (Wrien 1991) The Projecion of (and egp) Richard K. Larson (SUNY - Sony Brook) Recen versions of he Exended Sandard Theory (e.g., Chomsky 1981, 1986) have proposed ha synacic srucures are largely projecable from he properies of heir consiuen lexical iems - in paricular, from heir hemaic properies. In his paper I argue for an exension of his approach o nominal srucure. My focus will no be on nominals relaed in an obvious way o verbal projecions (1a), bu raher on ones involving simple deerminers, possessives and resricive modifiers (1b-d): (1) a. The enemy s desrucion of he ciy b. Every flower c. John s book d. The man ha I me The proposed analysis draws crucially on hree lines of semanic and synacic research: (a) he relaional view of deerminers, under which elemens like some, every, he, ec. correspond o binary relaions beween properies or conceps; (b) he " Hypohesis", under which noun phrases (so-called) are acually projecions of heir consiuen deerminers; and (c) he view of consiuen srucure proposed in Larson (1988), according o which X-bar projecions assume a uniform, recursive ransiive form. As I show, his accoun illuminaes a variey of familiar quesions abou he form of noun phrases, and suggess a reurn o cerain "classical" ransformaional proposals abou he relaion beween nominals and modifiers. In secion 1, I skech he relaional view of deerminers, and in secion 2, I show is basic implicaions for noun phrase srucure under he "head raising" synax in Larson (1988,1989). In secion 3, I examine nominals conaining resricive relaives and PPs, and argue, in effec, for a resoraion of he "Aricle-S" analysis of Smih (1964). According o he laer, resricive modifiers in are complemens of he deerminer and no he nominal. Secion 4 nex considers he srucure of possessive nominals, including possessives ha are inuiively linked o relaive consrucions (John s book/he book ha John has), and possessives ha involve argumenal relaions (he ciy s desrucion/he desrucion of he ciy). I sugges a derivaional analysis of such pairs analogous o he derivaional relaion holding beween oblique and double objec in examples like John gave Mary a book/john gave a book o Mary. Finally, in secion 5, I briefly explore an exension of hese views o he srucure of degree phrases and comparaives The Semanics of eerminers In logic exs, senences like hose in (2a,b) are sandardly represened in he forma of "unresriced quanificaion" shown in (3a,b) (respecively): 1 This paper was presened a MIT (1988), CUNY (1991), and in he 1991 Synax Colloquium series a Universiy of Indiana - Bloomingon. I am graeful o audience members for helpful commens and suggesions, including Bob Fiengo, Jane Fodor, Seve Franks Jerry Kaz, Richard Kayne, avid Pesesky, Ken Safir and Laurie Zaring.

2 (2) a. All whales are mammals b. Some man arrived (3) a. x[whale(x) mammal(x)] b. x[man(x) & arrived(x)] On his view, quanificaional deerminers correspond o operaors combining wih a single (possibly complex) unary predicae. Thus in (3a), combines wih he complex unary predicae if-a-whale-henmammal, ec. As is well-known, however, he forma of unresriced quanificaion appears unsaisfacory, on a number of couns, for represening naural language quanificaion. One problem is ha he synax depars sharply from ha of naural language: (3a,b) involve ruh-funcional connecives ha do no appear o be presen in (2a,b); moreover, (3a,b) have fundamenally he srucure of conjuncions, somehing no obviously rue of (2a,b). A second, more serious difficuly is ha senences involving cerain quanifiers can be shown o have no unresriced represenaion. Thus i can be proven formally ha firs-order represenaions analogous o (3a,b) simply canno be given for senences conaining mos, many and few, such as (4a,b): 2 (4) a. Mos people hink ha dinosaurs were cold-blooded b. Few cas rejec una fish Thus here is an expressive limiaion on unresriced quanificaion ha is apparenly exceeded by naural language. 1.1 The Relaional View of eerminers Given hese resuls, here has been considerable ineres in recen years in he analysis of naural language quanificaion as involving generalized or resriced quanifiers (Rescher (1962), Barwise and Cooper (1981), avies (1981), Higginboham and May (1981), Keenan and Savi (1983), Wiggins (1980)). The approach is based on a single, very simple idea ha may be saed as follows: Relaional View of eerminers: eerminers express relaions among predicae meanings. This idea descends from he Arisoelian radiion in logic, bu is also advanced by Frege (1953), who suggess ha in quanified examples like (2), he elemen all expresses a relaion beween beween conceps. In paricular, Frege proposes ha all expresses subordinaion of he concep whalehood o ha of mammalhood. On his view, he "logical form" of (2a) is somehing like (5a), where ALL corresponds o he subordinaion relaion. Tha is, (2a) is rue jus in case being-a-whale is subordinae o being a mammal so ha all individuals possessing he firs propery possess he second. In a similar way, (2b) may be aken o have he logical form in (5b), where SOME is he "nonexclusion" relaion. Tha is, (2b) is rue jus in case being-a-man and arriving are non-exclusive properies. (5) a. ALL( whalehood, mammalhood ) 2 See Barwise and Cooper (1981) for discussion of he non-firs order characer of mos and relaed quanifiers. 2

3 b. SOME( man, arrive ) According o he relaional analysis, hen, deerminers are semanically similar o ransiive predicaes such as ouch; bu whereas he laer express relaions beween wo individuals, such as Mary and John, he former express relaions beween wo conceps. 1.2 eerminers as Se-Relaions Frege s basic idea can be spelled ou precisely by consruing "propery", "subordinaion", "nonexclusion", ec. in se-heoreic erms. Suppose common nouns and verb phrases are viewed as corresponding semanically o ses of individuals: (6) a. whale => {x: x is a whale} b. mammal => {x: x is a mammal} c. man => {x: x is a man} d. arrive => {x: x arrives} Then deerminers can be inerpreed as expressing relaions of quaniy beween such ses. The "subordinaion" relaion ALL can be spelled ou in erms of he number of individuals in he common noun se (Y) ha are no in he verb phrase (X) (7a). Likewise he "nonexclusion" relaion SOME can be spelled ou in erms of he number of individuals ha are in boh (7b): (7) a. ALL(X,Y) iff Y X = 0 b. SOME(X,Y) iff Y X 0 Given (7a), All whales are mammals will be rue if and only he se of whales conains no members no in he se of mammals. Given (7b), Some man arrived will be rue jus is case he se of men and he se arrivers have a non-empy inersecion. These are he correc resuls. This general picure exends naurally o a variey of oher deerminers, including mos, which was problemaic for unresriced quanificaion: (8) a. NO(X,Y) iff Y X = 0 b. MOST(X,Y) iff Y X > Y X c. TWO(X,Y) iff Y X = 2 (and similarly for oher numeral deerminers) d. THE-TWO(X,Y) iff Y X = 0, where Y = 2 (similarly for oher numeral deerminers of he form he-n, for some n) e BOTH(X,Y) iff THE-TWO(X,Y) f. NEITHER(X,Y) iff Y X = 0, where Y = 2 g. THE(X,Y) iff THE-ONE(X,Y) In all such cases, he deerminer expresses a relaion of quaniy beween he exension of a common noun (Y), radiionally referred o as he resricion, and he exension of a verb (or oher predicae) phrase (X), radiionally referred o as he scope. specifies, in effec, how many hings saisfying he resricion Y are rue of he scope X. 3 3 The relaional view, which reas quanifiers as full-fledged lexical iems wih independen meaning, has a number of virues beyond expressive richness. Imporanly, i permis us o alk abou lexico-semanic properies of specific 3

4 2.0 Implicaions for Nominal Synax Under he relaional analysis, deerminers possess argumen srucure and lexical properies much like oher predicae expressions. This view has ineresing consequences for he synax of nominals under he Exended Sandard Theory of Chomsky (1981, 1986), which hypohesizes an inimae connecion beween argumen srucure and form. 2.1 The Relaional View and As we have noed, he relaional view of deerminers reas as a predicae, which selecs is siser noun much like a ransiive verb selecs an objec. This semanic analysis appears o fi naurally wih he heory of nominal synax proposed by Abney (1987) and Fukui and Speas (1986), wherein s are heads ha ake heir nouns as complemens (9a). In fac, he relaional view appears o fi much beer wih analysis han wih he radiional picure (9b), which expresses no selecion relaion beween and N: (9) a. b. e N he man he man Noneheless, despie he obvious aracions of connecing he relaional analysis and, he assimilaion is no compleely sraighforward. Abney (1987) classifies as a "funcional caegory," a group of forms bearing lile or no semanic conen on his view. He analogizes in o I in IP, along he lines shown in (10a,b), analyzing John as subjec of in (10a) jus as John is subjec of IP in (10b). (10) a. b. IP I John John I s compleion of he plan These proposals are simply no enable under he relaional analysis, however.firs, as we have seen, he relaional analysis does no ake s o be semanically empy; raher hey express relaions of quaniy beween ses. Furhermore, is no analogous o I under he relaional analysis, bu raher o a predicae caegory such as a V. Indeed, as well will discuss in deail below, he general semanics for s offers a naural classificaion of hese elemens ino monadic, dyadic and riadic forms, much as one finds wih verbs. Basic deerminers (every, some, he, ec.) correspond o dyadic predicaes (11a); pronouns (she, him, hey, ec.) correspond o monadic predicaes (11b), and complex s like more-han and every-excep correspond o riadic predicaes (11c,d): [TNS] complee he plan deerminers, such as (in)definieness and direcional enailingness, and universal properies of deerminers, such as conservaiviy. For more on his see Barwise and Cooper (1981), Keenan and Savi (1984) and Larson (1990b). 4

5 (11) a. EVERY(X,Y) iff Y - X = 0 "dyadic " b. HE n (X) iff g(x n) X "monadic " c. MORE-THAN(X,Y,Z) iff Y X > Z X "riadic " Finally, under he relaional view, an expression like John could no possibly consiue he subjec of in a semanic sense. Recall ha he exernal subjec argumen of (X) is semanically a predicae, one given by he synacic consiuen comprising he scope of he (usually he main predicae of he senence). This is no compaible wih a srucure like (10a), which makes a name he subjec of, and which makes no provision for he scope argumen of. In general, hen, alhough a joining of he relaional analysis and is aracive, his move doesn appear o be possible under he original inuiion of Abney (1987) and Fukui and Speas (1986) ha is a funcional caegory comparable o I. Raher he naural correspondence offered by he relaional semanics is beween and V. 2.2 A Theory of Srucural Projecion I wish o offer an alernaive accoun of projecion, one which is (I believe) more clearly compaible wih he relaional analysis of deerminers. This accoun is based on he heory of argumen projecion in Larson (1988), which includes following principles: (12) a. XP YP X b. X X ZP (13) If β is an argumen of α, hen β mus be realized wihin a projecion of α. (14) Roles deermined by a predicae are projeced according o he hemaic hierarchy Θ AGENT > Θ THEME > Θ GOAL > Θ OBLIQUE, such ha if Θ 1 > Θ 2, hen he argumen o which Θ 1 is assigned c-commands he argumen o which Θ 2 is assigned. (12) is a resriced version of X-bar heory embodying a "Single Complemen Hypohesis". Under he laer, maximal projecions are limied o one specifier and one complemen per phrase. (13) and (14) give principles for he realizaion of argumens vis-a-vis heir selecing head, specifying he locaion of hese argumens, and heir relaive hierarchical organizaion, respecively. To illusrae hese principles briefly, consider firs he ransiive verb kiss, which assigns an agen and a heme role (12)-(14) deermine a headed by kiss as in (15). This srucure conforms o he resriced X-bar heory; furhermore, all argumens of V are conained wihin a projecion of V; finally, he argumen bearing he agen role is projeced ino a posiion c-commanding he argumen bearing he heme role, in compliance wih (14) and he fac ha Θ AGENT > Θ THEME. 5

6 (15) V John V kiss Mary iransiive pu, which assigns agen, heme, and locaion represens a more complicaed case. Assuming Θ AGENT > Θ THEME > Θ LOC, we projec a minimal as in (16), conaining argumens corresponding o Θ THEME and Θ LOC, wih he former higher han he laer: (16) V sal V PP pu on he fish This srucure leaves Θ AGENT unassigned, and no posiion for is bearer. In Larson (1988) i is proposed ha his circumsance licenses he " shell" in (17a), which conains a higher specifier for he agen, and an empy verbal head posiion. The surface word order derives by raising he verb form o [ V e] (17b): (17) a. b. V V John V John V e V pu V sal V PP sal V PP pu on he fish on he fish In he laer configuraion, he requiremens in (12)-(14) are me. 6

7 2.3 Projecing The heory skeched above can be exended o s under he assumpion ha hey are projecions of heir consiuen deerminers. To do so, however, we mus firs sele some imporan preliminary quesions abou hemaic hierarchy, and he saus of he scope argumen for. We hen urn o he projecion of s of various ypes The Themaic Hierarchy in A crucial elemen in he heory of projecion given above is he assumpion of a hemaic hierarchy. The principle in (13) orchesraes he projecion of verbal argumens by mapping relaive prominence on he hemaic hierarchy ino relaive srucural prominence as defined by c-command. Larson (1988, 1989) assumes he specific hemaic hierarchy argued for by Carrier-uncan (1985), and Baker (1988) among ohers, viz.: 4 Θ AGENT > Θ THEME > Θ GOAL > Θ OBLIQUE Under he laer, agen phrases are always projeced ino srucure higher han oher argumens, hemes are projeced higher han everyhing excep agens, ec. 5 Whaever he correcness of his hierarchy for projecion of verbal argumens in, i should be clear ha i canno help us wih he projecion of. There simply is no sense in which he se argumens (X,Y) of under he relaional analysis can be hough of as playing roles like agen or heme in. These conceps seem o be irrelevan. Wha hen are he appropriae noions? Proposals in his area mus be regarded as highly enaive since he errain is almos enirely unexplored. To my knowledge, applicaion of hemaic heory o nominals has so far been confined enirely o nominal gerunds and derived nominals like John s desroying of he evidence and John s desrucion of he evidence, which show an obvious connecion o verbal forms (John s desroying he evidence, John desroyed he evidence). Noneheless, i is possible o reason by analogy o some exen. Canonically, verbs describe evens and noions like agen, heme, goal, ec. represen recurring semanic/funcional roles ha verbal argumens play in hose evens. Thinking analogically, we observe ha deerminers express quanificaion, and noions like resricion and scope represen wo main recurring semanic/funcional roles ha se argumens play in quanificaion. Semanically, he resricion ses he domain of quanificaion, whereas he scope deermines wha is rue of hose individuals. Synacically, resricion and scope are also plainly relevan in mapping he pars of. The former role is mapped o he complemen of. The laer role is associaed wih a main clause predicaion. 4 The equivalen is adoped in Relaional Grammar in he form of a Relaional Hierarchy (see Perlmuer 1981, Perlmuer and Rosen 1984). 5 The erm ΘOBLIQUE is a cover erm embracing he various adverbial elemens. In Larson (1990a) i is suggesed ha hese migh be furher ordered as: ΘMANNER > Θ LOC > Θ TEMP > Θ CAUSE/PURPOSE See secion 3 for discussion of adjunc projecion in. 7

8 Given hese poins, I sugges an approach employing he wo basic roles Θ SCOPE and Θ RESTRICT, which are ordered as such and play a par roughly similar o Θ AGENT and Θ THEME for V in a canonical. Thus, he scope argumen is projeced ino Spec, and he resricion argumen is projeced lower down, inside. Below we will inroduce various addiional "oblique" argumens of, including comparison phrases, excepion phrases, and various forms of adjuncs. Summarizing, hen, I propose he following hemaic hierarchy in, where "NOBLIQUE" sands for nominal obliques: Θ SCOPE > Θ RESTRICT > Θ NOBLIQUE I follows from his hierarchy ha he scope argumen will always be projeced highes, ha he nominal resricor will be projeced higher han everyhing excep he scope argumen, and so on The Scope Argumen X Having esablished a rudimenary hemaic hierarchy for deerminer argumen projecion, we mus reurn more carefully o he naure of he scope argumen. In our semanic discussion of examples like All whales swim, we have analyzed as relaing ses X, Y. The Y argumen was given by he noun and corresponded o he resricion on he deerminer. The X argumen was given by he predicae and corresponded o is scope. On his view, he main predicae appears o funcion direcly as one of he argumens of. (18) a. whales => {x: x is a whale} b. swim => {x: x swims} c. ALL(X,Y) iff Y - X = 0 d. All whales swim is rue iff {x: x is a whale} - {x: x swims} = 0 Consider now he wo candidae srucures in (19a,b) in he ligh of our principle (13) requiring all argumens of a predicae α o be realized wihin a projecion of α: (19) a. b. V swim swim all whales all whales Srucure (19a) correcly expresses he projecion requiremens of V: all whales, he agen of swim, is realized wihin a projecion of swim (). However, (19a) fails o express he projecion requiremens of. Alhough swim is an apparen argumen of under he relaional view, i fails o be realized wihin a projecion of. By conras, srucure (19b) has he opposie problem. Here he projecion requiremens of are correcly expressed: swim is an argumen of all and occurs wihin a projecion of all (). Bu, (19b) fails o express he projecion requiremens of V since all whales occurs ouside. The problem is clear cu. If is an argumen of V and is simulaneously an argumen of, how can we find a srucure ha mees heir join requiremens under he localiy consrain on hea-role assignmen (13)? 8

9 A relaed quesion arises wih senences conaining a quanified objec (20a). Here he scope argumen is a se of individuals ({x: John respecs x}) ha is no given by any surface consiuen (20b). A familiar view is ha he quanified undergoes cover raising and adjoins o a conaining caegory; he srucural residue of movemen (John respecs i ) hen deermines he scope argumen of (20c). (20) a. John respecs [ all whales] b. John respecs all whales is rue iff {x: x is a whale} - {x: John respecs x} = 0 c. [ i all whales] [John respecs i ] Here again we may ask how he localiy requiremens of are me given ha is apparen scope argumen (John respecs i ) does no occur wihin. Bu noice a furher quesion as well. Since all whales is a complemen of respec in (20a), i should consiue a phrase under X-bar heory, and hence should conain all is argumens wihin. Bu how can his be if he scope argumen of all is no presen unil afer all whales has undergone raising, as in (20c)? The answer o hese quesions I wish o sugges is ha he synacic scope argumen of is never in fac an over predicae in he clause - neiher he surface one given by (swim), nor a derived one creaed by movemen (John respecs i ). Raher, he scope argumen of is an independen, inaudible, pro-predicae elemen Pro, licensed by, and projeced in Spec of, under he hierarchy Θ SCOPE > Θ RESTRICT (21a). I sugges ha he semanic value of his Pro argumen is deermined configuraionally a he level of Logical Form. Specifically Pro ges is value from he derived predicae ha is he srucural siser of a LF (21b). (21) a. [ Pro [ ] ] Θ SCOPE Θ RESTRICT b. [ Pro [ ]] [ XP... i... ] _ GETS ITS VALUE FROM _ To illusrae hese ideas wih a concree case, consider again our example all whales swim. Under he proposals jus saed, his senence is projeced iniially wih he srucure in (22a) (irrelevan deails suppressed). conains all whales in specifier posiion, saisfying he local projecion requiremens of swim. By conras, conains he pro-predicae Pro in Spec posiion, saisfying he requiremens of all, and compleing is argumen projecion. subsequenly undergoes raising a he level of Logical Form as in (22b). A LF, Pro s value is idenified by he he TP i swim, he srucural siser of he raised (22c). Thus Pro comes o denoe he se {x: swims(x)}, he desired semanic resul. 6 6 The sandard Monague Grammar analysis of quanificaion akes (he equivalen of) o apply o is scope S as funcion o argumen (i) (ignoring inensions): (i) [[ [ i S ] ]] M,g = [[ ]] M,g (λx i [[ S ]] M,g ) Under he synax proposed here, Pro denoes a disinguished variable R over characerisic funcions, and we specify: (ii) [[ [ i S ] ]] M,g = 1 iff [[ ]] M,g = 1, where g is ha R-varian of g such ha g (R) = (λx i [[ S ]] M,g ) 9

10 (22) a. TP b. TP 1 TP V Pro Pro swim 1 V all whales swim all whales c. TP 1 TP Pro 1 V all whales swim {x: swims(x)} The same analysis applies sraighforwardly o examples wih a quanified objec, such as (20a). This accoun answers our wo quesions direcly. Wih example (18a), we see (conrary o iniial impressions) ha swim is no in fac an argumen of all, and hence no required o occur wihin a projecion of. Raher Pro is he rue scope argumen of ; swims simply idenifies Pro s value. Similarly in example (20a), John respecs i is no an argumen of all, and hence need no o be formed a he poin where he all whales is projeced. Raher, he scope argumen is Pro, which is presen when [ all whales ] is formed, bu whose value is only deermined a Logical Form, afer he quanified objec has raised and he derived predicae John respecs i has been formed. In essence, hen, Pro, answers our wo quesions by separaing he hemaic domains of and V, relaing hem only in an indirec way, hrough he assignmen of is value. The accoun also explains anoher, oherwise puzzling fac, wha we migh call he "caegorial neuraliy" of s scope argumen. Quanifiers have been argued o be able o adjoin o any caegory of phrase XP, aking XP as heir scope (Sowell 1982). On a heory in which XP consiues he direc argumen of, his implies ha any caegory of phrase can be he scope argumen of - in oher words, exercises no synacic selecion. This siuaion is a leas anomalous. Oher predicaes ypically do exer caegorial selecion on heir argumens, and iself limis is resricion argumen o. The analysis in (20) resolves his puzzle: does indeed exercise selecion on is subjec, consraining i o be Pro. The appearance of caegorial neuraliy arises from he fac ha various differen ypes of phrase can funcion as anecedens o Pro, fixing is value. 10

11 2.3.3 yadic s Earlier we suggesed ha deerminers, like verbs, can be divided semanically ino monadic (inransiive), dyadic (ransiive), and riadic (diransiive) forms, according o wheher hey ake 1, 2, or 3 predicae argumens. By far he mos common case seems o be he dyadic-ransiive one, illusraed by deerminer relaions like (23a-d). These ake a resricion argumen Y, and a scope argumen X, and map o he general srucure in (24), where he complemen denoes he former, and Pro in Spec denoes he laer: (23) a. ALL(X,Y) iff X Y = 0 b. SOME(X,Y) iff X Y 0 c. NO(X,Y) iff X Y = 0 d. MOST(X,Y) iff X Y > X Y (24) Pro all some no mos In his srucure, all posiions made available by he X-bar heory in (12) - Spec, head and complemen - are realized in a single projecion Monadic s The case of monadic, inransiive s is plausibly represened by he class of pronouns, which Monague (1974) analyzes (in effec) as resricion-less quanifiers. Monague assigns pronoun meanings according o a scheme equivalen o (25), which involves he single scope argumen (X). Under his scheme, he pronoun he 1, for example, is rue of hose ses conaining he individual g(x 1 ) under some assignmen g: (25) For any assignmen g, HE n (X) iff g(x n ) X Monague s semanics can be mapped o he synax in (26), where he pronoun is analyzed as a deerminer (following Posal (1969)), and where Pro consiues s sole argumen: (26) Pro he 1 11

12 Noe ha his srucure reas pronouns specifically as "unergaive deerminers" insofar as heir one argumen is an underlying subjec. 7 This poin migh lead us o expec parallelisms beween unergaive s and Vs. (27a-c) show ha unergaive verbs have he propery of licensing "cognae objecs", dummy complemens which (in bare form) add no ruh-condiional conen o, bu a mos serve o convey emphasis. Ineresingly, pronouns have he propery of licensing "emphaic reflexives", dummy anaphors ha also makes no ruh-condiional conribuion, bu serve o emphasize or inensify: (27) a. [ laughed [a laugh]] Cognae objecs b. [ coughed [a coughed]] c. [ smiled [a smile]] (28) a. [ he [ himself]] Emphaic reflexives b. [ she [ herself]] c. [ hey [ hemselves]] Larson (1988) noes he special saus of ransiive srucures under he X-bar heory in (12) and proposes ha cognae objec formaion represens a way of "filling ou" he basic ransiive frame wih complemen maerial. If his line of reasoning is correc, we migh expec parallel processes in oher caegories, wih oher unergaive heads. Emphaic reflexives are a poenial candidae in he domain of ; hey migh be analyzed, in effec, as cognae complemens of Triadic s Finally, consider riadic, or hree-argumen s. We suggesed complex deerminer consrucions like (29) and (30) as represenaives of his case: (29) a. more women han men b. *no/hree women han men (30) a. every boy bu/excep Bill b. no boy bu/excep Bill c. *each/some/hree/mos/many boy(s) bu/excep Bill As noed by Keenan and Savi (1984), examples like hese exhibi a dependency beween he boldfaced elemens. (29) shows ha he deerminer more licenses a comparison phrase following N, whereas oher deerminers do no. (30) shows ha he universal deerminers every, all and no license an excepion phrase following N, whereas oher deerminers (including universals like each) do no. The dependencies in (29) and (30) can be analyzed as arising ou of he basic semanics of he deerminers in quesion. Following Keenan and Savi (1984), a sraighforward analysis of (29) is ha more-han expresses he hree-place relaion in (31a), wih he se argumen Z provided by he hanphrase. (31b-c) illusrae how ruh-condiions wih more-han migh be compued in a simple case, where 7 I is naural o ask wheher here are also "unaccusaive deerminers". Presumably his would be a whose scope arose in he posiion of he resricive erm. To answer his quesion would require a beer undersanding han I have a presen of he hierarchy by which argumens are projeced. 12

13 he Z argumen is supplied firs (31b), followed by he resricion Y (31c), and he scope X (31d): (31) a. MORE-THAN(X,Y,Z) iff Y X > Z X b. More han women iff Y X > {z: woman(z)} X c. More men han women iff {x: man(x)} X > {z: woman(z)} X d. More men han women smoke iff {y: man(y)} {x: smokes(x)} > {z: woman(z)} {x: smokes(x)} Alhough I will no ry o defend he proposal in deail here, I sugges a similar approach o he excepive-consrucions in (30). Specifically, I propose ha he universal deerminers licensing excepionphrases are subjec o a lexical alernaion, which raises heir valence from wo o hree, and ha he excepion phrase supplies he hird argumen o he augmened deerminer. 8 The basic idea is skeched in (32). Thus, for cerain universal deerminers, including every, all and no (bu no each, boh, all hree, ec.), he grammar makes available an augmened hree-place relaion " -excep(x,y,z), where X is he scope se, Y is he resricion se, and Z is a se given by he excepion phrase. The semanics of " - excep (X,Y,Z) is given in (32b), where ± Cond(X,Z) is a relaion whose conen depends on wheher " is posiive (every, all) or negaive (no). For " posiive, Cond(X,Z) is Z X = 0. For " negaive, Cond(X,Z) is Z X 0. (33) and (34) show how ruh-condiions wih excepion phrase are compued in simple examples: (32) a. " = {every, all, no} b. " -excep(x,y,z) iff " (X, (Y Z )) and ± Cond(X,Z) (33) a. EVERY(X,Y) iff Y X = 0 b. EVERY-excep(X,Y,Z) iff (Y Z ) X = 0 and Z X = 0 c. every excep Bill iff (Y {bill}) X = 0 and {bill} X = 0 d. every boy excep Bill iff ({y: boy(y)} - {bill}) - X = 0 and {bill} «X = 0 e. every boy excep Bill smokes iff ({y: boy(y)} {bill} ) {x: smokes(x)} = 0 and {bill} {x: smokes(x)} = 0 (34) a. NO(X,Y) iff Y «X = 0 b. NO-excep(X,Y,Z) iff (Y - Z ) «X = 0 and Z «X π 0 c. no excep Bill iff (Y - {bill}) «X = 0 and {bill} «X π 0 d. no boy excep Bill iff ({boys} - {bill}) «X = 0 and {bill} «X π 0 e. no boy excep Bill smokes iff ({y: boy(y)} - {bill} ) «{x: smokes(x)} = 0 and {bill} «{x: smokes(x)} π 0 The semanic analyses in (31) and (32) make han-phrases and excepion-phrases argumens of heir associaed deerminer; hus he relevan s become hree-place. This in urn makes heir synacic 8 The dyadic-riadic alernaion proposed here may be viewed as roughly analogous o ha found wih he verb wrie. The laer has a dyadic form wrie(x,y) ha means (approximaely) x produces y by wriing characers. Bu wrie also has a riadic form wrie(x,y,z) ha means (approximaely) x communicaes wih z using y produced by wriing characers. This aleraion migh be analyzed as he resul of a regular lexical valence alernaion process. 13

14 projecion similar o ha of verbs like pu, discussed earlier. Assuming Θ SCOPE > Θ RESTRICT > Θ EXCEPT, we projec he minimal in (33a), conaining argumens corresponding o Θ RESTRICT and Θ EXCEPT, wih he former higher han he laer: (33) men PP more han women As in he case of pu, his srucure leaves a hemaic role unassigned (Θ SCOPE ), and no posiion for is bearer. We herefore license a " shell" in (34a), conaining a higher specifier for he scope argumen and an empy head posiion. The surface word order derives by raising more o [ e], sranding he han-phrase (34b): (34) a. b. Pro Pro 1 e more men PP men 1 PP more han women han women Noice ha his derivaion direcly accouns for he disconinuous dependency holding beween more and han. The former underlyingly governs he phrase headed by he laer, a relaionship ha is broken up by he subsequen raising of. A parallel derivaion can be given for excepive-consrucions under he semanics proposed above. Under he assumpion ha he excepion-phrase represens he firs argumen of, we creae an iniial as in (35a). The need o inegrae he scope argumen hen licenses a higher shell as in (35b). Finally, raises o he empy head posiion, sranding he excepion-phrase (35c): (35) a. [ boy [ every [ PP excep Bill ]]] b. [ Pro [ e [ boy [ every [ PP excep Bill ]]]] c. [ Pro [ every [ boy [ [ PP excep Bill ]]]]] 14

15 As above, he disconinuous dependency holding beween and PP is direcly accouned for under his derivaion Modifiers in The analogy beween and claimed in his analysis has ineresing implicaions for he synax of modifiers. 3.1 Adverbs and Adverbials as V-Complemens Larson (1988,1990) proposes a heory of verbal modifiers, including adverbs and adverbials, ha depars significanly from convenional views. A common proposal is ha -modifiers adjoin on he righ in, so ha righ-mos modifiers are highes (36a). Larson (1988,1990) proposes ha adverbs descend o he righ, so ha righ-mos modifiers are lowes (36b). 9 Lappin(1988) presens (ia-b) as arguing agains a disconinuous analysis of excepives like ha proposed here: (i) a. Bill saw no suden bu John, alhough Mary saw several. b. No sudens oher han he radicals paricipaed in he demonsraions, alhough many suppored i. Lappin analyzes several in (ia) as an ellipical conaining an N anaphorically dependen upon he N in no suden bu John. He saes: "The second conjunc of [(ia)] can only [my emphasis] be undersood as assering ha Mary saw several sudens oher han John." (p.987) Lappin concludes ha suden bu John mus be a consiuen in order o furnish an appropriae aneceden. Similar reasoning is applied o (ib). I have wo poins in response. Firs, I do no share Lappin s judgmens, and I have found no one else who does. For myself and every speaker I have consuled, he only reading of (ia) is one in which Mary saw several sudens, which may or may no have included John. In oher words, for hese speakers, and for me, (ii) is perfecly coheren. (ii) Bill saw no suden bu John, alhough Mary saw several, including John. Exacly he same resuls obain wih (ib). Many refers o "many sudens", no "many sudens oher han he radicals". The analysis proposed here predics hese judgmens; since suden bu John and sudens oher han he radicals are no consiuens, hey should no offer naural anecedens. Second, Lappin s analysis of he phenomenon in (ia,b) as ellipsis is highly quesionable, as opposed o he alernaive (by Hoeksema (1984)) ha several, many, none, ec. are pronominal deerminers, and he relevan relaion one of simple anaphora. Consider (iiia-c): (iii) a. There were men presen and women presen. Many were under he age of weny. b. I bough hree presens for Alice and wo gifs for Sue. Several were under $20 dollars in price. c. Several Korean nouns and Japanese adjecives were analyzed. None were dismissed. Clearly, many in (iiia) can be read as referring o "men or women", despie he fac ha is has no explici N aneceden ha picks ou his se. Similar remarks apply o (iiib,c). This argues for anaphora, no ellipsis. 15

16 (36) a. b. V V yeserday John V John V me V me Bill Bill V yeserday The srucure in (39b) follows from he hemaic hierarchy assumed, which ranks oblique phrases, such as manner, locaive and emporal modifiers, lower han agens, hemes or goals: Θ AGENT > Θ THEME > Θ GOAL > Θ OBLIQUE The low inserion for adverbs ypically riggers -shell projecion and subsequen verb-raising, as in (36b). An ineresing resul of his heory is ha verbal modifiers are acually complemens of he V head, and in fac closer complemens of V han subjecs, objecs or indirec objecs. Larson (1990a) offers several argumens for he low posiion for modifiers, including he fac ha adverbs on he righ ypically behave as if hey are in he domain of oher elemens, including objecs. For example, consider he facs ha adverbials conaining negaive polariy iems can be licensed by an affecive objec (37). Assuming a resricive heory of negpol licensing based on c-command, his resul follows under a srucure like (36b), where he objec c-commands he adverbial. I does no follow no under (36a), however. (37) a. John me few friends [any day his week] b. Alice speaks few languages [wih any fluency] c. Gwen does few hings [because anyone asks her o] Anoher argumen concerns he exisence of verb-adverbial idioms like (38a-c), which sugges a form of disconinuous dependency beween he boldfaced elemens: (38) a. [ rea John wih kid gloves] ("rea carefully") MANNER b. [ rub John he wrong way] ("boher") MANNER c. [ pu John on he spo] ("confron") LOCATION d. [ kill John wih kindness ] ("be very solicious oward") INSTRUMENT Such iems receive a very naural analysis in erms of V-raising, where he semanic uni consiued by he idiomaic elemens corresponds o an underlying synacic uni ha is broken up by subsequen movemen (39): 16

17 (39) [... e [ John [ V rea [ wih kid gloves ]]]] 3.2 Relaive Clauses as -Complemens The general head raising analysis, and he reamen of disconinuous dependencies, suggess a way of reviving some old, bu inuiively appealing views abou he grammar of relaive clauses. In he hisory of ransformaional grammar, here have been hree main approaches o relaive clause synax. One is he -S analysis of Ross (1967), according o which relaive clauses are adjoined o he maximal nominal phrase (40a). The second is he NOM-S analysis of Sockwell, Schacer and Paree (1970), according o which relaive clauses are adjoined o a smaller nominal phrase inside (40b). The hird, and oldes proposal is he Aricle-S analysis of Smih (1964), according o which relaive clauses are no modifiers of he noun a all, sricly speaking, bu are insead consiuens of he deerminer (40c). (40) a. The -S Analysis b. The NOM-S Analysis S e NOM e N ha I saw he N S he girl girl ha I saw c. The ARTICLE-S Analysis e N e S girl he ha I saw The nominal modifier analyses (40a,b) have so far received he wides suppor in he lieraure, wih exbooks (Baker 1978) and professional aricles (Paree 1976) framing he quesion of relaive clause srucure as a choice beween he wo. Among all he four srucures, cerainly he leas frequenly defended is he Aricle-S analysis. The reasons are seem fairly clear. The laer is in plainly he mos "absrac" of he hree accouns insofar as is srucure does no mach surface word order (in English, a any rae). This absracness also makes i he mos complex, since i necessiaes some kind of exra movemen operaion in order o derive he correc surface forms. Noneheless, he Aricle-S analysis also has a cerain aracion insofar as i appears o shed ligh on cerain ineresing daa ha are no easily accommodaed in he nominal modifier accouns. 17

18 Kuroda (1969) poins ou ha indefinie nouns like way can co-occur wih a bare demonsraive, bu no wih a bare definie aricle (41a,b). Ineresingly, when he aricle is accompanied by a resricive adjecive or a relaive clause, he resul improves dramaically (41c,d). In effec, he+modifier appear o "add up" o a deerminer like ha. Kuroda observes a similar dependency wih he pair in (42), where he presence/absence of negaion in he relaive correlaes wih he appropriaeness of an indefinie vs. a definie (respecively) (38a,b): (41) a. I earned i ha way b. *he way c. he old-fashioned way d. he way ha one should (42) a. He greeed me wih he/*a warmh I expeced b. *he/a warmh I hadn expeced Jackendoff (1977) makes virually he same poin wih proper nouns, observing ha alhough he laer rejec a bare definie aricle, a relaive clause or oher resricive modifier (AP, PP) renders he consrucion accepable (43): (43) a. *he Paris b. he old Paris c. he Paris ha I love d. he Paris of he wenies Wha such examples appear o show is a form of disconinuous dependency holding beween he deerminer and he resricive modifier, wheher relaive clause, aribuive adjecive or PP. The basic consiuency of he Aricle-S analysis provides a naural accoun hese kinds of dependencies in erms of selecion beween and is siser modifier. By conras, under he nominal modifier heories (40a,b), he explanaion mus presumably be more complicaed A -Raising Analysis The paern of dependency seen above wih and a relaive clause modifier resembles ha noed earlier wih V and an adverbial modifiers in our idiom cases (44): (44) a. RC b. V AdvP This suggess a similar approach. Suppose we rea relaive clauses (and oher resricive modifiers in ) as a form of deerminer complemen, governed by our hemaic hierarchy for, and insaniaing a 10 Bach and Cooper (1978) offer a Monague Grammar-syle composiional semanics for he -S relaive clause synax ha assigns deerminer inerpreaions conaining a variable R for he meaning of he relaive. (i) a. λqλp x[[q(x) & R(x)] P(x)] b. λqλp x[[q(x) & R(x)] & P(x)] This represens, in effec, an Aricle-S analysis since he underlying composiion is beween he deerminer and he relaive clause. For more see secion

19 lower hemaic role han Θ RESTRICT. For concreeness, I will label his role "Θ RMO ", for resricive modifier: Θ SCOPE > Θ RESTRICT > Θ RMO Then he inclusion of a relaive clause modifier in a headed by a dyadic will resul in he minimal projecion being filled by he argumens expressing Θ RESTRICT and Θ RMO. This will rigger shell projecion o accommodae he scope argumen (Pro), and subsequen -raising. The resuling srucure (45) is parallel o he adverbial case discussed earlier (39b). (45) (39b) Pro V John V he me V way CP Bill V ha one should yeserday This proposal essenially resurrecs he Aricle-S analysis insofar as and he relaive clause form an underlying consiuen ha excludes he noun. The head-raising analysis can accommodae facs originally aken o argue for he -S and NOM- S analyses. Consider he example in (46a), for insance. The former displays apparen conjuncion of a consiuen ha includes he noun and relaive clause bu excludes he deerminer (46b). The accepabiliy of such examples can be aken o argue for he NOM-S analysis, as discussed by Baker (1978). (46) a. All sudens who voed for Clinon and faculy who voed for Pero showed up. b. All [[sudens who voed for Clinon] and [faculy who voed for Pero]] Under he head raising analysis proposed here, his example can be analyzed as a case of inner conjuncion, wih across-he-board movemen along he lines indicaed in (47). 19

20 (47) Pro 1 all and sudens 1 CP faculy 1 CP who voed for Clinon who voed for Pero On his view, examples like (46a) become analogous o cases of apparen non-consiuen coordinaion of objecs and modifiers in, such as (48a). In Larson (1988) hese are analyzed as inner conjuncions, wih across-he-board V movemen (48b): (48) a. Max me Bill yeserday and Sue Tuesday. b. [ V me [ [ Bill [ V [ yeserday ]] and [ Sue [ V [ Tuesday ]]]]] Consider also he example in (49a), which displays apparen conjuncion of a consiuen ha includes he deerminer and noun, bu excludes he relaive clause. Is accepabiliy can be aken o argue for he -S analysis. (49) a. All sudens and many faculy who voed for Clinon showed up. b. [[All sudens] and [many faculy]] who voed for Clinon Under he head raising analysis, his example can be analyzed as a case of ouer conjuncion, wih Righ Node Raising (RNR) of he relaive clause o he righ edge of. 11 The analysis of RNR is conroversial, 12 however if his operaion is viewed as across-he-board movemen of, hen he represenaion of (49a) is approximaely as in (50), where raising has occurred separaely in each of he conjoined s: 11 The analysis of examples like (48a) as RNR consrucions is firs proposed (o my knowledge) by McCawley (1981). 12 See McCawley (1982) and McCloskey (1986) for discussion. 20

21 (50) CP1 and who voed for Clinon Pro Pro all all sudens CP 1 faculy CP1 In shor, hen, sandard conjuncion facs aken o argue for nominal modifier accouns are also compaible wih he -raising view Some Semanics The analysis of relaives clauses as -complemens invies naural quesions abou he compuaion of meaning for a srucure like (45). In giving a semanics, I will adop he general proposal of Keenan and Savi (1984) ha deerminer + relaive clause combinaions comprise a form of complex. Specifically, I will rea relaive clauses as combining wih dyadic deerminers o form new, complex dyadic deerminers. This view is made concree in he rule (51); a sample applicaion is given in (52a-e) for he senence every boy ha swims jogs: r: (51) Le be a deerminer projecion denoing a deerminer relaion (X,,Y), where Y has he role Θ RESTRICT. Le CP be a relaive clause denoing he se R. If is, hen [ CP ] denoes he relaion (X,Y), where Y has he role Θ RESTRICT and (X,Y) iff (X,(Y R)). If is, hen [ CP ] denoes he relaion (X,Y), where Y has he role Θ RESTRICT and (X,Y) iff (X,(Y R)). (52) a. EVERY(X,Y) iff Y X = 0 b. EVERY (X,Y) iff (Y R ) X = 0 c. every ha swims (Y {r: swims(r)}) X = 0 d. every boy ha swims ({y: boy(y)} {r: swims(r)}) X = 0 e. every boy ha swims jogs ({y: boy(y)} {r: swims(r)}) {x: jogs(x) = 0 Basic every expresses a dyadic relaion EVERY beween wo ses X, and Y, where Y is he resricion (52a). (51) enails, in effec, ha combining every wih a relaive CP creaes he new dyadic deerminer EVERY, defined as in (52b), where he resricion argumen of EVERY is specified as he inersecion of he relaive clause denoaion R wih he original resricion of EVERY. The relaive clause supplies he 21

22 value of R (52c); aferwards he nominal resricion and scope argumens combine, respecively (52d,e). 13 Alhough non-sandard, his analysis of relaive clauses has preceden. Bach and Cooper (1978) propose a Monague Grammar semanics for relaives based on deerminer ranslaions like (53b); he laer may be compared o he more sandard Monague Grammar ranslaion in (53a) (which ignores inensions): 14 (53) a. every ==> λqλp x[q(x) P(x)] b. every ==> λqλp x[[q(x) & R(x)] P(x)] The crucial feaure of (53b) is inclusion of a disinguished variable R, whose value is supplied by a relaive clause, and whose denoaion is inerseced wih ha of he resricion se Q. In effec, Bach and Cooper (1978) offer an Aricle-S semanic analysis, composing relaive clause denoaions wih deerminer denoaions, analogously o wha is proposed here. Relaive Clauses as Argumens? I is ineresing o compare a represenaion like (53b) wih one like (54), in which he resricion variable R is no only presen in he inerpreaion of, bu is also absraced over: (54) every ==> λrλqλp x[[q(x) & R(x)] P(x)] (53b) represens every as a binary deerminer ha always conains a resricion R on is quanificaional domain. The value or R is presumably deermined by conex or by an overly occurring resricion phrase, such as a relaive. By conras, (54) analyzes every as a rue ernary deerminer, which requires an addiional synacic resricor argumen o yield a binary. The correc analysis of a given as in (53b) or (54) is presumably a maer of wheher he deerminer in quesion genuinely requires a synacic resricor elemen. Some ineresing observaions by Vendler (1967) sugges ha inerpreaions like (54) may be jusified for definie deerminers. Consider (55) and (56), based on Vendler s examples. (55) a. I see a man. The man wears a ha. b. I see a man. The man I see wears a ha. c. I see a man. The man you know wears a ha. 13 A more radiional Monague Grammar version of (51) employing Monague s IL is given below, where (following owy e al (1981), he caegorial definiion of deerminers is T/CN, and where RC is he caegory of wh-relaive clauses: S RC If δ P T/CN and φ P RC, hen F 1000,n (δ,φ) P T/CN, where F 1000,n (δ,φ) = δ^φ T RC If δ P T/CN and φ P RC, and δ and φ ranslae ino δ and φ, respecively, hen F 1000,n (δ,φ) ranslaes ino λq[δ (^λx n [Q{x n } & φ ])] 14 Larson (1982) exends Bach and Cooper s accoun of deerminers and relaive clauses o he relaion beween enses and emporal adverbial clauses in analyzing cerain readings of Warlpiri adjoined relaives. 22

23 (56) a. I see a rose. The rose is lovely. b. I see a rose. The rose I see is lovely. c. I see a rose. The red rose is lovely. 15 (55a) conains a bare definie descripion ha is naurally undersood along he lines of (55b). Boh examples presen discourse ha is "coninuous" in Vendler s erms: he individual inroduced by he indefinie is undersood as he same one picked up by he definie. Ineresingly, as Vendler poins ou, (55c) is no coninuous in he same sense. The individuals picked ou wih he definie and indefinie are no naurally undersood as he same. The difference appears o be induced by he relaive clause you know in he second clause. Analogous poins apply o (56). Vendler inerpres hese resuls as supporing he view ha "he definie aricle in fron of a noun is always and infallibly he sign of a resricive adjunc, presen or recoverable..."(p.46) 16 In modern erms, a definie selecs a resricive modifier. (55a) is analyzed as conaining an ellipical or "deleed" relaive clause equivalen o (55b) allowing coninuiy. By conras, in (55c) he over relaive in effec "sauraes" he relaive clause required by he, hence (55c) canno be undersood equivalenly o (55b), hence no coninuiy. 17 The coninuiy phenomenon disinguishes he from oher quanifiers. Observe ha alhough (57a) is naurally read as coninuous, wih he linguiss referring o he linguiss I me, his is no rue in (57b). Mos linguiss is no naurally read as referring o mos linguiss ha I me. To obain his inerpreaion, an explici definie is required (57c): 18 (57) a. I me some linguiss. The linguiss were educaed in California. b. I me some linguiss. Mos linguiss were educaed in California. c. I me some linguiss. Mos of he linguiss were educaed in California. These observaions are very naurally inerpreed in our erms by saying ha whereas oher deerminers combine wih resricive modifiers via he rule in (50), he definie deerminer acually selecs a resricive modifier as argumen, as par of is basic lexical semanics. Tha is, he should be inerpreed via he relaion in (58a), which replaces our earlier (8g); (58b) is he equivalen in a convenional Monague- Grammar syle noaion: 15 The adjecive red is his example is o be undersood resricively. Under a nonresricive reading, he discourse can be coninuous. 16 Vendler s senence acually coninues "...aached o he noun." bu he seems o mean his only in he general sense of accompanying he noun, raher han as a definie synacic proposal of -RC consiuency. In any case, his observaions are mos compaible wih he view expressed in he ex. 17 Vendler saes: "[(54a)] is coninuous. The is he sign of he deleed clause (whom) I see. In (54c), he possibiliy of his clause is precluded by he presence of he acual clause (whom) you know. The in [(53c)] belongs o his clause and any furher resricive clauses are excluded. Consequenly, here is no reason o hink ha he man you know is he same as he man I see." (p.53) 18 Noe he alhough every differs from mos in allowing is resricion o be undersood via a preceding clause, i sill seems o differ from he insofar as an over relaive doesn produce disconinuiy. Thus i appears possible o undersand Every linguis you know in (ib) as referring o every linguis whom you know and whom I me: (i) a. I me some linguiss. Every linguis was educaed in California. c. I me some linguiss. Every linguis you know was educaed in California. To my knowledge, such "Vendler effecs" wih quanifiers have no been explored sysemaically in he lieraure. 23

24 (58) a. THE(X,Y,R) iff (Y R) - X = 0, where (Y R) = 1 b. he ==> λrλqλp y x[[[q(y) & R(y)] y = x] & P(x)] Thus Vendler s view ha he always occurs wih a resricive modifier, over or cover, fis in well wih he noion of relaive clauses as -complemens. 19 Ordering of Relaive Clauses and Excepion-phrases The analysis of relaives offered here yields an accoun of he ordering of relaive clauses in relaion o oher oblique elemens. Consider he facs in (59a,b), which show ha he excepion-phrase mus occur righ-mos (under a normal inonaion for he ). The reverse order is awkward o unaccepable. Since righ-mos phrases are lower and combine earlier wih in his framework, he srongly preferred order in (59a) suggess ha he excepion-phrase should projec lower and combine wih every before he wo combine wih he relaive (59c). (59) a. Every boy ha you saw excep John b.??every boy excep John ha you saw c. [ Pro [ every [ boy [ [ ha you saw [ [ PP excep Bill ]]]]]]] This resul follows under he view ha every-excep is a riadic deerminer, as proposed in (32)-(33). This enails ha every-excep mus firs combine wih is hird argumen before i can be subjec o he rule in (51), which combines relaives only wih s denoing dyadic deerminer relaions. Every-excep does no become dyadic unil afer is hird argumen is sauraed. So he excepion phrase mus combine firs, i.e., a lowes poin in he ree. There is one ineresing qualificaion of he facs in (59) concerning heaviness effecs. Consider he dialogue in (60), where he relaive clause is given heavy sress o convey conrasive emphasis; or consider he example in (61), where he relaive is quie long: 20 (60) A: Wha was every boy excep John wearing? B: Well, every boy excep John THAT I SAW was wearing a kil. (61) Every boy excep John [ha was roaming he highlands of Scoland] was wearing a kil In boh cases, he oherwise disfavored order: excepion-phrase followed by relaive clause seems o improve. 19 In a very general sense, he definie deerminer migh be viewed as analogous o a verb like word, which appears o ake a manner adverbial as is complemen, whereas oher verbs of similar meaning are merely compaible wih such a modifier: (i) a. John worded he leer *(carefully). b. John wroe he leer (carefully). In a similar way, he akes a resricive modifier as complemen, whereas oher deerminers are merely compaible wih such a modifier. 20 I am graeful o C. de Cuba for his example. 24

25 The view ha I wish o suppor, ulimaely, is ha (59a,b) represen he rue relaionship beween he relaive and he excepion phrase, and ha (60)-(61) represen forms derived by movemen from he equivalen of (59a). To moivae his proposal, we examine a broader, parallel phenomenon ha arises in he conex of muliple relaive clauses. 3.3 Muliple Relaives Muliple relaives receive very differen represenaions in he nominal modifier analyses versus he - raising accoun proposed here. On he former, examples like (62a) sack upward o he righ (62b), whereas on he laer hey branch downward (62c): (62) a. The woman who I liked who I invied came o he pary. b. c. Pro he CP 1 CP who I invied he woman who I liked woman 1 CP who I liked 1 CP who I invied Noe an apparen difference of scope wih respec o he wo relaives. In he former who I liked is he firs resricor of he nominal, wih who I invied resricing he resul. In he laer, who I invied resrics he nominal firs (according o our semanics), wih who I liked following. Under a sandard semanics using se-inersecion, such as he one adoped here, differences of relaive scope wih relaive clauses are no ruh-condiionally significan. This is because se-inersecion is a commuaive and associaive operaion. If we have a noun se N, and wo relaive clause ses R1 and R2, hen he resul of inersecing N wih R1 and he laer wih R2, will always be he same as he resul of inersecing N wih R2 and he laer wih R1: (63) (N R1) R2 = (N R2) R1 In many cases muliple relaives do no appear o show relaive scope, so ha invering he order of muliple relaives carries no difference in meaning. For example, when definie descripions like (64a,b) 25

26 are used referenially 21 and read wih neural inonaion, permuaion of relaives has no apparen effec; boh pick ou he same individual: (64) a. he man [ha you me] [ha I alked o] b. he man [ha I alked o] [ha you me] In oher cases, however, paricularly when emphasis is added, here does appear o be a scopal difference, even if no one ha is ruh-condiionally significan. Consider (65a,b), where capializaion indicaes sress or emphasis: (65) a. Every woman who enjoys books WHO REAS SHAKESPEARE (will like his) b. Every woman who reads Shakespeare WHO ENJOYS BOOKS (will like his) (65a) is naurally undersood as saying somehing like his: "among women who enjoy books, every one who reads Shakespeare will like his." By conras, (65b) conveys: "among women who read Shakespeare, every one who enjoys books will like his." The force of his difference is clear in quesionanswering conexs like (66), where he quesioner ses up he domain as women-who-enjoy-books, and hence he answerer mus qualify his same domain: (66) A: Which women who enjoy books will like his? B: Every woman who enjoys books WHO REAS SHAKESPEARE will like his! #Every woman who reads Shakespeare WHO ENJOYS BOOKS will like his! Summarizing informally, woman who enjoys books is perceived as a semanic consiuen in (65a), and woman who reads Shakespeare is perceived as a consiuen in (65b). When hese resuls are mached up agains he srucures in (62), he sacking heories of relaive clauses appear o be favored. (62b) seems o projec he correc consiuency relaions whereas (62c) does no. In fac, however, I hink ha an appropriae srucure can be assigned under he -raising, heory once anoher parallelism beween and is acknowledged Ligh Predicae Raising in Pairs like (67a,b), involving permuaion of an objec and oher maerial, have been widely analyzed in he lieraure as he produc of a movemen operaion ha shifs he objec righward (67c), adjoining i a he edge of. This operaion ypically involves phrases ha are phonologically "heavy" in relaion o he maerial shifed over, hence i is referred o as Heavy Shif": (67) a. John gave a picure of Mary o Bill. b. John gave o Bill a picure of Mary. c. John [ gave [ PP o Bill]] [ a picure of Mary] 21 The erm "referenial use" is adaped from Ludlow and Neale (1991), where i refers o he use of a descripion in which he speaker has singular grounds for his/her asserion, and where he proposiion ha he/she inends o convey is also singular. See Ludlow and Neale (1991) for deails. 26

27 Larson (1988) proposes an alernaive analysis of his phenomenon using shell srucures. The basic idea is ha examples like (67b) are no insances of righward movemen of a heavy nominal, bu raher lefward movemen of a ligh predicae. Accordingly, he phenomenon is rechrisened "Ligh Predicae Raising" (LPR). The key ingredien of his accoun is a reanalysis rule ha permis hemaically ransiive phrases - XPs wih wo unassigned hemaic roles - o be caegorially reanalyzed as X 0 s. Reanalysis allows he enire ransiive phrase o undergo head-raising. To illusrae, consider (68a), which is similar o our earlier (17a), bu conains a heavy objec all he sal he had. The senence receives he underlying shell srucure in (68b). Since he verb pu is diransiive, when i combines wih a locaion phrase he resuling V (pu on he fish) is hemaically ransiive, wih he wo hemaic roles Q AGENT and Q THEME unassigned. This enails ha V can undergo V -Reanalysis, as shown in (65c). Once pu on he fish is reanalyzed as a head, his elemen can raise around he objec, resuling in a righ-peripheral posiion for he objec (68d): 22 (68) a. John pu all he sal he had on he fish. b. [ John [ V e [ he sal he had [ V pu on he fish]]]]] c. d. V V -Reanalysis John V e V all he sal he had V PP pu on he fish V John V V PP V pu on he fish all he sal he had Alhough he exac naure of V reanalysis is somewha unclear in Larson (1988), one propery i clearly mus have is ha i does no yield X 0 s ha are opaque o furher synacic rules, including furher head raising. This is clear from need for V o raise ou of V / V 0 for inflecion in examples like John was puing on he fish all he sal he had. The progressive verb form (puing) mus assume a local relaion wih 22 Reanalysis is conceived as a consequence of he he X-bar heory in (13), which creaes an associaion beween he noions of ransiive predicae and head. The idea is ha phrases which are hemaic ransiives (i.e., having wo unassigned Θ-roles) can be reanalyzed as srucural ransiives (i.e., X 0 s). The Ligh Predicae Raising analysis is explored in deail in Larson (1989). 27

28 progressive be for agreemen. 23 V Reanalysis mus herefore be viewed as producing X 0 s only in he limied sense of iems ha can be reaed as unprojeced. Reanalysis clearly does no produce a morphological "word" Ligh Predicae Raising in? Alhough reanalysis and LPR were originally inroduced in he conex of, we have seen ha noions like monadic/inransiive and dyadic/ransiive can be carried over from o. A ransiive V-predicae is one wih Q-roles like Q AGENT and Q THEME o assign. A ransiive -predicae is one bearing Q-roles like Q SCOPE and Q RESTRICT. Accordingly, here seems o be no barrier o a more general noion of X - Reanalysis, allowing any ransiive X o reanalyze as X 0. To illusrae his exension, consider again our muliple relaive example every woman who enjoys books WHO REAS SHAKESPEARE. Suppose his example has he underlying form in (69a), in which he ouer relaive is projeced higher han he inner one. (69) a. b. Pro e woman e CP -Reanalysis WHO REAS SHAKESPEARE CP CP every who enjoys books every who enjoys books Every denoes a binary deerminer relaion and, under our semanics (51), so does he resul of combining every wih a relaive clause. Thus every who enjoys books also denoes a binary deerminer relaion. Since [ every who enjoys books] is hemaically ransiive, i is subjec o reanalysis as a head (69b). This allows i o raise as a uni around he CP o is lef (70): 23 The poin is even clearer in a pair like (ia,b). In Larson (1989), presenaional here examples like (ia) are analyzed as deriving hrough LPR: (i) a. There [ V /V was in he room] a all, dark sranger. b. Was here [ V /V in he room] a all, dark sranger? The froning of he verb in (ib) indicaes ha V (be) mus be able o raise ou of a reanalyzed V even in surface form. 28

29 (70) Pro e woman CP CP every who enjoys books WHO REAS SHAKESPEARE Finally, he head raises ou of he reanalyzed porion o he higher empy posiion (71), yielding he desired surface order every woman who enjoys books WHO REAS SHAKESPEARE: (71) Pro every woman CP CP who enjoys books WHO REAS SHAKESPEARE The same general derivaion will accoun for our earlier examples (60) and (61) (repeaed below) in which he order of he relaive clause and excepion phrase is opposie o wha we expec: (60) A: Wha was every boy excep John wearing? B: Well, every boy excep John THAT I SAW was wearing a kil. (61) Every boy excep John [ha was roaming he highlands of Scoland] was wearing a kil These examples can be analyzed as deriving by reanalysis of he lower conaining he deerminer + he excepion phrase. This complex is hen raised, wih he deerminer laer raising on is own (72): 29

30 (72) Pro every boy PP CP excep John THAT I SAW Prenominal Relaives I is naural o ask wha blocks he raising of he enire reanalyzed ino prenominal posiion, producing he ungrammaical (73). In fac, he same quesion arises wih he simples examples of relaives (74a). -reanalysis migh lead us o expec he ungrammaical (74b), where a reanalyzed raises around (74c): (73) *every who enjoys books woman WHO REAS SHAKESPEARE. (74) a. every woman who enjoys books b. *every who enjoys books woman c. [ Pro [ every who enjoys books] [ [ woman] ]] Examples like (73) and (74b) are plausibly ruled ou by he same general consrain on prenominal iems observed in cases like (75) and (76). I has been widely observed ha prenominal modifiers mus ypically occur head-adjacen o he nouns hey modify. Complemens of prenominal adjecives are hus excluded since hey preven adjacency (75ai-ci). Eiher he complemen mus be "exraposed" righward (75aii-cii) or else he enire Adj+complemen mus occur posnominally (75aiii-ciii). Similar remarks apply o he PPs in (76): 24 (75) a. i. *a similar o Bill man ii. a similar man o Bill iii. a man similar o Bill 24 Apparen prenominal PPs like (ia-c) are plausibly analyzed as some form of compound formaion, whose surface head-iniial srucure is no visible o he synax: (i) a. an under he couner deal b. his over he couner medicaion c. hree off he wall ideas 30

31 b. i. *a fun for children game ii. a fun game for children iii. a game fun for children c. i. *an unforunae for Max complicaion ii. an unforunae complicaion for Max iii. a complicaion unforunae for Max (76) a. i. *An a wo o clock meeing ii. a meeing a wo o clock b. i. *a nearby he park resauran ii. a nearby resauran (*he park) iii. a resauran nearby (he park) Relaive clauses are widely analyzed as CPs, wih heads iniial in heir phrase in a language like English (77a). Consider hen a raised, reanalyzed srucure like (77b). If he whole complex is considered as he prenominal elemen, hen i will clearly fail he requiremen of head-adjacency: is headed by every and he laer is no adjacen o woman. Similarly, if he relaive CP is considered as he prenominal elemen, i will also violae he requiremen since C is no of head-adjacen o woman: (77) a. [ CP who C [ IP enjoys books]] b. [... [ every [ CP who C enjoys books]] woman ] I follows, hen, ha raising of a complex ino prenominal posiion will always be excluded, and hence sranding derivaions of he kind in (71) and (72) will be required. 3.4 Oher -Modifiers The accoun of relaive clauses proposed above can be exended o oher caegories of posnominal nominal modifiers, including posnominal PPs and APs like hose illusraed in (78): There is an ineresing, well known quesions as o wheher posnominal PPs and APs are uniformly bare caegories or conain "hidden" relaive clause srucure, a leas in cerain insances (i): (i) a. he books prined (cf. he books ha have been prined) b. he books in prin (cf. he books ha have been in prin) One poenial consideraion (noed by Hudson 1973) is he fac ha hese elemens accep PP modifiers whose presence oherwise requires an aspecual verb (ii)-(iii): (ii) a. he books prined since 1980 b. he books in prin since 1980 (iii) a. *?hese books were prined/in prin since b. hese books have been prined/in prin since The need for have ilusraed in (iii) argues for he presence of a concealed have in (ii), hence a concealed relaive. 31

32 (78) a. he man [ PP a he podium] [ PP in a grey sui] b. hree women [ AP presen] [ AP capable of lifing a sofa] c. every book [ PP on he shelf] [ AP published since WWII ] These can be analyzed as projeced ino low posiions as complemens of, wih subsequen raising of he deerminer (79): (79) [ Pro [ every [ book [ [ [ PP on he shelf] [ [ AP published since WWII ]]]]]]] The inersecive semanics for relaive clauses can likewise be direcly exended o hese caegories. We simply generalize our rule o cover all predicaive XPs of his kind: (51 ) Le be a deerminer projecion denoing a deerminer relaion (X,,Y), where Y has he role Θ RESTRICT. Le XP be an AP, PP or relaive CP denoing he se R. If is, hen [ XP ] denoes he relaion (X,Y), where Y has he role Θ RESTRICT and (X,Y) iff (X,(Y R)). If is, hen [ CP ] denoes he relaion (X,Y), where Y has he role Θ RESTRICT and (X,Y) iff (X,(Y R)) Prenominal APs These resuls raise an ineresing quesion as o how pre-nominal adjecive modifiers should be analyzed. The adjecives in (80a-c) combine wih heir nominal in a way equivalen o ha of a relaive clause. All involve an inersecive semanics: (80) a. The all woman (cf. he woman who is all) b. Every beauiful house (cf. every house ha is beauiful) c. Three blind mice (cf. hree mice ha are blind) Under he general analysis pursued here, here appear o be few opions. If we aemp o rea prenominal adjecives as base-generaed in he -projecion, along he lines of (81), hen we mus analyze hem as iems ha can be combined wih beween he scope and resricion argumens: 32

33 (81) Pro 1 hree AP blind 1 mice Achieving his is no sraighforward, however. A crucial elemen in our approach o inersecive - modifiers is ha hey inerac semanically wih ransiive deerminer relaions - in essence, hey ake ransiive s and form larger, complex ransiive s (recall (51) and (51 ) above). This in urn requires ha he resricion phrase () no be combined wih a he poin where he modifer is added in. Evidenly, his requiremen is no me in (81); here has already combined wih a he poin where AP is encounered. If we canno uilize he equivalen of (51) or (51 ), he only obvious alernaive for generaing (81) is o rea prenominal APs as argumens of in heir own righ, assigning hem a hemaic role lying beween Θ SCOPE and Θ RESTRICT : Θ SCOPE > Θ X > Θ RESTRICT Bu his move is also problemaic. Prenominal adjecives are opional elemens, and alhough here is no problem making hem argumens of like relaive clauses, i does seem quie srange o locae an opionally assigned hemaic role (Θ X ) beween wo obligaorily assigned hemaic roles (Θ SCOPE, Θ RESTRICT ). Furhermore, i is well-known ha prenominal adjecives are ierable, so ha we can ge a number of such elemens ogeher (82): (82) a. hree German mice b. hree blind German mice c. hree grey blind German mice d. hree furry grey blind German mice e. hree small furry grey blind German mice f. hree excellen small furry grey blind German mice On he approach being considered, his would seem o enail expanding he hemaic heirarchy o include a number of roles beween Θ SCOPE and Θ RESTRICT, all of which mus be opional: 33

34 Θ SCOPE > Θ X1 > Θ X2 > Θ X3 > Θ X4 > Θ X5 > Θ X6 > Θ RESTRICT Noe ha none of hese issues arises in our approach o relaive clauses and oher pos-nominal resricive modifiers, such as PP and AP. The laer were no analysed as argumens of, and no par of he nominal hemaic hierarchy, bu raher as elemens ha were (recursively) added in by a process forming complex s. If a base-generaed approach o prenominal adjecives is problemaic, an aracive alernaive is o adop some version of he proposal by Smih (1964) and Jacobs and Rosenbaum (1968) ha prenominal adjecives originae as posnominal modifiers, and obain heir surface posiion by movemen (83): (83) Pro e mice AP hree blind Under his approach, he specific problems raised above for (81) disappear. However, a leas wo new issues arise. Firs, we require an accoun of he precise mechanism by which adjecives generaed posnominally are advanced o prenominal posiion. This accoun mus accommodae he familiar fac ha adjecives in prenominal posiion appear o obey cerain (universal) resricions on order of occurrence, which, for example, rule ou combinaions like hose in (84) (uered wih neural inonaion) (ixon 1977; Hezron 1978, Sproa, R. and C. Shih 1991): (84) a.?*hree blind small mice b.?*hree grey small blind mice c.??hree small blind furry grey mice d. *hree German small furry grey blind mice e.?*hree furry excellen small mice Second, such an analysis mus deal wih he fac ha no all prenominal adjecives have he inersecive semanics found wih relaive clauses, posnominal PPs and APs. Cases like (85)-(87) are familiar examples: 34

35 (85) a. Olga is an alleged dancer. (cf. * Olga is an dancer who is alleged) b. Alice is an imagined werewolf. (cf. Alice is a werewolf who is imagined.) c. Boris is a supposed perperaor of a crime. (cf. *Boris is a perperaor of a crime who is supposed.) (86) a. Olga was a relucan dancer. (cf. Olga is an dancer who is relucan.) b. Boris was a willing perperaor of a crime. (cf. Boris was a perperaor of a crime who was willing.) (87) a. Olga is a beauiful dancer. (cf. Olga is a dancer who is beauiful.) b. Kahrin is a skillful manager. (cf. Kahrin is a manager who is skillful.) c. Peer is an old friend. (cf. Peer is a friend who is old.) I believe ha boh of hese issues can be deal wih saisfacorily, and he piciure in (83) mainained; however jusifying his claim would require exensive addiional discussion, which I pu aside for developmen elsewhere. 4.0 Geniives The posulaion of a Pro subjec in all s has srong consequences for he analysis of prenominal geniive consrucions like (88a-d): (88) a. John s briefcase b. John s picure c. John s grandmoher d. John s compleion of he plan As noed earlier, Abney (1987) assimilaes he srucure of geniive s o clauses (IPs), wih he possessive elemen occupying a subjec -like posiion; recall (10a,b) (repeaed below): (10) a. b. IP I John John I s compleion of he plan [TNS] complee he plan 35

36 Szabolsci (1983) furher develops he clausal analogy wih Hungarian examples like (89), in which he possessive iem co-occurs wih a definie aricle. Szabolsci analyzes he laer as counerpar o a complemenizer; compare (90a,b): (89) (a) Mari kalap-ja-i (he) Mari ha-poss-pl-2sg Mari s has (90) a. b. CP Spec Spec C (N+I)P C IP (a) (N+I) ha I Mari-NOM kalap-ja-i [±poss] [(AGR)] Mary-NOM ran [±ense] [(AGR)] The analysis of srucure developed here does no compor wih he basic senenial analogy. As we have seen, on he curren accoun he highes argumen posiion in a quanified - is hemaic "subjec" - is always he scope argumen Pro. The possessive iem herefore canno be srucurally parallel o a subjec, and, by exension, he Hungarian definie deerminer canno be parallel o a complemenizer. In place of he senenial picure, a raher differen analogy suggess iself. 4.1 Possessive s as Triadic Predicaes Geniive s are familiar as definie nominals (McCawley 1988, Neale 1990). Suppose we view Hungarian as displaying he "rue shape" of he geniive, where he head is a definie deerminer, and where he geniive-marked possessor occurs below he definie. As a firs approximaion, we migh propose he analysis in (91) for John s briefcase, where he possessor (John) and possessed (briefcase) are boh argumens of a definie deerminer (THE) which raises: (91) a. [ Pro e [ John s [ THE briefcase]]] b. [ Pro THE [ John s [ briefcase]]] In Hungarian his definie deerminer would be phoneically realizable, whereas in English (as in many oher languages) i would be necessarily cover. Noice ha on his proposal, geniive nominals become a form of riadic consrucion, in which he wo lower argumens of (John and briefcase) sand in a possessive relaion. This siuaion is 36

37 ineresing given he general parallelism beween and ha arises in his heory. Larson (1988, 1990a, 1991) proposes an analysis very similar o (91) for a class of riadic consrucions in which he wo lower argumens of V sand in a possessive relaion: namely, double objec srucures. An example like Mary gave John a briefcase, for insance, ges an analysis approximaely as in (92a,b): (92) a. [ Mary e [ John [ V gave a briefcase]]] b. [ Mary gave [ John [ V a briefcase]]] Wha hese poins sugges, hen, is ha raher han viewing geniive nominals as clause-like, wih he possessor analogous o a subjec, and he definie deerminer parallel o C, we migh insead see hem as -like, wih he possessor analogous o an objec, and he definie parallel o V. Specifically, we migh analyze geniive nominals as he -equivalens of double objec consrucions in he verbal domain Preposiional aives and "aive Shif" in The analogy beween prenominal geniives and double objec consrucions can be developed furher, hrough a more refined view of he laer. Larson (1988, 1990a, 1991) proposes ha preposiional daives like (93a) involve a relaively ransparen source, in which he goal argumen is projeced lower han he agen and heme (in accordance wih he hemaic hierarchy in (14)), and where he daive verb raises (93b,c): (93) a. Mary gave a briefcase o John b. c. V V Mary V Mary V 1 e V gave V a briefcase V PP a briefcase V 1 PP gave P P o John o John By conras, double objec consrucions have a more complex derivaion, which involves a modern version of "daive shif". An example like (94a) is assigned he underlying in (94b), where he goal (John) is iniially projeced ino a low posiion, wihou he preposiion o ha would normally accompany i, and where he heme is a V-bar adjunc. 26 Absence of he case-marking provided by o riggers - 26 In Larson (1988) i is proposed ha he daive preposiion (o) normally accompanying he goal phrase is "absorbed" by give in he double objec consrucion, equivalenly o he case-absorpion ha occurs in a passive. Lack of case- 37

38 movemen. The resul is (94c) (where he verb has also raised o he higher V posiion). (94) a. Mary gave John a briefcase b. c. V V Mary V Mary V e V gave 1 V e V John V V a briefcase V 1 a briefcase gave John As discussed in Larson (1988, 1990a, 1991), his accoun respecs a srong heory of projecion, in which he hemaic hierarchy is direcly refleced in he relaive heighs of argumens. Hence he he goal argumen (John) sars ou lower han he heme argumen (a briefcase) in iniial srucure. Bu i also allows for he imporan observaion (due o Barss and Lasnik 1986), ha in a double objec consrucion he goal argumen appears o c-command he heme argumen a surface form. This resul is achieved by raising he goal o he higher posiion Posnominal Geniives and "Geniive Shif" in This analysis of preposiional daives, double objec srucures, and heir relaionships, can be exended direcly o posnominal and prenominal geniives, following he basic analogy suggesed above. Posnominal geniive consrucions like (95a) can be assigned a relaively ransparen iniial srucure as in (95b), where he geniive PP is reaed as an oblique modifier, and projeced lower han he scope and resricion argumens of, in accordance wih he hierarchy discussed earlier. The definie deerminer subsequenly raises (93c) yielding he correc surface order. 27 marking on he goal riggers -movemen. Afer raising of give, he lower V reanalyzes as V, yielding a srucure equivalen (up o he layered V-race) o (i): (i) [ Mary gave [ John [ V a briefcase ]]] 27 Posnominal geniives wih a definie deerminer, like ha in (95a), have he ofen-noed propery of requiring a resricive modifier (relaive clause, PP or posnominal adjecive) in order o occur smoohly (Lyons 1986). On an accoun where relaive clauses are argumens of definie s (recall secion 3.2.2), his migh be undersood as follows: whereas definie ypically allows he deicic deerminaion of is resricive argumen, he presence of he posnominal geniive blocks his possibiliy, forcing srucural realizaion. This proposal appears sensible given he semanics for hemaic geniives adoped below, wih he definie in geniives conains is own relaional variable R, whose value is deicically or srucurally deermined. In essence, srucural (non-deicic) deerminaion of R forces srucural (non-deicic) deerminaion of he resricive argumen as well. This leaves open he quesion of why prenominal geniives show he opposie requiremen: why an over resricive elemen is blocked in he laer case (*John s briefcase ha Alice los). I have no clear proposal o make a his poin. 38

39 (95) a. he briefcase of John s (ha Alice los/on he desk/aken) b. c. Pro Pro 1 e he briefcase PP briefcase 1 PP he P P of John s of John s Posnominal geniives hus become he -equivalens of he preposiional daive consrucions. By conras, prenominal geniives receive a more complex movemen derivaion, involving wha we migh call "geniive shif". Example (96a) is assigned he underlying in (96b), involving he cover definie deerminer THE. Here again, he possessor (John s) is projeced ino a low posiion, bu now wihou he preposiion of; he resricion argumen (briefcase) is projeced as a higher -bar adjunc. Absence of he case-marking provided by of once again riggers movemen. Finally, THE raises o he higher posiion, yielding (96c): (96) a. John s briefcase b. c. Pro Pro e THE 1 e John s briefcase 1 briefcase THE John s As in he double objec derivaion, his accoun mainains a sric mapping beween he hemaic hierarchy and he relaive srucural heigh of argumens. Thus he possessor argumen (John s) is projeced lower han he resricion argumen (briefcase) in boh he prenominal and posnominal consrucions. Bu his accoun also allows for he fac ha in he prenominal geniive consrucion, he possessor c-commands 39

40 he resricion argumen a surface form, as shown by examples like (97a,b) involving negaive polariy iems and anaphors. This resul is achieved by raising he possessor o he higher -spec posiion. (97) a. No one s picure of anyhing (cf. *Anyone s picures of no one) b. Their picures of each oher (cf. *Each oher s picures of hem) 4.2 Consequences and Comparisons The analysis skeched above enails ha prenominal geniives always achieve heir surface posiion by movemen, and ha he geniive always originaes as an (oblique) argumen of. Le us examine hese poins more carefully, considering wo basic classes of prenominal geniives idenified in he lieraure: so-called "lexical", or non-hemaic geniives, in which s plainly does no bear a hea-role assigned by N, and hemaic geniives, in which s a leas appears o bear a role assigned by N Non-Themaic Geniives Non-hemaic geniives include examples like (98a-d), where, in each case, he only hemaic role assigned by N (briefcase, arm, accessories, afernoon) is he usual one going o is exernal argumen. As many have noed, he exac relaion beween possessor and possessed is ypically vague in hese cases, and no confined o ownership. Thus John s briefcase can refer o one he owns, bu i can also refer o one near him, one he was alking abou, ec. Similarly Mary s arm can refer o her own limb, bu also one she is holding on o, or one lying before her on a dissecion able. And so on. (98) a. John s briefcase (is on he veranda) b. Mary s arm (is anned) c. Men s accessories (are in he nex aisle) d. Jill s afernoon (was hecic) On he presen accoun, examples like (98a-d) derive by movemen from he posiion of a posnominal geniive of-pp (99): (99) a. [ he briefcase of John s ] b. [ THE John s briefcase ] A movemen accoun of non-hemaic prenominal geniives is no new. Ross (1967,1981), Chomsky (1970), Sockwell, Schacer and Paree (1970), and McCawley (1988) all offer analyses ha include he equivalen of (99a,b) a some derivaional sage. For example, McCawley (1988) proposes he seps in (100a-c), where he geniive originaes as he predicae of a copular relaive clause (100a), ha is subsequenly reduced (100b), and where he geniive eiher combines wih of (100c.i), or frons o prenominal posiion (100c.ii). Sockwell, Schacer and Paree (1970) observe ha he predicae geniives 40

41 like (100a) show essenially he same range of readings found in (non-hemaic) prenominal geniives, hence a derivaional relaion beween hem seems semanically sound: (100) a. he briefcase [which is John s] b. he briefcase [John s] (from (100a) by Relaive Clause Reducion) c.i. he briefcase [of John s] (from (100b) by of-inserion) ii. John s briefcase (from (100b) by froning - s) McCawley s analysis anicipaes he one proposed here on several imporan poins. In McCawley s accoun, as in ours, he prenominal geniive derives by froning from he posnominal posiion of a geniive of-pp. Furhermore, for McCawley posnominal geniives occupy he same posiion as possessive relaives, a parallelism ha also holds in he our accoun, where relaive clauses (including possessive relaives) and possessive PPs like of John s are analyzed as oblique -argumens. The main divergence beween he proposals is he assumpion ha non-hemaic prenominal geniives lierally derive from possessive relaives. Tha apar, he analyses are very similar. I is ineresing o noe in his conex ha many languages show a formal similariy in he marking of relaive clause and geniive consrucions. The Ausralian languages yirbal and Gumbaingar, discussed by ixon (1966) illusrae his phenomenon. As (101a,b) illusrae, he nu-suffix appearing on he verb in yirbal relaives (101a) also occurs in yirbal geniives (101b): (101) a. yibi yara-ngu njalnga-ngu djilwa -nu -ru bura-n woman-nom man-erg child-erg kick -REL -ERG see-tns The man who had been kicked by he child saw he woman b. njalnga guda-ngu yara -nu ndjin-du badja-n child-nom dog-erg man -REL ERG bie-tns The man s dog bi he child ixon (1966) argues ha his fac is no coincidenal: ha yirbal and Gumbaingar possessive nominals acually derive from possessive relaives, and ha heir shared morphology reflecs his shared derivaional hisory. ixon s resuls (and he general convergence beween relaive clauses and geniives) also appear compaible wih he weaker proposal made here, ha geniive s are generaed in he same posiion as relaives (wihou being lierally derived from he laer) Anoher claim of he presen analysis is ha prenominal geniive s occupy an (indirec) objec-like posiion in. Consider he fac ha for many speakers (including myself) he verb award allows o-daives, double objecs, and wih-pps. In he laer, wih seems o be associaed wih he possession relaion holding beween Mary and he prize: (i) a. John awarded he grand prize o Mary. b. John awarded Mary he grand prize. c. John awarded Mary wih he grand prize. Rohsein (1988) observes ha wih-pps also appear o play a specifically possessive role wihin. She noes ha alhough (iia,b) look superficially similar, he locaive-pp can be paraphrased wih a copular relaive clause, bu he wih--pp requires a possessive relaive (iii): (ii) a. The plae on he able b. The plae wih he gold rim (iii) a. The plae ha is on he able b. The plae ha has he gold rim (cf. *The plae ha is wih he gold rim) 41

42 4.2.2 Themaic Geniives The "geniive shif" analysis appears more problemaic for hemaic geniives like hose in (102) and (103), where he possessive-marked appears o bear a hemaic role assigned by N. Thus, in (102), John seems o receive an agen role from N; and in (103) John appears o receive a heme role from N (a leas on one reading). (102) a. John s examinaion of he plan (cf. John compleed he plan.) b. John s selecion of he winner (cf. John seleced he winner.) (103) a. John s elecion (cf. They eleced John.) b. John s grandmoher (cf. The grandmoher of John) c. John s picure (cf. A picure of John) Such facs naurally sugges analyses in which John( s) is an underlying argumen of N. For example, Chomsky (1970, 1981), Anderson (1983/84), Kayne (1984) and Giorgi and Longobardi (1990) (among ohers) ake he possessive s in (102) o be base-generaed in he subjec posiion of he nominal, parallel o he subjec posiion of a clause (104). (103a-c) are aken o derive by movemen of he heme argumen of N o subjec posiion, much like wha occurs in a clausal passive (105). 29 (104) a. [John s selecion of he winner] b. [John seleced he winner] (105) a. [ he elecion of John ] b. [ John s elecion ] c. John was eleced Given his, i is ineresing o observe cerain possession/posiion alernaions in reminiscen of hose found wih award. For example, consider (iv), which exhibis an of-varian (iva), parallelling he o-daive, a prenominal varian (ivb), parallelling he double objec srucure, and a wih-varian (iv5c): (iv) a. The gold rim of he plae ha he plae has b. The plae s gold rim c. The plae wih he gold rim 29 Compare also analyses like Siegel (1974) and rescher and Hornsein (1979) in which he posnominal geniives are derived from prenominal forms by righward movemen. 42

43 On he analysis enerained here, hese proposals are no available, however. Assuming geniive s o be derived uniformly, none of he prenominal geniives in (102) or (103) is base-generaed; all undergo movemen. Furhermore, none originaes as an argumen of N; insead all are generaed iniially as oblique -argumens. Under our own assumpions abou localiy of hea-role assignmen, his view appears o enail ha wih relaional nouns and in nominalizaions, prenominal geniives do no in fac receive a hea-role direcly from N, bu raher via some oher, more indirec mechanism The Semanics of Themaic Geniives 30 Ineresingly, Grimshaw (1990) has drawn essenially he same conclusion. In a horough-going sudy of nominalizaions, Grimshaw argues ha, despie appearances, relaional nouns and nominalizaions never assign hemaic roles direcly o prenominal geniives, and hence possessives are never parallel o verbal argumens in his respec. Grimshaw erms such phrases "argumen adjuncs", a label expressing heir paradoxical saus as appearing o bear a hemaic role assigned by N, bu fully opional like unseleced adjuncs, and unlike rue subjecs. More recenly, Buron (1995) has advanced an aracive semanic proposal ha implemens Grimshaw s conclusions. Following a number of auhors, including Higginboham (1983), Paree (1983/97), and Williams (1985, 1987), Buron assumes ha possessives are headed by a definie deerminer conaining a free variable R over relaions. 31 In geniives conaining a non-relaional noun, like (106a), R is deermined deicically (106b). (106) a. John s briefcase b. [he x: briefcase (x) & R(x,John)] Leing he value of R vary wih conex, John s briefcase may hus denoe he briefcase ha John owns, he briefcase he is holding, he briefcase siing on he desk in fron of him, ec. By conras, in geniives conaining a relaional noun, like (107a), he value of R is deermined in one of wo differen ways. One way is hrough conexual deerminaion, as before. John s wife hus denoes he individual who is a wife of someone, and who John sands in some conexually given relaion o (107b). Alhough no he normal undersanding wih relaional nouns, his reading can be made pragmaically accessible. Imagine a diplomaic visi by a male digniary from a foreign counry allowing polygamy. uring he visi, each of he diploma s wives is assigned her own individual securiy agen. John is one of hese agens. In such a siuaion, John s wife migh naurally denoe he wife ha John has been assigned o proec; R is conexually fixed as he assignmen-relaion, ec This secion was wrien afer he remainder of he paper (including secion 5). I was no unil Buron (1995) (discussed below) ha a semanics became available o execue he conclusions drawn a he end of 4.2.2, viz., ha so-called hemaic geniives do no receive a Θ-role direcly from N. 31 Cooper (1979) develops a closely relaed proposal in which definie descripions conain a free propery variable Π, whose value is fixed by conex and ha can be elaboraed as a variable over relaions. Buron s analysis can be considered an exension of Cooper s general proposal o he specific case of possessive definies. 32 Noe ha reading (107b) is compaible wih he woman in quesion being John s wife. Imagine a bridge pary for married couples in which husbands and wives are paired as parners by drawing los. By chance John is paired off wih he woman o whom he is in fac married. In hese circumsances, he senence John s wife is his wife is no redundanly rue. 43

44 (107) a. John s wife b. [he x: y[wife(x,y) & R(x,John)] Along wih he deicic reading, here is also a (much more favored) "hemaic" reading where John s wife is undersood o denoe he individual ha John is married o. Buron derives his reading in an ineresing way, proposing ha i arises, in effec, by a form of "inner anaphora," in which he relaional noun serves as he aneceden of R (108a) and deermines i value (108b). 33 (108) a. [he x: y[ wife(x,y)] & R(x,John)] _ anecedes _ b. [he x: y[ wife(x,y)] & wife(x,john)] c. [he x: wife(x,john)] As Buron observes, he complex expression (108a) can be shown o be semanically equivalen o he simpler (108b), bu noe an imporan difference beween he wo. Under (108b), John is no a direc argumen of he relaional noun wife. Raher i is an argumen of he relaion R provided by he definie deerminer. John comes o be undersood as a semanic argumen of wife hrough an indirec chain: John is an argumen of R whose value is given by he relaional N wife. Buron s semanics appears o be fully compaible wih he synacic resuls derived above, according o which prenominal possessors are never direc argumens of N, even when N is relaional. Raher, he possessor is an argumen of he R-variable in, which ges is value hrough N. This proposal appears o be generalizable o all relaional nouns, and o nominalizaions as well, along he lines of (109c), employing he even semanics of avidson (1967). 34 (109) a. Nero s desrucion of Rome b. [he e: x[ desrucion(e,x,rome)] & R(e,Nero)] anecedes c. [he e: x[ desrucion(e,x,rome)] & desrucion(e,nero,rome)] d. [he e: desrucion(e,nero,rome)] Again (109c) is equivalen o he simpler (109b); bu once again under (109c), Nero is no a direc argumen of he nominalized form desrucion. Raher i is an argumen of he deerminer - specifically, is relaion R, whose conen is deermined by he noun. Under hese proposals, hen, he second major assumpion of he geniive shif accoun appears o be susainable. Indeed, he analysis seems o accord naurally wih Grimshaw s (1990) conclusions regarding he "argumen-adjunc" saus of prenominal geniives in he conex of relaional nouns and nominalizaions. 33 On reading (107c), he senence John s wife is his wife is redundanly rue. 34 In prose, (109c) may be read "he even e such ha e is a desrucion of Rome by some x, and e is a desroying of Rome by Nero". See also Higginboham (1983) for an alernaive version of his proposal. 44

45 5.0 egree Phrases and egp The general approach o srucure pursued here can be direcly exended o oher funcional caegories whose semanics is relaional and quanificaional. egree-modifiers of adjecival expressions are a poenial case in poin. 5.1 Comparaives as Quanificaional Predicaive adjecives are ofen analyzed in logic books as simple unary predicaes of individuals (110); however examples like (111) and (112) show his view o be simplisic. (111a) appears o aribue, no simply inelligence, bu a cerain degree of inelligence; ha is, he adjecive appears o relae individuals and degrees (111b). Likewise, (112a,b) appear o compare degrees of inelligence. (110) a. Leser is smar b. Smar(l) (111) a. Leser is ha smar b. Smar(l,d) (112) a. Leser is smarer han Kenon. b. Kenon is as smar as Leser. There is evidence ha cerain degree-modified adjecival expressions are quanificaional in naure. For example, equaive comparaives like (113a) can be direcly paraphrased wih expressions independenly argued o be quanificaional (Larson 1987), such as he free relaive consrucion in (113b). (113) a. Leser will grow as big as Kenon grew. b. Leser will grow however big Kenon grew. Furhermore, as firs noed by Russell (1905), comparaives appear o paricipae in familiar de dico/de re ambiguiies, widely aken o be a maer of scope. For example, (114a) is ambiguous beween wo readings, one of which is sensible (114b), and one of which is no (114b). This ambiguiy can be analzyed in erms of wo posiions for he degree-modified adjecival expression aller han he is, along he lines in (115a,b). In he firs, he adjecival phrase is scoped ouside he proposiional aiude verb, yielding he sensible reading. In he second, i ales scope inside hinks, yielding he nonsensical reading. 35 (114) a Maryann hinks Kenon is aller han he is. b. Kenon s heigh is such ha Maryann hinks Kenon is aller han ha. c. #Maryann hinks Kenon is aller han himself. (115) a. [ aller han he is ] i Maryann hinks [ Kenon is i ] b. Maryann hinks [ [ aller han he is ] i Kenon is i ] 35 See also Posal (1974) and rescher (1977). 45

46 Finally, comparaives exhibi he AC (Aneceden Conained eleion) phenomenon, widely associaed wih quanificaional srucures. Larson (1988) noes ha he boldfaced phrase in (116a), conaining an ellipical, is a complemen of he verb las. As discussed by Sag (1975), Williams (1997), and May (1985), proper recovery he elided maerial in such circumsances requires he null o escape he marix a Logical Form. This resul is achieved if he comparaive AP is quanificaional and undergoes raising (166b) wih subsequen reconsrucion of he missing (116c): (116) a. John s pary will [ las as long as Bill s pary will [ ]]. b. [ AP as long as Bill s pary will [ ]] John s pary will [ las ] c. [ AP as long as Bill s pary will [ las ]] John s pary will [ las ] RECONSTRUCTION Ross (1973), Seuren (1973) and Klein (1980) offer an aracive quanificaional semanics for comparaives and equaives, in which adjecives are undersood as applying o pairs of individuals and degrees. Comparaives are analyzed as involving exisenial quanificaion over he degree elemen (117)- (118), whereas equaives are analyzed as involving universal quanificaion (119): (117) a. Kenon is aller han Leser. b. d[ all(l,d) & all(k,d) ] (118) a. Leser is less all han Kenon. b. d[all(k,d) & all(l,d) ] (119) a. Leser as all as Kenon. b. d[ all(k,d) all(l,d) ] Noe crucially ha he noion of "degree" invoked in hese formulae is no simply "poin on a scale", bu raher "direced inerval" or "exen". So undersood, when an adjecival predicae is rue of an individual o a degree/exen d, i is rue of ha individual o all lesser exens/degrees d as well. For example, if Kenon, a chimpanzee, is all o degree/exen 5f., hen he is also all o degrees/exens 4.5 f., 4.0 f., 3.5 f., ec. 36 (120) 5 f. 4.5 f. If Kenon is all o he degree/exen 5f., hen he is all o all lesser degrees/exens as well. Kenon 36 This noion of degree is familiar in conexs like carnival rides, wih signs saying "you mus be his all o go on his ride". "This all" are undersood as expressing a degree or exen; anyone measuring ha heigh or higher is undersood o have ha exen of heigh. 46