Sluicing as Anaphora to Issues∗ Scott AnderBois
[email protected] SALT 20: UBC & SFU April 29-May 1, 2010
1
Introduction • Sluicing is an ellipsis process as in (1) by which an interrogative clause (subscript E) is deleted under identity with a prior antecedent clause (subscript A).
(1)
[Someone left]A , but I don’t know [who left]E • Merchant (2001) argues that the required identity condition is a semantic one: symmetric entailment between (the focus closures of) A and E.
While attractive in theory, symmetric entailment referring solely to truth-conditions faces several immediate empirical challenges: i. Doubly negated indefinites do not license sluicing • Despite truth-conditional equivalence to (1), (2)-(3) are ungrammatical. (2)
*[It’s not the case that no one left]A , but I don’t know [who left]E .
(3)
*[It’s not the case that John didn’t meet with a studentA , but Fred still wonders [who John met with]E
ii. Implicit passive agents do not either
Merchant (2001), Chung (2006)
• Again, despite truth-conditional equivalence, the implicit agent of a passive cannot serve as inner antecedent. (4)
a. b.
A: [The cake was eaten]A . B: *[Who ate the cake]E
iii. Disjunctions do license sluicing (5)
Chung et al. (1995)
[(Either) John or Fred left]A . Tell me [which (one) left]E
∗
Many thanks to Pranav Anand, Matt Barros, Adrian Brasoveanu, Sandy Chung, Jeroen van Craenenbroeck, Donka Farkas, Lyn Frazier, Robert Henderson, Jim McCloskey, Floris Roelofsen, Luis Vicente, and Matt Wagers for helpful and challenging comments on this work.
1
Sluicing as Anaphora to Issues • An account invoking only truth-conditional isomorphy can potentially capture (iii), . . . but would struggle with the other puzzles. • By referencing syntactic form, we could capture (i) and (ii), . . . but would struggle to account for (iii). • Semantic isomorphy over both truth conditions and discourse referents could capture (ii)-(iii), . . . but would make the wrong prediction on (3), which allows anaphora to the indefinite. – In addition to these old puzzles, we present novel data showing a previously unnoticed yet systematic restriction. iv. Overt indefinites inside appositives do not license sluicing (6)
Joe, who once killed a man in cold blood, doesn’t even remember who *(it was). • Such examples are problematic for any of the aforementioned accounts. • The would-be elided clause is identical to the antecedent in its syntax, truth-conditions, and discourse referents introduced.
Proposal: All four problems can be solved by a symmetric entailment condition under a notion of semantic content as comprising not only truth-conditional information, but also issues. • We develop such an account by making use of the semantics of issues as developed in Inquisitive Semantics (Groenendijk & Roelofsen (2009), Ciardelli (2009), AnderBois (submitted) inter alia) Road Map: §2 reviews the formal framework of inquisitive semantics; §3 develops the account of sluicing for indefinites, disjunctions, and doubly negated indefinites; §4 extends the account to implicit passive agents; §5 independently motivates a semantics of appositives; §6 shows that it makes the right predictions for sluicing; §7 preliminarily examines how the account might be extended to so-called ‘sprouting’ cases; and §8 concludes.
2
Sluicing as Anaphora to Issues
2
The semantics of issues
At its core, inquisitive semantics brings together two separate theoretical notions: i. Assertions are proposals to update the common ground. • Following Farkas & Bruce (2010) and others, we conceive of an assertion as a proposal to update the common ground.1 • Empirically, Farkas & Bruce (2010) present the possibility of responses like ‘yes’ and ‘no’ as in (7) as support for this view. (7)
Anne: Sam is home. Ben: Yes // Yeah, he’s home // No, he isn’t home
ii. Indefinites and disjunctions as alternative-evoking expressions. • Kratzer & Shimoyama (2002), Simons (2005), Alonso-Ovalle (2006) and others argue that a wide range of phenomena such as free-choice effects, disjunctive counterfactual antecedents, and quantificational variability effects are readily accounted for under this view. • Parallel phenomena do not seem to exist for conjunction and universal quantification. Combining (i) and (ii), sentences with widest-scope indefinites and disjunctions are: Proposals to update the common ground with a non-singleton set of alternatives, latently raising the issue of which one(s) holds. • That is, Hamblin alternatives are part of the Context Change Potential of sentences with widest-scope indefinites/disjunctions, not just part of subsentential composition.
2.1
Classical Expressions
• Formally, we capture this by treating assertions more like questions; both are of type stt • For an atomic formula, the denotation is simply the set containing the classical denotation as its sole member, as in S1. Atomic Formulas: S1: !ϕ"M,g,w = {
{ w! | ϕ(w) }
}
• Conjunction also returns a singleton set with the classical denotation as the sole member as in S22
1 While Stalnaker (1978) does not emphasize this aspect of assertion, the current conception is nonetheless consistent with his. Stalnaker writes “To make an assertion is to reduce the context set in a particular way, provided that there are no objections from other participants in the conversation.” The idea that assertion is a proposal to update the common ground, then, is a way of fleshing out Stalnaker’s ‘provided that’ clause. 2 The more complicated definition reflects the fact that each conjunct, in principle, contains elements other than atomic formulas (e.g. indefinites or disjunctions). This is orthogonal to our present concerns, see Groenendijk & Roelofsen (2009) for discussion of these issues.
3
Sluicing as Anaphora to Issues • Based on this, universal quantification can be defined as a conjunction of an unspecified number of conjuncts as in S3 (see AnderBois (submitted), Ciardelli (2009) for details) • Alt is Groenendijk & Roelofsen (2009)’s alternative closure operator which eliminates wouldbe alternatives which are non-maximal (i.e. contained within another alternative). Conjunction: S2: !ϕ ∧ ψ"M,g,w = Alt{α ⊆ W | ∃β ∈ !ϕ"M,g,w : α ⊆ β and ∃γ ∈ !ψ"M,g,w : α ⊆ γ} Universal Quantifier: S3: !∀uϕ"M,g,w = Alt{α ⊆ W | for all d ∈ De s.t. ∃β ∈ !ϕ"M,g[u/d],w : α ⊆ β}
2.2
Inquisitive elements
• The denotation for non-inquisitive elements differ from their classical counterparts only in semantic type (stt instead of st). • Where inquisitive semantics does real work for us is sentences with alternative-evoking elements: disjunctions, indefinites, and questions. • A sentence containing a disjunction will denote a non-singleton set of alternatives – one per disjunct – as in S4. • An indefinite can be captured in similar fashion as a disjunction of an unspecified number of disjuncts as in S5.3 Disjunction: S4: !ϕ ∨ ψ"M,g,w = Alt{α ⊆ W | ∃β ∈ !ϕ"M,g,w : α ⊆ β or ∃γ ∈ !ψ"M,g,w : α ⊆ γ}
Existential Quantifier: S5: !∃uϕ"M,g,w = Alt{α ⊆ W | there is some d ∈ De s.t. ∃β ∈ !ϕ"M,g[u/d],w : α ⊆ β}
• We treat a (widest-scope) disjunction or indefinite as a multi-alternative proposal to update the common ground. • Applying the logic of Farkas & Bruce (2010) for sentences which are classical, this semantics is supported by the availability of ‘secondary responses’ as in (8)-(9). • N.B. disjunctions and indefinites are still differentiated from questions since no response of this sort – or indeed of any overt response – is expected.
(8)
a. b.
Bill or Fred greeted Joe. It was Fred // Yeah, Fred // Fred // Probably Fred
(9)
a. b.
Someone greeted Joe. It was Fred // Yeah, Fred // Fred // Probably Fred
• Since alternatives in this system are part of the metalanguage semantics, the formulas denoted by these sentences do not show this difference. 3
Since we assume finite models throughout, we can stick to the simplified account of first-order quantifiers from AnderBois (submitted), avoiding complications discussed in Ciardelli (2009) for inquisitive quantification for models with infinite domains.
4
Sluicing as Anaphora to Issues • We can, however, schematize their metalanguage interpretations as follows: (10) (11)
!(8)" = {
{w! : bill greeted joe in w! }, {w! : fred greeted joe in w! }
}
!(9)" = { {w! : bill greeted joe in w! }, {w! : fred greeted joe in w! }, {w! : jane greeted joe in w! }, . . . }
• For models with very few possible worlds, we can fruitfully capture this metalanguage interpretation visually as in (12).4 (12)
!∃x.greet! (x,j)" = 11
10
11
10
01
00
∃x.ϕ(x) 01
2.3
00
Closing off alternatives with negation
The system thus far seems would suggest that any sentence with an indefinite or disjunction will be a multi-alternative proposal (i.e. will raise an issue in the discourse). • Both intuitively and according to the responses in (11)-(12), however, this issue-raising capacity only holds of widest-scope indefinites and disjunctions. • That is, (13), does not raise an issue on the narrow-scope reading of the indefinite (indeed, we would take the (b) responses to disambiguate in favor of a wide-scope reading). (13)
a. It is not the case that someone met Fred. b. #It was Bill // #Yeah, Bill // #Bill // #Probably Bill
• We capture this in S6 because negation quantifies over the alternatives in its scope, returning the singleton set containing the set of worlds where none of the alternatives in it hold. Negation S6: !¬ϕ"M,g,w = Alt{α ⊆ W | every β ∈ !ϕ"M,g,w is such that α ∩ β = ∅} • We see this illustrated in (14).
• Negation returns the maximal alternative which does not intersect with any positive alternative. 4
Circles indicate possible worlds, the number inside the circles indicate the truth values of two propositions in that world (e.g. bill greeted joe in w" ), and shaded boxes represent alternatives.
5
Sluicing as Anaphora to Issues
(14)
!¬∃x.greet! (x,j)" = 11
10
11
10
01
00
¬∃x.ϕ(x) 01
00
• One important result of this definition for negation is that double negation is no longer vacuous. • Double negation of a formula ϕ preserves truth-conditions, but eliminates issues within ϕ. • The non-vacuity of double negation is exactly what allows us to capture Puzzle (i). • Given this, we define the non-inquisitive closure operator (‘!’) as in (15). (15)
Non-inquisitive closure:
!ϕ = ¬¬ϕ
• Empirically, this semantics is directly supported by the contrast between (9) (repeated as (16)) and (17).5 (16)
a. b.
Someone greeted Joe. It was Fred // Yeah, Fred // Fred // Probably Fred
(17)
a. It’s not the case that no one greeted Joe. b. #Yeah, Fred // #Fred // #Probably Fred
• This is semantically captured as follows: (18)
!¬¬∃x.greet! (x,j)" = 11
10
11
10
01
00
¬¬∃x.ϕ(x) 01
00
5
It should be noted that in these cases, responses with VP-Ellipsis such as ‘Yeah, Fred did.’ seem to be felicitous. The secondary response diagnostic, then, applies only to secondary fragment responses.
6
Sluicing as Anaphora to Issues
2.4
Questions
• Like indefinites and disjunctions, questions also make an inquisitive contribution to a discourse. • An inquisitive assertion differs from a question, though, since it is also truth-conditionally informative • Specifically, it proposes to eliminate worlds from the context set where none of the alternatives holds (i.e. worlds shaded gray in the output). • A wh-question like (19), then, consists of an indefinite (the wh-word) made uninformative. (19)
Who greeted Joe?
Two paths to achieve uninformativity: 1. Groenendijk & Roelofsen (2009) posit a Qop which adds to the set of alternatives in the indefinite the single alternative where none of these holds. • This yields something similar in spirit to a Groenendijk & Stokhof (1984) semantics6 • We see this interpretation in (20). (20)
!(19)" = { {w! : bill greeted joe in w! }, {w! : fred greeted joe in w! }, {w! : jane greeted joe in w! }, . . . {w! : no one greeted joe in w! } }
11
10
01
00
Qop
! "# $ ∃x.ϕ(x) ∨¬∃x.ϕ(x)
6
11
10
01
00
The difference being that Groenendijk & Roelofsen (2009)’s semantics is not a partition since the alternatives can overlap. In order for the Groenendijk & Roelofsen (2009)-style question to entail the indefinite, we would first have to remove the ‘No one’ alternative introduced by the Qop . One obvious option to apply the current account under this question semantics would be to apply existential closure to the question (as Merchant (2001) does). Another possible option would be to compute symmetric entailment over the entire question other than the Qop .
7
Sluicing as Anaphora to Issues
2. AnderBois (submitted) proposes that the existential presupposition of the question renders the indefinite’s proposal uninformative. • This yields roughly a Hamblin (1973) semantics for wh-questions. • Specifically, (19) is given the following semantics: (21)
a. b.
Presupposition of (19): !∃x.greet! (x,joe) At-issue proposal of (19): ∃x.greet! (x,joe)
11
10
11
10
01
00
∃x.ϕ(x)
Presup: !∃x.ϕ(x) 01
00
We opt for the second option since it allows us to define symmetric entailment for the question itself, though this assumption is not crucial.
2.5
Entailment
• Entailment in inquisitive semantics can be uniformly defined across questions and assertions as in (22) from Groenendijk & Roelofsen (2009). (22)
Definition: ϕ ! ψ iff ∀α ∈ !ϕ" : ∃β ∈ !ψ" : α ⊆ β
• The key notion, inclusion of alternatives, reduces to the standard notion of entailment for formulas which are classical. • For questions, this definition mirrors Groenendijk & Stokhof (1984)’s, differing in that question denotations need not be partitions (since the relations they denote is not always transitive).
8
Sluicing as Anaphora to Issues
3
Semantic isomorphy and sluicing • Following Merchant (2001), we take sluicing to be subject to a symmetric entailment condition between the A and E clauses.7 • What is new is that our notion of entailment comprises not only truth conditional content, but also inquisitive content. • Since the E clause is always a question, symmetric entailment dictates that A clause must also be inquisitive.
Given the semantics in §2, this straightforwardly accounts for puzzles (i) and (iii) — disjunctions as inner antecedent and the effect of double negation. Indefinite: (23)
[Someone left]A , but I don’t know [who left]E
(24)
a. b.
!(23)A " = ∃x.leave! (x) !(23)E " = Presupposes: !∃x.leave! (x), At-issue: ∃x.leave! (x)
• As described in §2, !(23)A " denotes a proposal to update the common ground with the set of alternatives of the form ‘x left’.
• The at-issue content of !(23)E " has the same proposed output state. • !(23)E " differs only in its informative presupposition: !∃x.leave! (x). Therefore, !(23)A "
!(23)E "
⇔
11
10
11
10
01
00
01
00
Disjunction: • Disjunctions serve freely as inner antecedents to sluicing as noted by Chung et al. (1995). • In our account, this follows from the fact that we treat indefinites as disjunctions of an unspecified number of ‘disjuncts’. 7
There also, of course, other conditions on ellipsis more generally (e.g. focus parallelism domains). We ignore these in what follows as they are not particular to sluicing.
9
Sluicing as Anaphora to Issues • The only difference is that the wh-word in these cases must be the D-linked which, since this allows for restriction to a finite contextually salient set.8 (25)
[(Either) John or Fred left]A . Tell me [which (one) left]E
(26)
a. b.
!(25)A " = leave! (j) ∨ leave! (f) !(25)E " = Presupposes: !∃x.x ∈ {j,f} ∧ leave! (x), At-issue: ∃x.x ∈ {j,f} ∧ leave! (x)
Therefore, !(25)A " ⇔ !(25)E "
i.e. leave! (j) ∨ leave! (f) ⇔ ∃x.x ∈ {j,f} ∧ leave! (x)
• We also see cases like (27) where which is required perhaps not for D-linking, but because the alternatives in it are propositional rather than individual.
(27)
John danced or played trombone, but I don’t remember which
Double negation: • Despite truth-conditional equivalence, double negated indefinites as in (28) (or the more colloquial (29)) do not license sluicing. • Since double negation has a semantic effect in our account (non-inquisitive closure), we expect the contrast between (28)-(29) on the one hand and (23) on the other. (28) (29) (30)
*[It’s not the case that no one left]A , but I don’t know [who left]E . *[It’s not the case that John didn’t meet with a studentA , but Fred still wonders [who John met with]E a. b.
!(28)A " = !∃x.leave! (x) !(28)E " = Presupposes: !∃x.leave! (x), At-issue: ∃x.leave! (x) Therefore, !(28)A "
!
!(28)E "
11
10
11
10
01
00
01
00
8
It seems that some special indefinites also have this property. For example, Spanish alg´ un ‘some’ (unlike un) can serve as the inner antecedent for sluicing, but only with the D-linked wh-word cu´ al. Thanks to Luis Vicente for discussion of these facts.
10
Sluicing as Anaphora to Issues
4
Passive Agents and Sluicing
As noted by Merchant (2001), an implicit passive agent cannot serve as the inner antecedent to sluicing as in (31). (31)
*[The cake was eaten]A , and I want to find out [who ate the cake]E
• This does not seem to simply be a fact about ellipsis in general, however, since VPE is (at least marginally) permissible in such cases as Merchant (2007) shows. (32)
The system can be used by anyone who wants to use it.
• As Merchant (2007) notes, this plainly cannot be due to truth-conditional semantics. • He therefore concludes that the source of the asymmetry between (31) and (32) is syntactic. • We explore a third possibility: this pattern is driven by a (non-truth-conditional) semantic difference.
4.1
The semantics of the implicit passive agent
The passive voice is notorious for allowing the speaker to avoid discussion of the agent/causer/subject of the event being described as in (33). (33)
The vase was broken.
This property does not intuitively hold of sentences with an overt indefinite subject like (34) or (35) (whether passive or active). (34)
Someone broke the vase.
(35)
The vase was broken by someone.
• Unlike an inchoative, though, (33) clearly entails the existence of an causer/agent. • That is, the truth-conditional content of (33) is identical to that of (34). In inquisitive semantics, however, the overt indefinite in (35) makes another semantic contribution: it raises the issue of what individual broke the vase. • The absence of this inquisitive contribution, we claim, distinguishes (34) from (35) semantically as in (36)-(37). (36)
The vase was broken " !∃x. break! (x,vase)
(37)
The vase was broken by someone " ∃x. break! (x,vase)
• This intuitive characterization is supported by the availability of secondary responses as in (38)-(40). • Regardless of voice, the overt indefinite licenses such responses while the implicit agent does not.
11
Sluicing as Anaphora to Issues
(38)
a. b.
Someone broke the vase. It was Fred // Yeah, Fred // Fred // Probably Fred
(39)
a. b.
The vase was broken by someone. It was Fred // Yeah, Fred // Fred // Probably Fred
(40)
a. The vase was broken. b.#?It was Fred // #Yeah, Fred // #Fred // #Probably Fred
4.2
Sluicing and implicit passive agents
These patterns of secondary responses mirror the availability of sluicing in (41)-(43)9 (41)
*[The cake was eaten]A , and I want to find out [who ate the cake]E
(42)
[The cake was eaten by someone]A , and I want to find out [who the cake was eaten by]E
(43)
[Someone ate the cake]A , and I want to find out [who ate the cake]E
Just as for indefinites and double negation in §3, this follows from the proposed semantics: Overt indefinite passive agent: (44)
[The cake was eaten by someoneA , but I don’t know [who ate the cake]E
(45)
a. b.
!(44)A " = ∃x.eat! (x, cake) !(44)E " = Presupposes: !∃x.eat! (x,cake), At-issue: ∃x.eat! (x,cake) Therefore, !(44)A "
⇔
!(44)E "
11
10
11
10
01
00
01
00
Implicit passive agent: (46)
*[The cake was eatenA , but I don’t know [who ate the cake]E
(47)
a. b.
!(46)A " = !∃x.eat! (x, cake) !(46)E " = Presupposes: !∃x.eat! (x,cake), At-issue: ∃x.eat! (x,cake) Therefore,
9 It should be noted that apparently parallel examples to (i) are grammatical in Malagasy (Potsdam (2007)). Our prediction, then, would be that implicit passive agents in the language do not show this sort issue-suppressing semantics (i.e. that they are inquisitive existentials). This prediction should, in principal, be independently testable with secondary responses.
12
Sluicing as Anaphora to Issues !
!(46)A "
!(46)E "
11
10
11
10
01
00
01
00
The account thus far, predicts that examples like (48) (active A clause, passive E clause) will be semantically well-formed. (48)
*[Someone ate the cake]A , and I want to find out [who the cake was eaten by]E
Accounting for such examples relies upon a fuller understanding of ‘Sprouting’ more generally. See §7 for discussion.
5
The semantics of appositives • Having solved the three problems we started out with — double negation, disjunctions, and implicit passives — we turn now to a new puzzle which we claim has a related solution: – Overt indefinites inside of appositives do not license sluicing.
First, we provide an independent argument for the semantic status of appositives in discourse.
5.1
Appositive and at-issue content
• Since Potts (2005), it has been widely agreed upon that the content of appositives is, in some way separate from at-issue assertions. • (49)-(50) are truth-conditionally equivalent but do not have identical semantics. (49)
John, who nearly killed a woman with his car, visited her in the hospital.
(50)
John nearly killed a woman with his car and visited her in the hospital.
At the same time, there is growing evidence that the content of appositives and other similar elements is not wholly separate. • Presupposition — Nouwen (2007), Amaral et al. (2007), AnderBois et al. (this conference) • Anaphora — Nouwen (2007), AnderBois et al. (this conference) • NP and VP-Ellipsis - AnderBois et al. (this conference) Interim Conclusion: Appositive and at-issue content are fundamentally unidimensional (see Schlenker (to appear), Amaral et al. (2007), and AnderBois et al. (this conference) for discussion). 13
Sluicing as Anaphora to Issues
While appositive and at-issue content must be unidimensional, we still need to capture the separation of propositional content which motivated Potts (2005)’s multidimensional semantics. • The key intuition is that whereas an at-issue assertion proposes an update the Common Ground, appositives impose an update on the Common Ground. • Farkas & Bruce (2010) motivate this view of at-issue content empirically by the fact that assertions allow for the responses that polar questions expect (e.g. yes, no, maybe) • Appositives, though, do not readily allow for yes/no/maybe responses as seen by the following contrast: (51)
a. b. c.
A: Sonia, who is a terrible housemate, left the door unlocked last night. B: Yeah, but she is still a good housemate. B: No, but she is a terrible housemate.
(52)
a. b. c.
A: Sonia is a terrible housemate and she left the door unlocked last night. B: #Yeah, but Sonia is still a good housemate. B: #No, but she is a terrible housemate.
• AnderBois et al. (this conference) fleshes out this idea in terms of how the appositive’s content updates the common ground. • Here, we will pursue a related idea, that this difference in update procedures is closely related to their semantic content itself. Specifically, appositive content differs in that it is purely informational, not interacting with the questions/issues under discussion in the discourse. • An obvious illustration of this is that there are no appositive questions in spoken English. • Answering questions in an appositive is similarly odd. • This purely informative discourse status also predicts that appositive content is not subject to responses like ‘yes’ and ‘no’ in the same way that at-issue content is. Crucially for present purposes, by capturing this in terms of the semantics of issues, we can apply this notion to understand the restrictions on sluicing which appositives impose.
5.2
Appositives in Inquisitive Semantics
As discussed in §2 and in Groenendijk & Roelofsen (2009), one key impetus for inquisitive semantics is the idea that at-issue assertions are proposals. • The logical language captures this by modeling at-issue assertions, like questions, as being of type stt. • As we have seen, however, being of type stt is also the exact feature of the logic that allows for inquisitivity. • This is because inquisitivity arises because a formula denotes a set consisting of multiple alternatives of type st. 14
Sluicing as Anaphora to Issues
Result: Being a proposal and being potentially inquisitive are inextricably linked in Inquisitive Semantics. • Appositive are not proposals and they are not potentially inquisitive and therefore should be of type st; they are classical propositions. • Compositionally, we capture through the following comma operator: (53)
!comma(ϕ)" = {w | there is some α ∈ !ϕ" s.t. w ∈ α }
• Locating the appositive meaning in the comma intonation makes sense since this is the unique element in their form which distinguishes them. • Inquisitive content within an appositive behaves the same (e.g. embedded questions inside appositives are possible) • This type shift is motivated by the desire to understand the discourse status of appositives and given the inquisitive semantic framework we adopt, the lack of inquisitivity follows. In §6, we use this lack of inquisitivity combined with the symmetric entailment licensing condition to explain the inability of sluicing with inner antecedents in such environments.
6
Backgrounded content and sluicing
Anaphora and ellipsis in general can freely cross the appositive/at-issue boundary, as in (54) (Nouwen (2007), Amaral et al. (2007), AnderBois et al. (this conference)). (54)
a. b. c. d.
Jones, who graded each student’s final paper, gave them detailed feedback. John, who might give a presentation, would use slides. Bill would just use the board. Maryi , who doesn’t [help her sister]A , told Janej to [help heri/j sister]E . Jane played three [games of boggle]A with Jos´e, who also played two [games of boggle]E with his mother.
• The sluicing data below, then, contrast to other work on discourse parallelism constraints on ellipsis (e.g. Hardt & Romero (2004), Frazier & Clifton (2006)) which also affect VPE.
6.1
Sluicing and Appositives
While appositives do not impede anaphora and ellipsis in general, overt indefinites within appositive relative clauses, as seen in (55)-(57), cannot serve as the inner antecedent for Sluicing. (55)
a. *?Joe, who once killed a man in cold blood, doesn’t even remember who. b. Joe, who once killed a man in cold blood, doesn’t even remember who it was. c. Joe, who once killed a man in cold blood, doesn’t even remember who he killed.
(56)
a. *?The valiant knight, who once defeated a masked enemy, still wonders who. b. The valiant knight, who once defeated a masked enemy, still wonders who it was. c. The valiant knight, who once defeated a masked enemy, still wonders who he defeated.
15
Sluicing as Anaphora to Issues
(57)
a. *?Amy, who coined a new word during last night’s Scrabble game, forgot what. b. Amy, who coined a new word during last night’s Scrabble game, forgot what it was. c. Amy, who coined a new word during last night’s Scrabble game, forgot what word she coined.
• The (b)-(c) examples show that the problem is with the ellipsis itself, not some independent source. This contrast is further supported by naturally occurring examples like (58) (from the Corpus of Contemporary American English, www.americancorpus.org) which become infelicitous if we remove ‘it was’. (58)
My sister did not kill herself, which means someone else must have done it, and I intend to discover who *(it was).
We can also see the same contrast in examples where the clause containing the indefinite is embedded within the appositive as in (59). (59)
Elizabeth, who thinks that Joe murdered a man in cold blood, wants to find out who *(it was).
Generalization: Sluicing is ungrammatical if the prospective inner antecedent is in an appositive. • This generalization follows from the inquisitive semantic licensing condition on sluicing we have proposed . . . • . . . combined with the alternative-suppressing semantics for appositives. • The Comma operator ensures that the clause containing the indefinite will enter the CG via a purely informational update. • Given this, it cannot entail the inquisitive E clause and sluicing is predicted correctly not to be licensed10 : (60)
*Joe [, who once killed a man in cold blood, ]A , doesn’t even remember [who he killed in cold blood]E .
(61)
a. b.
!(60)A " = Comma(∃x. kill! (Joe, x)) !(60)E " = Presupposes: !∃x. kill! (Joe, x)), At-issue: ∃x.kill! (Joe, x))
Therefore, !(60)A " " !(60)E "
since Comma(∃x.kill! (Joe, x)) ! ∃x.kill! (Joe, x)
The condition on sluicing which we propose is in essence a condition on the anaphora-like retrieval of the issue which the inner antecedent introduces. • Since issues within the Comma operator do not exhibit any special behavior, we correctly predict that retrieving a sluice inside an appositive will be felicitous as in (62). In a technical sense, entailment as defined in §2.5 is not even defined for appositive content since they are of different types. We can fix this by defining entailment for elements of type st in terms of the entailment properties of the singleton sets containing them. 10
16
Sluicing as Anaphora to Issues
(62)
[Someone left the door open]A . Jamie, who wants to find out who [left the door open]E , is surveying the likely culprits.
• Contrasting the current account with the other potential accounts mentioned in the introduction, we see that none of the others can capture the ungrammaticality of (60). Syntactic: The appositive is straightforwardly identical in its lexical material and syntactic form. Truth-conditional: The truth-conditions of the appositive clause are the same as the wouldbe elided clause. Dynamic Semantic: Appositives allow indefinites within them to establish discourse referents just like at-issue content. Whereas the first three puzzles could be solved by reference to syntactic form and drefs (though neither one alone could account for all three), overt indefinites inside appositives require a semantic account referencing issues.
6.2
Issues in VPE
In the preceding section, we accounted for the ungrammaticality of Sluicing which finds its antecedent inside of an appositive relative clause. • The account given, however, only made indirect reference to Sluicing by referring directly to symmetric entailment over inquisitive content. • Since the elided clause in Sluicing is interrogative, this correctly predicts that Sluicing will be ill-formed. • It also, however, predicts that other ellipsis processes such as VPE cannot contain inquisitive content when their antecedents are inside of appositives. • That is, we predict that the indefinite in a case of VPE like (63), unlike ordinary overt indefinites, ought to not be inquisitive.11 (63)
Joe, who [murdered a man in cold blood]A , convinced Bill to [murder a man in cold blood]E too.
That is, we expect the semantics of the elided clause to be as in (64a) rather than (64b). (64)
a. b.
!(64)E " = λy.!∃x.murder! (y, x) !(64)E " = λy.∃x.murder! (y, x)
⇐Predicted
• Looking at our intuitions alone, it is not clear whether or not the elided indefinite is inquisitive. • However, there are two ways we can see that this prediction is upheld.
11
Thanks to Jeroen van Craenenbroeck for insightful discussion of this point and the data pertaining to it.
17
Sluicing as Anaphora to Issues
First, If we make the elided clause itself interrogative as in (65), we find that the results are still ungrammatical. (65)
*Joe, who once killed a man in cold blood, doesn’t even remember who he did.
(66)
*The valiant knight, who will defeat a masked enemy at sunrise, is trying to figure out who he will.
• One potential objection to this argument is that corresponding examples without an appositive are also ungrammatical as in (67). (67)
*Joe once killed a man in cold blood and he doesn’t even remember who he did.
• However, the standard explanation for cases like (67), is that they are ungrammatical because Sluicing would have been possible instead, . . . • . . . and a general principle, MaxElide, derives the ungrammaticality of (67) since more material could have been elided (Takahashi & Fox (2005), Merchant (2008), Hartman (2009) inter alia) In appositives, however, MaxElide does not seem to be a possible explanation since Sluicing itself is ungrammatical. Second, We can show that the elided VP of such examples is not a possible antecedent for Sluicing. • Before we can see this, we first must note that an elided VP with an indefinite is, in principle, a possible licensor of Sluicing as in (68). (68)
John met with a student yesterday. Jane did too, but she can’t remember who [she met with yesterday].
• In contrast, if the antecedent of the VPE is inside an appositive, the clause containing the elided VP can no longer serve as antecedent to Sluicing as in (69). (69)
*John, who met with a student yesterday, convinced Jane to too, but she can’t remember who [she met with yesterday].
Putting together these two pieces of evidence, we see that VPE is subject to the same inquisitive entailment condition as Sluicing. • Sluicing from an appositive is always ruled out by this principle since the E clause is necessarily interrogative. • For VPE, the ellipsis is ruled out only in case the elided clause is made inquisitive, as in the above examples. Summing up, the ellipsis data from appositives highlights the sense in which ellipsis truly is an anaphoric process. • An account that simply compared the logical form of the antecedent to the elided clause, whether syntactically or semantically, would be unable to account for such facts. • The content of the A clause is not what determines the ungrammaticality of these examples. • Rather, it is the fact that their material entered the common ground appositively which is responsible. 18
Sluicing as Anaphora to Issues
6.3
Beyond appositives?
We can use the ungrammaticality of Sluicing as a diagnostic for content which is backgrounded in the relevant sense (provided that the environment in question does license VPE). • According to this diagnostic, then, we find that the following environments to involve backgrounding: Absolute Adjuncts: — (70)
a. *Having defeated a masked enemy, the valiant knight wondered who. b. Having defeated a masked enemy, the valiant knight wondered who it was.
(71)
Having defeated a masked enemy, the valiant knight’s comrades were inspired to defeat a masked enemy too
Conditional Antecedents — (72)
a. *If a strange man comes into the bar, the bartender will wonder who. b. If a strange man comes into the bar, the bartender will wonder who he is.
(73)
If a strange man comes into the bar, his girlfriend probably will come into the bar too.
Unconditional Antecedents — (74)
a. #Whichever man comes to the party, John will know who. b. Whichever man comes to the party, John will know who he is.
(75)
Whoever comes to the party, Jane will want to come to the party too in order to meet them.
Are these prima facie reasonable predictions? Yes! • Consider, for example, Stalnaker (1968)’s two step dynamic theory of conditionals. • First the antecedent is added to the c.g. hypothetically, then the consequent is proffered relative to this updated c.g. • There is an asymmetry, however, between the two steps. • Whereas the consequent’s update can be of nearly any speech act type, the antecedent’s update cannot. Absolute adjuncts and unconditionals are similar enough to conditionals (as argued, e.g. by Rawlins (2008)) that they hopefully can be thought of along the same lines.
19
Sluicing as Anaphora to Issues
7
Preliminary thoughts on sprouting
In the preceding sections, we have considered the core cases of Sluicing, those where there is an overt indefinite/disjunction to serve as inner antecedent. • In this section, we present some preliminary thoughts on the subclass of Sluicing in (76), known since Chung et al. (1995) as ‘Sprouting’. • Such examples have a wh- word in the ellipsis clause, but, unlike the cases considered above, no corresponding inner antecedent. (76)
a. b. c.
John is coming to the party, but I don’t know {when/why/on which bus}. Fred ate, but I don’t know what. They’re jealous, but it’s unclear of who. Chung (2006)
• In (76a), the sprouted element is an adjunct; in (76b) and (76c), it is an optional internal argument. • Given the absence of an overt indefinite/disjunction in the A clause, such examples are perhaps surprising for the current account. While a full account of Sprouting is beyond the scope of the present work, I would like to sketch one direction such an account might go.12 • The account we have proposed relies crucially upon the presence of inquisitive material in the A clause. • It is perhaps intuitively plausible that an issue like when John is coming is somehow raised by the A clause in (76a). • It seems far less plausible that the issue of which bus John took is somehow highlighted by the sentence ‘John is coming to the party.’ While these specific issues individually are not raised, we would like to claim that a more general issue is plausibly raised. • When I utter a sentence like (77), I leave open (or at least do not explicitly close off) the issue of the details of the how the leaving event transpired. • One empirical confirmation of this are the secondary responses in (78) (77)
John left.
(78)
Yeah, yesterday. // Yeah, on a jet plane // Yeah, to go buy milk.
• One way to formalize this is with an inquisitively existentially quantified neo-Davidsonian event argument: 12 It should emphasized that such a semantic account does not necessarily need to be responsible for all of the fine-grained restrictions to which sprouting is subject. For example, Chung (2006) notes that the grammaticality of (76c) relies crucially on the presence of the preposition of. It may well be that accounting for such facts will require an account like Chung (2006)’s which contains both a semantic licensing requirement and a lexico-syntactic one.
20
Sluicing as Anaphora to Issues
(79)
∃e.leave! (e) ∧ Agent(e, john)
(As opposed to: !∃e.leave! (e) ∧ Agent(e, john))
Under this view, Sprouting is somewhat analogous to bridging in the realm of definite descriptions. • The issue raised by the antecedent clause is a very general one. • The remnant material in the elided clause picks out a subissue of this larger issue. • As in the case of bridging definite descriptions like (80), if the subissue is not contextually salient/plausible, Sprouting is less felicitous. (80) #A car drove by. The fuzzy dice were pink. (81) #John bathed, but I don’t know which child. • This additional bridging step in Sprouting is consistent with processing work which has found that Sprouting in general is more costly to process than Sluicing with an inner antecedent (Frazier & Clifton (1998) et seq.). • This processing penalty is present both in the case of optional arguments and adjuncts.
8
Conclusion
To sum up, we have examined four puzzles in Sluicing which pose a challenge to a truth-conditional isomorphy condition: (i) doubly negated indefinites, (ii) implicit passive agents, (iii) disjunctions, and (iv) indefinites inside of appositives. • We have argued that all four puzzles can be solved under a semantic licensing account which makes reference not only to truth conditions, but also to issues. • The account, then, provides an empirical argument that alternatives of the sort posited for indefinites by Kratzer & Shimoyama (2002) not only exist within the sentence, . . . • . . . but are part of the Context Change Potential of indefinites as Inquisitive Semantics holds. While discourse referents and issues are both part of a sentence’s CCP and are both tied to indefinites and disjunctions, neither one is reducible to the other. • Previous syntactic (e.g. Chung et al. (1995)) and semantic (e.g. Merchant (2001)) accounts hold that indefinites are inner antecedents because they introduce discourse referents. • The data from double negation and appositives suggest that this cannot be so. • Indefinites in either environment introduce discourse referents, yet do not license Sluicing. In the present work, we have argued that issues are the crucial property of inner antecedents. 21
Sluicing as Anaphora to Issues
References Alonso-Ovalle, Luis (2006) Disjunction in Alternative Semantics. Ph.D. thesis, University of Massachusetts. Amaral, Patricia, Craige Roberts, & E. Allyn Smith (2007) Review of The Logic of Conventional Implicatures. Linguistics and Philosophy 30: 707–749. AnderBois, Scott (submitted) Focus and Uninformativity in (Yukatek Maya) Questions, ms., UCSC. Submitted to NLS. AnderBois, Scott, Adrian Brasoveanu, & Robert Henderson (this conference) Crossing the Appositive/At-issue Meaning Boundary, presentation at SALT 20. Chung, Sandra, William Ladusaw, & James McCloskey (1995) Sluicing and Logical Form. Natural Language Semantics 3: 239–82. Chung, Sandy (2006) Sluicing and the Lexicon: The Point of No Return. In BLS 31: General Session and Parasession on Prosodic Variation and Change, 73–91. Ciardelli, Ivano (2009) Inquisitive Semantics and Intermediate Logics, mSc Thesis, University of Amsterdam. Farkas, Donka & Kim Bruce (2010) On reacting to assertions and polar questions. Journal of Semantics 27(1): 81–118. Frazier, Lyn & Charles Clifton (1998) Comprehension of sluiced sentences. Language and Cognitive Processes 13: 499–520. Frazier, Lyn & Charles Clifton (2006) Ellipsis and Discourse Coherence. Linguistics and Philosophy 29: 315–346. Groenendijk, Jeroen & Floris Roelofsen (2009) Inquisitive Semantics and Pragmatics. In Proceedings of SPR 09. Groenendijk, Jeroen & Martin Stokhof (1984) Studies on the Semantics of Questions and the Pragmatics of Answers. Ph.D. thesis, University of Amsterdam. Hamblin, C. L. (1973) Questions in Montague English. Foundations of Language 10: 41–53. Hardt, Daniel & Maribel Romero (2004) Ellipsis and the Structure of Discourse. Journal of Semantics 21(4): 375–414. Hartman, Jeremy (2009) The Semantic Effects of non-A! Traces: Evidence from Ellipsis Parallelism, handout from GLOW 32, Nantes. Kratzer, Angelika & Junko Shimoyama (2002) Indeterminate Pronouns: The View From Japanese. In The Proceedings of the Third Tokyo Conference on Psycholinguistics, 1–25. Merchant, Jason (2001) The Syntax of Silence. Oxford University Press. Merchant, Jason (2007) Voice and Ellipsis, retrieved http://home.uchicago.edu/ merchant/pubs/voice.and.ellipsis.pdf. 22
4/28/08
from:
Sluicing as Anaphora to Issues
Merchant, Jason (2008) Variable island repair under ellipsis. In Topics in Ellipsis, 132–153. Nouwen, Rick (2007) On appositives and dynamic binding. Research on Language and Computation 5: 87–102. Potsdam, Eric (2007) Malagasy sluicing and its consequences for the identity requirement on ellipsis. Natural Language and Linguistic Theory . Potts, Christopher (2005) The Logic of Conventional Implicatures. Oxford University Press. Rawlins, Kyle (2008) Unconditionals: an investigation in the syntax and semantics of conditional structures. Ph.D. thesis, UCSC. Schlenker, Philippe (to appear) Supplements within a Unidimensional Semantics I: Scope. In Proceedings of Amsterdam Colloquium 2009. Simons, Mandy (2005) Dividing things up: the semantics of or and the modal/or interaction. Natural Language Semantics 13: 271–316. Stalnaker, Robert (1968) A Theory of Conditionals. In Studies in Logical Theory, 98–112. Stalnaker, Robert (1978) Assertion. In Syntax and Semantics 9. Takahashi, Shoichi & Danny Fox (2005) MaxElide and the Re-binding Problem. In Proceedings of SALT 15.
23
