The structure of Mandarin embedded and matrix sluicing

. In Mandarin sluicing, the copula shi may precede the wh -phrase and it’s sometimes optional. Most existing analyses agree that Mandarin embedded sluic-ing with shi is an instance of pseudo-sluicing with an underlying copular structure. However, different analyses disagree on exactly when shi can be absent in Mandarin sluicing and by extension, the underlying structure of sluicing without shi . Many existing analyses assume that shi can be absent in embedded sluicing when the wh - phrase is complex. This study presents experimental findings showing that shi can only be absent in matrix sluicing and is obligatory in embedded sluicing, regardless of the wh -phrase. I then provide novel arguments in favor of the view that sluicing with shi is underlying copular, while sluicing without shi is not.

(1) Zhangsan Zhangsan An immediate question arises: how is (1) derived such that a copula may precede the wh-phrase?To complicate matters further, the copula may also sometimes be dropped.For example, it has been claimed that the copula may be dropped in matrix sluicing as in (2a) (Wang & Wu 2006), or when the wh-phrase is complex as in (2b) (Adams 2004;Wei 2004;Adams & Tomioka 2012). 1,2 A-not-A question as in (ia) or given a rising intonation as in (ib) (e.g., Hagstrom 2017).This issue does not arise in matrix sluicing such as (2a) and embedded sluicing whose matrix verb is non-factive such as (ii): since neither are polar questions, they cannot be turned into A-not-A questions and be given a felicitous rising intonation.I thus opt to always include the sentence-final particle in sluicing examples for consistency's sake throughout this paper. 1SFP n = sentence-final particle in the n-th class.Sentence-final particles in Mandarin Chinese are functional morphemes that appear at the end of a clause.These SFPs are traditionally grouped into three classes: SFP 1 , SFP 2 , and SFP 3 .SFP 2 are commonly treated as realizations of Force heads that perform clause-typing functions, and SFP 3 are typically analyzed as realizations of heads in a higher projection that encodes speaker or addressee attitude (Chao 1968;Zhu 1982;Paul 2015;Erlewine 2017). 2 To my ears, the sentence-final particles in both (2a) and (2b) are optional.It's important to note the following about this optionality.First, the sentence-final particle in (2b) is associated with the matrix clause and marks it as a polar question.If the particle is absent, in order to preserve its status as a polar question, (2b) is either turned into an 'Zhangsan hit some classmate.Do you know which classmate?' These facts raise a second question: exactly when can the copula be dropped?Several claims have been made in the literature regarding this question.
b. Claim 2: the copula can be dropped in matrix sluicing (e.g., Wang & Wu 2006).3c.Claim 3: the copula can be dropped when the sluice is not embedded under a nonfactive verb like think (e.g., Li & Wei 2023). 4inally, knowing the answer to 'when can the copula be dropped' raises yet another question: why can the copula be dropped, and if the copula can be dropped under exactly certain conditions, why those conditions?There are thus three questions needing to be answered: (4) a. Question 1: how are Mandarin SLCs derived s.t. a copula can precede the wh-phrase?b.Question 2: exactly when can the copula be dropped?c.Question 3: why can the copula be dropped and why only under certain conditions?Question 2 will be addressed experimentally through two naturalness rating experiments in §2 and 3.The experiments find that the copula can only be dropped in matrix sluicing.§4 then tackles Question 1.I present three novel arguments showing that that the copula shi can precede the wh-phrase in Mandarin SLCs because sluicing with shi is underlyingly copular.§5 finally addresses Question 3.While I do not make a positive proposal, I argue that unlike sluicing with copula, sluicing without copula is not underlyingly copular.§6 concludes.
2. Experiment 1. Claim 2 from the literature says that the copula can be dropped in matrix but not embedded sluicing.The goal of Experiment 1 is to determine whether this claim is correct.That is, its goal is to determine whether the type of sluicing (i.e., embedded vs. matrix sluicing) affect the naturalness of sluicing constructions with and without copula.2.1.DESIGN.4 target conditions were created by crossing the factors in (5). 3 minimal pairs of sentences were created for each target condition, where sentences in each minimal pair differed only in the presence of the copula, for a total of 24 sentences in 12 minimal pairs.Each target question consisted of one such minimal pair.
(5) Complexity of the wh-phrase: Type of sluicing: Animacy of the wh-phrase: complex embedded, matrix animate, inanimate In ( 5), the complexity of the wh-phrase is the control variable and is kept constant at complex wh-phrases (e.g.what book, which book).The type of sluicing is the independent variable being varied between embedded and matrix sluicing.The animacy of the wh-phrase is treated as a random variable that varied between animate (e.g.who) and inanimate wh-phrases (e.g.what).
All target sentences contained an antecedent clause followed by a sluiced sentence.All antecedent clauses had the structure (Adv-)NP 1 -V-Asp-NP 2 , where NP 1 is a proper name and NP 2 is an indefinite expression (e.g.someone, some student).For embedded sluicing, all sluiced sentences had the structure 2sg-V-(shi)-wh-SFP 2 , where V is the verb embedding the sluice and SFP 2 is a Class 2 sentence-final question particle (SFP).For matrix sluicing, all sluiced sentences had the structure (shi)-wh-SFP 3 , where SFP 3 is a Class 3 sentence-final question particle Sentences in each minimal pair were presented side-by-side in order to draw out relative preferences between the two forms, following Marty et al. (2020).Participants were asked to rate on a continuous slider how natural each sentence sounds, where 'natural' was defined in the instructions to mean something the participants would say themselves.Each minimal pair was also preceded by a one-to two-sentence context and a picture depicting that context.2.2.RECRUITMENT AND PROCEDURE.50 adult participants were recruited on Prolific Academic.Participants were prescreened on Prolific for their first language (Chinese, Mandarin).One additional participant was recruited through informal personal communication.Participants recruited on Prolific were compensated with an hourly rate of $12/hr.
Each participant saw 3 training questions prior to seeing the target questions.The training questions consisted of one question where both sentences in a minimal pair were very natural, one where both were very unnatural, and one where one is more natural than the other.The goal of the training questions was to familiarize participants with the task, encourage them to make use of the full slider scale, and discourage them from consistently rating one sentence higher than the other.
Each participant also saw 16 filler questions and 2 control questions in addition to the 12 target questions.Each filler and control question consisted of a minimal pair of two sentences that differed only in terms of one to a few words or word order between two words.
The order of presentation of all target, filler, and control questions were randomized.Sentences in each minimal pair were also randomly flipped.In order to prevent participants from failing to notice the minuscule differences between each sentence in a minimal pair due to the randomization, participants were asked explicitly in the instructions to pay close attention and never assume any two sentences to be the same even if they appear highly similar.
2.3.PREDICTIONS.Based on Claim 2 from the existing literature, participants are predicted to rate matrix sluicing without copula higher than embedded sluicing without copula.Matrix and embedded sluicing with copula are predicted to have similar ratings.
2.4.RESULTS.Ratings from 2 participants were excluded from analysis because they indicated in a demographic question that their native language is not Mandarin Chinese.Ratings from the remaining 49 participants were z-scored prior to analysis to accommodate for scale bias.Figure 1 (left and center) visualizes the distribution of ratings across target conditions.For sluicing with copula, participants tended to rate both matrix and embedded sluicing high.For sluicing without copula, participants tended to rate matrix sluicing significantly higher than embedded sluicing.Figure 1 (right) tells a similar story.The mean z-scored ratings for matrix and embedded sluicing with copula are similarly high, while the mean rating for matrix sluicing without copula is significantly higher than embedded sluicing without copula.
A simple linear regression model was used to analyze the dependence of z-scored ratings on the type of sluicing.The model revealed a significant effect of the type of sluicing on the ratings of sluicing constructions without copula, where embedded sluicing decreases ratings (β=-0.47,p<0.001).
2.5.DISCUSSION.Results from Experiment 1 confirmed Claim 2: the copula can be dropped in matrix sluicing, but not embedded sluicing.This difference between matrix and embedded sluicing in Mandarin SLCs is rarely discussed in the literature, but has important ramifications for any proposed answers to Question 1 and 3. §4 and §5 will address these ramifications in detail.
3. Experiment 2. Claim 1 from the literature says that the copula can be dropped when the whphrase is complex or an adjunct.The goal of Experiment 2 is to determine whether this claim is correct.Experiment 2 will however concentrate on wh-arguments and leave wh-adjuncts for future studies. 5The goal of Experiment 2 is then to determine whether the complexity of the whphrase (i.e.simple vs. complex wh-phrase) affect the naturalness of sluicing constructions with and without copula.
3.1.DESIGN.Experiment 2's design is the exact same as Experiment 1 except for the factors involved.Target conditions of Experiment 2 were creating by crossing the factors in (7) instead.
(7) Complexity of the wh-phrase: Type of sluicing: Animacy of the wh-phrase: simple, complex embedded animate, inanimate The complexity of the wh-phrase is now the independent variable being varied between simple and complex wh-phrases (e.g.what vs. what book).The type of sluicing is the control variable kept constant at embedded sluicing.The animacy of the wh-phrase is still treated as a random variable.A total of 24 target sentences in 12 minimal pairs were created, and each target question still consisted of one such minimal pair.
3.2.RECRUITMENT AND PROCEDURE.Similar to Experiment 1, 50 adult participants were recruited on Prolific Academic, and participants were prescreened on Prolific for their first language (Chinese, Mandarin).Participants recruited for Experiment 1 were also excluded from Experiment 2 using Prolific's prescreen tool.All participants were compensated with an hourly rate of $12/hr.Each participant in Experiment 2 saw 2 more filler questions than participants in Experiment 1.All other aspects of the experimental procedure remained the same.
3.3.PREDICTIONS.Based on Claim 1 from the literature, participants are predicted to rate sluicing without copula that have complex wh-phrases higher than those that have simple ones.Sluicing with copula is predicted to be rated high regardless of the complexity of the wh-phrase.
3.4.RESULTS.Ratings from 5 participants were excluded from analysis because they indicated in a demographic question that their native language is not Mandarin Chinese.Ratings from the remaining 45 participants were z-scored prior to analysis.Figure 2 (left and center) shows the distribution of ratings across target conditions.Participants tended to rate sluicing with copula high regardless of the complexity of the whphrase.Participants did tend to rate sluicing without copula that have complex wh-phrases higher than those that have simple ones, but the preference is not pronounced.
Figure 2 (right) repeats this pattern.The mean z-scored ratings for sluicing with copula that have simple wh-phrases and those that have complex ones are both high, while the mean rating for sluicing without copula that have complex wh-phrases is significantly higher than sluicing without copula that have simple wh-phrases.However, the mean ratings for (embedded) sluicing without copula as a whole are also significantly lower than (embedded) sluicing with copula, regardless of the complexity of the wh-phrase.
A simple linear regression model was used to analyze the dependence of z-scored ratings of sluicing constructions on the complexity of the wh-phrase.The model confirmed a significant effect of wh-phrase complexity on the ratings of sluicing constructions without copula, where complex wh-phrases increase ratings (β=0.43,p<0.001).
3.5.DISCUSSION.Results from Experiment 2 confirmed Claim 1: the copula can be dropped when the wh-phrase is complex or an adjunct, but not when it is simple.However, compared to the effect that the type of sluicing has on the ratings of sluicing without copula (found in Experiment 1), the effect of wh-phrase complexity was much less pronounced.I hypothesize that this preference for complex wh-phrases simply represented a general preference for complex whphrases not specific to sluicing.
In Experiment 2, participants also tended to rate non-sluicing sentences with complex whphrases higher than those with simple wh-phrases in the same minimal pair. 6Figure 3 shows the distribution of ratings for those sentences.A simple linear regression model was also used to analyze the dependence of z-scored ratings of non-sluicing constructions on wh-phrase complexity.The model found a significant effect of wh-phrase complexity on the ratings of non-sluicing constructions, where complex whphrases increase ratings (β=0.69,p<0.001), confirming the participants' preference for complex wh-phrases in general.
In conclusion, although results from Experiment 2 confirmed Claim 1, I argued that Claim 1 simply picked up on a general preference for complex wh-phrases, not a preference for complex wh-phrases specifically in sluicing. 7Taking Experiment 1 and 2 together, the results from these experiments thus show that the copula shi can only be dropped in Mandarin matrix sluicing.
4. The underlying structure of sluicing with copula.§4 now turns back to Question 1: how are Mandarin SLCs derived such that a copula can precede the wh-phrase?I follow much of the existing literature and propose below that sluicing with copula in Mandarin SLCs is underlyingly copular, hence the presence of the copula (Adams 2004;Wei 2004;Adams & Tomioka 2012;Gong 2019).The sentence in (8a), for example, is thus underlyingly (8b).( 8 The rest of §4 presents 3 novel arguments in support of this proposal.§4.1 makes a general argument against alternative movement-plus-deletion approaches.§4.2 argues based on the word order of dou 'all' and the copula shi that sluicing with copula is underlyingly copular.Lastly, §4.3 rejects the argument that idiomatic reconstruction effects in Mandarin SLCs favor a movement-plus-deletion approach. 4.1.AGAINST MOVEMENT-PLUS-DELETION. Roughly speaking, in a movement-plus-deletion approach to sluicing with copula, the wh-phrase is first moved to the left periphery, then the remaining clausal structure is deleted or left unpronounced.This section argues against movementplus-deletion approaches in general, by first arguing against movement ( §4.1.1)then against deletion ( §4.1.2). 4.1.1.AGAINST MOVEMENT.Mandarin, a wh-in-situ language, has two wh-ex-situ strategies: topicalization and focus-fronting (Cheung 2008;Pan 2014). 8However, both topicalization and focus-fronting obey constraints that sluicing with copula does not.
First, objects of action verbs encoding episodic eventualities (e.g.buy, find, hit) cannot focusfront (Pan 2014;Paul 2015).Compare (10a) and (10b), for example.(10a) encodes non-episodic 7 One may expect such a general preference to also be reflected in the ratings for sluicing with copula.However, Figure 2 (left) shows that this is apparently not borne out -sluicing with simple and complex wh-phrases are both rated high.I hypothesize that this results from a ceiling effect: the tendency to rate sluicing with copula high is already so great, that complex wh-phrases no longer have a noticeable effect. 8A fronted wh-phrase in Mandarin may optionally be preceded by the copula shi.Here I follow Pan (2014) and others and identify those preceded by shi with focus-fronting, and those not preceded by shi with topicalization.eventuality via the experiencer verb xihuan 'like' and focus-fronting the object wh-phrase is grammatical.(10b) encodes episodic eventuality via the action verb da 'hit' and focus-fronting becomes ill-formed.( 10 Since both wh-ex-situ strategies in Mandarin are constrained in ways that sluicing with copula is not, any movement-based approach to sluicing with copula must posit exceptional movements that are otherwise illicit.I take exceptional movements to be a priori undesirable if other options, e.g.pseudo-sluicing, are available.4.1.2.AGAINST DELETION.Evidence against deletion comes from the sentence-final particle laizhe.Laizhe may follow the wh-phrase in sluicing with copula; ( 14 'Zhangsan bumped into a friend.I don't/no longer remember who.' Sentence-final particles (SFPs) in Mandarin are traditionally grouped into 3 classes based on their structural heights (e.g., Chao 1968;Paul 2015).Laizhe belongs to the class of low sentencefinal particles SFP 1 in this taxonomy.Erlewine (2017) argues that SFP 1 are phase heads in the extended projection of VP based on several diagnostics.I repeat one below.Erlewine (2017) probes the position of low SFPs based on their scopal interactions with low and high negations.In Mandarin, the negations bu and bushi are known to have different structural heights.Here I follow Huang (1988) and treat bushi as a high negation located outside the extended projection of VP, and bu as a low negation located within the extended projection of VP.
If laizhe is a head in the CP periphery (cf.Paul 2015, among others), then it outscopes both bu and bushi and should thus interact with them in the same way.If laizhe is a phase head in the extended projection of VP, then it outscopes bu but not bushi; it may thus interact with them differently.
(15a) and (15b) show that the latter is the case: laizhe is compatible with and outscopes the low negation bu, but is entirely incompatible with the high negation bushi.Under a movement-plus-deletion approach to sluicing with copula, the wh-phrase moves to some TP-external position and the evacuated TP is deleted.No TP-internal element is thus predicted to survive deletion.Since laizhe is a TP-internal head, its occurrence in sluicing with copula is a counterexample to deletion-based approaches.
4.2.THE WORD ORDER OF dou AND shi IN THE SLUICE.The word order of the particle dou and the copula shi in the sluice shows that sluicing with copula must be underlyingly copular.This pattern is only found in copular constructions and not verbal constructions. 12In a pseudocleft like (18), for example, dou may precede shi but not vice versa.In a verbal construction like (19), it's the reverse pattern. 13Since the word order pattern in sluicing with copula matches only the pattern copular constructions, sluicing with copula must also be copular.
(  MANDARIN SLCS. Song & Yoshida (2017) argue in favor of a movement-plus-deletion approach to Mandarin SLCs with evidence from idiomatic reconstruction effects.The verb-object idiom chunk chi shei de cu 'eat whose vinegar' in (20a) has the idiomatic interpretation 'be jealous of whom'.This interpretation remains when the object wh-phrase topicalizes or focus-fronts as in (20b).'Who (is it that) Lisi is often jealous of?' Cheung (2008) Song & Yoshida (2017) cite the same effect in Mandarin SLCs: the wh-phrase shei de cu 'whose vinegar' in (21b) retains its idiomatic interpretation.Song & Yoshida (2017) take this as evidence that the wh-phrase reconstructed into silent material.By comparison, clearly copular constructions like (21c) with an overt copular subject lose the idiomatic interpretation.22b) are copular: (22a) is a copular sentence with a null pro subject and ( 22b) is a pseudo-cleft.Since they both retain idiomatic interpretations, (22a) and (22b) reject Song & Yoshida (2017)'s premise that idiomatic reconstruction effects are indicative of movement.4.4.INTERIM SUMMARY.This section proposed that sluicing with copula in Mandarin is underlyingly copular and presented 3 novel arguments in support.First, I dismissed movementplus-deletion approaches to sluicing with copula on grounds that exceptional movements are undesirable when pseudo-sluicing is a viable option.Second, I showed that sluicing with copula exhibits word orders only found in copular constructions.Finally, I objected to Song & Yoshida (2017)'s argument that idiomatic reconstruction effects in sluicing with copula entail derivation by movement-plus-deletion.The picture that emerges is that sluicing with copula is best accounted for if it is underlying copular.
5. The underlying structure of sluicing without copula.Now only Question 3 remains: why can the copula be dropped and why only in matrix sluicing?In other words, how is sluicing without copula derived such that the copula may be dropped in matrix sluicing?While I do not make a positive proposal here, I argue below that sluicing without copula must not be copular.5.1.SLUICING WITHOUT COPULA IS NOT COPULAR.§4 concluded that sluicing with copula is underlyingly copular; then the null hypothesis is that sluicing without copula, i.e., matrix sluicing, is also copular.However, I argue below that this hypothesis would contradictorily predict the copula to be obligatory in sluicing without copula.5.1.1.SLUICING WITHOUT COPULA IS NOT SPECIFICATIONAL.First, the copular structure underlying sluicing with copula is specificational rather than predicational.Here I use terms 'specificational' and 'predicational' in the sense of Mikkelsen (2005)'s taxonomy of copular sentences, where subjects of specificational copular sentences are predicative, property-denoting, while subjects of predicational copular sentences are referential, individual-denoting.Cheng (2021) shows that the subject of a specificational copular sentence in Mandarin can be replaced with na 'that' but not ta 's/he'.The same is true in sluicing with copula: in both (23a) and (23b), the (null) copular subjects can be replaced with na and not ta.Since sluicing with copula is specificational, the null hypothesis is that if sluicing without copula is copular, then it is specificational, too.However, the copula is obligatory in Mandarin specificational copular sentences (Cheng 2021).Cheng (2021) identifies 3 environments in which the copula is omissible in Mandarin: multi-clausal pair-lists (see (24a); this results in a gappinglike construction), the so-called 'nominal predicate constructions' (see (24b); Chao 1968;Zhu 1982), and embedded small clauses (see (24c)).( 24 Crucially, the copula is only omissible in these 3 environments if the sentence is predicational. (25) shows that the copula becomes obligatory when (24a-24c) become specificational.Thus if sluicing without copula is specificational, the copula must contradictorily be obligatory.
( 'Do you consider the idiot to be Zhangsan?' A specificational structure is also incompatible with a movement-plus-deletion approach: the post-copular wh-phrase cannot evacuate out of the copular structure15 in order for deletion to target only the null subject and the copula as in ( 26), since the post-copular nominal in a Mandarin specificational sentence cannot move (Cheng 2021). 16In conclusion, sluicing without copula must not be specificational if it is copular.'Who is that person?' Since pre-copular nominals in predicational sentences are free to move, this predicational structure lends itself to a movement-plus-deletion approach: the subject wh-phrase first moves to the left periphery; deletion then targets the evacuated copular structure.This is shown in (28).'Who 1 t 1 is that person?'However, predicational structures are not restricted to matrix sentences.Thus if sluicing without copula is predicational, then it must be available in both matrix and embedded sluicing.In other words, the copula is predicted to be droppable in both matrix and embedded sluicing, contrary to findings from Experiment 1 and 2. Sluicing without copula must then also not be predicational.
5.2.INTERIM SUMMARY.This section argued that sluicing without copula in Mandarin is not underlyingly copular.The argument went like this.First, following Mikkelsen (2005), I assume a three-way taxonomy of copular sentences: specificational, predicational, and equative.This section argued that if sluicing without copula is underlyingly copular, then it can be neither specificational nor predicational.Furthermore, I also follow Cheng (2021) and assume that Mandarin has no true equative copular sentences with the copula shi. 17Since sluicing without copula has no viable underlying copular source, I conclude that it must not be underlyingly copular.

Conclusion.
The goal of this paper has been to answer Question 1-3 about Mandarin SLCs.
(29) summarizes the proposed answers to each question.
b. Question 2: Answer: Exactly when can the copula be dropped?
The copula can only be dropped in matrix sluicing.
c. Question 3: Answer: Why can the copula be dropped and why only in matrix sluicing?Sluicing without copula is not underlying copular.
The contributions of this paper are potentially twofold.First, it clarifies the distribution of the copula in Mandarin SLCs.Since analyses of Mandarin SLCs have been centered around accounting for the distribution of the copula, a renewed understanding of this distribution inevitably affects existing analyses of Mandarin SLCs.Second, while many existing analyses of Mandarin SLCs have confined their scope to Mandarin embedded sluicing, this paper advocates researchers to pay more attention to Mandarin matrix sluicing as well.Based on new findings on the distribution of the copula from §2 and §3, I proposed a bifurcation of Mandarin SLCs into sluicing with copula and sluicing without copula: sluicing with copula are base-generated copular constructions, thus explaining the presence of the copula; however, sluicing without copula is derived differently and is confined to matrix sluicing.In order to fully under why the copula can be absent in Mandarin sluicing, we must thus look to matrix sluicing.
someone.But I don't know who.' you know which classmate?'

'
Which classmate do you think (it is)?'

(
Zhangsan likes someone i .But I don't know [who pro i is].' (8b) parallels the cross-sentential anaphora construction in (9), where the 3 rd person pronoun ta takes an indefinite antecedent from the previous sentence.I assume that pro is interpreted analogously in (8b)(Adams 2004; Adams & Tomioka 2012).
The Mandarin quantifier-distributor particle dou, roughly English post-nominal all, may precede the copula shi in Mandarin SLCs if the copula is present.(16) gives one example.Zhangsan bought a bunch of books.But I don't know what books (they all are).' Crucially, dou may only precede but not follow shi: 23) a. Zhangsan hit a classmate. . .