“VP” Adverbs without a VP: The syntax of adverbs in Tongan Douglas Ball TrumanStateUniversity Proceedingsofthe24thInternationalConferenceon Head-DrivenPhraseStructureGrammar UniversityofKentucky,Lexington StefanMu¨ller(Editor) 2017 CSLIPublications pages61–81 http://csli-publications.stanford.edu/HPSG/2017 Keywords: Tongan,Polynesian,Adverbs,AdjunctSyntax,HPSG Ball, Douglas. 2017. “VP” Adverbswithout a VP:The syntax ofadverbs in Ton- gan. In Mu¨ller, Stefan (Ed.), Proceedings of the 24th International Conference on Head-Driven Phrase Structure Grammar, University of Kentucky, Lexington, 61–81. Stanford,CA:CSLIPublications. Abstract ThePolynesianlanguageTonganappearstolacksurface-orientedmoti- vation for a VP constituent. Even so, adverbial elements appear in both a rightwards location and a leftwards location, superficially similar to the S- adverbsandVP-adverbsin well-studiedwesternEuropeanlanguages. This paper explores how the Tongan “VP-adverbs” (as well as others) can be analyzed in HPSG without a VP for those adverbs to attach to. Several kinds of analyses, representing different strands of research on the syntax of adjunctsin HPSG, are explored: a Adjuncts-as-Valentsanalysis, a VAL- sensitiveAdjuncts-as-Selectorsanalysis,anda WEIGHT-sensitiveAdjuncts- as-Selectorsanalysis. All suggestthat an analysisof the adverbswithouta VPispossible;aWEIGHT-sensitiveAdjuncts-as-Selectorsseemstohavethe fewestissues. 1 Introduction This paper examines the syntax of (mostly) single-word adverbial elements in the language ofTongan,alanguage oftheisland-nation ofTongaintheSouthPacific. Tonganisamemberof(fromsmallesttolargest)thePolynesian,Oceanic,andAus- tronesian language families. As one might expect, Tongan shares many syntactic properties withother members ofthese families, and, inparticular, itseems likely that many ofthe syntactic issues surrounding adverbial elements discussed herein are not just found in Tongan, but are widespread in other Polynesian languages and, at least, in other closely related Oceanic languages. However, the discussion below will focus on Tongan in order to ensure a thorough discussion and analysis foronelanguage. Asinmanylanguages, Tonganallows expressions functioning as modifiers of predicates – adverbial elements – in different places within the clause. In very broad strokes, the locations in Tongan are akin to Jackendoff’s (1972) two cat- egories for English: S-adverbs (more linearly leftwards) and VP-adverbs (more linearly rightwards). While it is not presently clear whether these two locations in Tongan have strong semantic motivations (enough to consider them “sentence †Thanks to Emily Bender, Rui Chaves, Berthold Crysmann, Petter Haugereid, Maksymilian Da˛bkowskiandtheanonymousreviewersofmyoriginalHPSG2017Conferenceabstractforhelpful suggestions,criticisms,andpointers.Theusualdisclaimersapply. Abbreviations usedinclude: ABS =absolutive case; CAT =category feature; CONT/CNT=con- tent feature; DEF = definitive accent; DEM = demonstrative; DEPS = dependents feature; DET = determiner; DU=dualnumber; elist=emptylist; ERG=ergativecase; ESS=essivecase; EXCL= exclusive; FUT =futuretense; HD=headfeature; IND=indexfeature; LOC =localfeature; MOD =modifieefeature;nelist=non-emptylist; PHON=phonologyfeature; PFT=perfectaspect; PL= plural number; PLUR = pluractional; PN = proper noun/pronoun; POSS.O = O-class(subordinate) possession; Pred/pred = predicate word class; Predmax = expression headed by a member of the predicatewordclass,requiringnofurthervalents;prep=preposition;PROX=proximal;PST=past tense;SorSG=singularnumber; SUorSUBJ=subjectgrammticalrelation;SYNSEM/SS=syntax- semanticsfeature;TAM=tense-aspect-moodword;TR=transitiveaffix;VAL=valencefeature;VP =verbphrase;XARG=externalargumentfeature 62 operators” and “predicate operators” like Thomason & Stalnaker (1973) propose forEnglish);primafacie,thereseemstobesimilarities. However,aswillbecomeclear, the“VP-adverb”location inTonganis,infact, quitesurprising. Thisisbecausethelocationcontravenestheconventionalwisdom that adjuncts appear further away from their heads than arguments do, and, given this location and conventional approaches to adverb syntax (see Pollock (1989) and Potsdam (1998) for some older, classic approaches), it would appear that the ′ TonganVP-adverbsdonothaveanobvious phrasalconstituent (thatis,eitheraX or XP category) to attach to. Thus, this class of adverb does not seem to have an obviousorstraightforward analysis,inbothconstraint-based lexicalistframeworks aswellasinmovement-based syntactic frameworks.1 Thus,thispaperlookstoexploreTonganadverbsfurther–especially the“VP- adverbs” – and works to develop the best analysis of them within the constraint- based lexicalist framework of Head-driven Phrase Structure Grammar (HPSG) (Pollard&Sag,1987,1994;Ginzburg&Sag,2000;Sagetal.,2003). Afterprovid- ing adiscussion ofthe foundations of Tonganclause structure, the paper willturn tothebasicsofadverbialsyntaxinTongan,notingthatthereare,infact,threeareas (orzones,astheywillbecalled)thatadverbialelementscanoccupyinTongan. The paper next considers the analysis of the adverbial elements in the two peripheral zones –Zones1and 3–inthecontext ofthetwodominant approaches toadjunct syntax within HPSG: Adjuncts-as-Selectors and Adjuncts-as-Valents. From this discussion, it seems as though the Adjuncts-as-Selectors approach seems to offer a slightly better analysis for the peripheral zones, and so the paper next considers how the Adjuncts-as-Selectors approach might handle the middle zone’s adver- bial elements. After sketching an analysis sensitive to valence, a problem for that analysis is pointed out. The paper then offers a final sketch analysis of another Adjuncts-as-Selectors approach – this one making use of grammatical weight – thatprovidesafixtotheproblemsfoundinthepreviousanalysis, beforewrapping upwithsomeconcluding remarks. The sketch formal analyses will employ the version of HPSG from Ginzburg & Sag (2000) – for concreteness – with one minor alteration: the VAL(ence) list willbeonesinglelist,ratherthansplitintoseparate SUBJECT andCOMPLEMENTS lists. A few further features not utilized in Ginzburg & Sag (2000) will be used in this paper, too, but they will be discussed as they become relevant. The choice to follow Ginzburg & Sag (2000) does not seem particularly confining and I am confident that the analyses presented herein could be fairly easily be ported into theframeworkoftheSign-BasedConstruction Grammar(SBCG)(Sag,2012).2 The analyses contained herein will sidestep the issue of whether these adver- bialswordsaretrulyaseparateclassofwords–whichonemightcalladverbs–or 1Massam (2010) provides a movement-based analysis for the adverbial elements in Tongan’s sibling language, Niuean, using “Roll Up Movement” of Cinque (2005). However, as Massam’s paperdiscusses,thisanalysisisnotentirelywithoutproblems,evenwithintheconfinesofMinimalist assumptions.Alsosee(Massam,2013)forfurtheranalysisofthisareaoftheclauseinNiuean 2Infact,Ball(2008),onwhichthispaperbuilds,isentirelycouchedwithintheSBCGframework. 63 have some other categorization. There is some reason to think that the adverbials wordsinTonganmightbeverbsoratleastcloselyalignedwiththem: theso-called adverbscanappearwithverbalderivational morphology andatleastsomeofthem can function as main predicates. However, the analyses would only be minutely different if the adverbial words are treated as verbs or as part of a distinct adverb class, soIwilldefault totreating them asadverbs (and will, henceforth, call them justbythatterm). 2 Basic Tongan Clause Structure 2.1 Empirical Basics ofClauses The morphosyntax of Tongan involves little-to-no inflectional morphology and, instead, uses a fair amount of function words. The phrases are strongly head- initial, with the aforementioned function words appearing at the left-edges of the relevantgroupings. Abasicsentence,whichillustratestheseproperties, isgivenin (1): (1) Na‘eta¯mate‘i‘e Te¯vita‘a Ko¯laiate. PST kill.TR ERG David ABS Golaith ‘DavidkilledGolaith.’ (Churchward, 1953, 67) Within (1), na‘e ‘PST’ is from a word class I will call TAM (tense-aspect-mood marker), aclass ofwords thatseemstofunction quite similar toauxiliary verbs in other languages. Ta¯mate‘i ‘kill’ is from a word class I will call predicate, a class thatIassumeincludes bothtraditional verbs aswellasadjectives. Thephrases ‘e Te¯vita ‘ERG David’ and ‘a Ko¯laiate ‘ABS Goliath’ are post-predicate argumental phrases related to this predicate; for concreteness, I will assume these are PPs. As the glosses in (1) indicate, the prepositions which signal predicate-argument relationships inTonganareergatively-aligned. Figure1givesaschematic viewof the clause in Tongan. Within the scheme of Figure 1, the TAM and predicate are TAM Predicate Argumental Phrases Figure1: BasicComponentsoftheTonganClause strictlyordered; theargumentalphrases, ontheotherhand,canbeflexiblyordered withintheirregionoftheclause, withinformation-structural import. While many arguments occur in the post-predicate location, not all do. Some argumentsarenot,infact,overtlyrealizedatall(thesewouldbeinstancesof“zero anaphora”). Anexampleofthisoccursin(2): (2) Na‘ehola. PST run.away ‘(He)ranaway.’ (Chung,1978, 39) 64 Afurtherclassofprincipledexceptionstothegeneralization thatallargumentsoc- cur after their predicate comes from certain arguments with pronominal meaning. Thesearerealizedbeforethepredicate, butaftertheTAM.Onesuchelementisku ‘1SG.SUBJ’ in(3): (3) Na‘áku manuatu’i ia. PST 1SG.SUBJ remember.TR 3SG ‘Irememberedhim.’ (Churchward, 1953, 66) As is evident from the glosses in (3), these elements are not ergatively-aligned; rather,theyindexthetraditionalsubjectcategory. These“preposed pronouns”(the term for them from Churchward (1953) that I will adopt3) do seem to play a role indetermining thebest analysis oftheTongan “VP-adverbs,” asIwillreturn toin section5. 2.2 AnalysisofBasicClausesinTongan In(1),aswithanyVSOordering, theverb(orpredicate) andtheobject(orpatien- tive argument) are realized discontinuously, and this is regularly possible in Ton- gan. This raises a question: should a constituent of a verb + object (alternatively, a predicate + its non-subject arguments) – what I will call a VP – be recognized? This question is considered in-depth in Ball (2008, ch. 3), and the finding there is that there is no strong surface-oriented motivation for recognizing a VP for Ton- gan. To summarize the motivations for Ball’s (2008) conclusion: (1) no auxiliary orotherverbclassobviouslysubcategorizesforaVP;(2)VP-coordination‘over’a subject isnotpossible; (3)“VP-fronting” ispossible, but seems toinvolve anom- inalized construction, so it could be seen as just a subspecies of NP-fronting; and (4)ellipsisispossible, but“VP-ellipsis”seemsliketheelidedelementisbetteran- alyzed(again)asanominalized construction. Furthermore, “VP-ellipsis” doesnot alwaysclearlypickoutjustthepredicate anditsnon-subject arguments. However, does that mean that there is no immediate constituents in Tongan betweentheclauseasawholeandthephrasesservingasarguments? Thisquestion is considered in-depth in Ball (2008, ch. 4) and the finding there is that a unit consisting ofthepredicateandallofitsargumentsdoesappeartobeaconstituent. max IwillinformallycallsuchaconstituentPred . Tosummarizethemotivationsfor max Pred inBall(2008): (1)bothTAMsandotherverbs,termed“quasi-auxiliaries,” max max doappeartosubcategorize foraPred and(2)Pred coordination ispossible. With these constituency ideas as a backdrop, let me next sketch the analy- sis presented in Ball (2008) for the Tongan clause, which incorporates these con- stituents. On this analysis, the Tongan clause is principally put together with the Head-All-Valents Rule,givenin(4):4 3Inspiteoftheconnotationsofthisterm,theredoseemtobecompellingreasonstoviewthe“pre- posedpronouns”assuffixesontheTAMs;seethediscussioninBall(2008,ch.4). Thediscussion thatfollowsdoesnotcruciallyhingeonhowexactlythe“preposedpronouns”relatemorphophono- logicallytotheTAMword,andsoIleavetheissueasidehere. 4BoldedHstandsforthehead,inbothrulesandtreestructures. 65 (4) Head-All-Valents Rule VAL h i → H VAL h 1,..., n i 1 ... n h i h i Theruleallowsaheadwithsomenumberofvalentstocombinewithallitsrequisite valentstoformaconstituent. Everyinstance oftheHead-All-Valents Rulecreates a“saturated”constituent, onewherenofurtherargumentsarerequiredtocomplete theunit. Although (4)is quite similar toprevious HPSG(or SBCG)proposals for verb-initial structures—including Schema 3 from Pollard & Sag (1994, 40), sai- ph from Ginzburg & Sag (2000, 36), aux-initial-cxt from Sag (2012, 188)—the rule in (4) is slightly different in one key way: its head is left underspecified for word/phrase distinction, for reasons that will be crucial for the later analysis of adverbs. To build a canonical Tongan clause, such as the one from (1), two instances of (4) are all that is needed. Looking at the structure bottom-up (though nothing intrinsically requires this), the predicate head ta¯mate‘i ‘killed’ is combined with itstwoarguments, ‘eTe¯vita‘ERG David’and‘aKo¯laiate ‘ABSGoliath’via(4)to max formaPred . Then,tomakethesentence(theTAMphrase),theTAMheadna‘e max ‘PST’ combines with the aforementioned Pred via (4). The resulting structure isasinFigure2. PHON A ⊕ B ⊕ C ⊕ D  HEAD 5  SS  "VAL h i#      PHON Ahna‘ei PHON B ⊕ C ⊕ D H HEAD 5tam   HEAD 4  SS SS 1  "VAL h 1 i#  "VAL h i#         PHON C h‘eTe¯vitai PHON D h‘aKo¯laiatei PHON B hta¯mate‘ii  prep   prep  H HEAD 4 pred  HEAD HEAD SS SS 2 "CASEerg# SS 3 "CASEabs#  "VAL h 2, 3 i#  VAL h i   VAL h i        Figure2: TreeStructureforExample(1) 3 Basics of Adverbial Syntax in Tongan With the Tongan clause basics established, I turn to the basics of the syntax of adverbs in Tongan. Following in the footsteps of the clear discussion of French adverbs in Bonami et al. (2004), it seems useful to talk about adverb locations in 66 terms of pre-analytical zones. In his seminal grammar of Tongan, Churchward (1953) suggests thattherearejusttwoadverbzones: “preposed” (before thepred- icate) and “postposed” (after the predicate). This division would seem to exactly line up with the S-adverb–VP-adverb distinction mentioned in the introduction. However, careful examination of adverb location suggests that there are actually at least 3 zones in which adverbs can appear in Tongan. Their positioning with respecttothelandmarksoftheTonganclausenotedinFigure1aregiveninFigure 3. Onthescheme in Figure 3, Churchward’s (1953) “postposed” adverbs are split Zone1 Zone2 Zone3 TAM ↓ Predicate ↓ ArgumentalPhrases ↓ Figure3: TheLocationsoftheAdverbZones betweenZone2andZone3. Whilethedifference betweenZone2andZone3can be slight (and, thus, Churchward’s distinctions were not without merit), there do seem to be some differences between the two, which will be highlighted further below. The ability of the same adverb to appear in different zones across sentences (a property of some English adverbs, as noted by Jackendoff (1972), and some French adverbs, as note by Bonami et al. (2004)) in Tongan is presently not well- understood. Preliminarily, potential positioning of a single adverb in multiple zones inTonganseemslike itisrare, ifnotimpossible. However, further research isneededtoclarifythisempiricalarea. Iturn now to considering the adverbs of each zone in slightly more depth. As Figure3indicated,theZone1adverbsarelocatedbetweentheTAMandpredicate. ExamplesofZone1adverbsinclude theitalicized wordsin(5)–(7): (5) Na‘etoutou fakama‘a‘e Te¯vita e faliki. PST repeatedly clean ERG (name) ABS.DET floor ‘Te¯vitacleanedsome(particular) floorrepeatedly.’ (Ball,2008, 65) (6) Na‘ekei kata ‘a e ongoki‘i ta‘ahine faka‘ofa‘ofá. PST stilllaugh ABS DET DU smallgirl beautiful.DEF ‘Thetwobeautiful girlswerestilllaughing.’ (Broschart, 2000,353) (7) Ko e tangata tonu pe¯ ia na‘á ne fa‘a fakakaunga¯me‘a ESS DET man exact very that PST 3S.SU habitually associate mo iá. with3SG.DEF ‘Thatwastheverymanwithwhomhehabitually associated.’ (Church- ward,1953,128) Itseemsplausible thatsomeadverbsinZone1(suchastheadverbsin(6)and(7)) have “high” scopal properties (maybe equivalent to semantically applying to the max wholeeventualitydenotedbythePred orotherwisehavesomesemanticaffinity 67 with the TAM), but how widespread this is remains to be empirically verified. It, however,isclearthatmultipleadverbsarepossibleinthiszoneatonetime,aswith fu‘u‘excessively’ andkei‘still’in(8):5 (8) ‘Oku fu‘u kei si‘i. PROG excessively stillsmall ‘Itisstilltoosmall.’ (Churchward, 1959,260) Zone 2 adverbs, as Figure 3 indicated, appear between the predicate and any and all of the argumental phrases. Examples include the italicized words in (9)– (13), many of which seem to come from semantically coherent subclasses of ad- verbs. Examples(9)and(10)illustratethatmanneradverbsappearinZone2: (9) Na‘efakama‘afakalelei ‘e Pita e faliki. PST clean well ERG (name) ABS.DET floor ‘Pitacleaned some(particular) floorwell.’ (Ball,2008, 49) (10) Na‘etali totoka ‘a Mele kihe pasi. PST waitcalmly ABS(name)to DET bus ‘Melewaitedcalmlyforsome(particular) bus.’ (owndata) AsubclassofadverbwidelyfoundinthePolynesianlanguagesiswhatPolynesian grammarians refer toasdirectionals. Directionals function toplace events intime orspace, sometimes inquite abstract oridiomatic ways. Example(11) showsthat directionals –atu‘forth’isone–likewiseappearinZone2inTongan: (11) Na‘efakatau atu ‘e Sione hono ‘u¯ su¯. PST transact forth ERG (name) 3SG.POSS.O PL shoe ‘Sionesoldhisshoes.’ (Ball,2008, 87) YetanothersubclassofadverbsiswhatChurchward(1953)dubbed theadverbsof uncertainty. Nai‘maybe’,amemberofthissubclass,appearsin(12),exemplifying thattheadverbsofuncertainty appearinZone2,too: (12) Na‘ekainai ‘e Sione e mata‘iika lahi? PST eat maybe ERG (name) ABS.DET piece fishbig ‘Sioneatesomebigpieceoffish?’ (Ball,2008, 87) Finally, theabovesubclasses arenotanexhaustive listofthekindsofadverbsthat are potentially able to appear in Zone 2. Stillother adverbs can appear in Zone 2, as(13)shows: (13) Na‘eto¯ ‘anefe¯ ‘e Sione ‘a e manioke? PST plantwhen.PST ERG (name) ABS DET cassava ‘WhendidSioneplantsomecassava?’ (owndata) 5OntheassumptionthattheZone1adverbfurthertotheleftshouldapplytothelargersemantic domain,thetranslationprovidedbyChurchwardfor(8)isabitsurprising.Itremainsforfuturework todeterminehowanomalous,ifatall,(8)mightactuallybe. 68 Impressionistically, of the three zones, it appears that Zone 2 has more diverse collection ofmembers,astheabovediscussion suggests. Certainly, the presence of manner adverbs in the Zone 2 seems unsurprising, asmanneradverbs seem intuitively tohaveastrong semantic affinity forthemain predication itself(suggesting thesemantically-based approach toadverb syntaxof Ernst (2002) may also be applicable to Tongan). However, as with the Zone 1 adverbs, an exhaustive study of the unity of the semantic subclasses awaits future research. Itisclear, however, thatmultiple adverbs arepossible in thiszone atthe same time,asexemplifiedin(14): (14) Na‘efakama‘afakalelei ma‘upe¯ ‘e Sione ‘a e faliki. PST clean well always ERG (name) ABS DET floor ‘Sionealwayscleanedsome(particular) floorproperly.’ (owndata) Furthermore, from example (14), it seems reasonable to conclude that, at least within this zone, the more rightwards an adverb is, the larger semantic domain it appliesto. Lastly,wecometotheadverbsofZone3,which,asFigure3indicated, appear clause-finally, afteranyargumental phrases. Adverbs appearing inZone3include theitalicized wordsin(15)and(16): (15) Na‘áne faieni ‘aneafi. PST 3SG.SUBJ do PROX.DEM yesterday ‘Hedidthisyesterday.’ (Churchward, 1953, 66) (16) ‘E ha‘u ia kiate kimautolu ‘apongipongi. FUT come 3SG to.PN 1PL.EXCL tomorrow ‘Heiscomingtoustomorrow.’ (Chung,1978,148) Asexamples(15)and(16)clearlyindicate, Zone3seemstobethespotfordeictic temporal adverbs. Whether other kinds of adverbs are found here remains to be empirically discovered. Iftherearenoovertargumental phrases ofagivenpredicate, Zone2andZone 3adverbslookliketheyappearinthesamelocation. Anexampleofthisisin(17), wheretheadverbagainisitalicized: (17) ‘E fai‘apongipongi. FUT do tomorrow ‘Itwillbedonetomorrow.’ (Churchward, 1953,197) Otherexamples(like(16))clarifythat‘apongipongi ‘tomorrow’doesseemtopat- tern as a Zone 3 adverb; but from just (17), that conclusion is not so clear. So, in theveryleast, thedistinction betweenZone2andZone3adverbs can betricky to tease apart; at worse, the difference is a false dichotomy. Even if the difference is not a false dichotomy, it does seem that these two kinds of adverbs share some affinities; any good analysis should group the adverbs ofthese zones together ina principled fashion. 69 4 Approaching An Analysis 4.1 Approaching Adjuncts inHPSG A vast majority of HPSG analyses of structure-building crucially rely on depen- dencies between theelements combining tolicense the said structure. Thisistrue forthe syntax of adjuncts (adjectives, adverbs, and others) asmuchother kinds of syntactic relationships. However, there have been two styles of approaches to the syntaxofadjuncts inHPSGandIwilloutlinethemhere. Ontheclassic HPSGapproach tothesyntax ofadjuncts –aspresented inPol- lard & Sag (1994, 55–57) – the adjunct is viewed as imposing requirements on the head that it goes with (in line with some observations about the nature of se- mantic restrictions on the head-adjunct relationship; for example, see Muehleisen (1997) forthediscussion ofthese kindsofrelationships withthedomainofadjec- tives in English). I will refer to this style of analysis as the Adjuncts-as-Selectors approach. Theseadjunct-mandated requirements, inHPSGanalyses, aremediated viatheMODIFIED(MOD)feature: thevalueofMODisadescriptionofthesyntactic (and semantic) expression that the adjunct goes with. Thus, adjunct combination inHPSGcangenerally beseenasinvolving thefollowingrule:6 (18) Head-AdjunctRule(underspecified version) VAL C → H1 VAL C , MOD 1 h i h i h i Thecommabetweenthetwodaughters(ontheright-sideoftherule)indicatesthat, atthislevelofabstraction, thedaughterscouldbeineitherorder,subjecttofurther constraints that a given language, combination, or syntactic item might impose. Despite what differences in notation might lead one to believe, the rule in (18) actually is very close to the usual phrase structure grammar approach to adjuncts (used byawidevariety offrameworks): itallow ahead elementand anadjunct to together form a phrase. However, unlike in the standard X-bar approach, the rule in(18)doesnotstipulatethattheheadmustbeofcategoryX′;instead,theadjunct is free to make its own requirements. This flexibility will be of great help in the analyses insection5. The alternative style of analysis (pursued by Pollard & Sag (1987); Bouma et al. (2001); Levine & Hukari (2006); Bonami & Godard (2007); Sato & Tam (2008), among others), flips the selection relationship (though not the headedness relationship). Onthisstyleofanalysis,theheadselectsfortheadjunct,justashead selects for arguments. Thus, syntactic analyses involved this style (which I will dub Adjuncts-as-Valents approach) manipulate adjuncts in a head-driven fashion. Inmanyversions ofthisstyleofanalysis, theadjuncts areadded tothe VAL listof theheadbyanargument-extending lexicalrule,suchasthe(generic) onein(19): 6This rule is highly comparable to Schema 5 of Pollard & Sag (1994, 58), the head-adjunct- phraseofGinzburg&Sag(2000),andtheHead-ModifierRuleofSagetal.(2003,146). 70