1. Introduction

Clausal ellipsis involves movement out of the ellipsis site plus deletion of a clause (Merchant 2004; van Craenenbroeck & Merchant 2013). Common types of clausal ellipsis are sluicing, fragment answers and bare argument ellipsis (also known as “stripping”). In sluicing, the constituent that moves out of the ellipsis site and survives ellipsis (the remnant) is a wh-phrase (who in (1)). Fragment answers are direct answers to questions (Pat in (2)). Stripping reduces the second conjunct in a coordinated structure, and requires a remnant (Chris in (3)) and one more element (usually not or also). The element that coordinates structures (usually but or and) is optional.

    1. (1)
    1. Someone here speaks Wampanoag – guess whoi [TP ti speaks Wampanoag].                       Sluicing
    1. (2)
    1. A:    Who speaks Wampanoag?
    2. B:    Pati [TP ti  speaks Wampanoag].                                                                            Fragment answer
    1. (3)
    1. Pat speaks Wampanoag, (but) not Chrisi [TP ti speaks Wampanoag].                                  Stripping

This squib focuses on a property well-known to clausal ellipsis – that is, the movement of the remnant in these ellipses can sometimes appear to be insensitive to island constraints (first observed by Ross 1969). I will show that stripping offers unique insights into why clausal ellipsis can sometimes appear to be island-insensitive, due to its unique properties. First, (4a–c) show the apparent island violations in clausal ellipsis. (4a–b) are examples from the literature, to which I add my own (4c):

    1. (4)
    1. a.
    1. They hired someone who speaks a Balkan language – guess which!
    2.                                                                                                 (Sluicing; Merchant 2001:209)
    1.  
    1. b.
    1. A: Does Abby speak the same Balkan language that someone in your class speaks?
    2. B: Yeah, Charlie.                         (Supplemental fragment;1 Griffiths and Liptak 2014:193)
    1.  
    1. c.
    1. They hired someone who speaks French yesterday, not German.                              Stripping

If we follow the standard analysis of clausal ellipsis, the remnant moves out of a relative clause, which is generally considered to be an island, and yet the result is fine (5a). Without ellipsis, it is not possible to move a wh-phrase out of a relative clause (5b).

    1. (5)
    1. a.
    1.   … guess whichi [tp they hired [Relative Clause  someone who speaks ti]]!
    1.  
    1. b.
    1. *… guess which they hired someone who speaks!

There have been many proposals for the seemingly exceptional island-violating behavior of clausal ellipsis (e.g. Ross 1969; Lasnik 2001 & 2005; Merchant 2001 & 2004; Fukaya 2007; Szczegielniak 2008; Abels 2011; Griffiths & Liptak 2014; Barros 2013; Barros et al. 2014; Yoshida et al. 2019). This squib presents evidence from stripping that supports Barros et al.’s approach, which was built on Merchant (2001), Fukaya (2007) and Abels (2011), and was originally proposed for sluicing. I extend their claim to stripping and clausal ellipsis in general, and add novel evidence based on new contexts (e.g., contradictory contexts). Following Barros et al., I argue that the apparent ability of stripping to violate islands is an illusion created by the evasion strategy – a parse that does not involve any islands. If all types of clausal ellipsis have the same underlying operations, the evasion strategy available to stripping should also be available to clausal ellipsis generally.

Section 2 provides four novel arguments that the evasion strategy saves island violations in stripping. Yoshida et al. (2019) explicitly rejected the evasion strategy as a source available to stripping. Subsection 2.5 responds to this critique, and argues that Yoshida et al.’s examples don’t allow the evasion strategy for independent reasons. Once these extraneous factors are controlled for, the evasion strategy is in fact available to stripping and can save island violations. Section 3 concludes the squib.

2. Arguments for the evasion strategies

Barros et al. (2014) have proposed many different evasion strategies. This squib argues for one type specifically – the short source, which was first proposed by Merchant (2001) for sluicing. The short source involves ellipsis of a subpart of the antecedent clause that doesn’t include the island, and creates the illusion of relative clause island violation in sluicing. Under this analysis, (4a) is fine because it has a parse whose elided phrase does not contain any island (6). In this parse the wh-remnant still moves out, but it does not cross any island. This parse contrasts with (5a), whose elided phrase is isomorphic to the antecedent. I call the isomorphic parse the long source. Furthermore, Merchant and Barros et al. (2014) suggested that the elided subject in the short source is an E-type pronoun in the sense of Evans (1980). This E-type pronoun refers to the indefinite in first conjunct, but is not in the scope of the indefinite.2

    1. (6)
    1. They hired someonei who speaks a Balkan language, guess which [TP shei speaks]]!

Assuming the short source is also available to stripping, the stripping sentence (7) should have two potential parses: the long source (8a), whose remnant German crosses a relative clause island, and the short source (8b), whose remnant does not cross any island.

    1. (7)
    1. They hired someone who speaks French yesterday, not German.
    1. (8)
    1. a.
    1. … not Germani [TP they hired [Relative clause someone who speaks ti]].                      Long source
    1.  
    1. b.
    1. … not Germani [TP she (= the person they hired) speaks ti]].                                   Short source

Barros et al. argued for the evasion strategies in sluicing by showing that when the evasion strategies are blocked (e.g., by replacing the indefinite someone with a negative quantifier or an NPI indefinite), ellipsis cannot appear to be island-insensitive any more. These arguments can be replicated here, but I will not repeat them here due to space. Instead, I will present four new arguments by taking advantage of some properties of stripping that sluicing and fragment answers do not have.3 While the long source and the short source for sluicing often have the same meaning ((5a) and (6)), I will show that stripping offers an opportunity to disambiguate them: once we manipulate the second conjunct carefully, we can get the short source to take on a different meaning from the long source. Crucially, in those cases, we get the meaning of the short source, suggesting that the short source is available.

All the judgments in this squib come from two informal surveys. In each survey, I asked native speakers to rate sentences on a scale of 1–7, and if they found them grammatical, I further asked them which reading(s) they could get. The first survey involved seven speakers, and later I carried out a follow-up survey with five out of those seven speakers. The order in which I present the sentences in this squib does not necessarily follow the temporal order in which I carried out the surveys. Thus, to be clear, for each sentence I will report the total number of speakers I asked and their judgments.

2.1. Evidence 1: Availability of the short source when there is no island

The discussion so far focuses on sentences that contain islands, specifically relative clause islands, which might lead us to think that the short source is only relevant to these sentences, but it does not have to be. This section argues that stripping sentences can have the short source, even when they do not contain any island. Consider (9).

    1. (9)
    1. They said fewer than three students speak French yesterday, and also German.

Following are the two possible parses of (9). The short source (10b) has a different meaning from the long source (10a).

    1. (10)
    1. a.
    1. … and also Germani [TP  they said fewer than three students speak ti].                      Long source
    1.  
    1. b.
    1. … and also Germani [TP  they speak ti].                                                                           Short source

The long source is compatible with a scenario where they talked about two different groups of students, French speakers and German speakers. The short source, on the other hand, is not compatible with this scenario, and means that they mentioned the same group of students who speak both French and German. Two out of my five consultants could get the short source reading, suggesting that at least for those speakers, the short source is available even when there is no island involved.

As to the long source reading, three out of my five consultants could get it. Though the long source is interesting in its own right, it is beside the point here. The point is to show that the short source reading is available, but not that the long source reading is.

The long source is generally available, but may require certain contexts, as Reinhart (1991:374) showed. For example, (11a) means that Lucie will not admit that she stole the car, not that Lucie will admit that she did not steal the car, suggesting that the remnant the car must have moved across the finite clause boundary (not [the car]i Lucie will admit that she stole ti).

    1. (11)
    1. a.
    1. Lucie will admit that she stole the diamonds if you press her, but not the car.
    1.  
    1. b.
    1. Lucie did not admit that she stole anything, when we pressed her, except the little red book.
    1.  
    1. c.
    1. More people said they will vote for Bush, in the last poll, than for Dukakis.           (Reinhart 1991:374)

I leave it to future research exactly which contexts bring out the long source reading easily, but instead focus on the availability of the short source in this squib.

2.2. Evidence 2: Availability of the short source when there is an island

This subsection continues the logic of the previous one by showing that when the short source has a different reading from the long source, we can get the reading of the short source. Consider (12), which differs from (9) minimally in that (12) contains an island.

    1. (12)
    1. They hired someone who speaks French yesterday, and also German.

I put the two potential parses of (12) below, which have different truth conditions:

    1. (13)
    1. a.
    1. … and also Germani [TP  they hired [RC someone who speaks tiyesterday]].                           Long source
    1.  
    1. b.
    1. … and also Germani [TP she (= the person they hired) speaks ti]].                                  Short source

The island-violating parse (13a) is true if they hired two monolingual speakers, a French speaker and a German speaker. The short source (13b) is only true if they hired a bilingual speaker who speaks both French and German.

Six out of seven speakers rated (12) 7 on a scale of 1–7. I further asked them how many people were hired. Everyone said only one bilingual person was hired, but one consultant said “maybe another interpretation is available, too,” where two monolinguals were hired, but this reading is harder than the interpretation of hiring a bilingual person.

Thus, nearly all the speakers can get the short source reading of (12), an indication that the short source is present. Because (7) is very similar to (12), we may infer that (7) also has the short source.

2.3. Evidence 3: Contradictory contexts

This subsection takes advantage of the observations in the previous subsections: when the sentence has no island, both the long source and the short source readings are available; but when the sentence contains an island, as in (12), the long source reading disappears for most speakers. This subsection blocks the short source with a contradictory context. In this context, the long source should be the only possible reading for a sentence that doesn’t contain any island. The sentence that contains an island can no longer appear to be island-insensitive.

Recall that the short-source readings in the previous subsections talk about bilingual speakers. We can thus block them with a contradictory context by using monolingual in the first conjunct. All my five consultants reported that the sentence that doesn’t contain any island (14) only has the long source (15a), presumably because the short source (15a) is contradictory.

    1. (14)
    1. They said fewer than three monolingual students speak French yesterday, and also German.
    1. (15)
    1. a.
    1. … and also Germani [TP  they said fewer than three monolingual students speak ti].                    Long source
    1.  
    1. b.
    1. #… and also Germani [TP  they speak ti].                                                                                              Short source

In contrast, the following sentence which contains an island was found to be ungrammatical by all my seven consultants:

    1. (16)
    1. *They hired a monolingual who speaks French yesterday, and also German.

Example (16) is bad because neither of its potential parses (17a–b) is possible. The long source (17a) is not available (for different reasons for different speakers: for those who could not get the long source reading for (9) and (14), because the long source reading is somehow not available to begin with, even if there is no island; for those who could get the long source reading for (9) and (14), due to the presence of the island). The short source (17b) is blocked by a monolingual in the antecedent.

    1. (17)
    1. a.
    1. *… and also Germani [TP  they hired [RCa monolingual who speaks ti]].                         Long source
    1.  
    1. b.
    1. #… and also Germani [TP she (= the person they hired) speaks ti].                               Short source

The following sentences make the same point but with different lexical items and PP objects.4

    1. (18)
    1. a.
    1. They need to find a monolingual speaker of Ch’ol for this course, and also Nahuatl.
    1.  
    1. b.
    1. They want to drive along the only road to Goose Bay on this trip, and also Nordkapp.

Example (18a) can only mean that they need to find two different monolingual speakers, and (18b) can only mean that they want to drive along two different roads. Compare them with (19), which contains the relative clause island, and should be ungrammatical because the island blocks the long source. All five of my consultants confirmed that there is a contrast between (18a) and (19a), and between (18b) and (19b).

    1. (19)
    1. a.
    1. *They need to find a monolingual who speaks Ch’ol for this course, and also Nahuatl.
    1.  
    1. b.
    1. *They want to drive along the only road that leads to Goose Bay on this trip, and also Nordkapp.

2.4. Evidence 4: Russian

This subsection makes a similar argument, but with a different language, Russian. Stripping in Russian can appear to be island-insensitive, but only with the short source reading. When the short source is spelled out overtly, it has the same intonation as the intonation of the stripping sentence, suggesting that the short source is indeed the reason why stripping can appear to be island-insensitive in Russian.

Russian has the relative clause island when there is no ellipsis, just like English, but the remnant of stripping can appear insensitive to the relative clause island (20). (20) only has the reading that they hired a single bilingual speaker, suggesting that only the short source is available. Also, (20) is only fine if a significant intonational break precedes a, and another shorter break may follow takzhe.5

    1. (20)
    1. Oni
    2. They
    1. vchera
    2. yesterday
    1. nanyali
    2. hired
    1. kogo-to,
    2. who-EI
    1. kto
    2. who
    1. govorit
    2. speaks
    1. po-francuzski,
    2. at-French,
    1. a
    2. and
    1. takzhe
    2. also
    1. po-nemecki.
    2. at-German.
    1. ‘They hired someone who speaks French yesterday, and also German.’

The meaning of (20) already suggests that its appearance of island-insensitivity is due to the short source. In addition, when the short source is spelled out overtly (21), it also requires the same intonation, where a significant break must immediately precede a, and a shorter break may follow takzhe.

    1. (21)
    1. Oni
    2. They
    1. vchera
    2. yesterday
    1. nanjali
    2. hired
    1. kogo-to,
    2. who-EI
    1. kto
    2. who
    1. govorit
    2. speaks
    1. po-francuzski,
    2. at-French,
    1. a
    2. and
    1. takzhe
    2. also
    1. ètot
    2. this
    1. chelovek
    2. person
    1. govorit
    2. speaks
    1. po-nemecki.
    2. at-German.
    1. ‘They hired someone who speaks French yesterday, and this person also speaks German.’

Since (20) only has the short source reading, and both (20) and (21) require the intonational breaks, the short source (21) must be the reason why (20) is grammatical.

2.5. Yoshida et al.’s (2019) rejection of the evasion strategy

Yoshida et al. (2019) presented experimental results showing that the evasion strategies can’t be the reason why stripping can violate islands. They constructed stripping examples where the remnant is an R-expression that is co-indexed with the matrix subject in the antecedent, as in (22).

    1. (22)
    1. Joe: While Joe was singing, shei noticed the student who met with Bill.
    2. Bill: *No, with Maryi.                                                            (Yoshida et al. 2019:1536)

If we follow the strict definition at the beginning of this squib that stripping has to occur in coordination, then Yoshida et al.’s examples are not stripping. Because with Mary in (22) adds information to the polar answer, I call it supplemental fragment (see also footnote 1). But exactly what to call (22) may not matter to us, if we assume that supplemental fragments, like stripping and other types of clausal ellipsis, involve movement of the remnant (i.e., with Mary) and deletion of a clause. If supplemental fragments involve the same types of syntactic operations as stripping, then the evasion strategies available to stripping should be available to supplemental fragments as well.

Returning to the analysis of (22), (23) spells out the possible derivations for the response in (22). In the long source (23a), the matrix subject she c-commands the trace of the remnant, incurring a Condition C violation. The short source (23b), if available, avoids such a violation because it does not include the matrix subject.6 (22) received low ratings from speakers, suggesting that the short source is not available.

    1. (23)
    1. a.
    1. Bill: No, with Maryi [TP sheinoticed the student who met ti].                         Long source
    1.  
    1. b.
    1. Bill: No, with Maryi [TP  {he / the student} met ti].                                          Short source

Yoshida et al. had two controls. In the first control, they tested another dialog with a pronoun instead of an R-expression in the remnant, which would avoid the Condition C violation. The control sentences received higher ratings, suggesting a real Condition C effect for (22). Then they did another control experiment to show that the effect is not due to the implausibility of cataphoricity (where a pronoun linearly precedes the expression it refers to), but to c-command. In this experiment, they kept the R-expression in Bill’s sentence, but varied the position of the pronoun in Joe’s sentence: the pronoun either appears in the adjunct clause (which would not c-command the R-expression in the long source), or in the main clause (which would c-command the R-expression in the long source). Dialogs where the pronoun appears in the adjunct clause receive higher ratings, suggesting that even for cataphoric pronouns, c-command still plays a role.

I agree that the short source is not available to (22); but when the short source is spelled out overtly, it already sounds odd:

    1. (24)
    1. Joe: While Joe was singing, shei noticed the student who met with Bill.
    2. Bill: #No, {he / the student} met with Maryi.

I speculate that (24) is odd because no can only deny Joe’s main assertion here, but not the content in the relative clause, which is presupposed. Thus, the badness of the short source may be due to the setup of this particular discourse. I leave the exact reason for future research, and simply point out that the short source is not available to (22) to begin with. When the short source is available, it does save an island violation.

Besides arguing against the short source, an important claim of Yoshida et al. is that the long source is available to stripping (or what I call supplemental fragments), even when an island is involved. In other words, stripping is genuinely island-insensitive. This seems to directly contradict my findings in subsections 2.2 and 2.3, where most of my consultants reject the long source reading of the stripping sentences that contain an island ((12) and (16)).

I want to briefly discuss two possible reasons why Yoshida et al.’s examples generated apparently contradictory results to mine. One possible reason has to do with the utterance-finality of the correlate (i.e., the phrase in the antecedent that corresponds to the remnant, such as with Bill in (22)). Yoshida et al.’s sentences all involve utterance-final correlates, whereas none of mine does. Barros et al. (2014) and Griffiths & Liptak (2014, fn. 10 and 28) have observed that the long source is exceptionally possible when the correlate is in the final position in the antecedent.

The second possible reason is that supplemental fragments may be more island-insensitive than stripping. In the informal surveys I elicited judgments of not only stripping, but also supplemental fragments. The short source is generally available to supplemental fragments, but I did not report those judgments in this squib because there were additional complications with the long source of supplemental fragments. Since the long source is not directly relevant to the claim of this squib, I left out supplemental fragments altogether to avoid distraction, and instead report some of the judgments here.

In the follow-up survey, I asked the five consultants who could only get the short source of (12) about (25B), the supplemental fragment counterpart to (12).

    1. (25)
    1. A:   They hired someone who speaks French yesterday.
    2. B:   Yeah, and also German.

All five of them accepted (25B), and all of them could get the short source reading (13b), where only one bilingual person was hired. Interestingly, four consultants could also get the long source reading (13a), where two monolinguals were hired. Note that the correlate in (25) French is not utterance-final. Thus, when an island is involved, these speakers seemed to find the long source to be more acceptable with supplemental fragments than with stripping. This suggests that supplemental fragments may be genuinely island-insensitive, which could also explain Yoshida et al.’s findings.

3. Conclusion

This squib has shown that evasion strategies, in particular the short source, are available to stripping, and that the short source must be the reason why the stripping examples discussed in this squib appear to be island-insensitive. Since stripping involves the same underlying operations as other types of clausal ellipses such as sluicing and fragment answers, the evasion strategies that are available to stripping should be available to clausal ellipsis more generally.

Notes

  1. Griffiths & Liptak (2014) called the entire sentence in (4B) fragment answer. If we follow the definition that fragment answers are direct answers to questions, then strictly speaking, only Yeah is the fragment answer. I therefore consider Charlie in (4bB) a fourth type of clausal ellipsis called supplemental fragment. It does not really matter to the claims of this squib whether Charlie in (4bB) is a fragment answer or supplemental fragment because the scope of this squib is clausal ellipsis generally, and the claims I make should apply to all types of clausal ellipsis. [^]
  2. This configuration satisfies the licensing condition on ellipsis that has been proposed in the literature. See Merchant (2001), for example, for a discussion on how it satisfies his semantic identity condition. [^]
  3. I will not provide any traditional evidence based on connectivity effects for the short source here, which a reviewer asked about. Evidence involving case, P-stranding and selection does not apply here because the remnant in the short source and the remnant in the long source are selected for by the same verb, and therefore have the same case and P-stranding profile. Evidence involving binding is notoriously unclear (see Barros et al. 2014 for a discussion and their failed attempt to replicate Lasnik’s 2001 judgments). Sentences involving NPI-licensing are confounded because they involve entities that do not exist, and E-type pronouns, which are used in the short source, cannot refer to non-existent entities. [^]
  4. I’m grateful to a reviewer for providing these examples, which I made some modification to. [^]
  5. As a reviewer pointed out, the correlate (i.e., the phrase in the antecedent that corresponds to the remnant, which is po-francuzski in (20)) is in the utterance-final position in (20) (my consultants preferred to put the temporal adverb immediately after the subject), which might make the long source reading exceptionally available in English (see the discussion of the utterance-finality effect in section 2.5). The fact that the long source reading is missing for (20) suggests that if the utterance-finality effect is what saves the long source in English, then this effect is not active in Russian. [^]
  6. Yoshida et al. (2019) actually focused on another evasion strategy called the cleft source, though they mentioned that their evidence would argue against the short source as well. Following Barros et al. (2014), I assume that the cleft source is generally not available to contrastive remnants. Since Yoshida et al.’s examples all have contrastive remnants, I assume the cleft source is not available, and do not discuss it. [^]

Acknowledgements

I am grateful to Danny Fox, David Pesetsky and the audience at MIT Ling-Lunch for helpful comments and feedback. I would also like to thank the three anonymous reviewers and editor Michael Yoshitaka Erlewine for detailed and helpful reviews. I am also thankful to Rafael Abramovitz, Neil Banerjee, Christopher Baron, Yadav Gowda, Cora Lesure, Elise Newman and Frank Staniszewski for providing the English judgments, and to Tanya Bondarenko, Anton Kukhto and Mitya Privoznov for providing the Russian judgments.

Competing interests

The author has no competing interests to declare.

References

Abels, Klaus. 2011. Don’t repair that island! It ain’t broke. Paper presented at the Islands in Contemporary Linguistic Theory conference, Vitoria-Gasteiz.

Barros, Matthew. 2013. A non-repair approach to island-sensitivity in contrastive TP ellipsis. Proceedings of Chicago Linguistics Society 48. Chicago, IL.

Barros, Matthew & Elliott, Patrick & Thoms, Gary. 2014. There is no island repair. Ms., Rutgers University.

Evans, Gareth. 1980. Pronouns. Linguistic Inquiry 11. 337–362.

Fukaya, Teruhiko. 2007. Sluicing and Stripping in Japanese and Some Implications. Ph.D. dissertation, University of Southern California.

Griffiths, James & Liptak, Aniko. 2014. Contrast and island-sensitivity in clausal ellipsis. Syntax 17: 189–234. DOI:  http://doi.org/10.1111/synt.12018

Lasnik, Howard. 2001. When can you save a structure by destroying it? In Kim, Minjoo & Strauss, Uri (eds.), North East Linguistics Society (NELS) 31, 301–320. Amherst: GLSA.

Lasnik, Howard. 2005. Review of the Syntax of Silence by Jason Merchant. Language 81. 259–265. DOI:  http://doi.org/10.1353/lan.2005.0025

Merchant, Jason. 2001. The syntax of silence: Sluicing, islands, and the theory of ellipsis. Oxford studies in theoretical linguistics. Oxford: Oxford University Press.

Merchant, Jason. 2004. Fragments and ellipsis. Linguistics and Philosophy 27. 661–738. DOI:  http://doi.org/10.1007/s10988-005-7378-3

Reinhart, Tanya. 1991. Elliptic conjunctions–non-quantificational LF. In Kasher, Asa (ed.), The Chomskyan Turn, 360–384. Cambridge: Basil Blackwell.

Ross, John Robert. 1969. Guess who? In Binnick, Robert I. & Davison, Alice & Green, Georgia M. & Morgan, Jerry L. (eds.), Chicago Linguistic Society (CLS) 5: 252–286. Chicago: Chicago Linguistic Society.

Szczegielniak, Adam. 2008. Islands in Sluicing in Polish. In Abner, Natasha & Bishop, Jason (eds.), Proceedings of the 27th West Coast Conference on Formal Linguistics. Cascadilla Proceedings Project, 404–412.

van Craenenbroeck, Jeroen & Merchant, Jason. 2013. Ellipsis phenomena. In den Dikken, Marcel (ed.), The Cambridge Handbook of Generative Syntax, 701–745. Cambridge: Cambridge University Press. DOI:  http://doi.org/10.1017/CBO9780511804571.025

Yoshida, Masaya & Potter, David & Hunter, Tim. 2019. Condition C reconstruction, clausal ellipsis and island repair, Natural Language and Linguistic Theory 37. 1515–1544. DOI:  http://doi.org/10.1007/s11049-018-9433-0