Integrating Linguistic Dimensions: The Scope of Adverbs Olivier Bonami Danie`le Godard Universite´ Paris-Sorbonne LLF Proceedingsofthe14thInternationalConferenceon Head-DrivenPhraseStructureGrammar StanfordDepartmentofLinguisticsandCSLI’sLinGOLab StefanMu¨ller(Editor) 2007 CSLIPublications pages25–45 http://csli-publications.stanford.edu/HPSG/2007 Bonami, Olivier, & Godard, Danie`le. 2007. Integrating Linguistic Dimensions: The Scope of Adverbs. In Mu¨ller, Stefan (Ed.), Proceedings of the 14th Interna- tional Conference on Head-Driven Phrase Structure Grammar, Stanford Depart- ment of Linguistics and CSLI’s LinGO Lab, 25–45. Stanford, CA: CSLI Publica- tions. Abstract Threedistinctionsseemrelevantforthescopepropertiesofadverbs:their function(adjunctsorcomplements),theirprosody(incidentalorintegrated) andtheirlexicalsemantics(parentheticalornonparenthetical). Wepropose ananalysisinwhichthescopeofFrenchadverbsis alignedwiththeirsyn- tactic properties, relying on a view of adjuncts as loci for quantification, a linearization approachto the word order, and an explicit modelling of dia- logue. 1 Introduction Adverbsingeneralarescopalelements.1 Theycontrastwithotherscopalelements such as quantified NPs in the way their scope properties interact with other lin- guisticdimensions: syntax,prosody,lexicalsemanticsandpragmatics. Sincethese properties are not strictly correlated, a formalism which relies on one type of dis- tinction, such as dominance (e.g. Dik (1997), Cinque (1999)), fails to do justice to the complexity of the data. The HPSGarchitecture, where the different dimen- sions are both distinguished and articulated in feature structures, offers a chance forstating suchinteractions. Inprevious work,after pulling apart the prosodic properties ofadverbs, which interact directly with their syntax and compositional semantics, from their prag- maticproperties, whichdepend crucially ontheirlexicalsemantics (Bonamietal., 2004), we proposed HPSG analyses of parenthetical adverbs, that is, adverbs which do not contribute directly to the main content of an utterance (Bonami and Godard, inpress, a,b). Hereweconcentrate onmodelling the interaction between prosody, syntax and scope, improving on the proposals ofBonami and Godard (2003). We show that a linearization-based approach to adverb placement eases the modelling of the observed syntax-semantics interface constraints. We use a conservative, STORE-based HPSG approach to quantifier scope, in the style of Ginzburg andSag(2000),but nothingcrucialhingesonthischoice. We follow a solid tradition in distinguishing a number of semantic classes (for French, see (Molinier and Levrier(2000), Bonami et al.(2004)): connectives (donc ‘therefore’), speech act adverbs (franchement ‘frankly’), evaluatives (mal- heureusement ‘unfortunately’), modals (peut-eˆtre, ‘perhaps’), sentential agentives (intelligemment ‘intelligently’ inIl a intelligemment refuse´ de re´pondre ‘He intel- ligentlydeclinedtoanswer’),habitualadverbs(ge´ne´ralement ‘generally’), domain adverbs (syntaxiquement ‘syntactically’), frequency adverbs (souvent ‘often’), du- rationadverbs(longtemps‘foralongtime’)temporallocationadverbs(re´cemment ‘recently’), aspectual adverbs (de´ja` ‘already’), manner adverbs (intelligemment ‘intelligently’ in Il a re´pondu intelligemment ‘He answered intelligently’), degree 1Some adverbs, in particular manner adverbs, are often said to be scopally inert. This lexical semanticpropertyisdebatable,and,inanycase,doesnotchangethescopalcharacterofthecategory asawhole;see(Parsons,1972;Peterson,1997;Scha¨fer,2005). 26 Figure1: Pitchtrackofacanonical incidental realization adverbs(beaucoup‘alot’,intense´ment‘intensely’),andassociativeadverbs(seule- ment ‘only’). We also follow common practice in regrouping the first six classes, whichsharesomeproperties, undertheterm‘sentence adverb’. Ouranalysis takes into account all classes, except for connectives and associative adverbs, which have special interface properties linked to their relational semantics. It is based on French adverbs, but should apply to other languages; that is, although the de- tails of the behavior are different (for instance, as is well known, the syntax of adverbs isdifferent inFrench andEnglish), thedifferent dimensions andthe types ofinteractions thatarerelevant areexpected tobesimilar. 2 What is incidentality? The distinction between incidental and integrated constituents correlates prosodic properties of realizations of constituents with constraints on their syntactic posi- tions.2 In the case of adverbs, it also correlates crucially with scope, as we will see below. Incidental constituents are usually set apart by commas in French or- thography, although usage is far from being consistent on this point. For clarity, we explicitly mark incidentality in the examples by adding the symbol ‘ ’ at the ↑ boundaries ofincidental constituents. ‘( )’signals optional incidentality. ↑ 2.1 Incidental vsintegrated adverbs: Aprosodicproperty Existingstudiesofincidental constituents inFrench(Fagyal,2002;Mertens,2004; Delais-Roussarie, 2005) state that they are prosodically ‘autonomous’, and are set 2Notethatincidentalityisnotapropertyspecifictoadverbs,nortoadjuncts. Afewexamplesof incidentalconstituentsare: dislocatedphrases, topicalizedphrases, vocatives, interpolatedclauses, appositions,somerealizationsofcomplements(Ila a` sonfre`re donne´ a` lireProust!‘Hehas,to ↑ ↑ hisbrother,givenProusttoread’). 27 Incidental Integrated Independent Independent Partofa Intonational Phrase Phonological Phrase Phonological Phrase (“commaintonation”) Figure2: Realizations ofincidental andintegrated prosody apartfromtheirenvironment byanumberoffactors,illustrated inthetypicalpitch track in Fig. 1: optional pauses, lengthening of the last syllable preceding the in- cidental, of the last syllable of the incidental, F modification at the boundaries, 0 register change. However none of these manifestations of incidentality appears to be categorically necessary, as confirmed by an ongoing study by Bonami and Delais-Roussarie on the speech corpus ESTER (Galliano et al., 2006). This sug- gests that the distinction is phonological rather than phonetic, and, accordingly, that neutralization phenomena make the distinction opaque in certain cases. As Fig. 2 illustrates, in terms of familiar prosodic categories (Selkirk, 1984), we ob- servethreetypesofrealizations, oneofwhich(Independent Phonological Phrase) iscompatible bothwithincidental andintegrated status. 2.2 Incidentalityand Adverb Classes Mostadverbscanoccurwitheitheranincidental oranintegrated prosody,asillus- trated in(1)withafewexamples,although there aresomeconstraints. (1) a. Paula( )heureusement ( )bienre´pondu. (evaluative) ↑ ↑ ‘Paulhasfortunately answered well.’ b. Paulavait ( )habituellement ( )unavistranche´. (habitual) ↑ ↑ ‘Paulhadusually aclear-cut advice.’ c. Paulavait ( )souvent ( )unavistranche´. (frequency) ↑ ↑ ‘Paulhasoftenaclear-cut advice.’ d. Paula( )silencieusement ( )quitte´ lapie`ce. (manner) ↑ ↑ ‘Paulhassilently lefttheroom.’ The dual prosodic realizations in (1) show that incidentality is a property of occurrences, not of lexemes per se, although some adverb classes (or subclasses) are specified regarding their prosody: degree adverbs are not incidentals, speech actadverbsarealwaysincidentals; light(Abeille´ andGodard,2001)andresultative (Geuder,2000) manneradverbscannot beincidentals. 28 2.3 Incidental adverbsand Position Thereareconstraints ontheprosodicrealization ofadverbsdepending ontheirpo- sition. Consider the following schema, where the potential position for the adverb is noted –px–. We distinguish between 4 positions: the adverb can occur initially (–p1–), before the verb (–p2–), between the auxiliary verb and the past participle (–p3–), andaftertheparticiple (–p4–).3 (2) –p1–Paul–p2– a–p3–envoye´ –p4– sesvoeux –p4–a` unvieilami–p4–. ‘Paulhassenthisbestwishestoanoldfriend.’ The generalizations are as follows. First, adverbs are normally incidental in –p1–,withafewexceptionsthatweleaveasideforthepurposesofthispaper.4 We illustrate theproperty withbothsentential (3a,b) andnon-sentential (3c)adverbs: (3) a. Franchement celan’envautpaslapeine. ↑ ‘Frankly, itisnotworthit.’ b. Malheureusement/ Naturellement/ Officiellement/ Habituellement/ In- telligemment nous allonsaucine´ma. ↑ ‘Unfortunately/ Naturally/ Officially/ Usually/ Intelligently we go to themovies.’ c. Re´cemment/ Souvent/Lentement ilestalle´ a` l’ope´ra. ↑ ‘Recently/ Often/Slowlyhewenttotheopera.’ Second,adverbs areincidental in–p2–iftheverbisfinite(4),butintegrated if theverbisinfinitival (5): (4) a. Paul malheureusement/naturellement/officiellement/habituellement ↑ nepeutpass’enpasser. ↑ ‘Paulunfortunately/ naturally/ officially/ usually cannotdowithout it.’ b. Paul souvent pre´fe`re rester chezlui. ↑ ↑ ‘Pauloftenprefers tostayhome.’ (5) a. Paulsepromettait desouvent alleraucine´ma. ’Paulpromised himselftooftengotothemovies’ b. Pauldisait habituellement alleraucine´maledimanche. ’Paulpretended tousually gotothemoviesonSundays’ Third,adverbsmaybeeitherincidentalorintegratedin–p3–and–p4–(1),with twoconstraints. Lightadverbsdonotoccurin–p4–(6)(Abeille´ andGodard1997), 3Thereisnoevidencefordistinctionamongpositionsforconstituentsaftertheparticiple. 4Nonincidental adverbs are found in –p1– in two constructions: the reinforced assertions con- struction discussed below and thecomplex cliticinversion construction, which iscompatible only withafewadverbs(e.g.Peut-eˆtrePaulviendra-t-il‘PerhapsPaulwill.come-he’.Inaddition,subject NPinversiondisallowsrealizinganutteranceinitialadverbasanindependent IP(e.g. alorsarriva Paul ‘then arrived Paul’). It remains to be seen whether the adverb is integrated in this case, or whethergeneralprosodicfactorsdisfavoranIPrealization. 29 and sentential adverbs can appear in –p4– only if incidentals (7). This pattern showsthatincidentalsarenotoutsidetherealmofsyntax,contrarytowhatisoften assumed, since they are sensitive to syntactic position. Note, however, that there is no complementary distribution: sentential adverbs occur either as integrated or incidental in –p3–, and the others occur either as integrated or incidental in –p3– and–p4–. (6) a. Paulamalre´pondu a` laquestion. ‘Paulbadlyanswered thequestion’ b. Paulare´pondu a` laquestion *(tre`s) mal. ‘Paulanswered thequestion (pretty) badly.’ (7) a. Paulare´pondu *( )force´ment *( )a` laquestion. ↑ ↑ ‘Paulnecessarily answered thequestion.’ b. Paulare´pondu /re´pondra a` laquestion *( )force´ment. ↑ We formalize the distinction with the feature INCID , which is a syntactic ± HEAD feature, with a prosodic correlate. The reason why we need a HEAD fea- ture (pending a more elaborate conception of phonological properties) is that an incidental expression can be a phrase, such as a modified adverb (Paul fort mal- ↑ heureusement aoublie´ lecadeau,lit. ’Paul,mostunfortunately, hasforgotten the ↑ gift’). 3 Scope, Syntactic Functions and Incidentality Adverbs may have four distinct functions: they can be heads of a clause, fillers, adjuncts orcomplements. Wediscuss adjunct and complement adverbs below. As headsofaclause,adverbsoccurwithaclausalcomplementwhichtheyscopeover, although a quantifier in the complement may outscope the adverb (see Probable- ment que tu as vu un de mes e´tudiants, lit. ’Probably that you have met one of my students’). Non-wh adverb fillers are found in two constructions. First, in adverb topicalization, as illustrated in (8a).5 In such cases the adverb receives in- cidental prosody, and takes its scope at the extraction site—in (8a), the extracted adverb re´cemment scopes below in-situ suˆrement. Second, in the reinforced as- sertion construction, where a clause initial adverb receives a special prosody, the 5Notethatclause-initialincidentaladverbsmaybeeitheradjunctsorfillers.Thatthetwoanalyses arepossibleisshownbytheadverbs,suchasfrequencyadverbs,thatcannotbefillers,butdooccur clauseinitially(i-ii).SeeBonamiandGodard(2007)fordetails,andMaekawa(2006)forananalysis ofparalleldatainEnglish. (i) #Fre´quemment,jesaisqu’ilvaa`Paris (intended)‘IknowhefrequentlygoestoParis.’ (ii) Fre´quemment,ilvaa`Paris ‘HefrequentlygoestoParis.’ 30 rest of the sentence being deaccented. The construction signals that the speaker amendsaproposition inthecommongroundwithrespecttothatpartofthepropo- sition whichisexpressed bythefiller.6 Itoccurs onlyinrootclauses, anddoes not involve the sameclasses ofadverbs astopicalizations. Weleave aside the analysis oftheseconstructions, although astandard viewofextraction andquantifier scope clearly predicts thecorrect scopal properties. (8) a. Re´cemment,jepense qu’il asuˆrement e´te´ authe´atre. ‘Recently, Ithink hecertainly wenttothetheater.’ b. Prudemmment,il m’avaitpromis qu’ il parlerait ! Prudently he promisedthathewould-talk 3.1 Integrated adjuncts Westartwiththecaseofintegrated adjuncts, although itisstatistically lesspromi- nent,becauseitismoststraightforwardly accounted for. Inouranalysis,integrated adjunct adverbs are found mostly to the left of infinitival VPs (not of finite VPs). They have scope over an adverb included in the VP (9), but they are not scopally ordered withrespect toquantified NPs(10). (9) a. Ilsesouvenait de[longtemps [s’eˆtre souvent retire´ chezsesparents]] (longtemps>souvent,*souvent>longtemps) ‘Heremembered having oftenretired tohisparents’ house.’ b. Ilsesouvenait de[souvent [s’eˆtre longtemps retire´ chezsesparents]] (souvent>longtemps,*longtemps>souvent) (10) Ilsepromettait de[souvent [lireunjournal]] (souvent>un,un>souvent) ‘Hepromised himselftooftenreadanewspaper.’ The data concerning the two adverbs is taken care of by the usual constraint on head-adjunct phrases: the content of the phrase is identified with that of the adjunct,whichtakesasitsargument thecontent ofthehead,andthecontent ofthe headVPisidentifiedwiththatoftheintegratedpostverbaladverb(seesection3.2). (11) hd-adj-ph CONT 2 → ! " H HEAD MOD 1 1 | # $ #CONT 2 $ Thedata concerning the quantifier NPshowsthat the adjoined adverb mustbe considered as a locus for quantification. Ginzburg and Sag (2000) analyzes only heads as such loci: they inherit the store of their arguments, and either transmit their store to the construction they head, or interpret the scopal elements (some or 6See Godard and Marandin(2006) on a syntactically different, but pragmatically similar, con- structionofItalian. 31 CONT 6 #STORE $ {} H MOD 5 CONT 3 # $ SS 5  QTS 4   #STORE 4 $ CONT 6 # $ { }  #NUC often(3)$       STORE {}  H   souvent ARG-ST 1 [IND x],2 a-rel  + ,   IND y  CONT 4 QTS CONT 3#NUC #r$ead(x,y)$  RES newspaper(y)           - . STORE 4  STORE 4   { }   { }      lire unjournal Figure3: (9a)withnarrowscopeforsouvent all),puttingtheminthevalueoftheirQUANTS. Weextendthisanalysistoadjoined constituentswiththefollowingconstraint,whichsaysthatthestorecomesnotonly fromarguments, butalsofromamodifiedconstituent. HEAD MOD [STORE 0] |  /0 12  (12) a. ordinary-lexeme ARG-ST [STORE 1],...,[STORE n]  →  STORE +((0) 1 n) S ,  ∪ ∪···∪ \    CONT QUANTS order(S)   |    CONT 1 b. quantifier-lexeme →#STORE 1 $ { } Accordingly, a quantifier such asun journal in (9a) can be scoped at the verb, that is put in its QUANTS, in which case the adverb souvent has scope over it (see Fig. 4). Alternatively, the quantifier remains in the store of the verb and the VP, and is scoped at the adjunct. In this case, it has scope over the adverb, because it is not part of the content of the VP, which the adverb takes as its argument. (see Fig.3). 32 CONT 6 #STORE $ {} H MOD 5 CONT 3 # $ SS 5  QTS   #STORE 4 $ CONT 6 #$ { }  #NUC often(3)$       STORE {}  H   souvent ARG-ST 1 [IND x],2 a-rel  + ,   IND y  CONT 4 QTS 4 CONT 3#NUC #rea$d(x,y)$  RES newspaper(y)           - . STORE  STORE 4   {}   { }      lire unjournal Figure4: (9a)withwidescopeforsouvent 3.2 Complements Asis largely accepted in HPSGanalyses of various languages, wetreat integrated post-verbal adverbs or adverbials as complements (e.g.Miller, 1992; Noord and Bouma, 1994; Abeille´ and Godard, 1997; Bouma et al., 2001). We adopt such a treatment mainly for coherence with existing HPSGaccounts of French grammar, in particular the grammar of pronominal affixation and extraction. Locative ad- verbials can be pronominal prefixes on the verb like complements (as in Paul l’y a rencontre´e, P. CL-CL has met, ’Paul has met her there’). Similarly, many ad- verbs can be extracted. Thus, if we assume that only valents can be extracted or realized as pronominal affixes, adverbs must be valents at least in some of their uses. Sincepostverbal integratedadverbshavethesamedistribution asargumental complements, itisnatural toanalyze themascomplements.7 The particular analysis we assume here relies on a lexical rule (13), which includes a modifier into the argument structure, and updates the content, to be the same as that of the modifier.8 The rule can be applied several times, the iteration beingconstrainedbythelexicalsemanticsoftheadverbs. Forinstance,ifamanner 7Infact,ouranalysisismostlyorthogonaltothedebatebetweentraceless,adverb-as-complement andtrace-based,adverbs-as-adjunctsanalyses,sincethefunctionofintegratedadverbsplaysnorole indeterminingtheirpositionortheirscope. 8This lexicalrule provides the same effects asthe version of Argument StructureExtension in BonamiandGodard(inpress,b),withouttheoverheadofanMRS-basedsemantics. 33 andamodaladverbareaddedintheargumentstructureofthesameverb(asinvenir probablement rapidement ’to probably come rapidly’), the rule must apply first to the manner adverb, since it cannot have scope over the content of amodal adverb. Asanillustration, (14) showsthelexical entryobtained byapplying therule twice totheverbvient‘comes’,whichisthenusedintheanalysisofasentenceinFig.5. word   INCID − ARG-ST 2 MOD #1$  (13) arg-extension-lr→  ⊕3CONT 3 4      STORE 4            CONT 3      STORE 4      word SS 1 ARG-ST 2   ! " (14) Theruleapplied twicetotheverbvient ‘comes’: INCID − INCID  −    come-rel A-S 3NPi,MCOONDT33CONT|NUC #ACT i $4,MCOONDT/!C5ONT 3"24     CONT 5      The main fact regarding the scope of postverbal integrated adverbs in French is its correlation with order: an adverb to the left has scope over an adverb to the right. For instance, the lexical semantics of souvent ‘often’ and longtemps ‘for a long time’ are such that either one can take scope over the other. Thus, the adverb on the left has scope in (15a,b). On the other hand, the lexical semantics ofprobablement ‘probably’ andsilencieusement ‘silently’aresuchthatthesecond cannot takescope overthefirst. Henceoneordering only isgrammatical. (15) a. Pauls’est souvent longtemps retire´ chezsesparents. (1>2,*2>1) 1 2 ‘Pauloftenretired tohisparents’ homeforalongtime.’ b. Pauls’est longtemps souvent retire´ chezsesparents. (2>1,*1>2) 2 1 c. Paulaprobablement silencieusement quitte´ lapie`ce. ‘Paulprobably silently lefttheroom.’ d. *Paulasilencieusement probablement quitte´ lapie`ce. The segregation of scopal material under the features QUANTS and NUCLEUS allows us to model this constraint directly as an order rule. Quantifiers may scope 34