West Coast Conference on Formal Linguistics (WCCFL) (38th : 2020)

Scrapping clauses : an anaphor based approach Grishin, Peter 2020-03-06

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata


73804-Grishin_Peter_Scrapping_WCCFL38_2020_handout.pdf [ 115.32kB ]
JSON: 73804-1.0389853.json
JSON-LD: 73804-1.0389853-ld.json
RDF/XML (Pretty): 73804-1.0389853-rdf.xml
RDF/JSON: 73804-1.0389853-rdf.json
Turtle: 73804-1.0389853-turtle.txt
N-Triples: 73804-1.0389853-rdf-ntriples.txt
Original Record: 73804-1.0389853-source.json
Full Text

Full Text

Scrapping clauses: an anaphor based approach∗Peter GrishinWCCFL38 March 6, 20201 Introduction˛ Consider the following English sentences:1(1) a. This building is bigger than I thought ∆!b. It started raining at the exact hour they predicted ∆.c. The shipwreck is located where the archaeologist claimed ∆.d. The queen arrived after we expected ∆.§ Here, wehave nonpronunciation of clausal complements to clause embedding verbs:think, predict, claim, and expect.§ It’s possible (for some speakers) to do this in (at least) comparatives (1a), relativeclauses (1b-1c), and temporal adjuncts (1d).˛ These sentences look like Null Complement Anaphora (NCA, Hankamer and Sag 1976,Depiante 2000, 2001); compare (2):(2) a. Sohla said that Chris made these sandwiches, but I’m not sure ∆.b. Gaby asked Brad to clean the counter, but he refused ∆.c. Priya says green peppers are delicious, and I agree ∆.˛ Despite this surface similarity, I’ll argue that (1) and (2) are genuinely different con-structions with different structures.∗Many thanks go to Patrick Elliott, Danny Fox, Sabine Iatridou, David Pesetsky, Roger Schwarzschild,and anonymous reviewers forWCCFL38 andGLOW43 for their valuable feedback and comments, andmanythanks are also due to the many native English speakers I bothered with some very difficult judgments. Asalways, any errors are mine.1A quick disclaimer before I proceed further: I’ve found that there’s quite a lot of idiolectal variationhere. The judgments presented in this paper aremine, and they have been cross-checkedwith native Englishspeakers who accept the baseline sentences in (1).1§ I’ll be calling the construction in (1) scrapping—Sentential Complement Reductionin ACD Positions—for reasons that will become apparent. The gap left behind isscrapping is a scrap.˛ Scrapping seems to have largely escaped the ellipsis literature, to the best of my knowl-edge, with the notable exceptions of Bresnan and Grimshaw (1978) (under the name“pseudo-free relatives”), Napoli (1983) (though she assumes it’s NCA), Wold (1992),Moltmann (1992, 1993) (in comparatives under the name “clausal comparative dele-tion”), Kennedy and Merchant (2000a), and Kennedy and Merchant (2000b).2§ Bresnan and Grimshaw (1978) offer a cursory analysis of scrapping in free relatives,suggesting that the embedding verb is “syntactically intransitive” and the relativehead is base-generated high.§ Moltmann (1992, 1993) discusses “clausal comparative deletion” as part of her ar-gument that comparative deletion involves LF copying rather than PF deletion.§ Kennedy and Merchant (2000b) look at scrapping in comparatives, and argue thatthe gap is a DP whose semantic value is “a propositional expression that contains afree variable over degrees (or amounts)”.§ I’ll argue that none of these analyses is quite sufficient to capture the full range offacts, though my analysis will share certain core similarities with Moltmann (1993)and Kennedy and Merchant (2000b).˛ The proposal: the gap contains a lowMod(al)P-sized anaphor (in the binding-theoreticsense) that must be c-commanded by an antecedent ModP, and an operator (e.g. a de-gree, relative clause, or temporal operator) must be adjoined to it: [Op PROModP].§ I’ll assume that PROModP gets its denotation from its c-commanding antecedent byLF-copying (Moltmann 1992, 1993).3§ I’ll be assuming that this lowModP can host root modals, and also crucially that it’sfirst merged below negation (e.g. Iatridou and Zeijlstra 2013, a.o.).§ Scrapping thus provides further evidence for the existence of anaphoric routes toellipsis/surface anaphora (e.g. Chao 1987, Hardt 1999, Schwarz 2000, a.o. in thedomain of VP ellipsis).2As far as I’m aware, this is an exhaustive list! Let me know if there’s any literature I’m missing.3Though I don’t think anything crucial hinges upon this, as I’ll discuss later. If you’rewilling to play a fewcompositional tricks, you can get PROModP to receive its denotation via regular semantic binding (thoughit needs to be binding of a predicate variable). It’s worth noting that, if you adopt the LF copying approach,scrapping would be an instance where syntactic and semantic binding come apart.2Roadmap:1. Scrapping is not NCA2. Scraps must be contained within their antecedents3. Scraps are ModP sized4. Scraps must be c-commanded by their antecedents at LF5. Enter PROModP6. Conclusion2 Scrapping is not NCA˛ While they look similar on the surface, scrapping cannot be reduced to a special case ofNCA. They display certain irreconcilable differences:§ Scrapping and NCA are licensed by different sets of predicates (Moltmann 1992,Kennedy and Merchant 2000b).§ Scrapping requires moving an operator out of the gap, whereas NCA bans anymovement out of its gap (Depiante 2000, 2001).§ Scrapping can’t take its antecedent from the discourse context, but NCA can. Fol-lowing Hankamer and Sag (1976), scrapping is surface anaphora and NCA is deepanaphora.2.1 Licensing˛ The set of scrapping predicates is not equal to the set of NCA predicates, nor is one asubset of the other.˛ Scrapping: yes; NCA: no.§ claim, predict, expect, think:(3) a. This building is bigger than I thought ∆! Scrapping: yesb. It started raining at the exact hour they predicted ∆.c. The shipwreck is located where the archaeologist claimed ∆.d. The queen arrived after we expected ∆.3(4) Amiel baked a cake… NCA: noa. *…and Rhoda thought ∆.b. *…and Rhoda predicted ∆.c. *…and Rhoda claimed ∆.d. *…and Rhoda expected ∆.˛ Scrapping: yes; NCA: yes.§ guess, agree, insist:(5) a. Jade left exactly when we agreed ⟨she should leave t⟩. Scrapping: yesb. Julia got married exactly where we guessed ⟨she would get married t⟩.c. The queen arrived after we insisted ⟨she should arrive t⟩.(6) a. You said Jade should leave, and I agree ∆. NCA: yesb. Q: Did Julia get married? A: I guess ∆.c. Have some cake—I insist ∆!˛ Scrapping: no; NCA: yes.§ disagree, try, be hopeful:(7) a. *Jade left when we tried ⟨to leave t⟩. Scrapping: nob. *Julia got married exactly where we disagreed ⟨she would get married t⟩.c. *The queen arrived after we were hopeful ⟨she would arrive t⟩.(8) a. Joe thinks Jade should leave, but I disagree ∆. NCA: yesb. Carla isn’t sure that this recipe will work, but I’ll try ∆.c. Lucy doesn’t think the basketball team will win, but I’m hopeful ∆.˛ These different classes of verbs would be difficult to account for under an analysis thatreduces scrapping and NCA to the same source.42.2 Extraction˛ Scrapping and NCA differ in their extraction properties: scrapping is characterized bymovement of an operator out of the gap, but NCA is known to disallow extraction (De-piante 2000, 2001).4What makes something a scrapping or an NCA predicate is an interesting question, and one I do nothave any firm answers for. I suspect that it’s purely idiosyncratic selectional properties of certain lexicalitems that makes them NCA predicates, but I suspect there is more to say about scrapping predicatez. Atthe very least, there seem to be certain patterns: for instance, negative predicates (like disagree and doubt),aspectual and implicative predicates (like try, manage, begin), and adjectival predicates (like be certain andbe hopeful) all seem to not license scrapping.4§ NCA disallows whmovement (9a), Quantifier Raising (QR; 9b), and clitic climbing(9c) (Depiante 2000, Depiante 2001).(9) a. *Roger agreed to review the movie, but I don’t remember when he agreed∆.b. A doctor volunteered to visit every patient, and a nurse also volunteered.∃ » ∀, *∀ » ∃c. * JuanJuanlasthem.fquierewantsver,to.seeyandMaríaMaríatambiénalsolasthem.fquierewants∆.‘Juan wants to see them, and María also wants (to see them).’d. JuanJuanquierewantsverlas,to.see.them.FyandMaríaMaríatambiénalsoquierewants∆.‘Juan wants to see them, and María also wants (to see them).’§ On the other hand, scrapping allows (10) and even requires (11) extraction out ofthe gap.(10) a. This building is bigger Op than I thought ⟨it was t-big⟩!b. It started raining at the hour Op they predicted ⟨it would rain t⟩.c. The shipwreck is located where she claimed ⟨it was located t⟩.d. The queen arrived before Opwe expected ⟨she would arrive t⟩.(11) a. *It started raining at the hour Op they predicted t ⟨it would rain⟩.b. *The shipwreck is located where she claimed t ⟨it was located⟩.c. *The queen arrived before Opwe expected t ⟨she would arrive⟩.‚ Wemight expect to get readings in (11) that relate the rain-timewith the prediction-time (11a), the shipwreck location with the claiming-location (11b), and thearrival-time with the expectation-time (11c)—but these are unavailable read-ings.‚ Weonly get the the readings associatedwith operatormovement out of the scrap(10).˛ Scrapping obligatorily involves the movement of some kind of operator out of the gap,and NCA obligatorily disallows such movements. This poses a severe challenge to anyunified analysis of scrapping and NCA.52.3 Discourse antecedents˛ NCA, being a kind of deep anaphora (Hankamer and Sag 1976), is able to be licensedby the discourse context:(12) Context: You are in a supermarket in the cereal aisle, and see a small child pushing boxesof cereal onto the floor. You utter:a. I disapprove ∆.(13) Context: You are in a supermarket in the cereal aisle, and try to pick up five cereal boxesat once. You manage briefly for a while, before they all come tumbling to the ground. Youutter:a. Well, at least I tried ∆.˛ Scrapping, in contrast, doesn’t seem to be able to be licensed by the discourse context.§ If we construct a context where the discourse makes salient a set of propositionscontaining a free variable of the correct type in the correct place, we might expect tolicense scrapping. Evidently, this is impossible.(14) Context: Alex is planning on attending a birthday party, but he doesn’t like mostof the people on the guest list. However, he knows that Claire is baking the birthdaycake, and he knows that Claire is an excellent baker. So, he has a plan: he wants toarrive right when the cake gets brought out, in order to minimize the amount of timespent at the party while still being able to get cake before it all gets eaten. You’re notsure when exactly the cake will be brought out, but you know that Alex has a guessas to when that’ll be, and that he’ll arrive at that time.a. #Alex will arrive when he predicts ⟨the cake will be brought out t⟩.§ Here, I’ve tried to make salient the set of propositions of the form the cake will bebrought out at t, but this isn’t enought to license scrapping.§ It seems like we need an overt linguistic antecedent, suggesting that scrapping issurface anaphora, rather than deep anaphora.˛ If NCA and scrapping are given a unified analysis, it’s unclear how we can derive thisdifference.3 The ACD generalization˛ Scrapping is subject to anAntecdent-Contained Deletion generalization (ACD): scrapsmustobligatorily be contained within their antecedents.˛ Assuming that the antecedent for the scrap is a clause:6(15) a. A[John got married exactly where he claimed E[⟨that he got married⟩]].b. A[Bill left exactly when we said E[⟨that he would leave⟩]].c. A[Mary arrived after we predicted E[⟨that she would arrive⟩]].d. A[John read fewer books than he claimed E[⟨he would read⟩]].(Wold 1992)§ I’ll show that scraps behave as if they’re inside vP/VP, using various vP/VP con-tituency tests.§ I’ll show that constructing non-ACD examples of scrapping results in ungrammati-cality, even when closely controlling for syntactic and semantic parallelism.3.1 Constituency tests˛ VP preposing, VP ellipsis (VPE), and do-so replacement all show that scraps are (or atleast can be) adjoined somewhere around vP:(16) VP preposinga. John said he would get married exactly where he claimed ∆, and [vP get mar-ried exactly where he claimed ∆] he did.b. Bill said he would leave exactly when we said ∆, and [vP leave exactly whenwe said ∆] he did.c. Mary said she would arrive after we predicted ∆, and [vP arrive after we pre-dicted ∆] she did.d. Mary said that John read fewer books than he claimed ∆, and [vP read fewerbooks that he claimed ∆] John did.(17) VPEa. John got married exactly where he claimed ∆, and Bill did [vP ∆] too.b. Bill left exactly when we said ∆, and John did [vP ∆] too.c. Mary arrived after we predicted ∆, and Jill did [vP ∆] too.d. John read fewer books than he claimed ∆, and Bill did [vP ∆] too.(18) do-so replacementa. John got married exactly where he claimed ∆, and Bill did so too.b. Bill left exactly when we said ∆, and John did so too.c. Mary arrived after we predicted ∆, and Jill did so too.d. John read fewer books than he claimed ∆, and Bill did so too.§ Assuming that the antecedent for scraps is larger than vP, this forces scraps to beinside their antecedents—a case of antecedent-containment.7§ Eventually, I’ll argue that scraps are actually (low)ModP sized, and take a ModP astheir antecedent. Assuming that ModP dominates vP, and assuming that these testsdiagnose vP-hood, my analysis preserves the notion that scrapping is only licensedunder ACD.3.2 Non-ACD scrapping is ungrammatical˛ We can take the scrap out of its antecedent—for instance, by putting the antecedent andthe scrap in different conjuncts of a conjunction—and the result is ungrammatical.(19) a. *I know that A[John got married somewhere], but where did he claim E⟨he gotmarried t⟩?b. *I know that A[Bill left sometime during the dinner], but when did you sayE⟨he left t⟩?c. *I know that A[Mary read some number of books], but how many did sheclaim E⟨she read t⟩?§ Note that here I’ve tried to configure the antecedent and the ellipsis site in a waythat mimics sluicing, with the wh expression matching with an indefinite correlatein the antecedent.§ This isn’t enough to license scrapping.˛ But what if this is telling us something about the parallelism conditions on scrapping?Maybe scrapping is subject to different, perhaps stricter parallelism conditions thansluicing is?˛ However, no amount of isomorphism will save non-ACD scrapping.§ We can try to ensure perfect isomorphism using ATB movement, and we can seethat VPE and scrapping behave differently:(20) ATB with VPE good, ATB with scrapping bada. Howi did [Jo hope I would cook the potatoes ti] and [Lily fear I would⟨cook the potatoes ti⟩]?b. *Howi did [Jo hope I would cook the potatoes ti] and [Lily fear ⟨I wouldcook the potatoes ti⟩]?§ We can try to ensure perfect isomorphism by having the antecedent be an embeddedquestion with a trace in the correct position, and we can see that VPE and scrappingbehave differently:8(21) I wonder wheni Harry left for the theater ti.a. Well, I asked Sally wheni she claimed he did ⟨leave for the theater ti⟩, butshe didn’t say. VPEb. *Well, I asked Sally wheni she claimed ⟨Harry left for the theater ti⟩, butshe didn’t say. Scrapping˛ In sum: no amount of parallelism can save non-ACD scrapping. Scraps must be con-tained within their antecedents.4 Scraps are small˛ Scraps are small: I argue that they’re the size of a lowModP that’s base generated belownegation.§ Scraps show similar temporal orientation properties as infinitival clauses, whichsuggests they’re smaller than TP (following Wurmbrand 2014, a.o.).§ Scraps cannot contain sentential negation, suggesting they’re smaller than NegP.§ Scraps can contain root modals, suggesting they’re at least ModP-sized.(22) The size of the scrapCPC TPT NegPNeg ModPMod vPv VP4.1 Temporal orientation˛ Someverbs, like expect, showdifferent temporal orientation propertieswhen they embedinfinitives versus fully finite clauses. For expect, there’s a future orientation requirementwith infinitives that doesn’t exist for finite clauses.9(23) Obligatory future orientation with infinitive complementsa. I expect him to run tomorrow.b. *I expect him to run yesterday.(24) No obligatory future orientation with finite complementsa. I expect that he will run tomorrow.b. I expect that he ran yesterday.§ Wurmbrand (2014) argues that these particular kinds of infinitives (“future irrealisinfinitive”) are tenseless, both syntactically (i.e. being smaller than TP) and seman-tically (i.e. not containing an independent tense specification).˛ How can scraps be interpreted under expect? It turns out that the show the same futureorientation requirement:(25) Obligatory future orientation with scrappinga. Harry will arrive earlier than Sally expects ⟨he will arrive⟩.b. *Harry arrived earlier than Sally expects ⟨he arrives⟩.§ If Wurmbrand is right, then this means that scraps are smaller than TP.4.2 Negation˛ We need to be wary of negative/inner islands. Here’s a baseline example that isn’t sub-ject to inner islands:(26) Context: Tanya’s friend Anton is planning a move. Tanya knows that Anton is a cheap-skate, so she expects that he will not move to Boston or Cambridge, because rent is ratherexpensive in those cities. She predicts he will move to Medford. However, unexpectedly,it turns out that Anton moved to Boston. So…a. Anton moved to a city wh Tanya didn’t expect he would ⟨move to t⟩.b. ?Anton moved to a city wh Tanya expected he wouldn’t ⟨move to t⟩.c. Anton moved to a city wh Tanya didn’t expect ⟨he would move to t⟩.§ The relative clause operator canmove acrossmatrix negation (26a), it canmarginallymove across embedded negation (26b), and it can move over matrix negation out ofa scrap (26c).§ So let’s try the logical fourth option: can we move over embedded negation, out of ascrap?10(27) Context: Tanya’s friend Anton is planning a move. Tanya knows that Anton is acheapskate, so she expects that he will not move to Boston or Cambridge, becauserent is rather expensive in those cities. She predicts he will move to Medford. And,exactly as Tanya predicted, it turns out that Anton didn’t move to either Boston orCambridge, but rather to Medford. So…a. ?Anton didn’t move to a city wh Tanya expected he wouldn’t ⟨move to t⟩.b. #Anton didn’t move to a city wh (that) Tanya expected ⟨he wouldn’t moveto t⟩.‚ Weneed to change the sentence a bit, to ensure that the antecedent has negation,and we also need to change the context to accommodate that.‚ Once we do so, the contrast between (27a) and (27b) shows that negation isunavailable inside the scrap.5˛ So it seems that scraps are not only smaller than TP, but also smaller than NegP.4.2.1 Constituent negation˛ Note that my analysis only predicts that sentential negation—that is, negation that occu-pies NegP on the clausal spine—can’t be inside a scrap, since a scrap is too small.˛ But what about constituent negation?˛ Interestingly, it seems like constituent negation is in fact able to be interpreted inside ascrap, in striking contrast to sentential negation!(28) Context: Your friend Mary has had digestive issues for several years now, and she finallywent to the doctor to figure out what’s going on. The doctor recommended that she not eatwheat, dairy, and beans. You know that Mary always follows the doctor’s orders, so youexpect that she’s going to stop eating those foods. And later, sure enough…a. #Mary has not been eating the foods I expected ∆.b. Mary has been not eating the foods I expected ∆.§ By the position of not, we can tell whether it’s sentential negation (28a) that followsthe first auxiliary that has raised to T, or constituent negation (28b) that directlymerges with and negates the VP.§ In this context, the sentential negation sentence (28a) is infelicitous, precisely be-cause the only reading available is one without negation in the scrap: Mary has notbeen eating the foods I expected she would eat.5It’s true that the sentence in (27a) is degraded. However, all of my consultants have agreed that (27b)is significantly worse in this context.11§ In contrast, the constituent negation sentence is acceptable (28b), showing that theunavailability of sentential negation in scraps really is due to the size of the scrap(i.e. it being smaller than NegP), rather than some independent constraint that bansnegation altogether inside the scrap.4.3 Root modals˛ Scraps are, however, big enough to contain root modals.˛ You can get deontic necessity modals:(29) Recently, the department changed the requirements for getting a PhD. Before, the require-ment was to write two qualifying papers. Now, the requirement is only to write one.However, the students haven’t realized this yet, and they still think they need to writetwo. One of the professors, noting this fact, says to another:a. The students have to write fewer generals papers than they think ∆.§ The interpretation we get here can be paraphrased as follows: the students thinkthat, in view of their obligations as PhD students, they must write n-many generalspapers; however, they actually must write fewer than n-many generals papers.‚ The number of papers they must write in the actual world is less than the num-ber of papers they must write in their belief worlds.§ We can express this reading with a simple maximality semantics for comparativeoperators as follows:(30) a. max(λn.□write(n-many papers)(s)) <max(λm.think(s)(□write(m-many papers)(s)))b. The maximal number n such that the students must write n-many papers is lessthan the maximal number m such that the students think they must write m-many papers.‚ Note that not only is there a modal interpreted inside the scrap, but also theDegP than they think scopes above matrix have to.6‚ This is expected: if the scrap contains a modal, then scoping it under the matrixmodalwould result in the scrap being irreconcilably antecedent-contained at LF.6Or else we’d derive a funny, nonexistent reading where there’s some sadistic requirement that the stu-dents always be wrong about however many papers they’re writing/they’ll write:(i) (a) □[max(λn.write(n-many papers)(s)) <max(λm.think(s)(write(m-many papers)(s)))](b) It is necessary that: the maximal number n such that the students write n-many papers beless than the maximal numberm such that the students think they writem-many papers.12‚ We need to QR the constituent containing the scrap out of the matrix ModPin order to escape antecedent-containment, so the DegP will inevitably end upscoping above the matrix modal.˛ You can get deontic possibility modals:(31) There’s a very exclusiveVIP party happening at a club, with very strict entry times. Mollythinks they’re only letting people in at 10:00, 10:15, and 10:30. However, unbeknownstto her, they’re also letting people in at 10:45.a. Molly can get in 15 minutes later than she thinks ∆.˛ You can get circumstantial possibility/ability modals:(32) Colin has a low opinion of his candy-eating abilities, and thinks he can only eat a fewcandies in a single sitting. However, I have greater faith in his candy-eating abilities, andthink he can eat many candies in a single sitting. I say to him:a. You can eat more candies than you think ∆!˛ So scraps have to be big enough to contain root modals. Taking all these facts into ac-count, I conclude that scraps are low ModPs.5 The C-command generalization˛ At LF, scrapsmust be c-commanded by their antecedents. In order to escape antecedent-containment, the constituent containing the scrap QRs to adjoin to the antecedent, andgoes no further.§ I’ll show that scraps cannot scope above sentential negation—QRing above senten-tial negation would result in the scrap no longer being c-commanded by its an-tecedent ModP.§ I’ll show that de re and de dicto readings for scraps correlate with whether the scrap’santecedent is large (thematrix clause) or small (the embedded clause). This directlyfalls out from the c-command generalization, given a scope theory of de re/de dicto.5.1 Negation (again)˛ The c-command generalization would predict that only one of the following LF config-urations will be possible:13(33) NegPNeg ModPXP…ModPS …ModPAMod vP… tXP …(34) NegPXP…ModPS …NegPNeg ModPAMod vP… tXP …§ I’ve underlined the scrap and boxed the antecedent ModP.§ In (33), ModPA c-commands the scrap; in (34), it doesn’t. The c-command general-ization predicts that only (33) should be possible.§ In other words, sentential negation must scope over XP.˛ This prediction is verified. Let’s look at what happens when XP is an existential.7§ Let’s go back to Tanya and Anton and look at the sentence Anton didn’t move to a cityTanya expected.§ ¬≫ ∃ is available:(35) Context: Tanya’s friend Anton is planning a move. Tanya knows that Anton is acheapskate, so she expects that hewill move to either Brighton,Medford, or Somerville,since rent is relatively cheaper in those cities. Later, it so turns out that Anton endedup moving to Watertown. So…a. Anton didn’t move to a city Tanya expected ⟨he would move to⟩.b. ¬≫ ∃§ But ∃ ≫ ¬ is not!7There are problems with this, given the empirical possibility of exceptional wide-scope indefinites andthe theoretical possibility of choice functional analyses (Fodor and Sag 1982, Abusch 1993, Kratzer 1998,Matthewson 1999, Schwarzschild 2002, a.m.o.). For the purposes of this paper, I’ll be assuming that exis-tential quantifiers headed by a(n) take scope by QR, and we’ll see that indeed, they will not be able to scopeabove negation in scrapping. This issue deserves much more thought than I have room to devote here. Un-fortunately, it won’t be very helpful to test universal quantifiers, since ∀ ≫ ¬ entails ¬ ≫ ∀, and thus lowscope of the universal is compatible with high scope contexts.14(36) Context: Tanya’s friend Anton is planning a move. Tanya knows that Anton is acheapskate, so she expects that hewill move to either Brighton,Medford, or Somerville,since rent is relatively cheaper in those cities. Later, it so turns out that Anton endedup moving to Medford. So…a. #Anton didn’t move to a city Tanya expected ⟨he would move to⟩ (for in-stance, Brighton).b. *∃ ≫ ¬§ Note, crucially, that both readings are available with VPE, so this really is aboutscrapping and not independent restrictions on the original sentence or on ellipsis/surface anaphora:(37) Anton didn’t move to a city Tanya expected he would.a. ✓¬≫ ∃b. ✓∃ ≫ ¬˛ The c-command generalization makes the correct prediction here, assuming that theexistential takes scope via QR.5.2 De re/de dicto˛ The c-command generalization acts as a “wall” for how high an expression can QR inscrapping.˛ We find similar “wall” effects in the domain of intensional predicates, with respect tothe de re/de dicto distinction. I’ll be assuming a simple scope theory of de re/de dicto.˛ The conclusion: there’s a direct correlation between the size of the antecedent (eitherbeing both the intensional predicate and the embedded clause, or just the embeddedclause), and the height of QR (QR above the intensional predicate, or QR within theintensional predicate) which results in either de re or de dicto readings.(38) low antecedent high antecedentlow QR ✓ ∗high QR ∗ ✓§ A low antecedent forces low QR and thus a de dicto reading, and a high antecedentforces high QR and thus a de re reading.8˛ Low antecedent, low QR: available8A similar pattern has been observed in the domain of stripping in temporal adverbials—sentences likeWhiskers woke up before Fido—by Overfelt (2019). It remains to be seen whether his cases can be subsumedunder my analysis of scrapping, or vice-versa.15(39) Context: Carla has the mistaken belief that Andy thinks I will arrive late today. Un-beknownst to Carla, Andy knows that I’m a very punctual person. As he (actually)expected, I arrived on time.a. Carla believes that I arrived earlier than Andy thought ⟨I would arrive⟩ (butAndy actually knew I would arrive on time).b. believe≫ DegP§ Here, wemust read theDegP than Andy thought relative to Carla’s beliefs, and not thespeaker’s beliefs, because it’s only the case that Andy arrived earlier than the timethat Carla thought Andy believed I would arrive, not the time that Andy actuallybelieved I would arrive.§ Thus, we have a de dicto reading, meaning low QR, and a low antecedent.˛ High antecedent, high QR: available(40) Context: Brad believes that Molly wants to own a chihuahua, but Molly actually wantsto own a great dane.a. Mollywants to have a bigger dog thanBrad thinks ⟨Molly wants to have a d-big dog.⟩b. DegP≫want§ Here, Molly doesn’t have any desires about her dog being bigger than any other dog,so the DegP is scoping above want.˛ High antecedent, low QR: unavailable(41) Context: Molly and Brad are in a rivalry, and Molly wants Brad to always be wrong.Brad knows that Molly is planning on getting a dog, and he has been guessing about whatkind of dog and how big it’ll be. Molly doesn’t actually really care about the size of thedog that she gets, as long as it’s bigger than what Brad thinks she wants.a. #Mollywants to have a bigger dog thanBrad thinks ⟨Molly wants to have a d-big dog.⟩b. *want≫ DegP§ Here, Molly has a desire about how big her dog is relative to what Brad thinks shewants, so the DegP is scoping below want.§ Wecanunderstand the unavailability of this reading as a case of irreconcilable antecedent-containment: the the DegP containing the scrap has failed to QR out of the scrap’santecedent.˛ Low antecedent, high QR: unavailable(42) Brad has just moved in to a new house in the neighborhood. He knows his neighbor isnamed Molly, but they haven’t properly met yet. Molly doesn’t know who Brad is. Oneday, Brad hears some dog yapping in a high-pitched voice in Molly’s house. He forms thebelief that Molly owns a very small dog. However, in actuality, Molly doesn’t own any16dog—she was just taking care of her friend’s chihuahua. Molly is a dog lover, and wantsto own a dog—in particular, she wants to have a great dane.a. #Molly wants to have a bigger dog than Brad thinks ⟨she has a d-big dog.⟩b. *DegP≫want§ Here, Molly doesn’t have any desires about her dog being bigger than any other dog,so the DegP QRs above want.§ Perhaps unexpectedly, we can’t combine this high QR with a low antecedent. Thisis unexpected under standard theories of ellipsis licensing.§ In fact, this reading becomes available with VPE:(43) Brad has just moved in to a new house in the neighborhood. He knows his neighboris named Molly, but they haven’t properly met yet. One day, Brad hears some dogyapping in a high-pitched voice in Molly’s house. He forms the belief that Mollyowns a very small dog. However, in actuality, Molly doesn’t own any dog—she wasjust taking care of her friend’s chihuahua. Molly is a dog lover, and wants to own adog—in particular she wants to have a great dane.a. Molly wants to have a bigger dog than Brad thinks she does ⟨she has ad-big dog.⟩b. ✓DegP≫want˛ This pattern follows straightforwardly from i) needing to license ACD; and ii) the c-command generalization.˛ The scrap must end up outside of its antecedent, so the QR height must be at least aslarge as the antecedent, but it can’t move too far, or else the scrap will no longer be c-commanded by its antecedent at LF.6 Enter PROModP˛ To recap: any theory of scrapping must account for the four properties I’ve just laid out:§ Scrapping must involve operator movement out of the scrap.§ Scrapping must occur in ACD configurations.§ Scraps are ModP sized.§ Scraps must be c-commanded by their antecedents at LF.˛ Analyzing the scrap as containing a bound ModP-sized anaphor, PROModP, captures allthese facts.176.1 The analysis(44) Jen saw every painting Masha expected Op PROModP.∀xpainting(x)∧expected(m)(ˆsaw(x)(j)) →saw(x)(j)Jenjλy.∀xpainting(x)∧expected(m)(ˆsaw(x)(y)) →saw(x)(y)λ1 ∀xpainting(x)∧expected(m)(ˆsaw(x)(g1)) →saw(x)(g1)λQ.∀xpainting(x)∧expected(m)(ˆsaw(x)(g1)) → Q(x)everyλP.λQ.∀xP(x) → Q(x)λx.painting(x)∧expected(m)(ˆsaw(x)(g1))paintingλx.painting(x)λx.expected(m)(ˆsaw(x)(g1))OpλA.Aλx.expected(m)(ˆsaw(x)(g1))λ2 expected(m)(ˆsaw(g2)(g1))Mashamλx.expected(x)(ˆsaw(g2)(g1))expectedλp.λx.expected(x)(p)saw(g2)(g1)t2g2PROModPλx.saw(x)(g1)λx.saw(x)(g1)λ3 saw(g3)(g1)t1g1λy.saw(g3)(y)sawλx.λy.saw(x)(y)t3g3⇝ by PM⇝ by IFA⇝ antecedent18˛ (44) provides a derivation of the sentence Jen saw every painting Masha expected.§ I’m assuming a very simple extensional semantics augmented with the Montagueup operator ˆ.§ I’ve boxed the antecdent ModP and PROModP; PM stands for Predicate Modification;IFA stands for Intensional Function Application.˛ The object every painting Masha expected QRs to adjoin to the antecedent ModP, gettingthe scrap out of its antecedent ModP, leaving behind a lambda binder λ3.˛ The node that c-commands PROModP is the node that immediately dominates λ3 and theoriginal ModP—the node with the denotation λx.saw(x)(g1).˛ Assuming some version of LF copying, this antecedent node copies over its denotationonto PROModP.9˛ Note that, since the node that c-commands PROModP is one that’s newly-created by QR,and thus contains a lambda binder, this node will always be of a predicate type.§ Which predicate type depends on the kind of movement involved. Here, we havean et predicate; if this were a comparative, we would have a dt type (assuming adegree quantifier analysis, Heim 1985, 2000, a.o.); if this were a temporal adjunct,we would have an it type derived by movement of a temporal operator (Geis 1970,Larson 1987, a.o.).˛ In order to compose PROModP, which will always receive a predicate type, with the em-bedding predicate expect, we’ll need to first saturate the first argument of of PROModP.˛ Thus, wemustmerge an operator of the requisite type, perhaps as a last-resort operationto ensure the derivation composes.§ This operator then later feeds operator movement; here, we QR it to create an etpredicate that can be composed with painting by PM.˛ The rest of the derivation then composes straightforwardly to derive the desired truthconditions.9If you don’t like the mechanism of LF copying, there are a few compositional tricks you can play to getproper semantic binding. One option is to give PROModP the semantics of a trace/pronoun of type σt—i.e. a predicate type—and base generate a lambda abstractor as a sister to every painting Masha expected thatabstracts over this trace/pronoun, converting it into a function that can take in λx.saw(x)(g1) as its argument.Alternatively, you could have the antecedent ModP do a short step of QR, leaving behind a lambda binder,with the same end result. Other options are also conceivable. For simplicity’s sake, I assume LF copyinghere.196.2 Deriving the four properties˛ The size of the scrap comes from stipulating that PROModP has the label ModP.10˛ The c-command requirement comes from stipulating that PROModP is syntactically ananaphor, and thus displays the characteristic syntactic properties of anaphorhood, suchas being c-commanded by its antecedent.˛ The obligatoriness of operator movement comes from PROModP being forced to receivea predicate denotation, which results in needing to merge an operator to ensure that thederivation composes.˛ I actually don’t derive the strong form of the ACD generalization—namely, that scrap-ping can only be licensed in ACD configurations.§ I derive something a bit weaker: that the scrap cannot be contained within a con-stituent that is first merged above its antecedent ModP—in this configuration, it’simpossible to get the antecedent to c-command the scrap, since our system doesn’tallow downwards movement.11§ In my system, the scrap can either originate below the antecedent ModP, in whichcan you have an ACD configuration (like all the examples we’ve looked at here), orit could potentially be first merged adjoined to the antecedent ModP.§ However, I’mnot aware of any construction that requires first-merging a phrasewithModP.127 Conclusion˛ I’ve introduced a little-studied elliptical construction I called scrapping, which involvedeliding the clausal complement of a clause-embedding verb in operator movement con-texts.10I have no answer yet to the deeper question of why it’s ModP in particular. How are children even ableto acquire this?11If you could A-move the antecedent ModP above the scrap, you should be able to license scrappingin a constituent first-merged high. But I’m not aware of any reason to believe that that kind of movementhappens in English.12Conditionals might be one such case, if we take seriously the idea that conditionals are modifiers ofmodals/modal bases. Interestingly, scrapping can be licensed in conditionals—further investigation is re-quired:(i) (a) You can sit here if you want ⟨to sit here⟩.(b) You must sit here if Rashida asks ⟨you to sit here⟩.20˛ I’ve argued for five empirical generalizations:§ Scrapping is not NCA;§ Scrapping obligatorily involves operator movement;§ Scraps must be contained within their antecedents;§ Scraps are ModP sized;§ Scraps must be c-commanded by their antecedents at LF.˛ I’ve shown that an analysis of the gap as containing an operator adjoined to a null ModPanaphor PROModP accounts for all these facts, providing further evidence for the possi-bility of anaphoric pathways to ellipsis.Thank you for listening!21ReferencesAbusch, Dorit. 1993. The scope of indefinites. NaturalLanguage Semantics 2:83–135.Bresnan, Joan, and Jane Grimshaw. 1978. The Syn-tax of Free Relatives in English. Linguistic Inquiry9:331–391.Chao, Wynn. 1987. On Ellipsis. Doctoral Disserta-tion, University of Massachusetts Amherst.Depiante, Marcela A. 2000. The syntax of deep andsurface anaphora: A study of null complementanaphora and stripping/bare argument ellipsis.Doctoral Dissertation, University of Connecticut.Depiante, Marcela A. 2001. On null complementanaphora in Spanish and Italian. Probus 13:193–221.Fodor, Janet D., and Ivan Sag. 1982. Referential andQuantificational Indefinites. Linguistics and Philos-ophy 5:355–398.Geis, Michael L. 1970. Adverbial subordinate clausesin English. Doctoral Dissertation, MIT, Cam-bridge, MA.Hankamer, Jorge, and Ivan Sag. 1976. Deep and Sur-face Anaphora. Linguistic Inquiry 7:391–426.Hardt, Daniel. 1999. Dynamic Interpretation of VerbPhrase Ellipsis. Linguistics and Philosophy 22:185–219.Heim, Irene. 1985. Notes on Comparatives and Re-lated Matters. Manuscript, University of Texas,Austin.Heim, Irene. 2000. Degree Operators and Scope. InProceedings of SALT 10, ed. Brendan Jackson andTanya Matthews, 40–64. Ithaca, NY: Cornell Uni-versity.Iatridou, Sabine, and Hedde Zeijlstra. 2013. Nega-tion, Polarity, and Deontic Modals. Linguistic In-quiry 44:529–568.Kennedy, Christopher, and Jason Merchant. 2000a.Attributive Comparative Deletion. Natural Lan-guage & Linguistic Theory 18:89–146.Kennedy, Christopher, and Jason Merchant. 2000b.The Case of the ‘Missing CP’ and the Secret Case.In Jorge Hankamer WebFest, ed. Sandra Chung,James McCloskey, and Nathan Sanders. UCSCLinguistics Department.Kratzer, Angelika. 1998. Scope or Pseudoscope? ArethereWide-Scope Indefinites? InEvents andGram-mar, ed. Susan Rothstein, volume 70 of Studiesin Linguistics and Philosophy, 163–196. Dordrecht:Springer.Larson, Richard K. 1987. “Missing prepositions” andthe analysis of English free relative clauses. Lin-guistic Inquiry 18:239–266.Matthewson, Lisa. 1999. On the interpretation ofwide-scope indefinites. Natural Language Seman-tics 7:79–134.Moltmann, Friederike. 1992. Coordination and com-paratives. Doctoral Dissertation, MIT.Moltmann, Friederike. 1993. The Empty Element inComparatives. In Proceedings of NELS 23, 319–333.Amherst: UMass GLSA.Napoli, Donna Jo. 1983. Comparative Ellipsis:A phrase structure analysis. Linguistic Inquiry14:675–694.Overfelt, Jason. 2019. The eliminative effect of ellip-sis on the distribution of temporal adverbs. URLlingbuzz/004830, manuscript.Schwarz, Bernhard. 2000. Topics in Ellipsis. Doc-toral Dissertation, University of MassachusettsAmherst.Schwarzschild, Roger. 2002. Singleton Indefinites.Journal of Semantics 19:289–314.Wold, Dag. 1992. IP deletion. MIT Squib.Wurmbrand, Susi. 2014. Tense and aspect in Englishinfinitives. Linguistic Inquiry 45:403–447.22


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items