Small Clause Reconstructuring Keir Moulton McGill University – A very drafty draft, March 2010 –

Small clause subjects do have narrow scope construals (contra Williams 1983), but only if the small clause predicate is itself quantificational. A paradox is revealed that is resolved by a new-ish semantic implementation of Chomsky’s (1955/1975) complex predicate analysis of small clauses. In fact, the proposal is the only one that resolves the paradox. Predictions about small clause subjects, such as their ability to take low scope and bleed Condition C, their interaction with other quantifiers, and their relation to copy raising subjects, are born out by the analysis.

1 A Small Clause Scope Paradox Williams (1983), as one of his arguments against analyzing the complement in (1a) as a Stowellian small clause (SC), showed that small clause subjects (SCSs) can’t scope below the predicate that embeds the small clause, whereas subjects of infinitives can (1b).1,2 (1) Given the fact that a seat is empty in our otherwise crowded classroom. . . 1 I will throughout use the terms “small clause” and “small clause subject” even though the ultimate analysis does not strictly recognize a small clause here. 2 The original sentences used the indefinite someone:

(1)

a. Someone seems sick. someone ≫ seem; *seem ≫ someone b. Someone seems to be sick. someone ≫ seem; seem ≫ someone

(Williams 1983, 293:(40a))

Complications arise here because of the “epsitemic” nature of someone. Examples in this squib use indefinite a.

1

a. A student seems sick today. ∃ ≫ seem; *seem ≫ ∃ b. A student seems to be sick today. ∃ ≫ seem; seem ≫ ∃ (1b) can report that, given the evidence, it must be the case that a student is sick, but no particular one. This interpretation is not available for the small clause in (1a), which is why that sentence is odd in the context given.3 Small clauses subjects, then, do not appear to reconstruct, a fact we’ll label SCS Freezing, for small clause subject freezing. Further evidence for SCS Freezing comes from Heycock (1995) and Stowell (1991). The subject of the infinitive in (2b) may receive a reading that puts it in the scope of consider, a “cardinal” reading. The small clause subject in (2a) may only receive a presuppositional reading. (2)

a. She considered many drugs available. (Heycock 1995, 234:(40b)) b. She considered many drugs to be available.

What (2b) can report that (2a) cannot is that what was considered (true) was that the number of drugs available was high. The only thing conveyed by (2a) is that for a large number of drugs, it is considered that those particular drugs are available. The contrasts in (3) tell a similar story. (3)

a. Several new books seem available. b. Several new books seem to be available. (Heycock 1995, 233:(36a,b))

SCS Freezing gave Williams evidence against small clauses. And it has given small clause proponents a puzzle to solve (Stowell 1991; Johnson and Tomioka 1998; Matushansky 2002; den Dikken 2008). But the following examples provide evidence that neither position is quite right. Small clause subjects can enjoy a scope position within the the small clause. The examples in (4) are easily interpreted with the indefinite in the scope of the modal adjective necessary. 3 Terminology: by ‘wide’ and ‘narrow’ I mean wide-scope, transparent and low-scope, opaque, respectively. Indefinites, of course, allow a third reading: low-scope, transparent (Fodor 1970), showing that there are at least two notions to be separated when looking at indefinites in opaque contexts. We, too, properly distinguish these when relevant. The role that transparency/opacity plays will become useful in teasing apart certain scopes in section 2.2.

2

(4)

a. She considers a new fridge unnecessary. unnecessary ≫ a new fridge; a new fridge ≫ unnecessary b. A stove seems to me much more necessary. necessary ≫ a stove; a stove ≫ necessary

I can use (4b) to convey that what seems to me is that in all worlds that satisfy some (contextually relevant) needs, there is a stove - but not necessarily the same stove across those possible situations or worlds where those needs are satisfied. This kind of modal variation for the indefinite is confirmed by (5), (5) A Green Party senator seems necessary if we are ever to make progress on climate change. which does not commit us to there being a senator so described. We’ve arrived at a paradox. A SCS can scope below the predicate inside the small clause but not, apparently, merely below the embedding predicate. Moreover, the embedding predicate has the SC predicate in its scope: the necessity of a new fridge in (4a) is attributed to the opinions of the matrix subject.4 We might then expect, by transitive reasoning, that the SCS should itself be able to scope below seems and consider. The facts suggest, however, that this scope can only be achieved if the small clause itself contains a scope bearing operator. That is, the difference between (1a)-(3a) and (4a)/(5) is that only in the latter cases is the embedded predicate itself quantificational.5 That is, the difference reduces to the difference between sick, available vs. necessary. It simply makes no sense to ask the relative scope of an indefinite with respect to happy.6 I now propose and will subsequently defend the following generalziation: (6) Small Clauses Scope Generalization A narrow scope interpretation for a SCS (with respect to the embedding predicate) is available only if the embedded predicate is itself a scope bearing element. 4 What we appear to get is “double modality.” Epistemic modals embedded under epistemic/belief verbs (as in John believes Fred must be happy) do not give rise to double modality. But modal operators like necessary under believe do (as in I believe that a stove is necessary). 5 Strictly speaking I do not need to say it’s quantification, only that it selects for anything other than individual types. 6 In section 2.2 we will see that merely having another scope bearing term in the SC is not enough. It has to be the case that the embedded predicate is itself one that can take things other than non-individual types.

3

We derive (6) by a particular interpretation of Chomksy’s (1955) original proposal for small clauses, coupled with an implementation of the semantics for how the embedding verbs compose with their complements. I will introduce both these ingredients in the next section, and present arguments supporting the analysis in the remainder of the squib (arguments relating to Condition C effects (section 2.1), the (non)interaction of SCSs with other lower quantifiers (section 2.2), and copy raising constructions (section 3)).

2 The complex predicate analysis Chomsky (1955/1975) proposed that SCs (among other constructions) are complex predicates in which the non-verbal predicate and the embedding verb form a constituent to the exclusion of the SCS. I will defend a version of this approach by giving (7a) the representation in in (7b). (See Williams and di Sciullo (1987), Larson (1988), and Neeleman (1994) for further syntactic arguments for this approach.7 ) (7)

a. Michael considers a fridge necessary. vP

b.

v’

DP Michael

v considers

VP DP

V’

a fridge

V

AP

t

necessary

We combine this apporach with an an analysis of seems and other SCtaking predicates that allows them to “connect” the embedded predicate with the SCS in a way such that the scope possibilities of the SCS are determined by the embedded predicate. To see this, we first need on the table a denotation for a modal adjective like necessary. It allows both a wide and narrow scope interpretation for an indefinite subject. 7 The placement of the verb, as moving to v differs, of course, from how some of the other proponents of complex predicate analysis derive the correct word order.

4

(8) A fridge is necessary necessary ≫ a; a ≫ necessary While a reduction may be possible in deriving both interpretations from one necessary, it is the narrow scope (opaque) interpretation that is of interest here, and so I offer (9) as a denotation for necessary that achieve this, in which it takes a property type argument.8 (9) J necessary Kw = λPe,st .∀w’ ∈ Nec(w) [∃x [P(x)(w’)]] where Nec(w) = {w’ such that what’s necessary in w is satisfied in w’ } This will combine with property denotations for indefinites such as a fridge, yielding a narrow scope interpretation. (10) J necessary (a fridge) Kw = 1 iff ∀w’ ∈ Nec(w) [∃x [fridge(x)(w’)]]

≈ In all worlds w’ where what’s necessary in w is satisfied, there is a fridge in w’ Now, the Small Clause Scope Generalization informs us that the SCS is outside the scope of the embedding predicate, but it can be smuggled into it if there is a scope-bearing embedded predicate. So to capture the low scope reading in (4b), for instance, we let small clause-embedding predicates pass up the denotations of the embedded predicate. To do this, I will adopt for small clauses a proposal made by Jacobson (1992) for infinitival raising complements. Jacobson uses function composition to put seem-type verbs together with their complements.9 Function composition effectively takes two functions and applies one to the other, allowing a “delay” in the composition of the inner function with its argument. (11) Function Composition A ◦ B = λg.A(B(g)) where B is a function taking g and A a function taking B(g) 8 As stated, I do not intend to claim that this is the definitive analysis of the narrow scope interpretation of (8), but whatever does account for it will carry over to the present analysis. 9 Jacobson argues that her analysis applies to infinitival raising constructions (e.g. A unicorn seems to be approaching) and not to small clauses. It appears better suited, though, to small clauses and not infinitives for the simple reason that subjects raised from infinitives can scope below seem in the absence of a lower scope-bearing predicate, unlike SC seem.

5

Take seems as the representative case. It is interpreted as a propositional operator, and so (aside from its world argument) takes a propositional argument. (As a stand-in for a real account of the nature of the epistemic modality of seems, we simply instruct seems to quantify over the set of seems-accessible worlds from w.) (12) J seem Kw = λp.λw.∀w’ ∈ seem(w) [p(w’)] We can now see how function composition puts together the meanings delivered to it by the complex predicate analysis of small clauses: (13) The formation of a complex predicate by FC 10 J seem Kw ◦ J necessary Kw = λP. J seem necessary(P) Kw This can then combine with the SCS, if the SCS is a property-denoting DP. The interpretation that results is given below: (14) J [ seem necessary ] ( fridge ) Kw = ∀w’ ∈ seem(w) [ ∀w” ∈ Nec(w’) [∃x [fridge(x)(w”) ]]]

≈ It seems that in all words w” where what is necessary is satisfied, there is a (possibly different) fridge in w” The result is that a fridge is in the scope of necessary and in the scope of seems. If the embedded predicate is not one that itself selects for anything but individual types, there is no way to “smuggle” a quantificational SCS below the scope of seems. The small clause in (15a) is just another example, like (1a), where a wide-scope SCS is obligatory (compare to the infinitive in (15b).) (15)

a. A citizen seems unhappy with your decision. a citizen ≫ seem; *seem ≫ a citizen b. A citizen seems to be unhappy with your decisions. a citizen ≫ seem; seem ≫ a citizen

10 I assume without implementation something like intensional functional composition - analogous to intensional functional application in Heim and Kratzer (1998) so that the world parameter of seems and necessary are identified.

6

The adjective is defined in (16a) and is function composed with seem in (16b) (16)

a. J unhappy Kw = λx. unhappyw (x) b. J seem Kw ◦ J unhappy Kw = λx. J seem unhappy(x) Kw

The complex predicate seems+unhappy inherits the argument-taking properties of unhappy. And since this complex predicate is a function calling for inviduals, a generalized quantifier will always scope above it, and therefore seems. The reader can verify that this will give wide scope to the other quantificational expression in (1a)-(3a). (17)

a. J seem Kw ( J unhappy Kw ) = λx.∀w’ ∈ seem(w) [ unhappyw’ (x ) ] b. J A citizen Kw ( J seem unhappy Kw ) = ∃x.citizen(x) & ∀w’ ∈ seem(w) [ unhappyw’ (x) ]

The scope paradox is resolved since the scope options for the SCS are entirely determined by the embedded predicate. When the embedded predicate selects for things like properties, the SCS gets not just narrow scope with respect to the embedded predicate, but the matrix as well. No other account, so far as we can tell, could be sensitive to the properties of the embedded predicate in this way. Allowing reconstruction below seems would allow narrow scope regardless of the nature of the embedded predicate. Ruling out reconstruction altogether would fail to capture the meaning of (4a).

2.1 Condition C Predictions Scope reconstruction feeds condition C in A-movement (Fox 1999; Romero 1998). Since our analysis lets SCSs can take narrow scope without ever being syntactically present in the embedded clause, we predict that no condition C violation will arise. And, indeed, no such interaction is found with SCSs that are interpreted with narrow scope.11 The relevant test sentences are reported in (18). In (18a), the SCS can scope below necessary and no disjoint reference effect is observed. (18b) is the control in which the SCS is nonquanitficational - and therefore in no need of recosntuction. The acceptability of the co-reference indicated is the same for both. 11 One must be careful in employing binding-theoretic diagnostics in propositional attitude reports, which is what seem with an experiencer is (see Moulton (2009) and Kuno (2004)).

7

(18)

a. b. c.

Another nude picture of the modeli ’s toes seemed to heri rather unnecessary. This nude picture of the modeli ’s toes seemed to heri rather unnecessary. ? A nude picture of the modeli ’s toes seemed to heri to have been rather unnecessary.

(18c) is the infinitival case, for which a disjoint reference effect is reported to obtain (Fox 1999). While worse than (18b), the marginality of (18c) may indicate that infinitives have the option of forming a complex predicate with the matrix verb too, a possibility consistent with our analysis.12 In sum, if scope reconstruction feeds Condition C, it must be the case that narrow scope interpretations for SCSs are not delivered by reconstruction. This is what we would expect on the present implementation of the complex predicate analysis.

2.2 No interaction with other quantifiers A further prediction of the analysis is that a SCS cannot take narrow scope in the presence of just any quantifier that forms part of the embedded predicate. For instance, we predict that in sentences of the form, (19) QP1 seems [ PRED [PP . . . QP2 . . . ]] QP1 - the SCS - will not be able to scope below QP2 unless QP2 itself QRs higher than QP1 . But in that case, QP1 will still outscope seems. And if PRED selects only for individual types, then the complex predicate, seems+PRED will be scope-neutral: it will preclude a quantifier (or DP of any type) from being interpreted anywhere but in its surface position above seem+PRED. That this prediction is correct is demonstrated by the fact that (23b) has one reading fewer (less?) than (20b). The relevant scopes are indicated. (20)

a. A woman seemed upset in every picture. i. ok ∃ ≫ seems ≫ ∀ ii. ok ∀ ≫ ∃ ≫ seems

12 The relevant test, then, would be whether a Condition C violation obtains in raising infinitives if there is no other quantificational element to “smuggle” the indefinite into the scope of seems: A student of David’si seems to himi to be at the party-which resembles more closely the original sentence upon which judgments were formed-does appear to exhibit a stronger disjoint reference effect than (18c). I leave this to future research.

8

iii. * ∀ ≫ seems ≫ ∃ b. A woman seemed to be upset in every picture. i. ok ∃ ≫ seems ≫ ∀ ii. ok ∀ ≫ ∃ ≫ seems iii. ok ∀ ≫ seems ≫ ∃ The scope permutations are hard to tease apart, particularly (ii) and (iii). In both, the indefinite falls under the universal, and so the women may vary from photo to photo. At stake, however, is whether the indefinite can fall within the scope of the modal operator seems. We have a way to see whether it does or not is to inspect whether an opaque reading is possible for the SCS.13 A narrow scope, opaque indefinite is compatible with a continuation in which the speaker is not committed to the descriptive content of the indefinite. This continuation is natural after the infinitival construction in (21a). It is not compatible with the small clause (21b). This is what we expect if the SCS cannot be interpreted opaquely (Both sentences strongly support narrow scope of the indefinite with respect to the universal, which shows that there can be a wide-scope, transparent but co-varying interpretation of a woman.) (21)

a. In every picture, it seems that a woman is upset, . . . but I cannot quite make out for sure if it’s a woman or a boy. b. In every picture, a woman seems upset, # . . . but I cannot quite make out for sure if it’s a woman or a boy.

Our original test sentences in (20) pattern likewise. The SCS subject cannot scope both below seems and below the universal. (22)

a. A woman seems to be upset in every picture, . . . but I cannot quite make out for sure if it’s a woman or a boy. b. A woman seems upset in every picture, # . . . but I cannot quite make out for sure if it’s a woman or a boy.

Another strategy for showing the same thing is to trap the lower quantifier low. Below I use an NPI, which must scope below negation. This means for the SCS to scope below the quantifier it would have to “reconstruct” there. That’s not possible given our analysis for small clauses, but it is possible for raising infinitives. 13 Narrow scope, opaque readings require that in addition to the quantificational portion of the indefinite being interpreted low, the NP (restrictor, descriptive) portion must be as well. This means the whole damn DP is low. I’ll assume that narrow, opaque readings always entail narrow scope-that is, that there is no “fourth reading” for indefinites.

9

(23)

a. A book didn’t seem fully upright on any shelf. i. ∃ ≫ seems ≫ ∀ ii. ∀ ≫ ∃ ≫ seems iii. * ∀ ≫ seems ≫ ∃ b. A book didn’t seem to be fully upright on any shelf. i. ∃ ≫ seems ≫ ∀ ii. ∀ ≫ ∃ ≫ seems iii. ∀ ≫ seems ≫ ∃

Only the complex predicate approach predicts the sensitivity of narrow scope to the nature of the predicate within the small clause.

3 Beyond compositional tricks The analysis trades on a compositional technique - or worse, a compositional slight of hand, one that may not seem particularly justified independently. Our seems takes a proposition, not a predicate. But it combines with a predicate denotation because function composition allowed it to accept the predicate’s promise of becoming a saturated proposition.14 One wonders why seem and consider can overcome this type-mismatch and function compose with a predicative complement but other proposition-taking embedders cannot. That is, we would expect lots of type-mismatches to be remedied in this way–without constraints, this approach opens the doors too wide to compositional havoc. One possibility is that we take “complex predicate” formation in these cases as signaling a distinct mode of composition. (Morzycki (2005) argues that function composition is in fact a strategy available to sub-lexical composition. It might be a stretch to call our complex predicates “sub-lexical”, but the comparison is often made Stowell (1991).) Further hints that our “compositional trick” may rest on some linguistic ground comes from copy raising constructions. Landau (2009) and Asudeh and Toivonen (2010) have argued that copy raising constructions require that seems combine first with a predicate and then a (non-thematic) argument, the raised copy (24). Ultimately, we want to understand the real reason why verbs like seems allow this method of composition. But it’s worthwhile to see that 14 We could have reduced small clauses to control, but that would have been wrong since control doesn’t allow even our “smuggled” narrow scopes: A linguist wants to be necessary. The situation we’ve encountered with SCSs is, indeed, one that requires tools beyond those that distinguish control from raising.

10

the the compositional strategy suggested for small clauses, while perhaps a stand-in for a better theory, is nonetheless a stand-in that is independently needed. (24) Charliei seems like/as if somebody’s just mocked hisi/?j accent. (Landau 2009, 4:(10a)) Two features unify small clauses and copy raising. The first is that the complement of seems in copy raising has been argued to denote a predicate. (Derived by special properties, one assumes, of clauses headed by like/as if ; see Heycock (1994), Landau (2009), and Asudeh and Toivonen (2010).) The second unifying characteristic is that, in certain cases, both copy raising and small clauses require the subject to be the very thing that provides the “evidence” on which the prejacent of a seems-report is founded. Matushansky (2002) shows that (25a) can be true if one believes the squire to be sick based on the presence of medicine bottles and other paraphernalia suggestive of illness. A small clause cannot: it must be facts about the squire himself that indicate something about his condition (25b). (25) Given that I saw medicine bottles, kleenexes, and smelled a foul, sickly stench, a. The squire seemed to be sick. b. # The squire seemed sick. (Matushansky 2002, 225:(14)) Similar evidential requirements are found for copy raising constructions (Asudeh and Toivonen 2010). The following is a version of Asudeh and Toivonen’s “absent cook” examples, modified for the above scenario: (26) Given that I saw medicine bottles, kleenexes, and smelled a foul, sickly stench, a. It seemed as if the squire was sick. b. # The squire seemed as if he was sick. Explaining the evidential dimension of copy raising and small clauses is beyond the scope of this paper (but see Landau (2009) on whether this is a persistent characteristic of copy raising and see Moulton, in prep.). But case of the Sick Squire suggests that the SCS (and copy raised subjects) can enter into the semantic calculation that delivers the relevant modal alternatives for seems. And that might place our “composition trick” on firmer ground. Despite some straining to avoid this conclusion in Asudeh and Toivonen (2010), SCSs may be a real argument of the matrix verb. SCSs constitute the evidence on which a seems-construction reports a conclusion. The argument structure of epistemic modals has received some renewed attention of late (Kratzer 2009). Seems may, after all, be dyadic. It has an “evidence” argument, but evidence for this argument has just been hard to see in a still largely GB world. * 11

References Asudeh, Ash, and Ida Toivonen. 2010. Copy raising and perception. Manuscript, Univeristy of Ottawa. Chomsky, Noam. 1955/1975. The logical structure of linguistic theory. Chicago University Press. den Dikken, Marcel. 2008. Small clauses, phases, and phase extension. In 31st GLOW Colloquium. Fodor, Janet Dean. 1970. The linguistic description of opaque contexts. Doctoral Dissertation, MIT. Fox, Danny. 1999. Reconstruction, binding theory, and the interpretation of chains. Linguistic Inquiry 30:157–196. Heim, Irene, and Angelika Kratzer. 1998. Semantics in generative grammar. Malden, MA: Blackwell. Heycock, Caroline. 1994. Layers of predication. Doctoral Dissertation, University of Pennsylvania. Heycock, Caroline. 1995. The internal structure of small clauses: New evidence from inversion. In Proceedings of the North East Linguistic Society 25, ed. Jill Beckman, 223–238. University of Pennsylvania: Graduate Linguistic Student Association. Jacobson, Pauline. 1992. Raising without movement. In Control and grammar(48), ed. Richard K. Larson, Sabine Iatridou, Utpal Lahiri, and James Higginbotham, 149–194. Dordrecht, The Netherlands: Kluwer Academic Publishers. Johnson, Kyle, and Satoshi Tomioka. 1998. Lowering and mid-size clauses. In Proceedings of the 1997 Tubingen ¨ workshop on reconstruction, ed. Graham Katz, Shin-Sook Kim, and Winhart Haike, 185–206. Tubingen, Germany: Sprachte¨ oretische Grundlagen fur ¨ die Computer Linguistik. Kratzer, Angelika. 2009. Context and content. Lecture slides, Institut Nicod, Paris. Kuno, Susumu. 2004. Empathy and direct discourse perspectives. In The handbook of pragmatics, ed. Laurence R. Horn and Gregory Ward, 315–343. Blackwell. Landau, Idan. 2009. Predication vs. aboutness in copy raising. Downloaded from lingBuzz (lingBuzz/000835). Larson, Richard. 1988. On the double object construction. Linguistic Inquiry 19:335–392. Matushansky, Ora. 2002. Tipping the scales: The syntax of scalarity in the complement of seem. Syntax 5:219–276. Morzycki, Marcin. 2005. Size adjectives and adnominal degree modification. In Proceedings of SALT XV, ed. Effi Georgala and Jonathan Howell, 116–133. Cornell University, Ithaca, NY: CLC Publications. Moulton, Keir. 2009. Natural selection and the syntax of clausal complementation. Doctoral Dissertation, Doctoral Dissertation, University of Massachusetts, Amherst. Neeleman, Ad. 1994. Complex predicates. Utrecht: Onderzoeksinstituut voor Taal en Spraak.

12

Romero, Maribel. 1998. Focus and reconstruction effects in wh-phrases. Doctoral Dissertation, University of Massachusetts, Amherst. Stowell, Timothy. 1991. Small clause restructuring. In Principles and parameters in comparative grammar, ed. Robert Freidin, 182–218. Cambridge, Massachusetts: MIT Press. Williams, Edwin. 1983. Against small clauses. Linguistic Inquiry 14:287–308. Williams, Edwin, and Maria di Sciullo. 1987. On the definition of word. Cambridge, Massachusetts: MIT Press.

13