Handout. - University of Massachusetts Amherst

16 Pages·2013·0.19 MB·English

∗
Fragment answers and the Question under Discussion
AndrewWeir,UniversityofMassachusettsAmherst
[email protected]
NELS44,UConn,October182013
1 Introduction
Iinvestigatesentencefragments,inparticularanswerstoquestions.
(1) a. WhatdidJohneat? —Chicken.
b. Wholeftearly? —Mary.
Keyquestion: arethesecovertlyclausal(i.e.elliptical),asbelow,orbase-generated‘bare’constituents?
(2) a. WhatdidJohneat? —Johnatechicken.
b. Wholeftearly? —Maryleftearly.
• Covertly clausal and elliptical: Merchant 2004. Base-generated ‘bare’ constituents: Stainton 1998,
2005,2006,Ginzburg&Sag2000,Jacobson2013.
• In particular Jacobson 2013 raises worries for an elliptical account of fragments given that in
some cases, short answers have different properties from full clausal answers, and from answers
containingVPellipsis.
(3) Whichmathematicsprofessorleftearly?
a. Jillleftearly,butshe’snotamathematicsprofessor.
b. Jilldid,butshe’snotamathematicsprofessor.
c. #Jill,butshe’snotamathematicsprofessor.
∗I’dliketothankmyadvisorsKyleJohnson, JeremyHartmanandEllenWoolfordforcommentsonthismaterial, aswell
asthreeanonymousNELSreviewersandtheparticipantsintheIdentityinEllipsisworkshopatLeidenUniversity,September
2013.Allerrorsaremine.
1
AndrewWeir FragmentanswersandtheQUD
UniversityofMassachusettsAmherst NELS44,UConn,October182013
• I propose that these data are in fact not problematic for the clausal ellipsis view of fragments if
we assume that the identity conditions on clausal ellipsis make reference to the Question under
Discussion.
2 Clausalellipsisaccountoffragments: Merchant2004
• InMerchant2004’sanalysisoffragments,afocusedconstituentraisestoaleft-peripheralposition.
Therestoftheclausethenelides.
• Syntactic implementation same as proposal for sluicing (Merchant 2001): the left-peripheral head
whichattractsthefragment(wh-wordinsluicing)toitsSpecisendowedwithan[E]-featurewhich
elidesitscomplement.1
(4) a. WhatdidJohneat? —Chicken.
b. FP
DP FP
Chicken F TP
[E]
DP TP
John T vP
atet
• This‘movement-plus-ellipsis’approachhasalsobeenadoptedforother‘remnant’cases,likewhy-
stripping (John ate chicken. Why chicken?) by Yoshida et al. 2013 (see also Weir to appear) and for
so-called ‘non-constituent coordination’ (John met with Mary on Tuesday and Bill on Wednesday) by
Sailor&Thoms2013
• Merchant2004amassesconsiderablesyntacticevidenceforthisapproachtofragments.
– Case connectivity: the case of a fragment answer is the same as the case which it would bear
in a full, non-elliptical utterance (parallels Ross 1969’s demonstration of the same facts for
sluicing).
1ThisisaslightoversimplificationofMerchant’ssyntaxforfragments. InfactMerchanthasthefragmentmoveagaintoa
positionhigherthantheSpecofthe[E]-bearinghead.ThisisnotgermaneheresoIabstractawayfromit.
2
AndrewWeir FragmentanswersandtheQUD
UniversityofMassachusettsAmherst NELS44,UConn,October182013
(5) Greek(Merchant’s(45,46))
a. Pjos idhetin Maria? —O Giannis. /*TonGianni.
who.NOMsaw theMaria —theGiannis.NOM/the Giannis.ACC
‘WhosawMaria? —Giannis.’
b. Pjon idhei Maria? —*O Giannis. /TonGianni.
who.ACCsaw theMaria? —theGiannis.NOM/the Giannis.ACC
‘WhodidMariasee? —Giannis.’
(6) German(Merchant’s(49,50))
a. Wem folgt Hans? —Dem Lehrer. /*Den Lehrer.
who.DATfollowsHans —the.DAT teacher/the.ACCteacher
‘WhoisHansfollowing? —Theteacher.’
b. Wen suchtHans? —*Dem Lehrer. /Den Lehrer.
who.ACCseeksHans —the.DAT teacher/the.ACCteacher
‘WhoisHanslookingfor? —Theteacher.’
– Movement-basedconstraints: onlyconstituentsthatareindependentlyobservedtomoveina
language can be fragment answers in that language. E.g. languages that require pied-piping
of prepositions in general systematically require this in fragment answers (the P-stranding
generalization;seealsoMerchant2001).
(7) (Merchant2004’s(72,78))
a. WithwhomwasPetertalking? —WithMary. /Mary.
b. Mit wem hat Annagesprochen? —Mit demHans. /*DemHans.
withwhomhasAnnaspoken withthe Hans /the Hans
‘WhodidAnnaspeakto? —Hans.’
• TheP-strandinggeneralizationisparticularlystrongevidenceforamovementstrategy(andhence
covertclausalstructure)intheseexamples: ‘todate, no-onehasevenhintedathowtoaccountfor
thesefactswithoutusingatheoryofpreposition-stranding,andno-onehaseverproposedatheory
ofpreposition-strandingthatdistinguishesGermanfromEnglishonanythingbutmorphosyntactic
grounds’(Merchant2010)
3
AndrewWeir FragmentanswersandtheQUD
UniversityofMassachusettsAmherst NELS44,UConn,October182013
3 Andyet...
3.1 Jacobson’schallenge
Jacobson 2013 points out that when the question contains a presupposition, short/fragment answers
havedifferentpropertiesfromfull/clausalanswers:
(8) (adaptedfromJacobson)
Whichmathematicsprofessorleftthepartyatmidnight?
a. Jillleftthepartyatmidnight...butshe’snotamathematicsprofessor.
b. Jilldid...(butshe’snotamathematicsprofessor.)
c. Jill...(#butshe’snotamathematicsprofessor.)
Thefragmentanswer(8c)commitsthespeakertotheviewthatJillisamathematicsprofessor, inaway
thatthefullanswer(8a)–andalsotheexamplewithVPellipsis,(8b)–doesnot.
Jacobson’s examples require a particular prosody (contrastive topic marking) in (8a,b), but the below
examples (due to Jeremy Hartman, p.c.) show that this is not the culprit for the badness of (8c); even
withfalling/focusprosody,thecontrastremains.
(9) WhichBronte¨ sisterwroteEmma?
a. JANE AUSTENwroteEmma(youfool).
b. JANE AUSTENdid(youfool).
c. #JANE AUSTEN(youfool).
The answer in (9c) is inappropriate because it commits the speaker to the view that Jane Austen was a
Bronte¨ sister. Note that again the VP ellipsis case is OK. Similar facts can be seen in cases where the
answerisaquantifier:
(10) Whichstudentsweredancinginthequad?
a. SomeGermansweredancinginthequad...(buttheyweren’tstudents).
b. SomeGermanswere...(buttheyweren’tstudents).
c. SomeGermans(#buttheyweren’tstudents).
The short answer in (10c) commits the speaker not just to the proposition that some Germans were
dancing,butthatsomeGermanstudentsweredancing.
Constraints that have been proposed for ellipsis, such as e-GIVENness (Merchant 2001 et seq.), do not
predict this behavior of fragment answers, or the difference between fragment answers and VP ellipsis
(ifwethinkthesameconstraintisinvolvedinboth).
4
AndrewWeir FragmentanswersandtheQUD
UniversityofMassachusettsAmherst NELS44,UConn,October182013
3.2 Jacobson’ssolution
Jacobson takes the contrast between the VP ellipsis cases and the short answer cases as indicative that
ellipsis is not involved in deriving the short answers. In Jacobson’s proposal, short answers directly
compose with the meaning of an antecedent question, understood as a (possibly partial) function from
entitiestopropositions:
(11) a. Whichstudentsweredancinginthequad?
= λx ∈ student : dancingInQuad(x)
b. SomeGermans. = λP∃x : german(x)&P(x)
c. Composition: ∃x ∈ student : german(x)&dancingInQuad(x)
Thisderivesthecontrastsabove,but:
• Jacobson’s proposal requires an antecedent to be found of the correct semantic and syntactic
category, that of a question (semantic type and syntactic category being tightly connected in the
categorial-type syntax of Jacobson’s Direct Compositionality approach). But not only questions
licensefragments: indefinitesandfocusedconstituentscantoo:
(12) a. Someoneleftearly. —Yeah,John.
b. JOHNleftearly. —No,Mary.
• Theabovesuggeststomethatweshouldmakeouraccountsensitivenotjusttoexplicitquestions,
asJacobsonsuggests,butalsotoimplicitquestions,suchastheQuestionunderDiscussion(Roberts
2012/1996).
• Inaddition,fragmentsshowsyntacticconnectivityeffects. Jacobsondoesproposewaysofderiving
binding connectivity effects, but it is less clear how a Direct Compositionality approach captures
theP-strandingcross-linguisticgeneralization,forexample.
4 Theconstraintonclausalellipsis
• Jacobson argues from the difference between VP ellipsis and short answers that ellipsis is not
involved,andthatproponentsoftheclausalellipsisapproachwouldhavetoclaim:
– (a)thatclausalellipsisworksdifferentlyfromVPellipsis;and
– (b)thattheantecedenceconditionsontheclausalellipsisinvolvedinfragmentanswers(and,
presumably, other forms of clausal ellipsis) makes reference to questions. (Intuitively, short
answershavetobe‘real’answerstothequestionposed.)
• Ibitethebullet: bothofthesearetrue.
5
AndrewWeir FragmentanswersandtheQUD
UniversityofMassachusettsAmherst NELS44,UConn,October182013
• Evidence has been building up that this is the case. In particular, the suggestion that (clausal)
ellipsis makes reference to questions in its antecedence conditions has been proposed for gapping
byReich2007andsluicingbyAnderBois2010.
• Iproposemyversionofthisquestion-basedconstraint,andshowhowitcapturestheproblematic
cases.
4.1 Questions
• I assume Groenendijk & Stokhof 1984’s view of questions as functions from worlds w to the
propositionwhichisthetrueanswertothequestioninw.
(13) Wholeft = λwλw(cid:48) : [λx : xleftinw] = [λy : y leftinw(cid:48)]
(cid:74) (cid:75)
(14) Toy example: in w only John left, in w only Mary left, in w both left, and in w no-one
0 1 2 3
left.
 
 (cid:104)(cid:104)ww0,,λλww(cid:48)(cid:48) ::JMoharnyleleftftininww(cid:48)(cid:48)aannddnnoo-o-onneeeelslseeleleftftininww(cid:48)(cid:105)(cid:48)(cid:105) 
1
Wholeft =
(cid:74) (cid:75)  (cid:104)(cid:104)ww23,,λλww(cid:48)(cid:48) ::JNooh-nonanedleMftainrywl(cid:48)e(cid:105)ftinw(cid:48)(cid:105) 
• Iwillsometimeswrite‘thatJohnleft’torefertotheproposition[λw(cid:48) :Johnleftinw(cid:48)]. Sowecould
rewrite(14)as:
 
 (cid:104)(cid:104)ww0,,tthhaattMJoharnyleleftftaannddnnoo-o-onneeeelslseeleleftf(cid:105)t(cid:105) 
1
(15) Wholeft =
(cid:74) (cid:75)  (cid:104)(cid:104)ww23,,tthhaattJnooh-onnaenldefMt(cid:105)aryleft(cid:105) 
• Thisfunctioncanbeseenas‘partitioning’thedomainofworlds: allthepossibleanswers(noneof
whichoverlap)tothequestionWholeftaregivenin(15),andthefunctionthatthequestiondenotes
takesaworldandmapsittothepropositionthatisthe(complete)trueansweratthatworld.
• The appropriate discourse move to respond to a question is to indicate which partition the actual
worldbelongsto,bygivingtheanswerthat(thespeakerbelieves)istrueintheactualworld.
(16) Wholeft? —Johnleft.
• Expressing that ‘John left’ is the answer tells your interlocutor that the actual world is in one of
thosepartitionsofworldsinwhich‘Johnleft’istrue(inourtoymodel,w orw ).
0 2
6
AndrewWeir FragmentanswersandtheQUD
UniversityofMassachusettsAmherst NELS44,UConn,October182013
• Sometimes the actual response given is not one that is in the Groenendijk and Stokhof denotation
ofthequestion,butpragmaticreasoningwillneverthelessletusconcludefromtheresponsewhich
partitiontheactualworldbelongsto.
(17) WhichBronte¨ sisterwroteEmma? —JaneAustenwroteEmma(youfool).
• ThequestionWhichBronte¨ sisterwroteEmmarepresentsthefollowingpartition:2
(18) WhichBronte¨ sisterwroteEmma
(cid:74)  (cid:75) 
 (cid:104)(cid:104)ww0,,tthhaattnCohaBrrloontttee¨wsirsoteterwEmromteaE(cid:105)mma(cid:105) 
1
=
 (cid:104)(cid:104)ww23,,tthhaattEAmnnileywwrrootteeEEmmmmaa(cid:105)(cid:105) 
• ‘That Jane Austen wrote Emma’ is not in this partition. However, the response Jane Austen wrote
Emma lets us conclude that the answer to the question (in the above technical sense) is ‘that no
Bronte¨ sisterwroteEmma’;such‘indirect’responsescanthereforestillbeunderstoodasgivingan
answertothequestion.
4.2 Questionsandshortanswers
• How will this help the problem with short answers? Well, consider the question Which math
professorleftearly?. Assume,forsimplicity,that(inallworlds)JohnandMaryaremathprofessors,
andno-oneelseis. Thenthisquestionhasthedenotation:
(19) Whichmathprofessorleftearly =
(cid:74) (cid:75) 
 (cid:104)(cid:104)ww0,,tthhaattJMoharnyleleftfteeaarrlylyaannddnnooooththeerrmmaaththpprroofefessssoorrleleftfteeaarrlyly(cid:105)(cid:105) 
1
 (cid:104)(cid:104)ww23,,tthhaattJnoohmnaatnhdpMroafreyssloerftleeaftrleya(cid:105)rly(cid:105) 
• Notably, Jill appears in no answer to the question Which math professor left, even if she did in fact
leave,becauseJillisnotamathprofessor.
• We can use this to encode Jacobson’s intuition that short answers have to be ‘real’ answers to the
question,i.e. thatthebelowisinfelicitous.
(20) Whichmathprofessorleftearly? —#Jill,butsheisn’tamathprofessor.
2Iignoreworldsinwhichtwoormoreofthesisterscollaborated.
7
AndrewWeir FragmentanswersandtheQUD
UniversityofMassachusettsAmherst NELS44,UConn,October182013
(21) Conditiononclausalellipsis
GivenaclauseE(interpretedataworldw(cid:48))whichdenotesapropositionp,andaQuestion
underDiscussionQUD (understoodasafunctionfromworldstopropositions):
EllipsisofEislicensediffthetrueanswertotheQUDattheworldofevaluationentails E ,
(cid:74) (cid:75)
i.e. QUD(w(cid:48)) ⇒ p.
• We can encode this as a presupposition that has to be met if the ellipsis-licensing [E ] feature is
C
used. (The is for ‘clausal’, to remind us that this feature does not participate in e.g. VP ellipsis,
C
butonlyclausalellipsis.)
(22) Presuppositionaldenotationfor(clausal)E-feature:
[E ] QUD = λpλw : p(w),iffQUD(w) ⇒ p;otherwiseundefined.
C
(cid:74) (cid:75)
• I assume focus reconstruction. I assume that the movement of the focused constituent is
fundamentally forced by the need for focused material to escape the domain of ellipsis (Yoshida
etal.2013).3
(23) WhatdidJohneat? —Thecake.
a. [ ThecakeFoc [ Johnatet]]
FocP [EC] TP
b. Afterreconstruction: [ Foc [ Johnatethecake]]
FocP [EC] TP
(24) a. QUD = λwλw(cid:48) : [λx :Johnatexinw] = [λy :Johnatey inw(cid:48)]
 
 (cid:104)(cid:104)ww0,,tthhaattJJoohhnnaatteetthheeccahkicekaenndanndotnhointhgienlgseel(cid:105)se(cid:105) 
1
=
 (cid:104)(cid:104)ww23,,tthhaattJJoohhnnaatteetnhoethciankge(cid:105)andthechicken(cid:105) 
b. QUD(w ) = λw(cid:48) :Johnatethecakeandnothingelseinw(cid:48)
0
c. Johnatethecake QUD = λw :Johnatethecakeinw
(cid:74) (cid:75)
d. [[E ]Johnatethecake] QUD = λw :Johnatethecakeinw
C
(cid:74) (cid:75)
Presupposition: QUD(w) ⇒ Johnatethecake
(cid:74) (cid:75)
• The presupposition that [E ] introduces is that the true answer to the question at the world of
C
evaluationmustentailthemeaningoftheelidedclause(‘Johnatethecake’),whichistriviallytrue
here.
• Thisexampleissomewhattrivialbutlet’sseewhereitgetsuswiththemoreproblematicexamples.
3Iremainneutralabouthowwepickthematerialthatbecomesfocusedinthefirstplace,assumingthatthiswillfollowfrom
somethinglikeSchwarzschild1999’sproposalofGIVENness.
8
AndrewWeir FragmentanswersandtheQUD
UniversityofMassachusettsAmherst NELS44,UConn,October182013
4.3 HandlingtheJacobsoncases
• TheproblematicJacobsonexample:
(25) Whichmathematicsprofessorleftearly?
a. Jilldid(butJillisn’tamathprofessor).
b. #Jill(butJillisn’tamathprofessor).
(26) QUD = λwλw(cid:48) : [λx ∈ mathProf : xleftearlyinw] = [λy ∈ mathProf : y leftearlyinw(cid:48)]
• Inourtoymodel, onlyJohnandMaryaremathprofessors(andthisisconstantacrossworlds), so
thequestion’sdenotationis:
(27) Whichmathprofessorleftearly =
(cid:74) (cid:75) 
 (cid:104)(cid:104)ww0,,tthhaattJMoharnyleleftfteeaarrlylyaannddnnooooththeerrmmaaththpprroofefessssoorrleleftfteeaarrlyly(cid:105)(cid:105) 
1
 (cid:104)(cid:104)ww32,,tthhaattnJoohmnaatnhdpMroafreyssloerftleeaftrleyar(cid:105)ly(cid:105) 
• Nowlet’sseewhatourclausalellipsisfeaturemakesofthis:
(28) [E ]Jillleftthepartyearly QUD =λw :Jillleftthepartyearlyinw
C
(cid:74) (cid:75)
Presupposition: QUD(w) ⇒ [λw(cid:48) :Jillleftthepartyearlyinw(cid:48)]
• The answer to the question in w , that is QUD(w ), is ‘that John left early and no other math
0 0
professorleftearly’. Thisclearlydoesnotentail‘thatJillleftthepartyearly’. Sothepresupposition
in(28)isnotmet: andtheanswerisinfelicitous.
• ThisalsohandlestheBronte¨ sistercases:
(29) WhichBronte¨ sisterwroteEmma?
a. JaneAustendid(youfool).
b. #JaneAusten(youfool).
(30) QUD = λwλw(cid:48) : [λx ∈ Bronte : xwroteEmmainw] = [λy ∈ Bronte : y wroteEmmainw(cid:48)]
(31) [[E ]JaneAustenwroteEmma] QUD =λw :JaneAustenwroteEmmainw
C
(cid:74) (cid:75)
Presupposition: QUD(w) ⇒ [λw(cid:48) :JaneAustenwroteEmmainw(cid:48)]
• QUD(w ) = ‘that no Bronte¨ sister wrote Emma’. This does not entail ‘that Jane Austen wrote
0
Emma’,sothepresuppositionof(31)isnotmet.
9
AndrewWeir FragmentanswersandtheQUD
UniversityofMassachusettsAmherst NELS44,UConn,October182013
• Thebelowcaseismoreinteresting.
(32) Whichstudentsleftearly? —SomeGermansleftearly.
(33) QUD = λwλw(cid:48) : [λx ∈ student : xleftearlyinw] = [λy ∈ student : y leftearlyinw(cid:48)]
 
 (cid:104)(cid:104)ww0,,λλww(cid:48)(cid:48) ::ss11+lesf2tlienftwin(cid:48) awn(cid:48)d(cid:105)nootherstudentleftinw(cid:48)(cid:105) 
1
=
 (cid:104)(cid:104)ww23,,λλww(cid:48)(cid:48) ::sn2olsetfutdinenwt(cid:48)laenftdinnow(cid:48)o(cid:105)therstudentleftinw(cid:48)(cid:105) 
(34) a. [SomeGermans[E ][leftearly]]
C
afterreconstruction: [[E ]SomeGermansleftearly]
C
b. [E ]SomeGermansleftearly QUD =λw :someGermansleftearlyinw
C
(cid:74) (cid:75)
Presupposition: QUD(w) ⇒ [λw(cid:48) :someGermansleftearlyinw(cid:48)]
• For(34)tobefelicitous, weneedthepropositionQUD(w ), i.e. ‘thats1+s2leftearly’, toentailthe
0
meaningoftheelidedclause,‘thatsomeGermansleftearly’.
• This entailment doesn’t go through on its own, but it would if we ensured that in the domain of
worldsthatweareconsideringinthecontext,s1ands2areGermans.
• I propose that this restriction on the contextually relevant domain of worlds is accommodated
on hearing the answer in (34), to make the entailment go through. So (34) is understood as
presupposing that s1 and s2 left and that s1 and s2 are Germans (in all the worlds that are
consideredcontextuallyrelevant).
4.4 Fragmentslicensedbynon-explicitquestions
• Fragmentsarenotjustlicensedbyexplicitquestions/interrogatives: focusalsolicensesfragments.
(35) a. JohnlikesCHICKEN. —No,beef./Beef,too.
b. DoesJohnlikeCHICKEN? —No,beef./Yes. Beef,too.
• ThatfocusplacementaffectswhattheQuestionunderDiscussionis(forexample,in(35),theQUD
isWhatdoesJohnlike?) iswellknown,startingfromRoberts2012/1996.
• Theconditionproposedhereonclausalellipsispredictsthatfragmentsshouldbelicensedbyfocus;
theimplicitQUDbroughtaboutprovidesthepresuppositionforthe[E ]-feature.
C
10