Run-of-the-mill fragments, like (1B), have propositional meaning, containing a predicate that is identical to the one in the fragment’s antecedent. This correspondence is one of the crucial reasons why structural theories of ellipsis (like PF deletion or LF copying accounts) assume that the fragment should be related to its fully pronounced version in (1B′).

(1)

  • Who are you shouting at?

  • My sister.

  • I’m shouting at my sister.

This squib presents novel data where the same correspondence is not observed. In the novel data, the original predicate cannot form part of the fragment in its ordinary meaning, posing a puzzle for the representation of the ellipsis site. The relevant data are described in sections 1 and 2. Section 3 then shows that the puzzle cannot be solved with reference to a nonisomorphic underlier in the ellipsis site. Section 4 offers a solution in two steps. First, an account based on accommodation of the lexical content of the ellipsis site is sketched (section 4.1). This proposal is shown to be untenable in structural accounts of ellipsis due to violations of syntactic and lexical identity (section 4.2), and it is replaced by an account in terms of metalinguistic reference (mixed quotation) in the ellipsis site (section 4.3). A quotation-based structural account is also suggested to be more parsimonious than a nonstructural account of the data.

1 The Puzzle: Pom Pom Dialogues

In naturally occurring conversations, a fragment answering a wh-question can be followed by an adversative continuation that negates the verb. The negated verb is usually pronounced with focal stress and supplies a correction with respect to the original predicate.

(2)

    • Where are you running to?

    • To school, but I’m not running.

    • What are you devouring?

    • A pizza, but I’m not devouring it.

    • Who are you shouting at?

    • My sister, but I’m not shouting.

As the presumed nonelliptical versions of these fragments in (3) indicate, supplying the answers with the original predicate results in incongruence (indicated by #): the first conjunct asserts a proposition and the second conjunct denies it. This leads to contradictory statements.

(3)

    • Where are you running to?

    • #I’m running to school, but I’m not running.

    • What are you devouring?

    • #I’m devouring a pizza, but I’m not devouring it.

    • Who are you shouting at?

    • #I’m shouting at my sister, but I’m not shouting.

Importantly, while the answers in (3) contain incongruent responses to a question, the elliptical answers in (2) are not incongruent in the same way. They do not give the impression that the speaker is confused about the kind of event he was involved in or that he changes his mind halfway through his utterance.

Answers like those in (2) are natural responses to questions. As the Hungarian equivalent of (2a) is the recurring initial dialogue in the cartoon series Pom Pom meséi (Tales of Pom Pom, penned by István Csukás 1979/1980), this kind of question-answer pair will be referred to as Pom Pom dialogues. While the rest of the examples in this squib will mostly come from English, these dialogues are also natural in Dutch, Hungarian, Russian, Turkish, Japanese, and possibly other languages.

Pom Pom dialogues pose a puzzle for theories of ellipsis, as it is not evident what kind of predicate should be construed as part of the representation of the fragment. Assuming a structural approach according to which there is syntactic structure in the ellipsis site (a reason for this choice will be provided in section 3) and a lexicalist stance according to which structure is built from lexical items, this squib describes the puzzle and sketches a solution for it.

2 Properties of Pom Pom Dialogues

Pom Pom dialogues contain a question about an event that in the addressee’s view is being wrongly characterized by the questioner. They require a specific context that allows the addressee to identify the event the questioner is referring to.

The contested matter usually corresponds to the manner component of the predicate, which is indicated by the fact that speakers consulted consider Pom Pom answers natural with verbs that contain a manner component, such as manner-of-motion verbs (run, jog, rush, trot, stroll, march, hop), manner-of-speaking verbs (shout, cry, mumble, mutter, yell), and verbs of ingesting (devour, gobble, gulp, munch, nibble, gorge); see the examples in (2).1

Verbs without a manner component are perceived as somewhat less natural in Pom Pom dialogues, but perfectly acceptable given the right context and if the correct predicate is made explicit after the answer (such continuations are not needed in (2)). Virtually any kind of verb can be corrected this way.

(4)

  • Context: Speaker A sees someone lying on the floor motionless.

    • Who is sleeping there?

    • Frank, but he’s not sleeping, he’s dead.

  • Context: Speaker A, who is blind, hears a hissing/creaking noise.

    • What did you open?

    • The door, but I didn’t open it, I closed it.

  • Context: Speaker A hears speaker B talk negatively about some people.

    • Who do you hate most?

    • Sue. But I don’t hate her, I find her irritating.

A further condition on felicitous Pom Pom answers is that the antecedent verb needs to be backgrounded in the question; that is, it cannot form the focus of the utterance. Focused verbs cannot be given answers of this type.

(5)

  • The kids have eaten lots of things in different ways. What have they DEVOURED?

  • A pizza. #But they haven’t devoured it.

Note also that in addition to wh-questions, polar questions are suitable antecedents to contrastive Pom Pom fragments as long as they contain a contrastive element that is not the verb.

(6)

    • Are you running to SCHOOL?

    • No, to the PLAYGROUND. But I’m not running.

    • Are you shouting at your SISTER?

    • No, at YOU! But I’m not shouting.

Last, besides fragments, Pom Pom answers can contain VP-ellipsis (at least for the British English speakers consulted), provided the elliptical clause refers to a backgrounded verb.

(7)

  • Who has rushed in just now?

  • John has, but he hasn’t exactly rushed in.

3 It-Clefts Do Not Come to Our Rescue

Before providing a solution to the puzzle, we need to rule out the option that the fragments in (2) are well-formed because they underlyingly contain material that is not isomorphic to the antecedent clause. Ellipsis sites need not always be syntactically isomorphic to their antecedent; they may, for example, correspond to a copular clause containing a cleft structure (Van Craenenbroeck 2010, Barros 2012). Our puzzle, however, cannot be explained with reference to a nonisomorphic ellipsis site of this type. To start, paraphrasing the ellipsis site with a pseudocleft does not eliminate the problem of contradictoriness.

(8)

  • What are you devouring?

  • #What I’m devouring is a pizza, but I’m not devouring it.

The only theoretical possibility would therefore be to analyze the remnant as the pivot of a simple it-cleft, with no relative clause following it.2

(9)

  • What are you devouring?

  • A pizza

    graphic
    it is
    graphic
    .

However, this possibility can be eliminated as well, on the basis of the testimony of languages with overt case marking, like Hungarian or Russian. In these languages, pivots of clefts can only be nominative. The remnant in a Pom Pom answer, however, is always marked for the case of its correlate (the wh-phrase) and in cases where the correlate is accusative, the fragment cannot show up in the nominative (cf. (10B), (10B′)). In addition to this, Hungarian (just like Russian) has the property that it-clefts are not fully acceptable answers to wh-questions (thus the ?* mark in (10B″)), making it very unlikely that the source of the fragment is an it-cleft to begin with.

(10) Hungarian

  • Mit faltál fel?

    what.ACC devour.PST.2SG up

    ‘What have you devoured?’

  • Egy pizzát. De nem felfaltam, hanem

    a pizza.ACC but not up.devour.PST.1SG but

    szépen megettem!

    nicely PRF.eat.PST.1SG

    ‘A pizza. But I haven’t devoured it, I’ve eaten it nicely!’

  • #Egy pizza.

    a pizza.NOM

  • ?*Egy pizza volt az.

    a pizza.NOM was that

    ‘It was a pizza.’

That the remnant necessarily bears the case (accusative) assigned by a transitive verb is most readily accounted for by assuming that there is silent structure in the ellipsis site with a transitive lexical verb in it, as is done in structural accounts of ellipsis.3 Nonstructural approaches (also called direct interpretation approaches), not assuming syntactic structure and lexical material in the ellipsis site (see Stainton 1995, Culicover and Jackendoff 2005, Sag and Nykiel 2011, Ginzburg 2012), need a kind of case identity constraint to apply to the remnant and its correlate to derive case matching. As I consider structural accounts more parsimonious in this respect, in the rest of the squib I aim to develop a structural approach to Pom Pom fragments, in search of the lexical elements contained in the silent structure.

4 Possible Solutions

This section presents a solution in two steps. I first propose an account that is based on accommodation of the lexical content of the ellipsis site, which turns out to be untenable in structural accounts due to violations of syntactic and lexical identity. In light of this, I will opt for a solution in terms of mixed quotation in the ellipsis site.

4.1 Accommodation of Lexical Content

The accommodation solution is specifically designed for the most natural dialogues with manner verbs (recall (2)) and proposes that Pom Pom answers respond to an accommodated (in the sense of modified, repaired) antecedent, following observations that elided material can be reduced compared with its antecedent (first and foremost Thoms 2013, but see also, e.g., Johnson 2012, Rooryck and Schoorlemmer 2014, Den Dikken 2016). Thoms (2013) shows that elliptical material can ignore adjunct material in the antecedent, such as the manner adverb in (11a) or the expressive adjective in (11b).

(11)

  • Britain would blindly follow America into Afghanistan, without questioning why

    graphic
    Britain would follow America into Afghanistan
    graphic
    .

    (Thoms 2013:562, (9b), adapted)

In these examples, the italicized adjuncts are missing from the ellipsis site, due to a pragmatically induced accommodation step, which is limited by the requirement that the altered antecedent be logically entailed by the original one (see Thoms 2013 for details).

Following the spirit of Thoms’s analysis, we can design an account in which the manner component of a verb is ignored by the ellipsis in Pom Pom answers by assuming that these answers do not answer the explicit question being asked, but instead answer an implicit question under discussion (QUD; see Reich 2007 and Weir 2014 for the QUD anaphoricity of fragments). This QUD is reduced compared with the original question in that it contains a less specific verb that must be entailed by the original verb. For example, in (2a) the QUD is not Where are you running to? but Where are you going to?, which contains a modified (“weakened”) predicate as compared with the original one. The manner component is ignored because, by providing the answer to the weaker question, the addressee still satisfies the communication goal of the questioner, which is to find out about the value of the wh-variable in the question.

To formalize this intuition, I borrow the notions at-issue and non-at-issue content from the literature (see, e.g., Simons et al. 2010, AnderBois 2011, Tonhauser 2012, AnderBois, Brasoveanu, and Henderson 2013). At-issue content is content with which the speaker contributes directly to the common ground and to which she expects her communication partner to respond, via answering a question or acknowledging the content of an assertion. Non-at-issue content finds its way into the common ground indirectly, with no response expected.

For the Pom Pom dialogues in (2), it can thus be proposed that the hearer, B, dissociates the two kinds of content in the original question: the at-issue content that corresponds to the weakened verb (entailed by the original predicate) and the non-at-issue manner ingredient that corresponds to an implicature, as shown in (12) for (2b).

(12)

  • At-issue: λx.∃e. EAT (e) ∧ AGENT (e, you)

    THEME (e, x)

  • Non-at-issue: manner-of-devouring (e)

The at-issue content is the QUD and is answered by the fragment. The non-at-issue content is not part of the QUD, and this is what the correction after the fragment responds to.

(13)

  • Fragment: (a pizza) λx.∃e ∧ EAT (e) ∧ AGENT (e, I)

    THEME (e, x)

  • Correction: ¬ manner-of-devouring (e)

Crucially, the splitting of the two types of content is only permitted if the non-at-issue content is responded to in the utterance: without a correction, the fragment cannot be interpreted as a weaker proposition than the antecedent. In other words, accommodation must be signaled via linguistic means (Fox 2000). The linguistic items doing the signaling must minimally contain the negated verb and preferably a contrastive particle like but.

(14)

  • What are you devouring?

  • A pizza. #(But) I’m not devouring it.

In sum, the gist of this first account is that (a) the original verb’s denotation—specifically, the extra meaning component in devour as compared with eat, for example—is part of the non-at-issue content and (b) ellipsis can ignore this content. The latter statement is not new: Potts et al. (2009) state that non-at-issue content represented in the form of expressives can be ignored by ellipsis identity (see (11b)). Pom Pom dialogues make it seem that non-at-issue content can also be ignored by ellipsis if that content is not tied to an expressive.

4.2 Problems of This Account for Structural Approaches

While the proposal in section 4.1 links the facts to other types of accommodation observed in ellipsis, it runs into nonevident problems when it comes to the representation of the ellipsis site. Assuming that there is structure in the ellipsis site, and that structure is projected from lexical elements, the proposal boils down to using novel lexical material (e.g., eat instead of devour) in the syntactic representation of the ellipsis site. This brings several problems with it, as I demonstrate in this section. In some cases, the required verbs do not exist as independent lexical items. When they do, their argument structure, or their selectional/case properties, might be distinct from those of their antecedent. Moreover, all examples (sometimes multiply) violate the lexicosyntactic identity condition on ellipsis called No New Words.

To start the discussion with the last problem, No New Words is a condition on ellipsis identity banning novel lexical material from the ellipsis site, originally stated by Chung (2006:83, (29)) as (15).

(15) Every lexical item in the numeration of the sluice that ends up (only) in the elided IP must be identical to an item in the numeration of the antecedent CP.

Chung proposes (15) as a constraint ruling out sluicing with an elided novel preposition or a noun (e.g., *They are jealous, but it is unclear who

graphic
they are jealous of
graphic
, *She read, but we are not sure by which author
graphic
she read something
graphic
), stating that the condition might apply to ellipsis in general. In some works (see Merchant 2013a,b), this condition indeed figures as a general identity condition on ellipsis, and is used to explain the ungrammaticality of voice mismatches and argument structure mismatches inside ellipsis sites.4

How severely Pom Pom answers violate No New Words depends on the example. For (2a) and (2b), we can supply go instead of run and eat instead of devour, introducing only one new lexical item. In the case of (2c), however, the “weakened” verb—most likely talk or speak—differs in selectional properties from the original shout as talk and speak take a PP argument rather than a DP argument in English.

(16)

  • Who are you shouting at?

  • My sister

    graphic
    I’m talking/speaking to
    graphic
    , but I’m not shouting.

This violates No New Words twice, as both talk/speak and to are novel lexical items. This problem is aggravated in Hungarian, where supplying speak instead of shout would entail distinct cases on the internal argument: üvölt ‘shout’ assigns sublative, but beszél ‘speak’ assigns allative. Case connectivity, however, requires the fragment to be in the original case, sublative, ruling out the speak-type verb beszél (with allative) in the ellipsis site.

(17)

  • Ki-re üvöltöttél?

    who-SUBL shout.PST.2SG

    ‘Who were you shouting at?’

  • A húgom-{ra/*-hoz}. De nem üvöltöttem!

    the sister.POSS.1SG-SUBL/ALL but not shout.PST.1SG

    ‘My sister. But I wasn’t shouting!’

Hungarian in fact possesses no lexical verb with the right semantics and the right case-assigning properties to fill the elliptical gap.

Extending the account to verbs without a manner ingredient (see (4)) is problematic for various reasons. First, it is not immediately clear how to define the non-at-issue content in these cases, as these verbs are not specifying a particular manner (rather, they are specifying a result, a psychological state, or the like). Second, finding an appropriate replacement can be difficult. While hate can be replaced by dislike in (4c), the other two examples are problematic. In (4a), the at-issue content can possibly be defined by weakening sleep to lie,on the (questionable) assumption that sleep is a positional verb and entails lie.

(18)

  • Original question: λx.∃e. SLEEP (e) ∧ AGENT (e, x)

    At-issue (QUD): λx.∃e. LIE (e) ∧ AGENT (e, x)

Another option is to weaken it to the contentless copula be (A: Who is sleeping there? B: Frank

graphic
is there
graphic
, but he isn’t sleeping, he’s dead), but this would involve manipulating the original verb’s argument structure, not allowed in elliptical constructions elsewhere (e.g., Johnson 2001), as be, unlike sleep, is an unaccusative verb.

In (4b), we face another problem. Extracting the result ingredient from open leaves behind a highly bleached verb—some kind of activity verb with a theme, like transitive affect—in the at-issue content.

(19)

  • Original question: λx.∃e. OPEN (e) ∧ AGENT (e, you)

    THEME (e, x)

  • At-issue (QUD): λx.∃e. AFFECT (e) ∧ AGENT (e, you)

    THEME (e, x)

The problem with affect, however, is that it belongs to the technical/scientific register; it is not used in this meaning in colloquial English. It is unlikely that we can find the right lexical verb for this case, similarly to the problem noted for (17).

It must therefore be concluded that accommodation of the verb’s lexical content cannot provide the right solution within a structural approach to the data. Accommodation of the QUD brings with it several problems: difficulties with finding an appropriate replacement, argument structure mismatches and selectional/case mismatches between antecedent and ellipsis site (unattested in elliptical constructions elsewhere), and violation of the No New Words condition. Note that the accommodation account would be a viable solution in nonstructural approaches to fragments, where the interpretation of the fragment is the result of composing it with the semantics of the accommodated QUD. Since these accounts do not assume the existence of any lexical verb in the ellipsis site, they do not face the problems listed above.

4.3 Ellipsis with Mixed Quotation

It turns out that a perfectly viable account can also be designed within structural approaches to ellipsis if we allow the ellipsis site to contain the corrected material in its metalinguistic meaning.5 In other words, the corrected element appears in the ellipsis site with metalinguistic reference, quoted from the speaker of the antecedent question. The other elements in the clause are used in their ordinary (nonmetalinguistic) sense. The result is that the elliptical clause contains mixed quotation (so-called by Davidson (1979), who defines it as a mix of direct and indirect discourse syntactically). For (2b), we then have the following underlying representation, where quotation is indicated by quotation marks.

(20)

  • What are you devouring?

  • A pizza

    graphic
    I’m “devouring”
    graphic
    , but I’m not devouring it.

A welcome result of this analysis is that now the ellipsis site is syntactically completely identical to the antecedent question, as mixed quotation always fully shares the syntax of the quoted constituent, concerning syntactic category, argument structure, and case-assigning properties of the quoted verb as well.6

When it comes to semantic representation, the ellipsis site differs from the antecedent in that the semantic contribution of the quoted verb is not the verb’s original denotation; rather, using the free relative paraphrase offered by Maier (2014), it is whatever the questioner referred to with the verb in the antecedent. This has the effect that the speaker uttering the answer is not committed to the questioner’s assertion; that is, he does not endorse the questioner’s characterization of the event when it comes to the verb’s lexical content.

Applying Maier’s (2014) presupposition-based analysis of mixed quotation to the data (but see Potts 2007 for a different solution), we can say that the use of “devour” in the ellipsis site triggers the presupposition that the interlocutor used the quoted item to refer to a two-place relation R, while the assertion in the question, which Maier (2014) treats as the at-issue content, only contains reference to the relation R.

(21)

  • At-issue: What are you R-ing?

  • Presupposition: The interlocutor used the word devour to express R.

This formulation is advantageous, as it captures even the most extreme cases of Pom Pom dialogues, where the meaning of the original verb needs to be “stripped down” to express only some basic relation between the arguments of the verb (see the treatment of (4a) and (4b) in (18) and (19)).

The mixed-quotation account in fact goes one step further and predicts that the antecedent verb need not even have any evident meaning. Since the semantic interpretation of a quotation is computed not in terms of the constituent expression, but in terms of its phonological interpretation, uninterpretable or nonexistent expressions can be quoted as well. This is also the case in Pom Pom dialogues. Imagine that A sees B eating something with audible enjoyment and makes up the verb shmoltz on the spot to describe the event, as in (22A). B can answer this question without having ever heard this verb before, to refer to whatever the questioner intended with it. Crucially, B must be able to figure out from the context that A is referring to her eating something when asking the question.

(22)

  • What are you shmoltzing?

  • A pizza. But what do you mean by “shmoltzing”?

(23)

  • At-issue: What are you R-ing?

  • Presupposition: The interlocutor used the word shmoltz to express R.

The quotation account thus also neatly explains why contextual clues are so very relevant for the felicity of Pom Pom answers: as the at-issue content only contains reference to a relation, it is the context that allows the hearer of the question to identify what relation the question is about.

The proposal made here is tantamount to saying that ellipsis is a strategy that allows the utterer of a fragment not to take responsibility for the choice of words that appear in the ellipsis site. Ellipsis is not the only strategy to achieve this. The addition of an as-parenthetical, as in (24), can serve the same function in a nonelliptical answer. Note that this example (unlike (3b)) is noncontradictory.

(24)

  • What are you devouring?

  • I am—as you put it—devouring a pizza, but I’m not devouring it.

A core ingredient of the above account (similar to the accommodation proposal in section 4.1) is that the answerer dissociates two kinds of content in the original question: the at-issue content that asks for the identification of the wh-variable; and the non-at-issue content—namely, the presupposition introducing the quoted string. The at-issue component is answered by the fragment, and the presupposition is commented on by the correction. This kind of dissociation seems to be possible under ellipsis, but only if the antecedent verb is not the focus of the utterance, as mentioned with reference to (5). Why focus has this effect is not clear, but it can be noted that disavowing responsibility for particular words is very odd in contrastive focus contexts in general, even in nonelliptical answers (see (25)), something that supports the proposal (I thank an anonymous reviewer for noting this).7

(25)

  • The kids have eaten lots of things in different ways.

    What have they DEVOURED?

  • They have—as you put it—devoured a pizza. ?#But they haven’t devoured it.

On balance, the above-presented metalinguistic account for Pom Pom fragments is the most feasible account within a structural framework of ellipsis, because it assumes complete syntactic identity for the ellipsis site and makes use of metalinguistic reference (a phenomenon independently needed) only in its explanation. In this, the proposed structural account is arguably even more parsimonious than a nonstructural approach to the data would be, as the latter would need two (rather than one) assumptions to account for Pom Pom fragments: a case identity constraint (to explain facts like (10)) and some form of a QUD accommodation process, as described in section 4.1.

Notes

1 In an online judgment task, participants—25 British English and 36 Dutch native speakers—were instructed to evaluate the naturalness of the dialogues in (2) (or their counterparts in Dutch) on a scale from 1 to 5. On average, 70% of the participants considered the dialogues to be perfectly natural or only slightly unnatural (specifically, 72% of the British English–speaking and 68% of the Dutch-speaking participants gave the dialogues a rating of 4 or 5).

Speakers who do not readily rate these dialogues as natural are of two types. A small subset finds the answers strictly speaking contradictory. Another subset finds fragments like those in (2) odd, but accepts them if something mitigating is added to the correction, or if the original question contains more material, like a secondary predicate, as in (i) (I thank an anonymous reviewer for making me aware of these nuances).

    • Who was screaming from the rooftop naked earlier?

    • John, but he wasn’t screaming as such.

2 Here and below, the nonpronounced elliptical material is indicated with angled brackets. The precise position of the remnant being inconsequential, all examples follow displacement accounts in which the remnant moves out of the elided chunk (see Merchant 2004) without further arguments.

3 In addition to case matching, other types of connectivity, such as binding, scope connectivity, and voice connectivity, are also satisfied in Pom Pom dialogues. Similarly, the P-stranding generalization (Merchant 2004) is also observed in the languages I was able to check.

4 It must be noted that No New Words is not without exceptions: copulas and pronouns in nonisomorphic cleft clauses violate it, as do functional elements like infinitival to in examples such as Decorating for the holidays is easy if you know how

graphic
to decorate
graphic
(Merchant 2001). As a result, the condition has been criticized and replaced by other conditions in Barros 2014 and Barros and Vicente 2015. The reason why I consider violations of No New Words in Pom Pom answers as a problem here is that they represent cases in which a distinctly nonfunctional item (a lexical verb) is an elided new word in the ellipsis site. Such exceptions to the condition are not attested elsewhere.

5 I thank Andrés Saab and Lyn Frazier for suggesting this solution to me. See also Saab 2016, where this idea is applied to the example in (11b) and the semantic identity relation is characterized in terms of it.

6 An anonymous reviewer questions the metalinguistic character of the verb in the ellipsis site in this proposal, noting that quoted material normally receives focus-related accent and deleting focus is often said to be impossible. I do not agree with the premise of this assessment. First, quotation is not known to involve lexical focus on the quoted material (in addition, the verb in a Pom Pom answer is not new information focus, either, as it is given). Second, while a quotation (as an instance of reported speech) can be realized with pitch accent, such accenting is not necessary. Prosodic studies such as Klewitz and Couper-Kuhlen’s (1999) and Kasimir’s (2008) have shown that prosodic marking of a (mixed) quotation need not necessarily be done by pitch accenting. It can be done by a change in either speech rate or voice quality or by means of a leading/trailing pause (in addition to the option of no marking at all). Whether these aspects of prosodic marking are compatible with deletion has not been investigated to my knowledge.

7 Further support for the metalinguistic account comes from the observation that other parts of speech, such as nouns, adjectives, adverbs, and phrases formed by them can also be corrected in Pom Pom answers, since mixed quotations exist with such entities. This prediction is correct, as illustrated for a corrected noun phrase in (i).

  • Context: Speaker A missed a meeting and wrongly assumes it was held by the director.

    • What did the director announce yesterday in the meeting?

    • The new retirement plan. But it was the vice director.

I leave the investigation of these cases for future work.

Acknowledgments

I am indebted to Crit Cremers for numerous discussions on Pom Pom dialogues and for his valuable input on all aspects of the proposal. In addition, I thank two anonymous reviewers, the Squibs and Discussion editors of Linguistic Inquiry, as well as Lisa Cheng, Peter Culicover, Marcel den Dikken, Lyn Frazier, James Griffiths, Güliz Güneş, Jim McCloskey, Jason Merchant, Andrés Saab, Wataru Uegaki, Andrew Weir, and the audience at Fragments (Saarbrücken, 2016) for discussions, data, and comments on previous versions of the material. I am also grateful to Colin Ewen, James Griffiths, and Heidi Klockmann for sharing their intuitions about the data and to Charissa Jansen for her generous help with online data collection. All errors are mine. This research is funded by NWO (Netherlands Organisation for Scientific Research).

References

AnderBois,
Scott
.
2011
.
Issues and alternatives
.
Doctoral dissertation, University of California, Santa Cruz
.
AnderBois,
Scott
,
Adrian
Brasoveanu
, and
Robert
Henderson
.
2013
.
At-issue proposals and appositive impositions in discourse
.
Journal of Semantics
13
:
1
46
.
Barros,
Matthew
.
2012
. A non-repair approach to island sensitivity in contrastive TP ellipsis. In
Proceedings from the Annual Meeting of the Chicago Linguistic Society
, ed. by
Andrea
Beltrama
,
Tasos
Chatzikonstantinou
,
Jackson L.
Lee
,
Mike
Pham
, and
Diane
Rak
,
61
75
.
Chicago
:
University of Chicago, Chicago Linguistic Society
.
Barros,
Matthew
.
2014
.
Sluicing and identity in ellipsis
.
Doctoral dissertation, Rutgers University, New Brunswick, NJ
.
Barros,
Matthew
, and
Luis
Vicente
.
2015
. A remnant condition of ellipsis. In
Proceedings of the 33rd West Coast Conference on Formal Linguistics
, ed. by
Kyeong-min
Kim
,
Pocholo
Umbal
,
Trevor
Block
,
Queenie
Chan
,
Tanie
Cheng
,
Kelli
Finney
,
Mara
Katz
,
Sophie
Nickel-Thompson
, and
Lisa
Shorten
,
57
66
.
Somerville, MA
:
Cascadilla Proceedings Project
.
Chung,
Sandra
.
2006
. Sluicing and the lexicon: The point of no return. In
Proceedings of the Thirty-first Meeting of the Berkeley Linguistics Society
, ed. by
Rebecca T.
Cover
and
Yuni
Kim
,
73
92
.
Berkeley
:
University of California, Berkeley Linguistics Society
.
Craenenbroeck,
Jeroen van
.
2010
.
Invisible last resort: A note on clefts as the underlying source for sluicing
.
Lingua
120
:
1714
1726
.
Culicover,
Peter W.
, and
Ray
Jackendoff
.
2005
.
Simpler syntax
.
Oxford
:
Oxford University Press
.
Davidson,
Donald
.
1979
.
Quotation
.
Theory and Decision
11
:
27
40
.
Dikken,
Marcel den
.
2016
.
Three challenges for constituent ellipsis and their solutions
.
Paper presented at International Society for the Linguistics of English 4. Poznan, Poland, 19 September 2016
.
Fox,
Danny
.
2000
.
Economy and semantic interpretation
.
Cambridge, MA
:
MIT Press
.
Ginzburg,
Jonathan
.
2012
.
The interactive stance
.
Oxford
:
Oxford University Press
.
Johnson,
Kyle
.
2001
. What VP-ellipsis can do, and what it can’t, but not why. In
The handbook of contemporary syntactic theory
, ed. by
Mark
Baltin
and
Chris
Collins
,
439
479
.
Malden, MA
:
Blackwell
.
Johnson,
Kyle
.
2012
.
Towards a better E-givenness
.
Four invited lectures, Leiden University
.
Kasimir,
Elke
.
2008
. Prosodic correlates of subclausal quotation marks. In
ZAS papers in linguistics 49
, ed. by
Marzena
Zygis
and
Susanne
Fuchs
,
67
77
.
Berlin
:
ZAS
.
Klewitz,
Gabriele
, and
Elizabeth
Couper-Kuhlen
.
1999
.
Quote–Unquote? The role of prosody in the contextualization of reported speech sequences
.
Interaction and Linguistic Structure
12
:
1
34
.
Maier,
Emar
.
2014
.
Mixed quotation: The grammar of apparently transparent opacity
.
Semantics and Pragmatics
7
(
7
):
1
67
.
Merchant,
Jason
.
2001
.
The syntax of silence
.
Oxford
:
Oxford University Press
.
Merchant,
Jason
.
2004
.
Fragments and ellipsis
.
Linguistics and Philosophy
27
:
661
738
.
Merchant,
Jason
.
2010
. Three kinds of ellipsis. In
Context-dependence, perspective, and relativity
, ed. by
François
Recanati
,
Isidora
Stojanovic
, and
Neftali
Villanueva
,
141
192
.
Berlin
:
Walter de Gruyter
.
Merchant,
Jason
.
2013a
. Polarity items under ellipsis. In
Diagnosing syntax
, ed. by
Lisa Lai-Shen
Cheng
and
Norbert
Corver
,
441
462
.
Oxford
:
Oxford University Press
.
Merchant,
Jason
.
2013b
.
Voice and ellipsis
.
Linguistic Inquiry
44
:
77
108
.
Potts,
Christopher
.
2007
. The dimension of quotation. In
Direct compositionality
, ed. by
Chris
Barker
and
Pauline
Jacobson
,
405
431
.
Oxford
:
Oxford University Press
.
Potts,
Christopher
,
Ash
Asudeh
,
Seth
Cable
,
Yurie
Hara
,
Eric
McCready
,
Luis
Alonso-Ovalle
,
Rajesh
Bhatt
et al
, et al.
2009
.
Expressives and identity conditions
.
Linguistic Inquiry
40
:
356
366
.
Reich,
Ingo
.
2007
. Toward a uniform analysis of short answers and gapping. In
On information structure, meaning and form
, ed. by
Kerstin
Schwabe
and
Suzanne
Winkler
,
467
484
.
Amsterdam
:
John Benjamins
.
Rooryck,
Johan
, and
Erik
Schoorlemmer
.
2014
.
Mistaken identity: Ellipsis, mismatches and accommodation
.
Paper presented at the annual meeting of the Linguistic Society of America, Minneapolis, MN
.
Saab,
Andrés
.
2016
.
On bias vehicle change: Syntactic identity and semantic blocking in ellipsis
.
Ms., Conicet, Buenos Aires
. http://ling.auf.net/lingbuzz/002764
(accessed 17 July 2018)
.
Sag,
Ivan
, and
Joanna
Nykiel
.
2011
. Remarks on sluicing. In
Proceedings of the 18th International Conference on Head-Driven Phrase Structure Grammar, University of Washington
, ed. by
Stefan
Müller
,
188
208
.
Stanford, CA
:
CSLI Publications
.
Simons,
Mandy
,
Judith
Tonhauser
,
David
Beaver
, and
Craige
Roberts
.
2010
. What projects and why. In
Proceedings of Semantics and Linguistic Theory 20
, ed. by
Nan
Li
and
David
Lutz
,
309
327
.
Ithaca, NY
:
Cornell University, CLC Publications
.
Stainton,
Robert
.
1995
.
Non-sentential assertions and semantic ellipsis
.
Linguistics and Philosophy
18
:
281
296
.
Thoms,
Gary
.
2013
. Lexical mismatches in ellipsis and the identity condition. In
Proceedings of the 42nd Annual Meeting of the North East Linguistic Society
, ed. by
Stefan
Keine
and
Shayne
Sloggett
,
559
572
.
Amherst
:
University of Massachusetts, Graduate Linguistic Student Association
.
Tonhauser,
Judith
.
2012
. Diagnosing (not-)at-issue content. In
SULA 6: Proceedings of the Sixth Conference on the Semantics of Under-represented Languages in the Americas and SULA-Bar
, ed. by
Elizabeth
Bogal-Allbritten
,
239
254
.
Amherst
:
University of Massachusetts, Graduate Linguistic Student Association
.
Weir,
Andrew
.
2014
.
Fragments and clausal ellipsis
.
Doctoral dissertation, University of Massachusetts, Amherst
.