A Null Theory of Phrase and Compound Stress Author(s): Guglielmo Cinque Reviewed work(s): Source: Linguistic Inquiry, Vol.

24, No. 2 (Spring, 1993), pp. 239-297 Published by: The MIT Press Stable URL: http://www.jstor.org/stable/4178812 . Accessed: 10/05/2012 05:36
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to Linguistic Inquiry.

http://www.jstor.org

Guglielmo Cinque

A Null Theory of Phrase and CompoundStress

1 Introduction Since the publicationof Chomsky,Halle, and Lukoff 1956,it has generallybeen assumed that (surface) constituent structureis the fundamentaldeterminantof phrase (and sentence) stress. A naturalquestion is whether, in additionto syntactic constituency and principlesof Universal Grammar,we need some language-specific phonologicalrule as well. The variousgenerativetreatmentsthat have been proposedin the literatureall have, either explicitly or implicitly, claimed that we do in assuming some form of Chomsky and Halle's (1968) Nuclear Stress Rule. Here, I would like to explore the possibility that no language-specificproviso is necessary and that the (unmarked)patternof phrase stress can be entirely determined on the basis of surface syntactic constituent structure,given the word stresses and the generalprinciplesof gridconstructiondefinedin Halle and Vergnaud's(1987)refilnement of Liberman's(1975) metricalgrid theory. If correct, the argumentwill imply that there is no such thing as a Nuclear Stress Rule of English as distinct from a Nuclear Stress Rule of German-more generally, no such thing as a Nuclear Stress Rule. Any differencein the patternsof phrase, and sentence, stress between two languages should instead follow from their respective constituent structure,as determinedby purely syntactic parameterssuch as the head-initial or head-finalcharacterof their phrases.' The argumentwill be made on the basis of rather limited evidence, essentially a comparisonof English, German, and Italian. At this preliminarystage, a more careful analysis of a few syntacticallybetter knownlanguagesmay be safer, and more revealing, than a superficialsurvey of several typologicallydifferentlanguages,even though some suggestive typological data will be cited (see section 8).
For comments and suggestionson an earlierversion of this article, I am indebtedto WernerAbraham, Josef Bayer, PaolaBeninca,Gerhard Brugger, LuigiBurzio,AndreaCalabrese, AnnaMariaDi Sciullo,Gunther Grewendorf,Morris Halle, Joachim Jacobs, RichardKayne, Michael Kenstowicz, Joan Mascar6, Marina Nespor, Petr Sgall, and an anonymousreviewer. ' By the same token, no language-specific Compound Stress Ruleexists eitherif the nulltheoryof phrase stress extends to compounds,as I indeed suggest in section 9. 239
Linguistic Inquiry. Volume 24. Number 2. Spring 1993 239-297 ? 1993 by The Massachusetts Institute of Technology

240

GUGLIELMO

CINQUE

Given the crucial role that the metricalgrid theory plays in the argument,I begin by briefly sketching the theory in the form given to it by Halle and Vergnaud(1987). 2 The MetricalGrid Theory Within the metrical grid theory, which develops an idea of Liberman(1975), stress is represented in a separate autosegmentalplane, like tone. The autosegmentalline for stress is a sequence of abstract positions (conventionallymarkedwith asterisks) associated with the potentially stress-bearingpositions on the central line of phonemes, as illustratedin (1).
(I)
* __________*

t _o

As with other phonologicalentities, this formalismpermits, among other things, a local computationof phenomenathat appearnonlocal on the phoneme line. Not every potentially stress-bearingunit (e.g., a syllabic nucleus) represents a stressed position, on the phoneme line. One way to markthose that actually do is to set up an additionalline on the stress plane where only such units receive an asterisk, as illustratedin (2) with the word serendipity,whose first and thirdsyllabic nuclei only are stressed. (2)
* \ \ \e \* * *\ I ~~~~~~~~~~line

0 ~~~~~~~~~~~~~~~
yu

If one of these units carries a stress more prominentthan the others (as is the case with the third syllabic nucleus of serendipity),then it alone will receive an asterisk on a still higher line 2 (not indicatedin (2)). Since at most three degrees of stress (besides zero stress) are distinguishedamong stressed syllables in noncompoundwords, only three lines (besides line 0) will be needed to represent the main stress of individualwords (see Halle and Vergnaud1987). Within this basic formalism, Halle and Vergnaud show that, by recognizing the existence of constituents on each line, and their heads (markedon the next higherline), it is possible to "rationalize" the considerablevariety and apparentcapriciousness of the patternsof word stress in the languagesof the world. These can, in fact, be seen as arisingfrom differentsettings of the same few parameters rules of constituentboundand

NULL

THEORY

OF

PHRASE

AND

COMPOUND

STRESS

241

ary construction: whether a constituent on line L is bounded or unbounded(+ BND); head-terminal(in which case the further choice is between right-headednessor leftheadedness) or not (? HT); constructedfrom left to right or from right to left. By way of illustration, consider one of the various possibilities that follow from Halle and Vergnaud's parametrictheory.2 The stress pattern of Maranungku,where stress falls on all odd-numbered syllables countingfrom the beginningof the word, with the leftmost bearingmain stress, is obtainedby means of the parametersettings in (3), interactingwith the general principlesof grid constructionin (4). (3) a. b. (4) a. b. Line 0 parametersettings: + HT, + BND, left, left to right Line 1 parametersettings: + HT, - BND, left Constructconstituent boundarieson line L. Locate the heads of line L on line L + 1.

The representative stress pattern of the language is thus the metrical grid (5) (where indicationof the plane and the phoneme line is omitted).
(5)
* . . .

(* . * . *) (* *)(* *)(*)

line 2 line I line 0

The gridis obtainedby meansof the parameter settings(3a-b) andthe rulesthatconstruct constituentboundaries(4a-b) in the followingway. First, line 0 is constructedby marking with an asteriskall (potentially)stress-bearing elementsin the word(takinga five-syllable word as representative).Then, a constituentstructureis imposedon this line by applying rule (4a) in accordancewith the parametric values indicatedin (3a). In this way, bounded (binary)constituents are constructedleft to right(the last being a defective constituent). Given the positive value of the head-terminalparameterand the "left" value of the headedness parameter,a head for each constituentis located via (4b) on the next higher line (line 1) over the asterisk adjacentto the left boundaryof the constituent. Then, a constituent structureis imposed on line 1 by againapplyingrule (4a) in accordancewith the parametricvalues indicated in (3b). An unboundedconstituent is thus built, comprising all three asterisks on line 1. Given the positive value of the head-terminalparameter and the "left" value of the headedness parameter,a head for the constituent on line 1 is located on line 2 over the asterisk adjacent to the left boundary of the constituent in line 1. The correct representationof the stress pattern of Maranungku words, with the appropriatedegrees of stress, is thus derived. Different choices of the same parameters(in possible interactionwith the further choices mentionedin footnote 2, and others) give rise to the stress patternsfound in the
2 Abstractingfrom a numberof furtherchoices and refinementsrequiredby some languages, such as whetherthe initial (or final) segmentof the word is extrametrical; whetherthe heads of the constituentsin a lower layer are "predetermined" (e.g., all the vowels of closed or long syllables, as in Koya), or simply determinedby the parameters rules of constituentboundaryconstruction(as in Maranungku); so on. and and

242

GUGLIELMO

CINQUE

other languages.(For detaileddiscussion, see Halle and Vergnaud1987,Halle and Kenstowicz 1991.) In this approach,differences in degrees of stress are expressed in terms of the differentheights of the associated asterisk columns, as seen. Having briefly sketched Halle and Vergnaud's(1987)theory of the stress contours of single words, I next consider how they propose to extend the theory to the stress contours of phrases (and sentences). 3 The NuclearStress Rule: A MetricalVersion As is well known, when words are combined into phrases, the stress contours of the individualwords are largely unaffected, the effect of the combinationbeing merely the assignmentof greaterprominenceto the main stress of one constituentover that of the others. In English (and Italian) the constituent whose main stress is enhanced under normalconditions is the rightmost.This is essentially what Chomskyand Halle's (1968) Nuclear Stress Rule of English was meant to express. Halle and Vergnaud(1987:264)propose to incorporatethe rule into the formalism of their theory as follows (see their (80)): (6) Nuclear Stress Rule a. Parametersettings on line N (N - 3) are [-BND, + HT, right]. b. Interpret boundaries of syntactic constituents composed of two or more stressed words as metricalboundaries. c. Locate the heads of line N constituents on line N + 1. The effect of this rule will be to add new lines to the metricalgrid, one for each new phrase computed, from the most embedded one to the root sentence. The languagespecific proviso of the rule is representedby the parametersettings in (6a). By way of illustration,consider the derivationof the stress contourof the sentence Jesus preached to the people of Judea (Halle and Vergnaud1987:265):

(7)
(.

. .

line 6
line 5 line 4 line 3

(
*

..

*) *)
*)

(*

[Jesus [preachedto the [people of Judea]]] The first metrical constituent built on the basis of (6b) is that correspondingto people of Judea. In accordance with the principle of the cycle, (metrical) constituents that contain unerasedbrackets, such as [preached to the [people of Judea]], cannot be computed until after the innermostconstituentis computed(and its bracketserased).3
3 Also note that constituentsone of whose elementsis a stressless word, such as a prepositionor article, are systematicallyskipped(Halle and Vergnaud1987:264). certaincases prepositionscan be accented (see In Cruttenden and Faber 1991).

A NULL

THEORY

OF

PHRASE

AND

COMPOUND

STRESS

243

The head of the constituent is located on line 4 by applying (6c). Then a metrical constituent is created on line 4, and so forth. Noting that the procedureas such does not reflect the stress contour of the sentence correctly (with Jesus bearingmore stress than preached, and the latter bearing more stress than people), Halle and Vergnaud (1987:265)propose to supplementit with the following convention: (8) Stress EqualizationConvention When two or more constituents are conjoined into a single higher-levelconstituent, the asterisk columns of the heads of the constituentsare equalizedby addingasterisks to the lesser column(s). This has the effect of introducingasterisks in place of some of the dots of (7).4 In their theory, then, the stress contour of phrases and sentences is determinedby means of the same rules and parametersutilized for determiningthe stress contour of individualwords.

4 A Null Theoryof PhraseStress Halle and Vergnaud's extension of their theory of word stress to phrase stress raises two conceptual questions: (9) a. Are phrase stress systems as numerousand diverse as word stress systems, as the differentsettings of the [ BND, ? HT, left/right]parameterswould lead one to expect? Is it an accident that the Nuclear Stress Rule gives prominenceto the rightmost constituent of a phrase in languageslike English or Italian, which are essentially right-branching (i.e., have increasingdepth of embeddingto the
right)?5

b.

Suppose, for the sake of argument,that it is no accident that the Nuclear Stress Rule gives prominenceto the rightmostconstituent of a phrase in right-branching languages (the branchingdirection of the language actually implyingthe same direction in stress prominence).If this is so, there is an immediateimplicationfor the other question too. Only two general types of phrase stress systems should exist accordingto whether the languageis left- or right-branching. Right-branching languagesshould show the effects of the Nuclear Stress Rule, and left-branching languages should show the reverse (essentially, the effects yielded by the CompoundStress Rule of English,which gives prom4 There being no upper bound on depth of embedding,no upperbound exists on the numberof stress distinctionseither. As speakers may "make fewer distinctionsthan are providedby this procedure,"Halle and Vergnaud(1987:266)suggest a Grid Simplification Convention.On this question, also see Chomskyand Halle 1968:23. S Question(9b) in fact is not particular Halle and Vergnaud'stheory, but applies to any theory comto prising(some version of) the Nuclear Stress Rule.

244

GUGLIELMO

CINQUE

inence to the main stress of the leftmost constituent). Languageswith mixed branching should combine propertiesof the two "pure" systems.6 However, if the effects of the NuclearStress Rule (andof its reverse)dependentirely on the directionin which depthof embeddingdevelops, then the rules become redundant. They merely recapitulatewhat follows from purely syntactic parameters.Hence, they should be eliminated,at least if a way exists to link the directionin which stress promdirectionof the phrase. WhatI would inence is assigned withina phraseto the branching like to suggest is that such a link is implicit in Halle and Vergnaud'sprocedureof grid construction:an additionaldistinctive advantageof the metricalgrid notationover such alternativesas the metricaltree notation,which apparentlycannotderive the same result (see section 5 for some discussion). What is apparentlynecessary and sufficient is the combinationof Halle and Vergnaud's(6b), simplifiedas in (lOa);(6c), repeatedhere as (lOb);the principleof the cycle, (lOc), essentially in its originalformulation;and the condition that there be no gap in an asterisk column, (lOd). (10) a. b. c. d. Interpretboundariesof syntactic constituents as metricalboundaries.7 Locate the heads of line N constituents on line N + 1. Each rule applies to a maximalstringcontainingno internalboundaries. An asterisk on line N must correspondto an asterisk on line N - I.'

Let us consider first a couple of simplifiedabstractcases to illustratethe working of (10)-namely, (1la-b) (or (12a-b) in tree format),whereA, B, C are arbitrary syntactic maximalprojectionsand the asterisksindicatethe mainstress of the words that constitute their heads. (11)a. b.
[A * [B*
[A[B[[C

[C*]]]

*] *1

6 One may note, incidentally,that some of the parametric at options that seem appropriate the word stress level could in any event turnout to be irrelevantat the phrasestress level. So, for example, if Kayne's (1984)strict binary-branching hypothesisis correct,as seems likely, the ? BND parameter could be dispensed with along with the ? HT parameter. 7 (lOa)differsfrom Halle and Vergnaud's (6c) in that it makesreferenceto constituenthood, irrespective of whetherthe constituentcontainsonly one or morethanone stressedword (the most generalinterpretation, in any case). As will become clear, this simplification crucialto obtainthe correctresults. is One may also note that this formulationis entirely consonant with the notion of head and projection discussed by Halle and Vergnaudthemselves (pp. 8-9). 8 This conditionis nothingbut Prince's(1983:33) that "requirement a columnmust have entries at every 3 2 4 level up to its peak," which is instrumentalin accounting for why stress retracts in antique chair (< 2 3 2 4 4 3 1 antique chair) but not in antique dealer. As far as I can tell, it entails, thoughit is not equivalentto, Halle's (1985)LandingSite Constraint("When an asteriskis moved it lands on the highestcolumn of asterisksthat is in its domain. If there is more than one such columnit lands on the nearestof these"). This conditionalso appearsto accountfor the curiousexception to the RhythmRule noted by Halle and Vergnaud(1987:235,270), accordingto which the rule fails to applyin words where the potentiallandingsite

lacks stress (serene summers vs. fifteen summers < fifteen summers); for there is no way for the topmost asteriskof the second syllable of serene to move over the first syllablewithoutviolatingthe requirement that

there be an asteriskon the immediatelylower line.

NULL

THEORY

OF

PHRASE

AND

COMPOUND

STRESS

245

(12) a.
*

b.

A
*

B C

(1la)/(12a) and (1lb)/(12b) represent right- and left-branchingstructures, respectively, with each constituent nested in the next higherone. Applicationof (10) gives rise to the grids (13a-b) (lines below line 3 are omitted).
(13) a.

(
((

* . *

b.

* * .

.)

* )) (* (* (*)))

(( * .) .) (((*) *) *)

line line line line

6 5 4 3

Consider how. The combinationof (lOa) and (lOc), the principleof the cycle, dictates that the first metricalconstituent to be computedis the innermost(to the right in (13a), and to the left in (13b)). (lOb)then requiresus to locate the head of this constituent on the next higherline (line 4). Since the constituenthas only one position, the asterisk on line 4 can only be located in the same column as this position. On the next cycle, (lOb) will again demand that the head of this constituentbe located on the next line up (line 5). The innermost constituent of line 4 has two positions, but only one of them is an asterisk. So, by (lOd), we have no choice. The head (on line 5) of the constituent on line 4 can only be in the same columnas the single asteriskfoundon line 4. Reapplication of the same proceduregives rise to the complete grids (13a) and (13b). In this fashion, stress prominencein a phraseis a mere reflectionof depth of embedding. And the rightmostor leftmost location of the main stress is simply a function of the rightmostor leftmost location of the most deeply embeddedphrase (as determined by the direction of branching). If, as I will claim, the relation between two constituents of a phrase is always asymmetrical(in the sense that one of the two is necessarily more deeply embedded than the other), the direction of stress prominence,as in the Nuclear Stress Rule, need not be stipulated.The first constituentto receive an asterisk, whether on the left or the right, will "attract" all later asterisks. The procedurein (10) would seem to suffer from the same deficiency as Halle and Vergnaud'sNuclear Stress Rule (6), which called for a Stress EqualizationConvention to assign the appropriatedegrees of stress to a sentence such as Jesus preached to the people of Judea, but this is not quite so. Note that the sentence in question has two

246

GUGLIELMO

CINQUE

nonintersectingconstituents, the subject NP and the predicateVP (for simplicity, I now ignore all functionaland intermediateX-bar projections): (14)
[[NP

Jesus] [vp preached [pp to

[NP

the people [pp of

[NP

Judeal]]]]].

This means that the subject NP and the VP undergotwo parallelcycles before joining at the sentence level. In particular,this means that the NP Jesus will receive a line 4 asterisk-one more thanpreached andpeople, whichfail to receive one on line 4 because the innermost constituent [NP Judea] receives it first, thus attractingall later asterisks (those of lines 5 and 6, as well as that of line 7, after the whole sentence is computed):9
(15)

(.
(
.

. .

. . .

(.
.

( * ( ( . ))) (( * ) ( * ( * ( * )))) [[Jesus] [preached[to the people [of Judea]]]]

* * * *

) ))

line line line line line

7 6 5 4 3

The general consequence of this formal procedureis then that the first constituent to receive an asterisk will be the one to ultimatelyreceive greatest stress within a phrase. An immediateproblem would seem to be posed by those cases where the subject NP has more layers of embeddingthan the predicate(e.g., in a sentence such as [[The author [of many popular articles [on the effects [of senescence]]]] [[died]]]). Here the formalprocedurewould lead us to expect the most deeply embeddedconstituentof the subject NP to bear more stress than the verb in the VP. But this is not necessarily the case. I returnto this problembelow when discussingthe relationbetween the proposed formal procedure and the effects of the focus and presuppositionarticulationof the sentence (its "informationstructure").Also see the discussion at the end of section 7. Other problems stemmingfrom predictionsof this procedurethat appearto fail (given certain assumptions about the constituent structureof English and Italian)are also deferred until section 7, after some implicationsof this general approachhave been considered in more detail. As noted, underthe hypothesis to be exploredhere, no language-specific rule (such as the Nuclear Stress Rule) should be postulatedto determinestress prominenceat the phrase level. Rather, phrase stress should be entirely determinable(given the word stresses) from the independentprinciplesin (10) in interactionwith such purelysyntactic parametersas the head-initialor head-finalparameter(responsiblefor the direction of embedding).As observed, the generalpredictionof this hypothesisis thatin right-branching phrases the stress prominence should fall on the main stress of the rightmostconclaim that preachedhas more stress thanpeople. If this is a clear and 9 Halle and Vergnaud(1987:265) of principles prosodic in intuition,thenthe procedure (10)will need to be amended."Supplementary perceptible realization"(Prince 1983:24)are likely to superimposethemselveson the effects of the presentprocedureto give finer stress gradations.Concerningrhythmicprinciples,see Selkirk 1984,Dell 1984,amongothers.

A NULL

THEORY

OF PHRASE

AND

COMPOUND

STRESS

247

stituent (thus derivingthe effects of the Nuclear Stress Rule of English), whereas in leftbranchingphrases the stress prominenceshouldfall on the leftmost (to yield, in essence, the same effects produced by the CompoundStress Rule of English). Although some suggestive typological evidence in this direction does exist (see section 8), it may be useful to consider a specific case in some detail. Germanstands out as particularlyappropriateto the task. Its mixed branchingcharacterallows us to test the opposing predictions of the hypothesis within a single language. Furthermore, its syntactic structureand its accentual system are both ratherwell known.'O In the next section I begin with a brief excursus of the main features of German phrase stress, based in essence on the classical work of Kiparsky(1966).I then compare the language-specificapproachtaken there and in subsequent works with the null approach developed above and consider a numberof more subtle predictionsaffordedby the latter approachon the basis of what is now known about the syntactic structureof German. 5 Phrase Stress in German Kiparsky (1966:81)distinguishestwo differentclasses of phrases in Germanaccording to whetherthey receive stress prominenceon theirrightmostor leftmost constituent(for suggestionsalong the same lines, see Jacobs 1982).His termsNom and Satz, taken from Bierwisch's (1963) fragmentof Germangrammar,correspond to NP and CP, respectively. " The rendition of other terms is more problematicsince some of them do not even seem to correspondto constituentsin today's theory. So, for example, D essentially
10 The syntactic literatureon Germanproducedin the last decade is quite extensive, and a widespread consensus has now been reachedon the fundamental of structure the Germansentence(to be reviewedbelow). See, amongother works, Den Besten 1989,Thiersch1982, 1985,Grewendorf1988, 1989,and variouspapers in Abraham1983, 1985,Toman 1985,Haiderand Prinzhorn1986,andGrewendorf Sternefeld1990.Works and on Germanphrase and sentence stress include Kiparsky1966, Bierwisch 1966, 1968, Loetscher 1981, 1983, and Bresson 1983. On the relationbetween stress and focus in German,see Hohle 1982,Jacobs 1982,Selkirk1984:225-230, von Stechow and Uhmann1984, 1985,and Grewendorf1989:chap. 4. (ia-l) represent(partof) the fragmentof Germangrammar presupposedby Kiparsky.

(i) a. b. d. e. f. g. h. i.

Satz I D VP Vb Nom Aux d

-

(I) S W
impJ

c. S

- DVP
Nom (Nom) (Nom) (Adv) > (Satz) Vb Aux (Adv) V d (Satz) N (Md) Fin determinant modal verbs (sollen, wollen,

1. Md .. In additionto the rules in (i), Kiparsky,againfollowingBierwisch,assumestwo rulesapplyingto root clauses, one moving an arbitrary constituentto first position (corresponding currentXP-movementto the Spec of to CP), and anothermovingthe finite verb to second position(corresponding currenthead movementof V (to to I) to C).

248

GUGLIELMO

CINQUE

12 tradition. This comprises renders the notion of Mittelfeld in the Germangrammatical all the constituents found between the head of CP and the head of VP, namely, the subject NP and the possible adjunctand argumentXPs of the VP without the verb-a sequence that is apparentlynot a constituent (but see footnote 16). Analogously, Kiparsky's "VP" is used for a verbal group comprisinga verb plus an auxiliary (plus a complementCP if there is any), but excludingthe rest of the verb's complements:another nonconstituent under current assumptions, which analyze auxiliaries as heads (of an auxiliary VP) taking ordinary VPs (or AgrPs containing VPs) as their complements. Assumingthis partitionof the Germansentence and the fact thatNPs, CPs, and D receive final stress prominence,whereas the sequence S (= D + VP), correspondingto IP, and the verbal group "VP" receive initial stress prominence, Kiparsky manages to derive the intricacies of ordinarysentence stress in Germanwith remarkableaccuracy." Let us briefly consider how. The case of NPs is straightforward. They receive final

stress prominence (Endbetonung):

(16) a. b.

[die dicke Emma] the fat Emma [der Mann aus Rio] the man from Rio

ConcerningCPs, the stress contourof such simple root clauses as (17) is deriveddirectly under the "constituent" analysis in (18) (Kiparsky 1966:81). (17) a. Waldemarspielt Theater. Waldemarplays theater 'Waldemaris on the stage.' Die Katze lief weg. the cat ran away [Satz[l Waldemarspielt] [s Theater]]. [Satz[, Die Katze liefl [s weg]].

b. (18) a. b.

S has only one word (Theater, weg, respectively), so that word will receive the primary stress of the S cycle. Although Kiparskydoes not explicitly discuss the stress contour of I, it is reasonable to assume that in his system it would have left prominence. If so, the subject NP will receive the primarystress of the I cycle. Whenthe Satz cycle, which is subjectto the ordinaryNuclear Stress Rule, is reached,primarystress will be assigned to the most prominent stress to the right (Theater and weg, respectively), all other stresses being reduced by one at the same time. Kiparsky's ingenious procedure will
12 See Drach 1940. For recent critical comparisonsof this tradition with currentgenerativeanalyses of German,see Olsen 1982, Scherpenisse 1986:chap.1, and Grewendorf1988:20,amongothers. analogueof the) Nuclear 13 Phrasedin other words, NPs, CPs, and Ds would be subjectto the (German Stress Rule, whereas IPs and the verbalgroup "VP" would be subjectto a "Reverse Nuclear Stress Rule," domainof versus word-internal identicalin effects to the EnglishCompoundRule, modulothe word-external application.

A NULL

THEORY

OF PHRASE

AND

COMPOUND

STRESS

249

also derive the correct result in more complex cases such as (19), under the analysis indicated in (20). (19) Hans wird einem Kind ein Buichgeben konnen. Hans will to a child a book give be able 'Hans will be able to give a book to a child.' (20) [Satz[l Hans wird] [S[D einem Kind ein Buch] [vp geben konnen]]]. The verbal group "VP" and I are subject to the "Reverse Nuclear Stress Rule," so that the leftmost constituent (geben and Hans, respectively) will receive the primarystress of the cycle. D is subject to the ordinaryNuclear Stress Rule, so that the rightmost constituent ein Buch will receive primarystress. When the S cycle is reached, which is again subject to the "Reverse Nuclear Stress Rule," the most prominentstress of the leftmost constituent (that of Buch) will receive primarystress, thereby causing all other stresses of that cycle to lower. Finally, when the Satz cycle is reached, which is subject to the ordinaryNuclear Stress Rule, the most prominentstress to the right (once again that of Buch) will receive the primarystress, with the concomitantweakeningof all the other stresses. Despite its ingenuity and remarkableempiricalsuccess, Kiparsky'sanalysis raises certain questions about the constituent structureit must assume.14 But even if all such questions could be satisfactorilyanswered, the fact that certain Germanphrases take left prominence whereas others take right prominencewould still be treated as an accident. The theory could just as well accommodatethe opposite arrangement stress of
prominence.

A more interesting theory, it seems, would be one that derived the rightmost or leftmost prominenceof a certainGermanphrase as a necessary consequence of general and independentprinciples. The null theory sketched above appearsto qualify as such a theory. This is because, as predicted, the leftmost or rightmoststress prominenceof a Germanphrase appears to correlate exactly with the direction in which the phrase's depth of embeddingdevelops. Let us see how, beginningwith complements and delaying for a moment the examinationof adjuncts and other modifiers. Considerwhat value the head complement parametertakes in each phrase. NPs are head-initial.This means that their complements are found to their right: (21) a. die [N' Entdeckung[NP des [N' Impfstoffs]]] the of the vaccine discovery

"' It also faces certainempirical problems.So, for example,given that subjectsare includedin D and that D receives greaterprominencein S (= D + "VP"), the subjectratherthan the verb would be expected to bear primarystress in the followingcase, contraryto fact: (i) Da hat das Kind mit gespielt. it has the child with played 'With it the child has played.'

250 b.

GUGLIELMO

CINQUE

die [N' Landung[pp auf [NP dem [N' Mond]]]] moon on the the landing

Since the complement's head is more deeply embeddedthan the head N (in the sense that it is dominatedby more projectionsthan the head N), it will receive an asterisk on the first and second phrase cycles, the lower N' and NP, before the head N can receive one on the higherN' cycle. This, in turn,will meanthatthe mainstress of the complement will continue to attractall later asterisks, in accordancewith principle(lOd), ultimately bearingthe strongest prominencewithin the largest phrase, as desired. Except for a handfulof cases (such as (23)), PPs are also head-initial,and in fact their stress properties are analogous to those of the NPs in (21), with prominence on the main stress of the complementto the right: (22) a. b. c. auf den Tisch onto the table durch das Zimmer throughthe room unter den Linden under the lime-trees

The situationis reversed with postpositionalphrases, as expected. Greaterprominence is now on the left:'5 (23) a. entlang den FluJB the river along 'along the river' hinauf den Berg the mountainup 'up the mountain'

b.

Next, consider VPs, which in Germanare head-final.The rough generalizationis that the primarystress falls on the XP to the immediateleft of the verb, or verbalgroup (see von Stechow and Uhmann 1986:315,Grewendorf1989:sec. 4.3, among others):
(24) a.
...

daB Hans [ein Buch auf den Tisch gestellt] hat.

that Hans a book b.

c.
...
...

on the table put

gegeben] hat.

has has

daB3Fritz [einem Kind Geld

that Fritz to a child that Karl a book

money given

daB Karl [ein Buch mit Muhe

lesen] kann.

with difficultyread can

15 This fact also follows under Van Riemsdijk's(1990)interesting analysis of post- and circumpositional PPs in German.Accordingto this analysis,lexical Ps are alwayshead-initial, althoughtheirmaximalprojection is selected by a functionalhead-finalhead ([FP[PP P NP] F]). Postpositionsthen are prepositionsraisingto F, and circumpositions cases of Ps base-generated both P and F. are in

A NULL

THEORY

OF PHRASE

AND

COMPOUND

STRESS

251

d.

...

daB Hans [ein Buch interessant findet].

that Hans a book

interesting finds

This is in fact what the null theory predicts, since (as (25) illustrates)in a left-branching constituent whose head is V, the constituent to its immediateleft is in each case the most deeply embedded constituent of the VP. (25) NP
VP

IP I'
I

YP

VI

XP

This is true even if the configurationin (25) is not base-generated,but derived, with YP moved from a position between XP and V (see footnote 32). Note that the direct object in (26) has been moved to the left via scrambling,thus leaving the indirectobject as the most deeply embedded constituent of the VP.
(26) ... daB Bruno sein Geld oft [den Kindern gab].

that Bruno his money often to the childrengave If so, the basic stress contour of the VP is predictedwith no need for a special version of the Nuclear Stress Rule. The same predictionsfollow a fortiori if the verb raises to I, as is now standardly
16 assumed:

(27)

...

daB Bruno sein Geld oft [den Kindern tj] gabj.

Next, consider APs (not discussed in Kiparsky 1966).They can take prepositional complements on both sides, and Case-markedNPs to their left only. See (28) and (29), with the indicationof their unmarkedstress pattern. (28) a. Er ist [uber seinen Freundungehailten]. he is at his friend angry

16 See Giusti 1986, 1990, Den Besten 1989, Webelhuth1989,amongothers. Note that raisingof the verb leaves a constituentessentially corresponding Kiparsky'sD constituent,which thus looks less ad hoc, in to retrospect.

252

GUGLlELMO

CINQUE

Er ist [ungehaltenuber seinen Freund]. he is angry at his friend (29) a. Er war [dem Mann bose]. he was to the man nasty b. *Er war [bose dem Mann]. he was nasty to the man This is generallytaken to indicate their head-finalcharacter(with the head-complement orderderived via the independentrule of extraposition;see Giorgiand Longobardi1991: chap. 3, Tappe 1990): (30) a. b. Er ist [AP[A'[PP Er ist [AP[AP[A' uber seinen Freund]ungehalten]]. t ungehalten]][ppuber seinen Freund]].

b.

If this were the case, the null theory of phrasestress would face a serious problem,since it would predict the strongest stress to fall in (28a) and (29a) on Freund and Mann, the most deeply embeddedconstituents, ratherthan on the adjective, as is instead the case. Interestingly,there is evidence that more is involved-in particular,that preadjectival complements in Germancannot stay under A', but must move out of it, to adjoin at least to AP, whatever their D-Structuresource (also see Webelhuth 1989:chap.6). The evidence comes from the following peculiarityof the word order internalto AP: when the complement precedes the head and there is a lexical specifier or some other prehead modifier, one finds the order complement-specifier-head ratherthan the order For specifier-complement-head. example: Er ist uber seinen Freundsehr ungehalten. friend very angry he is at his b. *Er ist sehr uber seinen Freundungehalten. friend angry he is very at his (32) a. Er war dem Mann oft bose. he was to the man often nasty b. *Er war oft dem Mannbose. he was often to the man nasty The fact that the complement must precede the specifier sehr 'very'/oft 'often' is evidence that it cannot stay underA', but instead must adjoinat least to AP. As Henk van Riemsdijk(personal communication)has pointed out, there is in fact evidence that the complementmust adjointo some projectioneven higherthan AP, since the complement must precede the negation (which is outside the AP and is in fact taken to markthe left boundaryof VP; see Webelhuth 1990:55): (33) a.
...

(31) a.

daB er dem Mann nicht [VP[AP that he to the man not

[VP[AP

bose] ti] wari.

nasty

was was

b. *. . . daB er nicht

dem Mann bose] ti] wari.

that he not

to the man nasty

A NULL

THEORY

OF PHRASE

AND

COMPOUND

STRESS

253

(The stringin (33b)is-irrelevantly-possible if nicht and dem Mannform a constituent; that is, if nicht is a constituent ratherthan a sententialnegation.) The case of GermanAPs is methodologicallyinstructive.It shows how careful one must be in putting a hypothesis to test. Insufficientlyanalyzed structuresmay easily lead to incorrectconclusions concerningthe hypothesis to test. When the functionalcategories IP (or ratherAgrPs, TP, AgrPo of Chomsky 1991) and CP are taken into account, a more complete picture of Germansentence stress can be given. As (34) illustrates, German Agro and presumably T and Agrs take their complementsto the left, whereas C takes AgrPs to its right (see Grewendorf1988, Den Besten 1989:274,among others). (34) CP

C'

AgrPs
Agr's

TP T' AgrPo Agr'o

VP Agro

Agrs

V' V In root clauses (and in embedded clauses selected by certain predicates; Bader and Penner 1990, Vikner 1990, Cinque 1989), the V raises to C, and a maximalprojection

254

GUGLIELMO

CINQUE

must fill the Spec of CP (in declaratives). In non-V2 subordinate clauses, instead, the

finite V does not raise higherthan Agrs. This gives rise to a variety of differentcases, all sharingthe propertythat the greatestprominencefalls on the most deeply embedded (lexical) constituent, as predictedby the null theory (when not crucial, I conflate AgrPs, TP, and AgrPo into IP):'7 (35) a. [cp Waldemari[c spieltk[1P ti [VP Theatertk] tk]]]. Waldemar plays theater 'Waldemaris on the stage.' [cp Das Buchi [c findetk[1P er [vp t, interessanttk] the book finds he interesting 'He finds the book interesting.'

b.

tk]]].

c.

[cp[Trinken wollen]i [c, wirdk [1P sie ihn nicht VPi tk]II.

(36) a.

b.

(37) a.

want to drink will she it not 'She will not want to drinkit.' [cp Hansi [c hatk [IP ti [vP kein Buichgelesen]] tk].'8 Hans has no book read 'Hans has read no book.' [cp Deni [c hatk [Hans [t, gelesen] tk]]]. that has Hans read 'Hans has read that one.' Weil [1P Fritz [VP[NP viele T6rten] backen] kann] ... Fritz because bake can many pies
'Because Fritz can bake many pies . . .' Weil [lp Fritz [vp gut kochen] kann] ...

b.

because

Fritz

well cook
.

can

'Because Fritz can cook well .

Next consider (38a-b). (38) a. [cp Der Arzt [c' wird [AgrP[VP einen Patientenuntersuchen]l]]. the doctor will a patient examine 'The doctor will examine a patient.'

17 I abstractfor the momentfrom such cases as Ein Brief kaman 'A letterarrived'and Ottokommt'Otto is coming', which have primarystress on the subject(in the Spec of CP) in the unmarked case (see Kiparsky 1966:89,von Stechow and Uhmann 1984:253).An understanding such cases requiresa discussion of the of focus and presuppositionarticulationof the sentence, which I undertakein the next section. On the stress propertiesof separableprefixes in German,see Giegerich1985and the remarksin footnote 20 below. 18 For the apparently unexpectedstress contourin (i), due to Josef Bayer (personalcommunication), see the discussion in section 7, where it is suggestedthat right-branching structureon a left branchconstitutes a separatecycle, "invisible" to the main cycle. (i) Man hat den Mannilohne zu zogern] hingerichtet. one has the man withouthesitatingexecuted 'They executed the man withouthesitation.'

NULL

THEORY

OF

PHRASE

AND

COMPOUND

STRESS

255

b.

[cp Der Arzt [c' wird [AgrP den Patienteni[vp t1untersuchenl]]]. will the doctor the patient examine 'The doctor will examine the patient.'

Such contrasts as this one connected to the (in)definitecharacterof the object (Kiparsky 1966:91f.)fall into place if one assumes that only indefiniteobjects can remain in VP, definite ones being necessarily scrambledto some projectionof Infl (perhaps, Spec of AgrPO),as arguedfor on independentgroundsby Brugger(1990)(also see the discussion
in Moltmann 1990). 9

I close this section by briefly consideringthe position of a numberof specifiers and other modifiersin relationto the positioningof stress. In noun phrases with prenominal genitives and adjectives the main stress goes on the head N (Kiparsky 1966): (39) a. b. Peters Auto Peter's car die dicke Emma the fat Emma

Recent work on the structureof the noun phrase indicates that its internalstructureis more complex than previouslyassumed, with a projectionfor the determinerand at least two functional projections intermediatebetween D and the NP (see Ritter 1990 and Cinque 1990for evidence based on Hebrew and Romance, respectively). If prenominal adjectives are located in the Spec position of these intermediatefunctionalprojections (possibly, of agreement;Cinque 1990),the head N will qualifyas the most deeply embedded constituent, thus bearing primarystress accordingto the null theory (the case of prenominalgenitives requires,instead, a refinementthat will be discussed in section 7): (40)
[DP

die

[FP[FP

dicke

[F'

[NP[N'

Emma]]]]]]

'" The same conclusion is reachedfor Hindi by Mahajan (1991).The contrastreproducesitself with indefinite subjects(which in Germancan remainwithinVP), but apparently with indefiniteindirectobjects. not See (ia-b) and (iia-b), providedby GerhardBruggerand GuntherGrewendorf(personalcommunications), respectively: (i) a. ...daB3 diesen Baum ein Forsterfallte. that this tree a forester cut

b.

...da

diesen Baum der Forster fallte.

that this tree the forester cut (ii) a. ...daB3 das Buch dem Studentengehort. that the book to the student belongs b. ...daB3 das Buch einem Studdnten gehort. that the book to a student belongs QuantifiedNPs seem to go togetherwith definiteratherthan indefiniteobjects, as GuntherGrewendorf (personal communication) points out: (iii) . . daB der Arzt bereitwillig jeden Patientenuntersuchte. that the doctor willingly every patient examined

256

GUGLIELMO

CINQUE

Similarconsiderationshold for adjectivaland verbal specifiers:20 (41) a. b. sehr [A bose]] very nasty [weil [1p er [oft [kochen] kann]]] because he often cook can
[AP

When a postnominalgenitive or adverbialPP modifieris present, the main stress of the noun phrase is located on it: (42) a. b. die Ankunft von Ka'rl the arrival of Karl der Mann aus Rio the man from Rio

This follows from their being more deeply embedded than the head N under current assumptions (Giorgiand Longobardi1991, Haider 1991): (43) a. b.
[DP [DP

die [NP Ankunft [pp[p, von [DP Karl]]]]] der [NP Mann [pp[p' aus [DP Rio]]]]]

No doubt, other aspects of Germanphrase and sentence structuredeserve attention. The cases so far reviewed, however, alreadyconstitutesome evidence for the null theory of phrase stress. Dutch appears to provide analogous evidence, at least to judge from the inventory of Dutch phrase and sentence types given by Baart (1987:83-103)with an indicationof their unmarked,noncontrastivestress. Before we turnto some hithertoignoredquestionsconcerningthe informationstructure of the sentence, a brief comparisonmay be worthwhilebetween the metricalgrid theory and the metricaltree theory. As we have seen, the formerpermitsthe derivation of phrase stress from the word stresses and surface syntactic structurevia asterisk addition to each phrase (in conformitywith certain general conditions). The formalismof the latter appears unable to derive the same result without stipulation.This is because it is a purely interpretiveprocedurethat marks the two branches of a binary structure weak (W) and strong (S) in relation to each other, and independentlyof the mannerof
20 A differentinterestingcase is providedby German"separable(verb) prefixes," one of whose characteristics is to bear greaterprominencethan the verb (Helbig and Buscha 1984, Bresson 1983:vol. 1, 562). Since they separatefromthe verb, they cannotbe treatedas beingpartof a complexlexical unity. A plausible analysis, which capturesboth properties,wouldconsist in analyzingthemas headsof intransitive selected PPs by the V (see Van Riemsdijk1988):

(i) a.

Wann werden wir [vp[pp[p an]] [v kommen]]? 'When will we arrive?' b. Wann kommen wir [vp[pp[p' an]] [v t]]? 'When do we arrive?'

In this case the "prefix" would be more deeply embeddedthan the V in the VP (especially if this moves to
I even in infinitival clauses; Giusti 1986, 1990).

NULL

THEORY

OF

PHRASE

AND

COMPOUND

STRESS

257

its application(cyclic or not). Thus, it leaves no way to link the assignmentof S to the most deeply embeddedconstituent. Even if assignmentof S were somehow to be linked to depth of embedding,the link would not be principled.That is, it seeminglywould not follow as a necessary consequence of the formalism.If correct, this approachalso shows that at least certain phonological phenomena may be directly syntax-driven, without recourse to prosodic-theoreticnotions such as the phonologicalphrase or intonational phrase. 6 The Focus and Presupposition Articulation the Sentence of The study of phrase and sentence stress is not complete without considering the "information" articulationof the sentence into F(ocus) and P(resupposition)(Chomsky 1970),a distinctionthat recalls that found in other traditionsbetween "new" and "old" information(Halliday 1967/68),"rheme" and "theme" (Firbas1964and referencescited there), or "comment" and "topic" (Chomsky 1965:221,Dahl 1974).21 in Such distinctionspertainto discourse grammar that they determine"the relation
of the utterance to .
.

. utterances to which it is a possible response, and to other sen-

tences in the discourse" (Chomsky 1970:205).For example, in the context of a question like (44a), which introducesJohn into the discourse and shows ignoranceof his actions, an appropriateanswer will have John as part of the presupposition(or old information, or theme, or topic) and the VP as the focus (or new information,or rheme, or comment). Conversely, in (45b), the answer to (45a), John will be the focus and the VP the presupposition. (44) a. b. (45) a. b. What did John do? [John] [left]. P F Who left? [John] [left].
F P

The absoluteprominenceof the sentence falls in both cases on the phrasethat constitutes the focus, the VP left in (44b), and the subject NP John in (45b). Note that in neither case is any contrastiveor emphaticstress necessarilyinvolved. For this reason the stress contour of (45b) has occasionally been taken to be an exception to Chomskyand Halle's Nuclear Stress Rule (see Schmerling1976). But this is not really so. One must distinguishthe formal sentence grammarprocedure that determines where the prominenceof a phrase will be located (the Nuclear Stress Rule or the null alternativediscussed above) from the discourse grammarpro-

21 I don't mean to say that these dichotomiesexactly correspondto each other (see Gundel 1977:chap. 2 for a review of the varioustraditions),but they all seem to try to separatethe partof a sentence that provides information "assumedby the speakernot to be sharedby him and the hearer"from the part "assumedto be shared" (Jackendoff1972:230).

258

GUGLIELMO

CINQUE

cedure that determinesthat the prominenceof the phrasein focus will win out, in relative terms, over that of the presupposedphrase. That the two proceduresare indeed different,and must be distinguished,is shown by the fact that the formalprocedureis at work both in the phraseconstitutingthe focus and in the phrase constitutingthe presupposition,as we see from cases slightly more complex than (44) or (45) (where ""' is less prominentthan "" in absolute terms): (46) a. b. (Any news of John?) [NP Our poor child] [vp is in bed with a 'flu]. (Who's in bed with a 'flu?) [NP Our poor child] [vp is in bed with a 'flu].

Both the presupposition(the NP in (46a) and the VP in (46b)) and the focus (the VP in (46a) and the NP in (46b)) have a detectable prominence, determined by the formal sentence grammarprocedure, which applies blindly to each phrase. The fact that in both cases the prominenceof the phrase in focus will ultimatelybe higher than the prominenceof the phrase constitutingthe presuppositionis a different matter (see Jackendoff 1972:237).22 In this light, the sentence grammarprocedure of phrase stress assignment can be conceived of as a formal means for locating the main stress of a phrase (the most deeply embedded constituent under the null theory), and for markingthe relative degree of prominenceof the various stresses in the phrase (in terms of the respective numberof asterisksin the metricalgrid).The discourse grammar procedureinstead may be taken to impose the requirementthat the main stress of the phrasein focus be more prominentthanthe mainstress of the presupposition(in absolute terms). The well-knownambiguityin focus of a sentence like (47) (Chomsky 1970)-where,
22 As expected, when a VP is preposed, it is the rightmostconstituentin Italianand the constituentleftadjacentto the V in Germanthat receive VP-internal prominence,irrespectiveof whetherthe VP constitutes the focus or the presupposition:

P F (i) a. [vp Vendutola macchina]ancoranon ha. sold the car yet not he has 'Sold his car, he hasn't yet.' P F b. [vp Sein Auto verkauft]hat er noch nicht. his car sold has he not yet 'Sold his car, he hasn't yet.' P F (ii) a. [vp Venduto la macchina]pare che abbia. sold the car it seems that he has 'Sold his car, it seems he has.' P F b. [vp Sein Auto verkauft]hat er sch6n. his car sold has he already 'Sold his car, he has already.'

A NULL

THEORY

OF PHRASE

AND

COMPOUND

STRESS

259

as shown by the variety of answers in (48), any of the phrases indicatedcan be focusis a direct consequence of the interplayof the two procedures.(I follow here Chomsky's convention of capitalizingthe word bearingthe most prominentstress.) (47) Was he [warned[to look out for [an ex-convict [with a red [SHIRT]]]]]? (48) a. No, he was warned to look out for an ex-convict with a red [TIE]. b. No, he was warned to look out for an ex-convict [with a CARNATION]. c. No, he was warned to look out for [an AUTOMOBILEsalesman]. d. No, he was warned [to expect a visit from the FBI]. e. No, he was [simply told to be more CAUTIOUS]. The ambiguityarises from the fact that the most prominentstress of a phrase will be located by the formalprocedureon the most deeply embeddedconstituentof the phrase, and the fact that the noun shirt qualifiesas the most deeply embeddedconstituent of all of the phrases indicated in (47), each one potentiallyqualifyingas focus.23 If the main stress were on red, the ambiguitywould disappear,since that is the most deeply embedied constituent only of the dominatingAP, not of the NP containingit, nor of any other more comprehensivephrase. Hence, main stress on red would be compatibleonly with the AP being in focus. Depending on context, the most deeply embedded constituent of a focus phrase (where the formalprocedurewould predictthe main stress to fall) may happento be old information,thus qualifyingas part of the presuppositionratherthan of the focus. For example: (49) a. b. I'd give the money to Mary, but I don't TRUST Mary. (Schmerling1976: 59) Has John read TristramShandy? He doesn't READ novels.

In this case the constituent is destressed, the main stress falling on the most deeply embedded constituent left in the phrase that qualifiesas focus. Such destressingis possibly a consequence of the "marginalization" the presupposedconstituent(Antinucci of and Cinque 1977, Calabrese 1990), whereby it is removed from its base position and adjoinedto some highernode, thus ceasing to be the most deeply embeddedconstituent of the phrase. Certainelements, such as anaphoricpronouns and epithets, are inherentlyold information, so to speak-hence marginalized(50a), unless specially contrasted (SOb).
23 Certainlanguagesappearto be able to disambiguate cases like (47). In Polish, if the most deeply embedded constituentis stressed on the initialsyllableratherthanregularly the penult,only that constituentmay on be focus. See the contrastbetween (ia) and (ib), noted by Dogil (1980:225f.),who points out that only in the latter case can any of the phrasesindicatedbe focus. (i) a. [waznosc [komunikacji [SAmochodowej]]] b. [wa2nosc [komunikacji [samochoDOwej]]] the importance of traveling by car

260 (50) a.

GUGLIELMO

CINQUE

I'd give the money to JOHN, but I don't TRUST him/that bastard. (Schmerling1976:71) b. John insulted Mary, and then SHE insulted HIM. (Lakoff 1968)

Other possibilities exist, but the few remarksjust made should be sufficient to justify postulating the two different procedures for stress assignment. Failure to distinguish them has led certain authors to deny the existence of a formal means to predict the location of the most prominentstress of a phrase based on structuralprinciples.24 But we have just seen that their conclusion is not warranted. Nonetheless, their work provides importantinsights on the not always easy task of determiningwhat counts as focus in a certaindiscourse, and in out-of-the-bluecontexts. Concerningthe latter, for example, from Schmerling's(1976:41f.)discussion of the minimalpairof out-of-the-bluesentences in (51), one can surmisethatdetermination focus of and presuppositionmay depend on knowledge of the world's events. (51) a. TrumanDIED. b. JOHNSON died. As Schmerlingrecalls, when (Sla) was uttered PresidentTrumanhad been written and spoken of by the news mediafor some time because of his ill health;so it was appropriate to consider him as part of the presupposition,whereas the news was the terminationof his critical state. PresidentJohnson, instead, died somewhatunexpectedly. He was not on people's minds as Trumanhad been, so it would have been inappropriate take him to as part of the presupposition. Given that the entire event was new, one may wonder why stress on the VP could not have served as the unmarkedstress patternin this case too in which the entire CP is in focus. After all, V is more deeply embeddedthanthe subjectN, since it is dominated by its own projectionsplus at least the projectionsof T and Agr (Pollock 1989, Belletti 1990, Chomsky 1991). An answer may come from a comparisonwith those languages,like Italian,in which the subject may remainin its D-Structureposition. In Italianthe sentences appropriate to the above contexts are (52a) for (Sla), and (52b) for (Sib), not (52c), the word-byword translationof (Sib). (See Dezso 1982:118f.for similarremarkson Russian.) Trumane MORTO. (52) a. b. E' morto JOHNSON. c. (*)JOHNSONe morto. (52b), as a reportfor a totally new event, is expected if the subject is in the D-Structure
24 Bolinger (1972), Schmerling (1976), Ladd (1980), to some extent, Bardovi-Harlig (1986), Bing (1979a,b), Erteschik-Shir and Lappin (1983), Gussenhoven (1984), among others.

A NULL

THEORY

OF PHRASE

AND

COMPOUND

STRESS

261

objectposition. This is because it qualifiesthereas the most deeply embeddedconstituent of (the VP and) the entire CP.25 The inappropriatenessof (52c) in the same context would follow if a preverbal subject (in the Spec of Agrs) were necessarily part of the presupposition.This is in fact just what Gueron (1980) suggests on partly similargrounds. See, in particular,her distinction between predication sentences (like (51a)), in which the preverbal subject is presupposed, and presentation sentences (like (51b)), most common with verbs of appearance, in which "the subject is the (unmarked)Focus" (p. 659). (Also see Firbas 1964,Allertonand Cruttenden1979,Culicoverand Rochemont 1983:fn.34, Faber 1987). Given that no postverbal subject is possible in English (*Died JOHNSON), the subject must move to the Spec of Agrs. This, however, would give rise to a predication sentence in which the subjectis presupposedandthe predicateis focus-an inappropriate state of affairs in such contexts where the entire event is new. The way out apparently consists in markingthe least predictableelement in the event (the subject)as focus while treatingthe predicate as presupposed (dying is one of the possible accidents that may occur to someone), in a kind of weighing of relative predictability.26

25 Nothing changes even if propernames, as definite NPs, cannot remainin situ, but must raise to the postverbalVP subjectpositionof transitiveand unergativeverbs (Belletti 1988).(However, for evidence that propernames behave more like indefinitethan definite NPs, hence can remainin situ, see Beninca 1991:fn.

5, where the relevance of such cases as All'improvviso e entrato Giorgio dalla finestra 'Suddenly entered

Giorgiofrom the window' is noted.) This is because the V (whetherfinite or participial) moves out of VP, to the respective Agr position, thus leaving the subjectas the most deeply embeddedconstituentanyway. This can be seen with sentencescontaining verbsdenotingtypes of ordinary unergative happenings. for example, (i), can be utteredin out-of-the-blue contexts where the entire event is new (notjust Gianni). (i) Ha telefonato GIANNI. has telephonedGianni 'Giannihas telephoned.' (ii), however, is possible only if Giannialone is in focus and V plus object is the presupposition: (ii) a. Ha telefonato a me GIANNI. has telephonedme Gianni 'Giannihas telephonedme.' b. Ha telefonato la notizia IL NOSTROCORRISPONDENTE. has telephonedthe news our correspondent 'Ourcorrespondent telephonedthe news.' has 26 The focus/presupposition structure intransitive of sentenceswith unaccusativeverbs in Englishappears to be preservedunderembedding: (i) a. I just heardthat TrumanDIED. b. I just heardthat JOHNSONdied. The impossibilityin Italianof the word-by-word translationof JOHNSON died in the same context is presumablyrelated to the existence of the less markedoption (52b). It is possible that such unmarkedstress patternsas (iia-c), reportedby Schmerling(1976:21f.),shouldfall underthe same generalization. (ii) a. There is a CAR coming. b. I don't know what I'm going to do-I don't have any money and the RENT's due. c. You left the WATERrunning. Interestingly,the corresponding Italiansentences cannot have the same contour, but instead have rightmost prominence(or subjectinversion).

262

GUGLIELMO

CINQUE

Otherwell-knownminimalpairs, possibly susceptibleto similartreatmentvia computation of relative predictability,are (53a-b) and (54a-b), from Bolinger 1972. (53) a. I have a POINT to make. b. I have a point to EMPHASIZE. (54) a. The end of the chapter is reserved for various PROBLEMSto solve. b. The end of the chapter is reserved for various problemsto COMPUTERIZE. Indeed, Bolingernotes, "In phrases like.. . workto do, clothes to wear, . . , the verb is highly predictable: ... clothes are to wear, work is to do, .... Less predictable verbs are less likely to be deaccented. Whereone has lessons to learn, one will probably have passages to memorize" (1972:633-634).(For remarksalong similarlines, see Berman and Szamosi 1972:312.)27 In this context, the stress propertiesof a sentence like [[Theauthor[of manypopular articles [on the effects [of senescence]]]] [[died]]], mentionedin section 4 as a potential difficulty for the null theory, ceases to be problematic.The most prominentstress will fall either on the subject (in which case it will be located on its most deeply embedded constituent senescence) or on the predicate(died), dependingon which of the two constitutes the focus portion of the sentence. (In this connection, also see the discussion at the end of section 7.) Other cases undoubtedlydeserve specific discussion. Many English stress patterns are still poorly understood or unanalyzed, as are a number of cross-linguistic differences.28 However, it seems that, whatever the results turn out to be, they should not affect the main point of this section, namely, that a formalsentence grammar procedure
27 It is however not inconceivablethat such contrastshave a structural basis. I leave the questionopen. observedby Newman(1946),thatled Bresnan(1971, 1972)to propose (53a)is the same type of case, originally that the Nuclear Stress Rule applies at the end of each syntactic (NP and S) cycle ratherthan on surface some of Bresnan'sassumptionsinto the presenttheory, structure.Quiteapartfromthe difficultyof translating her suggestionis not without empiricalproblems,as Lakoff (1972)and Bermanand Szamosi (1972)pointed out. Bresnan (1972:332f.)in fact acknowledgesone such problem.Anotheris providedby the Italianfacts discussed here in footnote 28. For furthercriticaldiscussionof Bresnan'sproposal,see Selkirk 1984:239ff. Chomskyand Halle's (1968:15)insight All in all, it seems that there is no real motivationfor abandoning that "[stress] contoursare determinedin some mannerby the surfacestructureof the utterance."Of course, of articulation procedure,sensitive to the focus and presupposition what interferesis the discoursegrammar the sentence. 28 For example,it appearsthatItalian fromEnglishin generallynot allowingprimary differssystematically stress in nonfinalposition (in nonemphaticcontexts). Many of Bresnan's(1971)contraststhus appearto be neutralizedin Italian.For example: istruzioni di PARTIRE. (i) a. Ho I have instructionsto LEAVE b. Ho da LASCIARE. istruzioni I have INSTRUCTIONSto leave (also possible: Ho da lasciareISTRUZIONI.) (ii) a. John chiese cosa Helen avesse SCRITTO. John asked what Helen had WRITTEN b. John chiese quali libri Helen avesse SCRITTO. John asked what BOOKS Helen had written

A NULL

THEORY

OF PHRASE

AND

COMPOUND

STRESS

263

of phrase stress assignmentshould be carefullydistinguishedfrom a discourse grammar procedurethat favors the main stress of the phrase in focus over the main stress of the phrase constitutingthe presupposition. 7 Some ResidualQuestionsand a Refinement In this section I will first consider certain structuresdisplayinga stress pattern that is at first sight problematicfor the null theory. In each case we will see that alternative analyses exist in the literature,or appearto be plausible on closer scrutiny, which are indeed compatible with the null theory. One residualclass of cases will also point to a particularrefinementof the system proposed above. Consider, to begin with, (55a-b). (55) a. b. Loro stanno seguendo la lezione attentamente. They are following the lecture attentively.

In both Italian and English the greatest prominence of the sentence is, under normal conditions, on the adverbialphrase(AdvP),the rightmostconstituent.This is unexpected if the sentence structureis that shown in (56), the one assumed traditionally.
(56) . . . IP

NP loro
they

I' I stanno
are

VP V' AdvP NP la lezione the lecture attentamente attentively

seguendo following

Here, the most deeply embeddedconstituentis the object N (even abstractingfrom the extra DP projection)and consequentlyit, of all the VP constituents, should receive the greatest prominence. Given this structure,the only way for the AdvP to bear greatest prominence would be for it to be the only constituent in focus, with stanno seguendo
(iii) a. Giorgioha trovato qualcunoche vorrebbeche tu INCONTRASSI. someone he'd like you to MEET Giorgiofound b. Giorgioha trovato degli amici che vorrebbeche tu INCONTRASSI. SOME FRIENDS he'd like you to meet Giorgiofound

264

GUGLIELMO

CINQUE

la lezionelthey are following the lecture constitutingthe presupposition(as is the case in the context of a question like How are theyfollowing the lecture?). But this is clearly not necessary. (55b) can be an answer to Whatare they doing?, with the entire VP as focus. Fortunately, there is evidence that (56), the traditionalsyntactic analysis of such cases, is incorrect. Larson (1988:fns. 11 and 49) and Stroik (1990), extending Barss and Lasnik's (1986) analysis, provide evidence that objects asymmetricallyc-commandVP adverbials (at least) at S-Structure.Hence, a more accurate representationof (55a-b) would be somethinglike (57), where the AdvP indeedqualifiesas the most deeply embedded constituent of the VP.29
(57) .
.

. IP

NP loro they I stanno are

If VP V
V VP

seguendo following

NP la lezione the lecture V

Vf

AdvP attentamente attentively

29 Stroik(1990)derives this particular on c-command relationbetweenobjectsand VP adverbials the basis of object-adverbial asymmetriessuch as the following(p. 656):

(i) Negative polarity

a. John saw no one anywhere. b. *Johnsaw anyone nowhere.

(ii) Superiority

a. Who did you see where? b. *Wheredid you see who?

(iii) Bound pronouns

a. I saw everyone the day before he did. b. *1 see a man who plays Santa on it every Christmas. etc. The double VP structurewith movementof the V from head to head is Larson's.

NULL

THEORY

OF

PHRASE

AND

COMPOUND

STRESS

265

If so, the unmarkedstress patternof (55) with mainstress on the VP adverbialis precisely what the null theory predicts.30 If the first object of the double object constructionin English (the "dative") asymmetricallyc-commands (is higher than) the second (see Kayne 1984, Barss and Lasnik 1986, and Larson 1988), then the fact that the second object bears greatest prominence in the VP is expected under the null theory, since it is more deeply embedded. Analogously, if directobjects asymmetrically c-commandprepositional objects (Larson 1988), it is no longer surprisingthat prominencewill fall on the latter. Another potentially problematiccase is representedby heavy NP shift. Consider the following alternations: (58) a. Loro ricordaronol'appuntamento a Carlo. they reminded the appointmentto Carlo 'They remindedCarlo of the appointment.' b. Loro ricordaronoa Carlo l'appuntamento (che gli avevano dato). they reminded to Carlo the appointment(that they had given him) 'They remindedCarlo of the appointmentthat they had given him.' (59) a. Gianni incontro il figlio arrabbiato. his son angry Gianni met b. Gianni incontro arrabbiato figlio. il Gianni met his son angry (60) a. Carlo parlo a Maria di noi. Carlo spoke to Mariaabout us b. Carlo parlo di noi a Maria. Carlo spoke about us to Maria

In such cases too the greatestprominenceis on the rightmostconstituent, whateverthat

is.

This would follow once again from the null theory if we were to adopt Larson's (1988, 1990) general approach, with its uniformrightwarddownwardbranching: ... elements appearingon the right . . . are typically lower in the phrase markerthan elements to their left" (1990:591)-hence, we expect, will bear greaterprominence.3' For
30 If speaker-oriented adverbslike probablyare not underVP, butundersome higherfunctionalprojection (see Belletti 1990),thenthe nulltheorypredictsthatthey will not beargreatestprominence even whenrightmost in the sentence-a correct result. For example: (i) a. Giorgioe uscfto, probabilmente. (vs. *Giorgioe uscito probabilmente.) b. Giorgioleft, probably. (vs. *Giorgioleft pr6bably.) 31 As RichardKayne (personalcommunication) has noted, the predictionis not always fulfilled,though. For example, particles,in some contexts, cannotbearthe greatestprominence the VP (and sentence), even of if they are the rightmostconstituent: (i) John threw a b6ok away.

266

GUGLIELMO

CINQUE

such cases Larson suggests a rule of light predicate raising, which moves the V + XP sequence, reanalyzedas V, leftwardaroundthe object. This particularanalysis faces a problemin Romance, where finite Vs raise to Agr, across temporaland aspectual adverbs. These adverbscan separatethe V from the XP, and moreover cannot intervene between the putatively reanalyzed V + XP sequence and the heavy-NP-shiftedobject: (61) a. Marianon ricorda mai a Carlo gli appuntamenti lavoro. di Mariadoes not remindever to Carlo business appointments 'Marianever remindsCarlo of business appointments.' b. *?Marianon ricordaa Carlo mai gli appuntamenti lavoro. di il (62) a. Gianni non incontra piCu arrabbiato suo direttore. Gianni does not meet any longer angry his boss 'Giannino longer meets his boss angry.' b. *Gianninon incontraarrabbiato il suo direttore. piiu

All this suggests that the V raises alone across such adverbsand cannot in fact be treated as part of a complex V including the XP. If it is the case that the heavy-NP-shifted element must be the only constituentin focus (cf. Rochemont's(1978:33)termfocus NP shift), then there is no problem for the null theory even if the NP is right-adjoined to VP (hence is not the most embedded constituent). This is because, as noted, the main stress of the focus constituent is the most prominentstress. A possible alternative, compatible with Larson's general approach, would be to assume a light XP shift, which adjoins the oblique complements(or adjuncts)leftward across the NP object, much as is generallyassumedfor German,which would thus look more similarto English and Italian, modulo the rightwardraising of the V to Agr.32It
This cannot be due to their inabilityto bear main stress, as shown by the followingcases: (ii) a. John threw it away. b. Whatdid John throw away? Kayne (1985)providesevidence that these constructionsinvolve a "smallclause" particlephrase,where the object is in fact the subjectof an intransitiveparticle: (iii) John [vp threw [partp[a book] [Part'[Part away]]]]. If this is so, the possibility arises of extending to these cases the focus/presupposition analysis of English intransitiveclauses with verbs of appearance,discussed in footnote 26 and the corresponding text. The systematic difference with Italiannoted there is, suggestively, found here too: *John ha buttato[un libro via]
(unless un libro is contrasted) versus John ha buttato [via un libro].

Other cases where as expected the rightmostconstituentcannot bear the greatest prominenceof the sentence are, as alreadynoted, those with sentence-finalspeaker-oriented adverbs(*Johnleft pr6bably)and
other types of adjuncts. The well-known ambiguity of John doesn't beat his wife because he loves her, where

the adjunct can be either inside VP, within the scope of the negation, or outside it, appears to correlate systematicallywith two differentstress patterns,consistent with the null theory. 32 Note that the GermanVP is not a perfectmirror imageof Englishand ItalianVPs. As a matterof fact, the threelanguagesessentiallyshow the samedomainasymmetries, withdirectobjectshigherthancomplement PPs, locative PPs, and adverbials. For an interestingaccount of this apparentasymmetrybetween Englishand German(actuallybetween

A NULL

THEORY

OF PHRASE

AND

COMPOUND

STRESS

267

may be that such PP alternationsas those in (60a-b) are not to be treated as cases of heavy NP shift (or light XP shift). (For discussion, see Larson 1990.) Anotherquestion is raisedby coordinatestructures.As has occasionally been noted (by Kiparsky (1966:82ff.), among others), the last conjunct of a coordinate structure usually bears greatest prominence: (63) a. b. c.
(64)

[[Kafkas Werke] und [die moderne Novelle]] 2 3 2 4 [[Kafka's works] and [the modern short story]] 2 2 3 4 [[le opere di Kafka] e [la novella moderna]]
XP

This is unexpected underthe null theory if coordinationis symmetric,as often assumed:

XP

and

XP

However, in this case, too, there is some evidence for an asymmetricrepresentationof coordinationin which the last conjunctis more deeply embeddedthan the others. Ross (1967:162-167)presents both syntactic and phonologicalevidence that the coordinating conjunctionforms a constituentwith the following conjunct, so that a more appropriate representationof (63) would be (65) ratherthan (64). (65) XP

XP
and XP

Kayne's (1984) binary branchingrequirementfor syntactic structuresprovides an additional conceptual reason to prefer (65) over (64). He has, in fact, explicitly argued(in Kayne 1983)that coordinatingconjunctionsshould be treated as heads (in X-bar terms) of a maximal projection (for similarclaims, see Munn 1987 and Larson 1990:595f.).If so, coordinatestructuresare entirelycompatiblewith, andin fact support,the null theory of phrase stress. Admittedly, other questions that deserve attention remain open. Should the null theory withstanda wider and deeper scrutiny, then it would not be unreasonableto use it to question certain syntactic analyses that do not conform with its predictions.
VO and OV languages),see Haider 1991. Haider'sargumentsto the effect that UniversalGrammar permits rightward downwardbranching,but no leftwarddownwardbranching, providea straightforward explanation for the fact that in VO languagesit is the rightmostconstituentof the VP that bears greatest stress, whereas in OV languagesit is not the leftmost constituentof the VP, but the constituentto the immediateleft of the V. (See section 8 below.)

268

GUGLIELMO

CINQUE

There is still one class of facts that are apparentlynot reconcilable with the assumptionsgrantedso far, and thatpoint to a particular refinementof the presentanalysis. In section 6 we saw how the unexpectedly possible stress prominenceof the predicate
in a sentence like [[The author [of many popular articles [about the effects [of senes-

cence]]]] [died]] could be renderedcompatiblewith the null theory by taking into consideration an independent dimension of the sentence: its articulationinto focus and presupposition.The same accountis not available,however, for the comparablesituation found in NPs. Consider the following examples, pointed out by Richard Kayne and MorrisHalle (personalcommunications),respectively: (66) a. b. [the [man [from [Philadelphia]]]]'s hait [der [[von sieben [jungen[Italienern]]] entdeckte] Impfstoff] the by seven young Italians discovered vaccine

These phrases are not internally partitionedinto a focused and a presupposed part. Rather,they are usuallythemselves partof the focus or the presupposition the sentence of in which they appear. Yet in spite of the specifier's complexity, the main stress falls on the head. This suggests the need to refine the procedure of grid construction utilized above (see (10)). One way of drawing a principleddistinction between heads and complements on one side and specifiers on the other would be to capitalize on how the property of recursion (namely, the property of having a certain category dominatedby a category of the same type, ad infinitum)is realized within a phrase. It is well known that the complement, not the specifier, introducesrecursion, so that dependingon the relative position of the complementand the head a languagewill be right-recursive (say, Italian) or left-recursive (say, Japanese). (I abstractfrom the case of mixed languages, where the position of the complementwith respect to the head may vary fromphraseto phrase.) That there is a real asymmetrybetween the recursiveand the nonrecursivesides is shown by the fact that (as has often been noted) the side opposite to that of the complement has only limited, or selective, recursion, at least for the lexical phrases NP, AP, AdvP, VP, and PP (see Zwarts 1974, Emonds 1976:19,1985:130ff.,Williams 1982, and Longobardi1991a:95-100).The asymmetryin questioncan be roughlycharacterized as follows: When on the recursive side, recursionis possible either to the recursive or to the nonrecursive side. When on the nonrecursiveside, recursionis possible only to the same nonrecursiveside.33 The latterpropertyis exemplifiedby the followingcases, simplifiedfor convenience: (67) a. b. c. He is a [NP proud (*of his children)father]. He was [AP less (*thanus) sympathetic]. He walks [AdvP more (*thanus) rapidly].

as 3 Emonds(1985)refers to this restriction the "Recursion Restriction,"and Longobardi (1991b)refers to it as the "ConsistencyPrinciple."The notion of "recursiveside" is Emonds's (1976:19).

NULL

THEORY

OF

PHRASE

AND

COMPOUND

STRESS

269

d. He [vp specially (*for us) made the cake]. e. The boat was [ppthree miles (*furtherthan Sue's) off the coast]. As noted by Emonds and Longobardi,this restrictiondoes not hold for (certain)functional projections, free right recursion being allowed in the specifier positions of CP, AgrP, and DP. For example: (68) a. b.
C.

[cp On which day of the week [c are they coming]]? [AgrP The destructionof the documents [Agr' was deliberate]]. [DP the man from Philadelphia's[D' hat]]

Now, accordingto the null theory, a consequence of the generalprinciplesof grid constructionis that location of the main stress is simply a function of depth of embedding. Qualifyingthis notion, suppose we take it to involve considerationof the recursive side of a phrase in such a way that the relevant notion of embeddingis now limited to the continuous path unitingfrom the bottom all and only the nodes found on the recursive side and on the X-bar projectionline of a phrase up to the node that is expandedon the nonrecursiveside. A simple example, (69), will illustratethis idea. (69) X"

W" X' W'

W

l\-~

Y"
Y'

K"

I
K

I
Z'

One path of embeddingis that which unites Z to X" (Z, Z', Z", Y', Y", X', X"). It is a well-formedpath of embeddingsince all the nodes that it connects are either on the Xbaraxis (Z, Z', Z",Y', Y",X', X")or expandedon the recursiveside (assumingrecursion to be fixed to the right). Anotheris the path of embeddingunitingK to W".By contrast, the path of embeddingunitingK to X"is ill formed, since it contains at least one branch, (W",X"), that is neither on the X-bar axis nor on the recursive side. Furthermore,let the mainpath of embeddingbe that path which has the root of the constituent as one of its extremities (in (69), the path connectingZ to X"). Qualifyingthe notion of embeddingin this way, we obtain, as a consequence, that recursion of the specifier of CP, AgrP, and DP be considered a separate (minor) path

270

GUGLIELMO

CINQUE

of embedding.If so, it is not unreasonableto take the principlesof grid constructionto operate on this path on a separate cycle. Assume, further, that when a minor path of embeddingjoins the main path (i.e., when the minor cycle joins the main cycle), only the end result of the former is visible in the form of a single asterisk. This implies that no matterhow complex the specifier of CP, AgrP, and DP, it will never win over a complement, or, in the absence of one, over the head. Considerthe simplifiedstructures(70a-b), which underlie(71a)and (71b), respectively. (70) a. YP
*

XP X'
X
*
I*

b. YP
ZP

XP X'

x
*

z
(71) a. b. the man from Philadelphia'sletter to the dean the man from Philadelphia'sletter

In (70a), where the complementis present, the complementwill prevail over the head (andthe specifier), since its head will be the first to receive an asterisk (it being the most deeply embedded constituent on the main path of embedding).When no complementis present, as in (70b), the head of the phrase will qualify as the most deeply embedded constituent. This is because the internal structureof YP is visible at the level of the main cycle only in its end result, in the form of a single asterisk. Since the head receives an additionalasterisk at the X' level, it will also attractthe asteriskof the XP level. This refinementthus appearsto have the correctempiricalconsequencesfor the stress contour of the apparentlyproblematiccases (66a-b).34 It may also shed some light on the observation made in section 6 (following considerable tradition)that a sentence with a preverbal subject cannot count as a single
34 It also appears makethe correctprediction to for the stresscontourof conjunctsunderthe X-baranalysis of coordination discussed above. Accordingto thatanalysis, the conjunction the head of an X-barprojection is the taking(as a first approximation) two conjunctsas its specifierand head, respectively([XP [and XP]]). In head-initial languages,this will imply, then, that however complex the first conjunctis, the main stress will fall on the second, since the formeris on the nonrecursiveside. This indeed appearsto be true: (i) ffil primo [dei due autori [del libro [di poesie]]]] e [Mario]] the first of the two authors of the book of poems and Mario This refinementof the null theory has both some points in common (in enhancingthe role of complements) and some points of divergence(in allowingheads to win over specifiers)with Duanmu's(1990:chap. notion 4) of "nonheadstress" ("whereby in a head-nonhead relation, the stress is assigned to the nonhead";p. 174). In the present system, the cases cited by Duanmuwhere a specifierappearsto have more prominencethan the head could perhapsbe interpreted compounds,as suggestedby Selkirk(see Duanmu1991). as

A NULL

THEORY

OF PHRASE

AND

COMPOUND

STRESS

271

focus constituent (as opposed to one with a postverbalsubjectin an unaccusativestructure). Either the subject is focus and the predicate is presuppositionor vice versa. In this framework,what prevents the entire sentence from countingas a single focus is the fact that it is made up of two distinct paths of embedding.An unaccusative structure with an inverted subject in situ is different, being in fact made up of a single path of embedding.
8 Some Typological Evidence for the Null Theory

As we have seen, accordingto the null theory, a phrase's main stress is located on its most deeply embedded constituent. This is ordinarilythe innermostcomplementof the phrasehead;hence, the locationof the mainstress is expected to covary with the location of the innermostcomplement,as determinedby the head-complement parameter.Thus, in a VP, main stress should be to the right of V in VO languagesand to its left in OV languages. Caution is needed, however, since this is one of a numberof simplifications.For example, in section 5 we noted that the D-Structureinnermostcomplement of a head can be moved higher up in the tree so that another constituent ends up as the most deeply embedded constituent, consequently receiving the main stress. The responsibleway to proceed in checkingthe correctnessof the correlationwould be to reach a perfect understanding the S-Structureconstituency of the language,and of then consider its stress patterns-a hardlyfeasible task. Nevertheless, it is possible to find in the literaturesome (at least) suggestive evidence going in the direction of what the null theory predicts. Maling (1971) observes that a version of the CompoundRule (which gives prominence to the leftmost element of a constituent) "applies to all categories and nodes (except prepositionalphrase)" in Old English (p. 382) and that this appears to be connected to the verb-finalnatureof the language(p. 382, fn. 1). McCawley (1977:273)notes that the Nuclear Stress Rules of English and Japanese are "mirrorimages" of each other ("the rule that the first accent in a constituent predominates is the Japanese analogue to the English 'nuclearstress rule' (Chomsky and Halle 1968), according to which the last accent in a constituent predominates"). The correlationwith the head-finalcharacterof Japanesephrases (Kuno 1973)and the headinitial characterof English phrases seems hardlyaccidental. In a more comparativevein, Donegan and Stampe (1983:337),contrastingtypologically the Mundafamily of languageswith the Mon-Khmerfamily, make the important observation that phrase stress correlates with constituent order, so that in the Munda languages, whose phrases are head-final ("operator-first,"in their terminology), the phrasestress is phrase-initial (e.g., ['O V]), whereasin the Mon-Khmer languages,whose phrases are head-initial,the phrase stress is phrase-final([V '0]). Similargeneral observations are also found in the more traditionaltypological lit-

272

GUGLIELMO

CINQUE

erature, in particularin Dezso's work on the typology of theme-rhemestructure and sentence stress (see Dezso 1974, 1977, 1982).Basinghis workon datafromUralic, Altaic, and some Indo-Europeanlanguages, Dezso suggests that in SOV languages "the usual place of sentence stress and hence of rheme is the position immediatelypreceding the verb," whereas in SVO languages "the usual place of sentence stress is after the verb either in an immediatelypostverbal position or after an unstressed element" (1977:7). (Also see Dezso 1982:149f.)Although in his theoreticalframeworksentence stress is determinedby theme-rhemestructure(and only indirectlyby word order), in practice Dezso relates accentuationtypology to word order directly-a revealinginsight. Dezso's results are confirmedand furtherextended in Kim's (1988) study, where Dezso's word order-stress correlationis found to obtainin other languagefamilies. The focus (and sentence stress) falls SOV languagesexamined there in which the unmarked in the position immediatelyprecedingthe V are Telugu,LaccadiveMalayalam,andTamil of the Dravidianfamily, Dogri, Bengali, Gujarati,and Hindi-Urduof the Indo-European family,35 Sherpaof the Sino-Tibetanfamily, Mongolianand Turkishof the Altaic family, and Japanese and Korean. For work along the same lines, devoted to mixed languages (Hungarian,German, etc.), also see Harlig and Bardovi-Harlig1988. Another mixed language, apparently,is Afghan Persian, as discussed by Bing (1980). 9 On the Stress Patternof Compounds Since the dependence of compoundstress on the internalconstituency of the compound is (if anything)even more strikingthan that of phrases, it is temptingto try to extend the null hypothesis of phrase stress to compoundsas well. Ideally, one should be able to derive the stress pattern of compounds from the stresses of the component words, the internalconstituentstructureof the compound,and the metricalgridtheory, without rule. As with the stress of phrases, any cross-linguistic recourse to any language-specific difference should reflect the different structuralrelations into which the component words enter within the compound, possibly subject to parametricchoices (see footnote 39). In what follows, I will show that this appearsto be possible for the three languages analysis of the constituent structureof we have been examining, under a finer-grained compounds. Should the analysis prove correctmoregenerally,then no language-specific CompoundStress Rule would be needed anymore-undoubtedly a desirable result. In English two-word compounds, the classical CompoundStress Rule "assigns primary stress to the first of the two peaks, reducingall other stress levels by one" (Chomsky and Halle 1968:92):36 (72) a. b.
c

[[kitchen] [towel]] [[towel] [rack]]

[eachers] ulnio)n]]

36

3 For a differentpictureof Bengaliphrasestress, however, see Hayes and Lahiri 1991:sec.3. I assume here, as above, that 3 representsthe highestdegree of word stress.

A NULL

THEORY

OF PHRASE

AND

COMPOUND

STRESS

273

More interestingis the case in which one of the two componentwords is itself a binary compound: (73) a. b. [[[kitchen][towel]] [rack]] [[kitchen][[towel] [rack]]]

in Here, dependingon the directionof branching (left branching (73a)and rightbranching in (73b)), the stress contour changes.37 In the first case the most prominentstress falls on kitchen, the leftmost subcomponent of the leftmost element of the compound. In the second case it falls on towel, the leftmost subcomponentof the rightmostelement of the compound. The stress pattern of (73a) requires no particularmodificationof the classical rule handlingtwo-word compounds. It follows from a cyclic applicationof the rule. In the innermostconstituent, prominenceis assigned to the leftmost peak kitchen, and in the outer constituent it is assigned to kitchen as well, since this word is also the leftmost peak of the outer cycle. Matters are not as simple with the stress pattern of (73b). Right-branching compounds such as this one requirea special qualificationin Chomsky and Halle's (1968: 93) system, and have called for some special statementin all treatmentsof compound stress since then. In Libermanand Prince 1977,the iff clause of their CompoundStress Rule (8b), repeated here as (74), is motivated by just such cases. (74) In a configuration A B c], if C is a lexical category,B is strongiff it branches. [c In Halle 1985and Halle and Vergnaud1987,Liberman Prince'sif clause is rephrased and as a condition on the retractionof stress (my italics): (75) In a constituentC composed of two or more words, retractthe rightboundary of C to a position immediatelybefore the head of C, providedthat C is dominated by a lexical category and that the head of C is located in the last word of C. Although both (74) and (75) derive the correct stress patternof (root) compounds, one cannot help wonderingwhy the rule shouldcontainthose very conditionsand not others. No principledreason appearsto exclude that the conditionfor (74) be "iff A branches," or that of (75) "(provided)that the head of C is located in the penultimateword of C."' It seems that the correct theory of compoundstress should derive as necessary the fact that the most prominentstress falls on A in the left-branching structure(76a), and on B in the right-branching structure(76b).

which implies in one case that kitchenand towel form a 3 The differencein the directionof branching, unit and in the other that towel and rack form a unit, correlateswith a predictablesemanticdifference:'rack for kitchen towels' vs. 'towel rack in the kitchen' (see Chomskyand Halle 1968:93). In this and similarcases, syntactic (subword)structuremediatesthe correlationbetween stress patterns I and semanticinterpretations. will come back to this questionlater.

274

GUGLIELMO

CINQUE

(76) a. D A

E C B

b. A

E D B C

The null theory of phrase stress appearsto have precisely these consequences under an analysis of the internalstructureof compoundsthat in partmodifiesthe standardanalysis along lines that can be independentlymotivated. For compounds, the standardanalysis recognizes the existence of a head (the rightmost constituent in English;Allen 1978, Williams1981)and of a modifier,but takes the two constituents to be of equal bar level, X? (the same level, in fact, as the compound
itself):38

(77)

[N[N

towel] [N rack]]

Suppose, however, that Universal Grammar uniformlyforbids such symmetricrelations in the lexicon (in word-syntax), as it does in the syntax proper, by requiringthat the relationbetween a head (the governor)and its complementsor modifiers(the governees) be asymmetric, with the head an X? category and its complements or modifiers XP categories (Stowell 1981, Muysken 1982, Chomsky 1986, Baker 1988).This amounts to saying that the head status of a constituentis singled out structurallyratherthan being assigned by rule (as with the Right-HandHead Rule of Williams 1981).39 In this light, a more accurate representationof a compound such as towel rack would be as shown in (78). (78)
[N[NP

towel] [N rack]]

This modification,promptedby conceptualconsiderations,also has an empiricalpayoff. As often noted in the literature,the modifierconstituentof a compoundcan be phrasal (with certain restrictions):40 (79) a. b. a [[groundto air] [missile]] a [[green vegetable(s)] [shelfl]

38 See, for example,Toman1982,Selkirk1982,1984,Di Sciulloand Williams1987.This is trueof Chomsky and Halle's (1968)analysis as well, except for the fact that they also appearto admit cases like [N[NP ] N]:

for instance,

[N[NP

American history] teacher],

[N[NP

blackboard] eraser] (p. 21f.).

39 This is not to say that Williams'sinsight(for compounds) superfluous. is One mustin any event specify

whetherthe structurally representedhead is the rightmostor leftmostconstituentof the compound,a matter of parametric choice, given that Englishand the Germaniclanguagesin generallocate it to the rightwhereas the Romancelanguageslocate it to the left. For some discussion,see Giorgiand Longobardi1991:chap. and 3 Di Sciullo 1991.
40 I have already mentioned Chomsky and Halle's (1968:22, fn. 9) [N[NP American history] teacher]. Also

see Fabb 1984:136,145, 190, Sproat 1985:199ff., Roeper 1988:205f., Hoeksema 1988:124ff., Lieber 1988(from which some of the examples in (79) are drawn),Visch 1990:app.1, and Libermanand Sproat 1992:162ff.

NULL

THEORY

OF

PHRASE

AND

COMPOUND

STRESS

275

c. d. e. f.

an [[ate too much] [headache]] an [[over the fence] [gossip]] [[everyday] [life]] an [[I turn the wheel of the universe] [air]]

All such cases would be excluded if a word formationrule like (80) were assumed (as in Williams 1981, Toman 1982, Selkirk 1982). (80) X? -> [xo X? X0] But if we have to admit a structurelike [xo[yp ] X0] for at least some compounds, then uniformityconsiderationswould suggest extending it to the simple case of towel rack as well. (Also see Hoeksema 1988for argumentsthat the modifierof a compound is a maximalprojection.) Although I will soon introducetwo furtherimportantrefinementsof the structure with the nulltheoryof stress assignment,already of compounds,this analysis, interacting compounds proves able to derive the stress patternsof the basic left- and right-branching (76a)and (76b) as a necessary consequence. Let us considerhow. In (81), equal to (73b) with labels added in conformitywith the previous discussion (and ignoringintermediate bar levels), the most deeply embeddedconstituentis towel, not kitchen-hence its more prominentstress. The correspondingmetricalgrid is given in (82). (81) N

N NP N kitchen NP N towel N rak

On the basis of the ill-formedness cases like (ia-b), of (i) a. *a [[the Bible] lover] b. *an [[every animal]eater] it is sometimesclaimed that modifiersof compoundsmay be of categoryN' (X', more generally),but never Withinthe DP hypothesis, (iamaximalprojections(Fabb 1984;but see his p. 143, and Sproat 1985:202ff.). b) no longer warrantthis conclusion. If "this suggests that a referentialnoun phrasemay not occupy this in1991 :sec. 3.4), it could Giorgiand Longobardi compoundposition" (Fabb 1984:143) (also see Lieber 1988:206, simply be that only the D projectionis missing, owing to its role in the referentialstatus of a noun phrase (Stowell 1989, Longobardi1991b),not the maximalprojectionNP, D's complement.

276 (82)
*
. . .

GUGLIELMO

CINQUE

( . ( *

(( * ) (( * ) * [[kitchen][[towel] rack]]

) )) ))

line 6 line 5 line 4

line 3

Now consider (73a), with labels added as in (83). (83) NP N NP N kitchen N towel N rack N

Here the most embedded constituent is kitchen; hence, it is this element that attracts more asterisks, as shown in the correspondingmetricalgrid (84).
(84)

* * * *

.
. . .

.
.

* ((((* )* [[[[kitchen]towel]] rack]

) )) ))

. .

) ) ) )

line line line line

7 6 5 4

line 3

As it stands, the above analysis of compoundstructureraises one conceptualproblem and, relatedto it, a more seriousempiricalproblem,in connectionwith morecomplex cases such as (85), in which the most prominentstress is on the leftmost subcomponent of the rightmostconstituent (towel), as in the more basic case (81). This structure,and the procedurefollowed so far, would seem to predict an incorrect stress contour, since the most embedded constituent appears to be the N hotel, not the N towel. At an intuitive level, just as we would like to say that the topmost N counts as a line 3 (word) asterisk when the first syntactic cycle is reached, like any other noncompound word (the hotel kitchen towel rack's size), so we would like to say that the N

hotel kitchen in (85), irrespective of the internal derivation of its stress, begins with a line 3 asterisk, so that (85) reduces in effect to the more basic (81). Even if correct, this

A NULL

THEORY

OF PHRASE

AND

COMPOUND

STRESS

277

(85)
NP

N NP N N NP N towel

hotel

kitchen

N rack

II

intuitionruns into a technical problemif the structureof hotel kitchen towel rack is the one indicated in (85). The reason is that irrespective of the internal derivation of its stress, towel rack would also count as a single line 3 asterisk when it combines with hotel kitchen, since it too is dominatedby N. We would thus obtain the wrong result once again:the most prominentstress shouldgo on the modifierhotel kitchen(ultimately on hotel), since this is the moredeeply embeddedconstituentof the two (beingdominated by the extra node NP). However, the discussion so far has failed to note an importantcategorialdifference between the two constituents hotel kitchen and towel rack. This difference suggests a way to begin to solve the conceptual problem and derive the correct empiricalgeneralizations. The key lies in the notion "head." Suppose that the notion "head" is one and the same in the lexicon and in the syntax, any apparentdifferencebetween the two being a function of the differentlevel of representationin which it is employed (in the spirit of Borer's (forthcoming)notion of ParallelMorphology). Within X-bar theory, a head is such only in relation to a maximal phrase that it projects. For example, the N [hotel kitchen towel rack] is a syntactic head only if it projects to an NP (ultimatelya DP), as in [a [cheap [hotel kitchen towel rack]]]. This means that if rack is the head of the compounds(81) and (83) (both compoundsdesignate kinds of racks, locate any inflectionalending on rack, etc.), it will have to project to a maximal phrase, which in turn implies that the topmost N is maximalin lexical terms with respect to the N immediatelydominatingrack. Since the topmost N is an X? in syntactic terms, we can agree to express its maximalityby markinglower-level nodes
on the lexical side with successive negative integers
(-

1, - 2,

. . .).

This gives (86) as

a more accurate lexical representationof (85).41

41 This recalls similarproposalsmade in the literature (see Selkirk 1982,amongothers), but differsfrom them in that N'- and N-2 are not taken here as subwordentities (e.g., "stem" and "root"). N-2 and N` constitute, with No, the projectionline of the compoundhead (N -2), which dominatesa full-fledgedword.

278

GUGLIELMO

CINQUE

(86)

No

NP

N NP N' N hotel N-' N-2 kitchen NP N' N towel

N-'

N-2

1l

rack

Besides expressing in a more perspicuousway the notion "head of a compound," this analysis appearsto have the correct empiricalconsequences, in interactionwith the startswith one (line 3) asteriskat the topmost cycle. Towel hypothesis that hotelkitchen will receive the most prominentstress, since it will collect one asterisk more than the highermodifierhotelkitchen-that is, one at the N' level, one at the NP level, and one will receive only two, at the circled N' and NP at the N-' level (= 3). Hotel kitchen levels, so that when the common No cycle is reached, the final asteriskwill be attracted in correspondencewith the column over towel:
(87)

( (

.
*

* *

. .

(
*

* * *

) ))

) . )) )) * )) [[[[hotel kitchen]]] [[[[towel]]][rack]]]

(( ((( *

) ((

, )) (((

line 7 line 6 line 5 line 4 line 3

As indicatedin (87), the fact that syntactic heads (Ns projectingan NP) count as single words, starting with one (line 3) asterisk, implies that the cyclic derivation that has
Alternatively,one could utilize the usual X-bar schema [N"[N' N?]] and establishthe convention that in compound structuresN" counts as N? in the next higher (compoundor syntactic) cycle. Evidence will be presentedlater in this section that a more literalextension of X-bartheory to compoundsis in order, hence that X-bartheory shouldperhapsbe generalizedacross levels (wordstructure,compoundstructure,syntactic
structure, . . .).

A NULL

THEORY

OF PHRASE

AND

COMPOUND

STRESS

279

derived the stress contourof hotel kitchenis "invisible" in (87) (except for its end result in the form of a line 3 asterisk). This implies that there will be a "round" of stress assignmentfor each branching(compound)N-in (86), one at the N dominatinghotel kitchen, responsible for the relative prominenceof hotel over kitchen, and one at the matrixN, responsiblefor the relativeprominenceof towel over hotel kitchen(andrack). The previous account of the stress patternof left-branching compounds([[[kitchen] [towel]] [rack]]) must accordinglybe slightly altered, but the outcome is unaffectedby this refinement. Although [kitchen towel], as a syntactic head projectingan NP, will merely count as one (line 3) asterisk located over the first syllabic nucleus of kitchen in the first round of stress assignment, it will nonetheless be more deeply embedded than the head rack, thus ultimately receiving the most prominentstress. See the structure (88), and the correspondingmetricalgrid (89).42 (88) NP N'
N

No

N-'

N-2

kitchen towel
(89)

rack
. .

* *

(( *

(((*

) )) (

* *

) ) ))

line 6 line 5 line 4

line 3

[[[[kitchentowel]]] [[rack]]] Despite its apparentsuccess, this analysis of the internal structureof compounds requiresa furtherrefinementfor both conceptualand empiricalreasons. The refinement involves a more literalextension of X-bartheory, which exactly parallelsthe one familiar in syntax. Under this interpretation,one must recognize a complement dominatedby the first projection dominatingthe head and a specifier dominatedby the maximalprojection (modulo, in English, the different value of the head-complementparameterat the syntactic and the compoundlevels):

42 One can imaginevariousinterpretive algorithms that, fromthe metricalgridsconstructedat each round (cycle), compute the relative prominenceof the constituentsof the entire compound(see the discussion in Libermanand Prince 1977). Since the best form of the algorithmis not important the currentargument,I to will not attemptto determineit here.

280

GUGLIELMO

CINQUE

(90)

No

(specifier)

N N -2

(complement)

This has desirable empiricalconsequences. So far we have focused on N N compoundsin which prominencefalls on the nonhead, a very widespread(perhaps,the majority)pattern,but certainlynot the only one. As documentedin the literature,there are many cases where the most prominentstress falls on the head.43Here are some representativecases (also see (79a-f)): (91) a. b. c. d. NN kitchen table town haill woman doctor police investigaition

The generalizationthat emergesfromthe works cited in footnote 43 (in fact, madeexplicit in Selkirk 1984:244ff.)is that in N N compounds, as well as in other types, stress falls on the nonhead if this is an argumentof the head; otherwise, it falls on the head. This is particularlyclear in N A compounds, as shown by the contrasts between (92a) and (92b). (92) a. frost bitten b. lily white disease prone waist high blood thirsty dirt cheap germ resistant crystal clear

As Selkirkputs it,
What differentiatesthe (a) and the (b) cases, aside from the prominencepatterns, is the semanticrelationholdingbetween the head and its sister. In [the (a) cases], the head has an argumentto its left: bitten by frost, prone to disease, thirstyfor blood, resistant to germs. In [the (b) cases], the head's sister has either the characterof an adjunctmodifier(e.g., as white as a lily, as high as (the) waist, as cheap as dirt) or a locative force.... (1984:245)

The same generalizationis apparentlyat work in such well-knownprominencecontrasts in N N compounds as the following:44

43

See, in the more recent literature,Selkirk 1984, Zwicky 1986, Bates 1988, and Libermanand Sproat

1992:157ff.

" Anotherpossibilitywouldbe to claimthatthe (b) cases are not compounds phrases,hence patterning but in the expected way. This claim is occasionallymade (see Bloomfield1933:228,Bates 1985, among others),

A NULL

THEORY

OF

PHRASE

AND

COMPOUND

STRESS

281

(93) a.

apprenticewelder b. apprenticewelder toy factory toy factory wine drinker party drinker

If the nonhead is interpretedas a complementof the head as in (93a) ('one who welds apprentices', 'a factory producing toys', etc.), stress prominence falls on it. If it is as interpretedas a modifier/specifier in (93b)('a welder who is an apprentice', 'a factory that is a toy', etc.), stress prominencefalls on the head. This conclusion is reinforced when we examine Zwicky's (1986) careful classificationof stress prominencesin compounds. Accordingto Zwicky's material,wheneverthe nonheadbears a possessive (government commission), locative (kitchen table, town hall), temporal(summer holiday), attributive(womandoctor), or material(wood chest) relationto the head-that is, whenever the nonhead bears a specifier relationto the head-the nonhead never bears main stress. To take another example, this time from A A compounds, consider the contrasts between (94a) and (94b) (from Selkirk 1984:245ff.,Bates 1988:176ff.). (94) a. sick looking b. ever lasting nice seeming hard hitting strange sounding long suffering

If the first adjective is interpretedas a complementof the following deverbal adjective as in (94a) ('he looks sick', 'he seems nice', 'it sounds strange', etc.), it bears main stress. If its interpretation instead that of a modifier,as in (94b) ('it lasts forever', 'he is hits hard', 'he suffered for a long time', etc.), stress falls on the head, the deverbal adjective. Of course, as Selkirk (1984:246)notes, work must be done to render the notion "complement" relevant to compounds more precise. For example, it must be that the left-handN in such cases as steel warehouse 'a warehousefor steel', towel rack 'a rack for towels', wheat flour 'flour (made) from wheat', coal-tar product 'product (made) from coal tar' counts as a complement (bearinga goal, or source, 0-role) of the righthand N. Whatis relatively sureris that "when the left-handelement clearly has adjunct status (as a modifier, for example), the head is prominentand the adjunctmay not be" (Selkirk 1984:247).45
but it runs counterto the evidence that they indeed behave like compounds-for example, in takingprefixes

chap. 3 for in-depthdiscussion.) In any event, it does not seem possible to treat all collocationswith "afterstress" as phrases. As Zwicky (1986) points out, following Lees (1960), a sequence like legal document is ambiguousbetween the two readings'a documentthatconformsto the law' and 'a documentemployedwithin the context of the legal profession', where the first is a phrasalcompositionalreadingand the second a compound reading.But prominenceis on documentin both cases. 45 The stress patternof certain compoundsappearsto be indeterminate "within and among speakers" Perhapsthis is due to the possibility of analyzingthe left-handmemberas either a complement('built by slaves', 'washed by hand', etc.) or an adjunct('builtwith the help of slaves', 'washedwith hands', etc.). See Zwicky 1986:55for anothercase possibly interpretable this manner. in
(Bates 1988:177). Variation seems to exist in such cases as slave built, hand washed, pan fried, hand made.

(an ex apprentice welder vs. *an ex nice person). (See Levi 1978, Zwicky 1986, and especially Bates 1988:

282

GUGLIELMO

CINQUE

Despite certain idealizations and open questions, Selkirk's insight is illuminating. It captures what appears to be the fundamentalgeneralizationgoverning stress prominence within compounds (althoughthere may also be lexical idiosyncrasies). Suppose, then (in fact, departing from Selkirk'sfurtherconclusionthat "the location of prominence [in compounds] cannot be explained in purely structuralterms" (1984: 245)), that we assign a structuralcorrelateto the semantic complement/adjunct distinction (just as we do in syntax).46In other words, suppose we assume that compounds are no different from phrases in the relevant respect, with complements generated as sisters of the head, and specifiers as daughtersof the maximalprojection, as in (95). (95) X?

YP

ZP

X-2

(specifier)

(complement)

An X? is, in lexical terms, a maximalprojectionwith respect to X and X - 2. Therefore, YP and ZP should more properly be Y0 and ZO,althoughwe have seen that genuine phrasalcategories can exceptionally be internalto compounds(see (79)). A compound such as kitchen towel rack would thus receive the representationin (96). (96) No N-' No
-1

N?
-

N -2

Nl

N-2

N-2

kitchen

towel

rack

4 There are reasons to reject an approachthat institutesa directlink betweenthe semanticsof the commediatebetweenthe two. For one thing,suchan approach poundandits stress patternwithouthavingstructure makesit arbitrary complementsratherthanheadsor specifiersshouldreceive stress prominence.Nothing that in principlewould seem to exclude the opposite state of affairsfroma semanticpointof view. In the structural approach,instead, where the particularsemanticsof a compoundis given a structuraltranslation(as in the syntax), with complementsmore deeply embeddedthan heads and specifiers, the particularstress patterns that we find are forced by the null theory of stress assignment(the same theory that also works for syntactic phrases).

A NULL

THEORY

OF

PHRASE

AND

COMPOUND

STRESS

283

The analysis appearsto derive as a necessary consequence a generalizationabout compound structurenoted by Selkirk(1982:36ff.) still in need of a principledexplanation. and In her words, "All [nonsubject]argumentsof the head of a compoundmust be satisfied within the compound immediatelydominatingthe head." A compound like tree eater is ambiguousbetween a locative interpretation ('eater in a tree') and a theme interpretation ('eater of the tree'). But a compoundlike treepasta eater is no longer ambiguous, tree having only the locative interpretation; a compoundlike pasta tree eater, with and the complement higher than the locative adjunct, is impossible. This follows from imposing an X-bar structureon the compound, since this forces internalargumentsto be under X' (X- ) and nonargumentsto be in the specifier position under X" (X?). Since pasta cannot receive a locative interpretation, must occupy the complementposition, it closer to the head.47 The single structure(95), with the recursive possibility for YP (or Y?) and ZP (or Z?) each to be itself a compound, appearsto subsume all the differentpossibilities discussed in the literaturewhile bringingout the perfect identityof compoundsand phrases in the pattern of stress prominence. Recall the particularrefinementof the principles of grid constructionfor phrases motivated at the end of section 7, accordingto which only elements embedded on the X-bar and the complement axis are visible in their entirety for the principles of grid construction. The net effect of such a refinementwas that a complementwins over the head and the specifier and, in the absence of a complement, the head wins over the specifier, with specifiers counting only as a single asterisk, because they are on a nonrecursive branch. Compoundsare no exception to this generalization.In compounds, too, a complement wins over the head and the specifier (see (97a)), and, in the absence of a complement, the head wins over the specifier, in this case giving right-handprominence(see (97b)).48 This grounds Selkirk's generalizationin purely structuralterms, thus avoiding the problems noted above of a semanticallybased stress assignment.

The structural approachalso appearsto constitutea plausibleacquisitionmodel. Semanticconsiderations via Universal Grammarsingle out particularstructuralrepresentations, which in turn, again via Universal Grammar, determinestress properties. A second difficultyfor a purelysemantically basedapproach represented cross-linguistic is by differences. In this approach,presumablyone would not expect parametric choices in the respective orderof arguments and heads to affect the stress patternof the compound,but they do. Compare Englishtoyfactory andits Italian equivalentfabbrica giocattolo (lit. 'factorytoy') with the reverse stress pattern. 47 Selkirk " (1982:37) notes that"two [nonsubject] also arguments cannotappearin a compound. Therefore, one cannot have *babytoy handing, *tableboot puttingto render'the handingof toys to babies', 'the putting of boots on the table'. This may follow againfrom an X-barstructurelike the one proposedby Larson, with a singlespecifieranda singlecomplement,if it is assumedthatan abstractheadproviding the thirdargument for is absent from compounds. 48 For a compoundlike kitchentable, this analysispredictsthat if the interpretation table for kitchens' 'a is possible over and above the "locative" interpretation table that is in the kitchen', then kitchen should 'a be given prominence,since it comes to occupy the complementposition. For the speakersI have consulted, the predictionappearsto be confirmed:In the bathroom,we have a KITCHENtable.

284 (97) a. No

GUGLIELMO

CINQUE

/\N~-1

/
{

/\~~~N
}

- 2

kitchen

b.

TOWEL No

rack

N/-'

kitchen

RACK

More complex cases are instantiationsof the same single structure,with its recursivity potentials realized in differentways. This can be seen by inspectinga few cases, some of which are already well known from the literature.(As above, the constituent receiving the most prominentstress is set in capital letters. The dotted lines are meant to make conspicuous the fact that what is at stake in each case is a recursive expansion First consider of either the complementor the specifier of the same basic structure.)49 (98a) and (98b).
are and structure compoundstressin German essentiallythe sameas in English,as expected, "4 Compound given the same head-finalchoice for the relevant parameter(Giorgiand Longobardi1991:chap.3). See the following cases, drawnfrom Giegerich 1981, 1983, and from Grewendorf,Hamm, and Stemefeld 1987:sec. 3.4. (Also see Giegerich 1985, Doleschal 1988, and references cited there. For similarfacts in Dutch, see Langeweg 1987.) (i) a.
FUSS - ball - feld STRASSEN - bahn - linie 'foot - ball - field' 'street - car - line'

b.
Welt - SPAR - tag Landes - HAUPT - versammiung 'world - savings - day' region - general - reunion'

c.
Atom - waffen - SPERR - vertrag Blau - wangen - BIENEN - fresser 'atomic - weapons - ban - treaty' 'blue - cheek - bee - eater'

d.
Welt - NICHT - raucher - tag 'world - non - smoker - day'

A NULL

THEORY

OF PHRASE

AND

COMPOUND

STRESS

285

e.

Spiel - waren - AUSSEN - handels - gesellschaft 'toy - foreign - trade - society'

Certainapparentcounterexamples 'centralrailwaystation',HALLENschwimmbad (HAUPTbahnhof 'indoor swimmingpool', or DREIfarbstift'three-colorpencil' vs. DreiGROSCHENoper 'three-pennyopera') are explainedaway if Bahnhof,Schwimmbad, etc., are singleratherthancompoundwords, and if Dreigroschenoper differs structurally from Dreifarbstiftessentiallyin the same way that [N[NP[AP black] board] eraser] differs from [N[NP[N blackboard]] eraser] (see Giegerich1983).Josef Bayer(personalcommunication) pointsout that for him WELTspartag also possible alongsideWeltSPARtag, is is thoughWELTnichtrauchertag not. Perhaps Spartag, thoughnot Nichtrauchertag,can be optionallyanalyzedas a single word. Given the head-initial orderof Italian(moregenerally,Romance)compounds,we shouldexpect a perfect mirrorimage. The expectationappearsto be essentiallycorrectinsofaras it can be checked, given the severe limitations(in productivity,and perhapstype) on nominalcompounding Italian(for which see Zuffi 1981: in 17f. and fn. 20, where it is pointedout that N N compoundsin Italianare generallycoordinating ratherthan and subordinating, Giorgiand Longobardi1991:sec.3.5, especiallyfn. 25). Thus, we find cases like (iia)/(iiia) contrastingwith English(iib)/(iiib). (ii) a.
N -\
N2

N0
/ N?

b.

N"
NN" N-

settore
(iii) a.

VENDITE
N"

0
b.

SALES
N0

department

N N-'

\ \N/

N N-'

N2

4 XN2

N2

N-2

capo

settore

VENDITE

0 0

0 0

SALES

department head

But it is not clear whether compoundswith the structure(iv), corresponding structure(v) in Germanic, to exist. No examples of that sort are found in Zuffi 1981. (iv) (v) kitchen TOWEL rack Landes-HA PT-versammlung U Where (iv) would seem to be semanticallyappropriate, there are two main stresses, one on each modifier, separatedby a pause: [[tavola CALDA],PIZZERIA]'(lit.) hot table, pizzeria'.

286 (98) a. No

GUGLIELMO

CINQUE

~~N-1

/N?0\

N-2

N-2

0
b.

0 LAW
No

degree

requirement

/NO-

0 0

~~degree requirement changes LAW TEACHERS union president election

Here each N appearsto stand in a complementrelationto the N to its right ('degree in law', 'requirement (a degree in law)', 'changes of . .. ')-hence the prominenceon for law, the most deeply embedded element of the compoundin each case. Similarremarkshold for cases such as teachers unionpresident election. The contrast between (98) and (99) is instructive.

A NULL

THEORY

OF PHRASE

AND

COMPOUND

STRESS

287

(99)
--

No
_ -

No N-'
/\7-2

N-'

/--N-2 LANGUAGE requirement

law

degree

Here, law degree is not a complementof language, itself complementof requirement. Hence, it is no longer part of the recursivepath of embedding,as it was in (98b). It can in only be relatedto the head requirement a more indirectway, as a specifierof language requirement. Consequently, it will be language that qualifies as the most embedded element of the compoundand is given prominence.50 (100)-(102) show additionalvariationsof the same basic structure.51 (100) a. No N' 0

N\

\I

0
50

kitchen TOWEL

rack

deposit

This recalls Selkirk's (1984) discussion (modulo the noted difference in approach): In a compoundconfiguration [A[CD]A B[EF]B] it is impossiblefor A and B to standin an argument-head like relationto each other. (See Selkirk1982:ch.2, where it is shownthat a wordwith an open argument position-e.g. requirementmust have that argumentsatisfiedby a sister constituent(if it is to be satisfiedat all)-e.g. language. Since branching constituentslike language requirement not, on that theory, have open argument do positions, it follows that the sister to a branching constituentwill never be an argument with respectto it). (p. 250) 51 Note that in (102a-b) the most prominent stress falls on the middle constituent, although the two cases instantiate two different structures and interpretations: (102a) 'the president of the labor union's finance committee', (102b) 'the finance committee's president of the labor union'.

288 b.

GUGLIELMO

CINQUE

No

No
N0N

0 (101) a.

kitchen N?

TOWEL

rack

deposit

warden

N2N 2- 2

kitchen

RACK

deposit

Ratherthanin theircontour,the two possiblydifferin the locationof (virtual) pauses, with a biggerpause between committeeandpresidentin (102a)andbetweenunionandfinance in (102b),muchas in the ambiguous Americanhistory teacher (see Chomskyand Halle 1968:22,fn. 9). For a minimalpair in Germansimilarto (102a)and (102b), see Giegerich 1983:7and fn. 7, where the followingtwo structuresare assigned to Spielwarenaussenhandelsgesellschaft, with correspondingly different interpretations:

(i) a. [[A B] [[C DI El] b. [[[A B] [C DI] E]

A NULL

THEORY

OF PHRASE

AND

COMPOUND

STRESS

289

b.

No

N'N-'

0/

/~~~~~ kitchen

N -2

N-2

RACK

deposit

warden

(102) a.

No

0
/

0
/~~ ~~~

a NdNuoF
Nn-2

NN2

~labor union FINANCE committee

president

290 b.

GUGLIELMO

CINQUE

No

No

N-'

N2~~~~ N
/ D N

N
~- 2
/N

N
- 2

N-

0 labor union

0 FINANCE committee

president

A potentialproblemis providedby such A N compoundsas those in (103), in which stress falls on the head even if the adjective apparentlyintroducesa complementof the head. (103) a. b. c. lunarexploration 'explorationof the moon' stellar observaition'observationof the stars' presidentialelection 'election of the president'

If the adjective, as a consequence, were generated in the complementposition of the compound, then it should receive stress prominence;but it does not. Cases like (103) differ from the cases discussed above, in which an adjective stood in a complefnentrelationto a deverbaladjectiveand indeed received stress prominence within the compound(sick looking, etc.). In (103), for the adjectiveto stand in a proper complement relation to the N, it would have to receive the 0-role theme from the N. However, we know from the syntax of NPs that adjectives cannot receive a 0-role by beinggeneratedin complementposition, thoughthey can in specifierposition (see Kayne 1984:63).If an analogous propertyholds for compoundstructure,then such adjectives will only be generable in the specifier position of the compoundand will possibly bear a general relation to the head similarto the R-relationholding between a genitive and the N in NPs. This indeed appearsplausible,given such cases as nuclearprotest, where the relation between nuclear and protest is more similarto an R-relationthan to one of 0-role assignment. In sum, if such adjectivesaregeneratedin the specifierratherthanin the complement position, then as expected stress prominencewill go to the head."2A(P)s would thus
52 Cases like blackbird,bluefish, bluebookare only apparent in exceptions to the generalization the text if they are lexically reanalyzed as single Ns, as Libermanand Sproat (1992:148ff.)suggest. Perhaps the

A NULL

THEORY

OF PHRASE

AND

COMPOUND

STRESS

291

differ from N(P)s precisely in not being able to absorb a 0-role via governmentin complement position. It is interestingto note, then, that substitutingan N(P) for the A(P) in (103) makes stress prominence change systematically, in the expected way (Bates 1988:80f.): (104) a. b. c. caiveexploration stairobservation officer election

In N N compounds,there is a class of cases in which the head receives stress prominence in the apparentpresence of an argument: (105) a. student rebellion b. governmentfunding c. consumer spending d. enemy movements e. state hiring At first sight the argumentwould seem to be an "external" one (in Williams's (1981) sense)-hence, if anything, generated in specifier ratherthan in complement position of the compound. However, it is dubious that the nonhead Ns in (105) are genuine (external)argumentsof the head. A genuineexternalargumentcannot in general appear in the specifierposition of compounds:for example, *girlswimming,*kideating (Selkirk 1982:34).Furthermore,as Bates (1988:111ff.)notes, the head still retains the ability to assign the external 0-role (a student rebellion by Cambridgeundergraduates).Hence, as she indeed suggests, the nonhead is interpretedmore as a modifier ('typical of students') than as an argument.In either case, at any rate, stress prominenceis expected on the head.53 Many questions concerningEnglish compoundshave not been addressed here, nor has the importantquestionconcerningparameterization, whichdeserves a separatetreatment. Other questions have been addressed only briefly. I nevertheless hope that the present analysis will suggest possible ways to solve the remainingissues.

forestress of such cases as dpple cake (vs. apple pie) and Mddison Street (vs. Madison Avenue) should be

analogouslytreated. Zimmer(1971:C19),Schmerling(1971), Allen (1978:103ff.),and Libermanand Sproat (1992:157) note that manycases of the formertype in fact show individual differences.Some people say apple
pie, cream sauce, ch6colate cake, potato salad, and peanut butter; others say apple pie, cream sauce, chocolate cdke, potato salad, and peanut batter. This would be expected if lexicalizationis at issue.

The other street names, in English,behave like Avenue, not Street. In German,apparently,all but Markt
have stress on the first N (Ludwigweg, B&ggasse, Karl Platz, etc., vs. Hoher Markt), but interestingly, if

the first element of the compoundis an adjective, stress may go on the head (Danziger Stral3e).In Dutch there is apparently greatvariability both amongstreet namesandin the uses of a single streetnamein different cities (I thankGerhardBruggerand JeannetteSchaefferfor this information). 5 The asymmetrictheory of coordination discussed in section 7 predictsthat stress should fall on the second of the two membersof a stressed element of a compound.This predictionis correct:salt and pepper shaker (see Bates 1988:213ff. an interestingdiscussionof coordination for within compounds).

292 References

GUGLIELMO

CINQUE

Abraham, Werner, ed. 1983. On the formal syntax of Westgermania. Amsterdam: Benjamins. Abraham, Werner, ed. 1985. Erklarende Syntax des Deutschen. Tubingen: Narr.

Allen, MargaretReece. 1978. Morphologicalinvestigations.Doctoral dissertation, University of Connecticut, Storrs. Allerton,D. J., and Alan Cruttenden.1979.Threereasonsfor accentinga definitesubject.Journal
of Linguistics 15:49-53.

Antinucci, Francesco, and GuglielmoCinque. 1977. Sull'ordinedelle parole in italiano:L'emarin ginazione. Studi di grammaticaitaliana 6:121-146. [Reprinted GuglielmoCinque. 1991.
Teoria linguistica e sintassi italiana, 163-195. Bologna: Il Mulino.] Baart, Joan L. G. 1987. Focus, syntax, and accent placement. Dordrecht: ICG Printing.

Bader,Thomas,and Zvi Penner. 1990.Mainclause phenomenain embeddedclauses: The licencing of embeddedV2-clauses, Vl -clauses, andrelatedstructuresin Bernese Swiss German.Ms., UniversitatBern.
Baker, Mark C. 1988. Incorporation: A theory of grammaticalfunction changing. Chicago: Uni-

versity of ChicagoPress. Bardovi-Harlig,Kathleen. 1986. Discourse determinantsof English sentence stress. Distributed by IndianaUniversity LinguisticsClub, Bloomington. Barss, Andrew, and Howard Lasnik. 1986. A note on anaphoraand double objects. Linguistic
Inquiry 17:347-354.

Bates, Dawn. 1985. On phrases and compounds in English. In Proceedings of the West Coast Conference on Formal Linguistics 4, 23-34. StanfordLinguistics Association, Stanford University, Stanford,Calif. Bates, Dawn. 1988. Prominencerelationsand structurein English compoundmorphology.Doctoral dissertation, University of Washington,Seattle. Belletti, Adriana. 1988. The Case of unaccusatives.LinguisticInquiry19:1-34. Belletti, Adriana. 1990. Generalizedverb movement.Turin:Rosenbergand Sellier. Beninca,Paola. 1991.Su unanota sintatticadi CarloSalvioni.In Saggi di linguisticae di letteratura
in memoria di Paolo Zolli, 43-52. Padua: Antenore.

Berman, Arlene, and Michael Szamosi. 1972. Observationson sentential stress. Language 48: 304-325.
Besten, Hans den. 1989. Studies in West Germanic syntax. Amsterdam: Rodopi.

Bierwisch, Manfred. 1963. Grammatik deutschen Verbs. Berlin:Akademie-Verlag. des Bierwisch, Manfred. 1966. Regeln fur die IntonationdeutscherSatze. Studia Grammatica7:99201. Bierwisch, Manfred. 1968. Two critical problemsin accent rules. Journal of Linguistics 4:173178. Bing, Janet Mueller. 1979a.Aspects of Englishprosody. Doctoraldissertation,Universityof Massachusetts, Amherst. Bing, Janet Mueller. 1979b. One man's beat is anotherman's tone: Sentence stress reanalysed.
In The elements: A parasession on linguistic units and levels, 1-10. Chicago Linguistic

Society, University of Chicago, Chicago, Ill. structurein Afghan Persian. LinBing, Janet Mueller. 1980. Linguisticrhythmand grammatical
guistic Inquiry 11:437-463.

Bolinger, Dwight. 1972. Accent is predictable(if you're a mind-reader). Language 48:633-644. Borer, Hagit. Forthcoming.Parallel morphology.Cambridge,Mass.: MIT Press. Bresnan, Joan. 1971. Sentence stress and syntactic transformations. Language 47:257-281.

NULL

THEORY

OF

PHRASE

AND

COMPOUND

STRESS

293

Bresnan, Joan. 1972. Stress and syntax: A reply. Language 48:326-342.

Bresson, Daniel. 1983. Les phenomenes suprasegmentaux en allemand moderne. Paris: Didier.

Brugger, Gerhard. 1990. ObligatorischeElemente im "Restrictive Clause." M.A. dissertation, UniversitatWien. Calabrese,Andrea. 1990. Some informalremarkson focus and logical structuresin Italian. Ms., HarvardUniversity, Cambridge,Mass.
Chomsky, Noam. 1965. Aspects of the theory of syntax. Cambridge, Mass.: MIT Press.

In Chomsky, Noam. 1970. Deep structure,surface structureand semanticinterpretation. Studies

in general and Oriental linguistics, presented to Shiro Hattori on the occasion of his 60th

birthday,ed. RomanJakobsonand Shigeo Kawamoto,52-91. Tokyo: TEC. Chomsky, Noam. 1986.Barriers. Cambridge,Mass.: MIT Press. Chomsky, Noam. 1991. Some notes on economy of derivationand representation.In Principles
and parameters in comparative grammar, ed. Robert Freidin, 417-454. Cambridge, Mass.:

MIT Press. Chomsky, Noam, and MorrisHalle. 1968. The sound pattern of English. New York: Harperand Row. Chomsky, Noam, Morris Halle, and Fred Lukoff. 1956. On accent and juncture in English. In For Roman Jakobson, ed. Morris Halle, Horace Lunt, Hugh MacLean, and Cornelis H. van Schooneveld, 65-80. The Hague: Mouton. Cinque, Guglielmo. 1989. On embeddedverb second clauses and ergativityin German. In Sentential complementation and the lexicon: Studies in honor of Wim de Geest, ed. Dany

Jaspers et al., 77-96. Dordrecht:Foris. Cinque, Guglielmo. 1990. Agreementand head-to-headmovementin the Romance noun phrase. Paper presented at LSRL XX, University of Ottawa, Ottawa, Ont. Cruttenden, Alan, and David Faber. 1991.The accentuation prepositions.JournalofPragmatics of 15:265-286. Culicover, Peter, and Michael Rochemont. 1983. Stress and focus in English. Language 59:123165.
Dahl, Osten. 1974. Topic, comment, contextual boundedness and focus. Hamburg: Buske.

Dell, Fran9ois. 1984. L'Accentuationdans les phrasesen francais. In Forme sonore du langage, ed. Fran9oisDell, Daniel Hirst, and Jean-RogerVergnaud,65-122. Paris: Hermann.
Dezso, La'szlo. 1974. Topics in syntactic typology. In Linguistica generalia. Vol. I: Studies in

linguistic typology, ed. M. Romportl, V. Skalicka, L. Popela, and B. Palek, 191-210. Prague:CharlesUniversity. Dezso, La'szlo. 1977. Toward a typology of theme and rheme: SOV languages. In Wortstellung und Bedeutung, ed. Maria-Elisabeth Conte, Anna Ramat, and Paolo Ramat, 3-11. Tubingen: Niemeyer.
Dezso, Laszlo. 1982. Studies in syntactic typology and contrastive grammar. The Hague: Mouton.

Di Sciullo, Anna Maria. 1991. On the structureof deverbal compounds. Universityof Venice workingpapers in linguistics4/91. CentroLinguisticoInterfacolta,Universitadi Venezia. Di Sciullo, Anna Maria,and Edwin Williams. 1987.On the definitionof word. Cambridge,Mass.: MIT Press. Dogil, Grzegorz. 1980. Focus markingin Polish. LinguisticAnalysis 6:221-245. Doleschal, Ursula. 1988. Zum deutschen Kompositionsakzent: Tema con variazioni. WienerLinguistische Gazette 40-41:3-28.

Donegan, Patricia, and David Stampe. 1983. Rhythm and the holistic organizationof language
structure. In The interplay of phonology, morphology and syntax, 337-353. Chicago Lin-

guistic Society, University of Chicago, Chicago, Ill.

294

GUGLIELMO

CINQUE

Drach, Erich. 1940. Grundgedanken deutschen Satzlehre. 3d ed. Frankfurtam Main: Diesder terweg. Duanmu, San. 1990. A formal study of syllable, tone, stress, and domain in Chinese languages. Doctoral dissertation,MIT, Cambridge,Mass. Duanmu, San. 1991. Stress and syntax-phonologymismatches:Tonal domains in Danyang and
Shanghai. To appear in Proceedings of the West Coast Conference on Formal Linguistics

10, StanfordLinguisticsAssociation, StanfordUniversity, Stanford,Calif.

Emonds, Joseph. 1976. A transformational approach to English syntax. New York: Academic

Press.
Emonds, Joseph. 1985. A unified theory of syntactic categories. Dordrecht: Foris.

Erteschik-Shir, Nomi, and ShalomLappin. 1983.Understress. A functionalexplanationof English

sentence stress. Journal of Linguistics 19:419-453.

Fabb, Nigel. 1984. Syntactic affixation.Doctoral dissertation,MIT, Cambridge,Mass. Faber, David. 1987.The accentuationof intransitivesentences in English.Journal of Linguistics 23:341-358. Firbas,Jan. 1964.On definingthe theme in functionalsentence perspective. Travauxlinguistiques
de Prague 1:267-280.

Giegerich, Heinz J. 1981. Compoundstress in German:Some observationsand an exercise in metricalphonology. In Workin progress, vol. 14, 83-97. Departmentof Linguistics, University of Edinburgh. Giegerich,Heinz J. 1983. MetrischePhonologieund Kompositionsakzent Deutschen. Papiere im
zur Linguistik 28:3-25. Giegerich, Heinz J. 1985. Metrical phonology and phonological structure: German and English.

Cambridge: CambridgeUniversity Press.

Giorgi, Alessandra, and Giuseppe Longobardi. 1991. The syntax of noun phrases: Configuration, parameters, and empty categories. Cambridge: Cambridge University Press.

Giusti, Giuliana. 1986. On the lack of wh-infinitiveswith "zu" and the projectionof COMP in
German. Groninger Arbeiten zur germanistischen Linguistik 28:115-169.

Giusti, Giuliana. 1990. Zu-infinitivalsand the structureof IP in German.Ms., Universita di Venezia. [To appearin Rivista di linguistica, Turin:Rosenbergand Sellier.]
Grewendorf, Gunther. 1988. Aspekte der deutschen Syntax. Tubingen: Narr.

Grewendorf,Gunther. 1989.Ergativityin German.Dordrecht:Foris. Grewendorf,Gunther,Fritz Hamm, and WolfgangSternefeld. 1987.SprachlichesWissen. Frankfurt am Main: Suhrkamp. Grewendorf,Gunther,and WolfgangSternefeld,eds. 1990.Scramblingand barriers.Amsterdam: Benjamins. Gueron, Jacqueline. 1980. On the syntax and semantics of PP extraposition.Linguistic Inquiry 11:637-678. Gundel,Jeanette. 1977.The role of topic and commentin linguistictheory. Distributedby Indiana University Linguistics Club, Bloomington.
Gussenhoven, Carlos. 1984. On the grammar and semantics of sentence-accents. Foris. Dordrecht:

Haider, Hubert. 1991. Branchingand discharge. Ms., UniversitatStuttgart.

Haider, Hubert, and Martin Prinzhorn, eds. 1986. Verb second phenomena in Germanic lan-

guages. Dordrecht:Foris. Halle, Morris. 1985. Grids and trees in metricalphonology. In Phonologica 1985. Halle, Morris, and Michael Kenstowicz. 1991. The Free Element Conditionand cyclic versus noncyclic stress. LinguisticInquiry22:457-501.

A NULL

THEORY

OF

PHRASE

AND

COMPOUND

STRESS

295

Halle, Morris,andJean-Roger Mass.: MITPress. Vergnaud.1987.An essay on stress. Cambridge, Halliday, M. A. K. 1967/68.Notes on transitivityand theme in English.Journal of Linguistics 3 (1967): 199-244 and 4 (1968): 179-215. Harlig, Jeffrey, and Kathleen Bardovi-Harlig.1988. Accentuation typology, word order, and theme-rheme structure. In Studies in syntactic typology, ed. Michael Hammond, Edith Moravcsik, and Jessica R. Wirth, 125-146. Amsterdam:Benjamins. Hayes, Bruce, and Aditi Lahiri. 1991. Bengali intonationalphonology. Natural Language and
Linguistic Theory 9:47-96.

Helbig, Gerhard,and JoachimBuscha. 1984.Deutsche Grammatik. ed. Leipzig: VEB Verlag 8th Enzyklopadie. Hoeksema, Jack. 1988. Head-types in morpho-syntax.In Yearbookof morphology, ed. Geert Booij and Jaap van Marle, 123-137. Foris: Dordrecht. Hohle, Tilman. 1982. Explikationenfur "normale Betonung" und "normale Wortstellung."In
Satzglieder im Deutschen, ed. Werner Abraham, 75-153. Tubingen: Narr. Jackendoff, Ray. 1972. Semantic interpretation in generative grammar. Cambridge, Mass.: MIT

Press. Jacobs, Joachim. 1982. Neutralerund nicht-neutraler Satzakzentim Deutschen. In Silben, Segmente, Akzente, ed. Theo Vennemann, 141-169. Tubingen:Niemeyer. Kayne, Richard. 1983.Binarybranching.Paperpresentedat the workshop"Featuresin Syntax," University of Tilburg.
Kayne, Richard. 1984. Connectedness and binary branching. Dordrecht: Foris.

Kayne, Richard. 1985. Principlesof particle constructions. In Grammaticalrepresentation,ed. JacquelineGueronet al., 101-140. Dordrecht:Foris. Kim, Alan Hyem-Oak.1988.Preverbalfocussingand type XXIIIlanguages.In Studies in syntactic typology, ed. Michael Hammond,Edith Moravcsik,and Jessica R. Wirth, 147-169. Amsterdam:Benjamins. Kiparsky, Paul. 1966. Uber den deutschen Akzent. Studia Grammatica7:69-98.
Kuno, Susumu. 1973. The structure of the Japanese language. Cambridge, Mass.: MIT Press. Ladd, D. Robert. 1980. The structure of intonational meaning. Bloomington, Ind.: Indiana Uni-

versity Press. Lakoff, George. 1968. Pronouns and reference. Distributedby Indiana University Linguistics Club, Bloomington. Lakoff, George. 1972. The global natureof the Nuclear Stress Rule. Language 48:285-303. Langeweg, Simone J. 1987. Stress assignmentin Dutch compoundnouns. In Linguistics in the Netherlands 1987, ed. Fritz Beukemaand Peter Coopmans, 131-140. Dordrecht:Foris. Larson, Richard. 1988. On the double object construction.LinguisticInquiry19:335-391. Larson, Richard. 1990. Double objects revisited:Reply to Jackendoff.LinguisticInquiry21:589632.
Lees, Robert. 1960. The grammar of English nominalizations. The Hague: Mouton.

Levi, Judith N. 1978. The syntax and semantics of complex nominals. New York: Academic Press. Liberman,Mark. 1975. The intonationalsystem of English. Doctoral dissertation, MIT, Cambridge, Mass. Liberman,Mark. 1978. The intonationalsystem of English. Distributedby the IndianaUniversity LinguisticsClub, Bloomington. Liberman,Mark, and Alan Prince. 1977. On stress and linguistic rhythm. Linguistic Inquiry 8: 249-336. Liberman,Mark, and RichardSproat. 1992. The stress and structureof modified noun phrases

296

GUGLIELMO

CINQUE

in English. In Lexical matters, ed. Ivan A. Sag and Anna Szabolcsi, 131-181. Center for the Study of Languageand Information,StanfordUniversity, Stanford,Calif. interface. In Lieber, Rochelle. 1988. Phrasalcompoundsin English and the morphology-syntax
Papers from the Twenty-Fourth Regional Meeting, Chicago Linguistic Society, vol. 2, 202-

222. Chicago Linguistic Society, University of Chicago, Chicago, Ill. und Tonhohenbewegungim Standarddeutschen. Loetscher, Andreas. 1981. Satzakzentuierung
Linguistische Berichte 74:20-34. Loetscher, Andreas. 1983. Satzakzent undfunktionale Satzperspektive im Deutschen. Tubingen:

Niemeyer. Longobardi,Giuseppe. 1991a.Extractionfrom NP and the propernotion of head government.In Giorgiand Longobardi1991, 57-112. Longobardi,Giuseppe. 1991b.Propernamesand the theoryof N-movementin Syntaxand Logical
Form. University of Venice working papers in linguistics 9/91. Centro Linguistico Inter-

facolta, Universita di Venezia. McCawley,James. 1977.Accent in Japanese.In Studies in stress and accent (SouthernCalifornia Occasional Papers in Linguistics 4), 261-302. Departmentof Linguistics, University of SouthernCalifornia,Los Angeles. Mahajan,Anoop. 1991. On the natureof objective case assignment.Paperpresented at the University of Illinois, Urbana-Champaign. Maling,Joan. 1971. Sentence stress in Old English. LinguisticInquiry2:379-399. Moltmann,Friederike. 1990. Scramblingin Germanand the specificity effect. Ms., MIT, Cambridge, Mass. papers in linguistics Munn,Alan. 1987.Coordinatestructureand X-bartheory. In McGillworking 4, 121-139. Departmentof Linguistics, McGill University, Montreal,Quebec. the Muysken, Pieter. 1982. Parametrizing notion "head." Journal of Linguistic Research 2(3):
57-75.

Newman, Stanley S. 1946. On the stress system of English. Word2:171-187. Olsen, Susan. 1982. On the syntactic descriptionof German:Topologicalfields vs. X'-theory. In
Sprachtheorie und angewandte Linguistik, ed. W. Welte, Tubingen.

and Pollock, Jean-Yves. 1989.Verbmovement,UniversalGrammar, the structureof IP. Linguistic

Inquiry 20:365-424.

Prince, Alan. 1983. Relatingto the grid. LinguisticInquiry14:19-100. Paperpresentedat the Universityof MaryRiemsdijk,Henk van. 1988.Remarkson incorporation. land, College Park.
Riemsdijk, Henk van. 1990. Functional prepositions. In Unity in diversity: Papers presented to

Simon C. Dik, ed. H. Pinksterand I. Genee, 229-241. Dordrecht:Foris. Ritter, Elizabeth. 1990. Two functionalcategories in noun phrases: Evidence from Modern Hebrew. Ms., Universite du Quebec a Montrdal. Rochemont,Michael. 1978.A theory of stylisticrules in English.Doctoraldissertation,University of Massachusetts, Amherst. [Publishedby Garland,New York.] of Roeper, Thomas. 1988. Compoundsyntax and head movement. In Yearbook morphology,ed. Geert Booij and Jaap van Marle. Dordrecht:Foris. Ross, John R. 1967. Constraintson variablesin syntax. Doctoral dissertation,MIT, Cambridge, Mass.
Scherpenisse, Wim. 1986. The connection between base structure and linearization restrictions in German and Dutch. Frankfurt am Main: P. Lang. Schmerling, Susan. 1971. A stress mess. Studies in the Linguistic Sciences 1:52-66.

Schmerling,Susan. 1976. Aspects of English sentence stress. Austin, Tex.: University of Texas Press.

A NULL

THEORY

OF

PHRASE

AND

COMPOUND

STRESS

297

Selkirk, Elisabeth. 1982. The syntax of words. Cambridge,Mass.: MIT Press.

Selkirk, Elisabeth. 1984. Phonology and syntax: The relation between sound and structure. Cam-

bridge, Mass.: MIT Press. Sproat, Richard. 1985. On derivingthe lexicon. Doctoral dissertation,MIT, Cambridge,Mass. Stechow, Arnimvon, and Susanne Uhmann. 1984.On the focus-pitch accent relation.Groninger
Arbeiten zur germanistischen Linguistik 25:228-263.

Stechow, Arnimvon, and Susanne Uhmann. 1985. Some remarkson focus projection.In Topic, focus, and configurationality,ed. WernerAbrahamand Sjaak de Meij, 295-320. Amsterdam: Benjamins. Stowell, Tim. 1981. Originsof phrase structure.Doctoral dissertation,MIT, Cambridge,Mass. Stowell, Tim. 1989. Subject, specifiers, and X-bar theory. In Alternativeconceptions of phrase structure,ed. MarkBaltin and Anthony Kroch, 232-262. Chicago:University of Chicago Press. Stroik, Thomas. 1990. Adverbs as V-sisters. LinguisticInquiry21:654-661. Tappe, Thilo. 1990. Determinerphrases and agreementin German.Ms., UniversitatGottingen. Thiersch,Craig. 1982.A note on scrambling the existence of VP. WienerlinguistischeGazette and 27/28. Thiersch,Craig.1985.VP and scrambling the GermanMittelfeld.Ms., Universityof Connecticut in and Universitat Koln.
Toman, Jindrich. 1982. Wortsyntax: Eine Diskussion ausgewahlter Probleme deutscher Wortbild-

ung. Tubingen:Niemeyer. Toman, Jindrich,ed. 1985. Studies on Germangrammar.Dordrecht:Foris. Vikner, Sten. 1990.Verb movementand the licencingof NP-positionsin the Germaniclanguages. Doctoral dissertation,Universite de Geneve.
Visch, Ellis. 1990. A metrical theory of rhythmic stress phenomena. Dordrecht: Foris.

Webelhuth, Gert. 1989. Syntactic saturationphenomenaand the modern Germaniclanguages. Doctoral dissertation,University of Massachusetts,Amherst. Webelhuth,Gert. 1990. Diagnosticfor structure.In Grewendorfand Sternefeld 1990, 41-75. Williams, Edwin. 1981. On the notions "lexically related" and "head of a word." Linguistic
Inquiry 12:245-274.

Williams,Edwin. 1982. Anotherargumentthat passive is transformational. LinguisticInquiry13: 160-163. Zimmer, Karl. 1971. Some general observationsabout nominalcompounds. In Working papers on language universals5:C1-C21. LanguageUniversalsProject.Committeeon Linguistics, StanfordUniversity, Stanford,Calif. Zuffi, Stefano. 1981.The nominalcompositionin Italian:Topics in generativemorphology.Journal
of Italian Linguistics 6(2):1-54.

Zwarts, Frans. 1973. On restrictingbase structurerecursionin Dutch. Ms., University of Amsterdam.

Zwicky, Arnold M. 1986. Forestress and afterstress. In Interfaces: Papers by Arnold M. Zwicky.

Workingpapers in linguistics, 32:46-62. Departmentof Linguistics, The Ohio State University, Columbus.
Seminario di Linguistica Universita' di Venezia San Marco 3417, Venezia Italy cinque@iveuncc. bitnet