The relationship between Chinese zhiyou ‘only’ and cai: a matter of morphosyntax

This short article pursues two goals. First, it critically reviews some results arrived at by Sun (2021), an article that discusses the interplay of Chinese zhiyou ‘only’ with an empty exclusive operator while denying any interesting interaction between zhiyou and the scalar particle cai (against the view defended by Hole, 2004, 2017). Second, it develops a featural account to analyze the morphosyntactic interaction between zhiyou and cai. In sum, the gist of Sun’s and Hole’s views complement each other.

1 Introduction Sun (2021) presents a bipartite analysis of Chinese monoclausal zhiyou 'only' sentences in the tradition of Bayer (1996), Kayne (1998), Wagner (2006), Quek & Hirsch (2017) and others. An example is given in (1). 1 (1) Akiu [zhiyou niurou] cai chi t i Akiu only beef CAI eat 'Akiu eats only beef.' Zhiyou combines with XPs that are not (extended) verbal projections (here, the direct object niurou 'beef') and, according to this analysis, values features of an empty exclusiveness head by moving to, or through, its specifier. That's why we find the direct object in preverbal position, and not in its canonical postverbal position. The lowest position of this exclusiveness head is the vP edge. Higher positions are possible as scopally needed. This analysis denies a close relationship between zhiyou and the particle cai. A strand of research assumes a spec-head relationship between these two elements (Hole 2004(Hole , 2017 claims that cai always goes hand in hand with a scalar interpretation of the zhiyou focus, where a scalar interpretation presupposes or conventionally implicates the assessment of the focus value as little or insignificant. At the same time, an order of counting as more or less insignificant is imposed on the set of alternatives. Hence, for Hole, zhiyou and cai participate in the marking of scalarity with 'only' foci in Chinese, where Sun (2021) has zhiyou and zero interact to express 'only' with vP-external (frequently moved) foci.
Sun's syntactic proposal, which is based on Quek and Hirsch (2017), is summarized in (2). Note that it depicts the structure before movement of the 'only'focus (Sun 2021: 327). (

2) [ FocP [ Foc′ Foc 0 [iONLY(), EPP] [ vP/TP … [ QP Q 0 [uONLY(+)] XP F ] … ] ] ]
Foc 0 is the empty exclusive operator. It has an interpretable, though unvalued, ONLY feature, and an EPP feature. The EPP feature triggers movement of the lower QP. 2 QP, the zhiyou-phrase in Chinese, is headed by zhiyou/Q. Q bears an uninterpretable, though valued, ONLY feature. Its sister is the focus XP. After movement, QP is in Spec,Foc. In this position, it values the ONLY feature of Foc 0 . There's a lot of variance in the literature when it comes to the term "scalarity". I divide the empirical domain as follows. There is informational scalarity, a property of propositions to entail certain other propositions, and not entail certain others (Krifka 1995). If one says I ate only two apples this entails that I didn't eat three apples, and it entails that I ate one apple. There is scalarity proper (Guerzoni 2003: 173), which presupposes or conventionally implicates that the prejacent focus counts as little. If I say I ate only salad, then, in most contexts, this will not just mean that I ate nothing but salad, but that to eat salad and no more counts as little on some contextually determined scale. Finally, there is a frequently overlooked component of evaluation as bad that may accompany the use of 'only' words and other elements (Hole 2015). If someone says He's only a cadet, don't marry him, then the speaker will, in most contexts, want to express that being no more than a cadet is not just little or insignificant, but also bad. As already mentioned, I am concerned with scalarity proper here, the assessment of a focus value as little or insignificant. Note that monotonocity has certain effects on scalarity that sometimes obscure the assessment as little. There is no problem with sentences like She owns only two T-shirts. Things are different with Only a golden bracelet will do. Here it seems that a golden bracelet counts as a lot, seemingly reversing the scales.
However, I argue that this is still a case of normal scalarity proper: only as little as gold (and maybe platinum) will do, irrespective of the high value of gold.
Because (3a) excludes teenagers who are 17 or 19 years old. (3b) exludes teenagers who are younger than 18 and leaves it to context whether youngsters who are older than 18 may participate. (3b) necessarily involves scalarity, whereas (3a) needn't.
As mentioned above, Sun (2021) denies any interesting or morphosyntactic interaction between zhiyou and cai (cf. the obligatory use uf cai in (1)). This article takes issue with this and some other views as defended by Sun (2021). Specifically, I will demonstrate that the cai frequently co-occurring with zhiyou is not emphatic cai, as proposed by Sun (Sect. 2.1). Furthermore, I will show that cai defines a scope position that does have a presuppositional or conventional-implicature scalarity semantics as proposed by Hole (2015) (Sect. 2.2). 3 Section 2.3 is devoted to describing the tradition leading from Hole (2015) to Quek and Hirsch (2017) and then to Sun (2021), thereby pinpointing the close parallels between Hole's and Sun's accounts. In Sect. 2.4, I discuss some problematic data which Sun (2021) uses to argue in favor of an A′-movement account for zhiyou foci. I don't wish to deny the fact that A′-movement is involved, I would just like to insist on better data. Section 3 presents an account that combines Hole's and Sun's insights. In sum, I think both Hole (2017) and Sun (2021) may be right in that zhiyou phrases have morphosyntactic interactions with both an empty exclusiveness head and with the overt scalarity head cai.

A critical review of Sun (2021)
In this Sect. 1 present data and arguments which show that a reassessment of some of Sun's (2021) results is needed. I would like to make it clear at this early point that this leaves the core of Sun's proposal intact. Nevertheless, the reassessment is necessary to justify the addition to Sun's theory that I will present in Sect. 3.

2.1
The cai co-occurring with preposed zhiyou phrases is not emphatic cai Sun (2021) argues that the cai which frequently co-occurs with zhiyou is not a head of any sorts and that its use is not as obligatory with zhiyou phrases as Hole (2004Hole ( , 2017 makes it appear. (Hole 2017 claims it is a scalarity head, and an 'only'-focus must sit in, or move through, its specifier.) Instead, Sun identifies the adjunct emphatic use type of cai as the particle really co-occurring with zhiyou phrases. Examples of each use type, the debated one and the clearly emphatic one, are given in (4) and (5).
(Sun's 2021 (64) ≈'How ludicrous that movie is!' (Hou ed., 1998: 77) It is a welcome addition to our knowledge of Chinese to have it pointed out, more than it is in Hole (2004: 52), where the cai in question is not obligatory (after subject foci, with more than a bare verb following cai, and after some temporal adjuncts). However, I would like to point out three facts in defense of the head analysis. First, as Sun (2021: fn. 4) acknowledges herself, cai is absolutely mandatory after subordinate 'only-if'-clauses introduced by zhiyou (Wimmer 2022). Sun (2021) chooses not to incorporate these cases in her analysis. I doubt that this is a good strategy. We have subordinate zhiyou clauses with obligatory matrix cai, we have obligatory cai with many monoclausal foci, and we have perfectly similar mới in Vietnamese (Hole 2013(Hole , 2015. It is the core idea in Hole (2015) and in the present article, that the use of cai with 'only'-foci invariably indicates a scalar interpretation of the focus. Like this it becomes possible to account both for its obligatoriness in many cases, and for its optionality in many cases, at least in principle. 4 (We will see in Sect. 2.2 below, as we saw in the introduction, that under certain circumstances 4 Note in passing that the similar pair zhiyao…jiu used in conditionals with foci in the protasis (Wimmer 2022) is likewise attested in a variant without jiu in colloquial language, while no-one contests the general obligatoriness of jiu in this construction. (i) presents an attested example, where the variant without jiu is the attested one (Hou ed., 1998;Hole 2004: 72 true minimal pairs with and without cai do exist which feature a stable contrast in meaning as well.) Second, and as has been known at least since Alleton (1972), the cai in (4) cannot be emphatic cai, because emphatic cai only occurs in exclamations and is typically, but not always (Hou ed., 1998: 78, Hole 2004, followed by the sentence-final particle ne; cf. (5). 5, 6 Neither of these two generalizations holds for the cai co-occuring with zhiyou phrases. What is more, emphatic cai sits higher in the tree, a fact which is evidenced by the contrast between (4′)  According to Alleton (1972: 147), emphatic cai may be followed by zhen 'really' without ever influencing acceptability. This may not be true (Alleton only has few pertinent examples in her corpus), but the difference in acceptability judgments between (4′) and (5′) is robust. A generalization that probably holds without exception is that emphatic cai may always be replaced by zhen without leading to unacceptability. Hole (2017) and Bross and Hole (2017) locate the scalarity projection spelled out in (1), (4) and (4′) right above T, whereas zhen 'really' is an evidential device and, hence, above the scalarity projection (Cinque 1999). Ironically, the informal semantics that Sun (2021: 344) provides in (6) characterizes an exclusive-though corrective-focus particle with a scalar semantics, precisely what Hole (2017) claims for the interaction between zhiyou and (parametric) cai.
5 Lai (1999) and Sun (2021) base their generalizations concerning emphatic cai on a single (the same) example. Biq (1984Biq ( , 1988 lists three examples, and Alleton (1972) has 17 examples in her corpus. My treatment of emphatic cai rests on Alleton (1972) and Hole's (2004: 18-9) rendering of the tradition. 6 Liu Mingming (p.c.) points out to me that to him (5b) sounds more like a structure involving a contrastive topic. '[This game] CT was soso. [That game] CT , on the other hand, was CAI great.' It has indeed been observed by Alleton (1972) that emphatic cai comes in two flavors, one associating with a gradable property, and one which associates with something that is similar to a clefted constituent. I take it that Liu read (5) as such. The distributional generalization to be discussed next in the main text holds for both use-types of emphatic cai. 7 An anonymous reviewer points out that (5) contains a scalar predicate whereas (4) does not, and that this may account for the difference in acceptability. (i) features a scalar predicate, contains scalar cai and still disallows zhen.
(i) Zhangsan cai (*zhen) wu-sui. Zhangsan CAI really 5-CL:year.of.life 'Zhangsan is a mere 5 years old.' I take this to mean that the cai that Hole and Sun mainly investigate is not the emphatic cai of Alleton (1972) and others.
(6) cai(p) is used in a context whena. p and its alternatives are ranked in terms of what standard they indicate for a context-salient property G; b. The prejacent p is uttered to correct a potential alternative p′ in the previous discourse to indicate a higher standard of G.
All in all, Sun (2021) rejects Hole's semantic and distributional analysis of parametric cai, only to reintroduce a device with an almost identical function as a replacement, claiming that this is a discourse marker, and not a device involved in focus marking.

2.2
Cai defines a scope position of its own interacting with zhiyou foci Sun (2021: 339, fn. 11) uses the contrast between (7a) and (7b) to argue against a fixed scope position of the 'only' foci involved.
(7) a. Zhiyou zhe-zhong qufeng Yuehan keneng cai jieshou. only this-kind music.style John likely CAI accept 'Only this kind of music style is John likely to accept.' b. Yuehan keneng zhiyou zhe-zhong qufeng cai jieshou. John likely only this-kind music.style CAI accept 'It is likely that John only likes this kind of music style.' The truth conditions of these two sentences differ (albeit only in a superset fashion: the truth-conditions of (7a) are a superset of those in (7b), i.e. (7a) is true in fewer situations than (7b)). Hence, Sun argues, it cannot be the specifier of a cai phrase where the zhiyou focus has its obligatory site of interpretation. I grant that much, but this is not what Hole (2015) claims. What he claims is that cai defines the scope of the scalarity operator LITTLE. It is instructive in this context to see what happens in a configuration where the interacting operators have different monotonicity properties and the truth-conditions differ more dramatically than in (7) With zhiyou scoping above bukeneng 'impossibly' as in (8a), cai is deviant. With zhiyou scoping below bukeneng as in (8b), it is fine. If we spell out the scalarity component of cai, we see why this is the case.
(8) a. # 'Only for this music style is it impossible that John accepts as few music styles as that.' b. 'It's impossible that John accepts only as little as this music style.' The reason for the impossibility of cai in (8a) lies in the fact that it is implied by cai in its scoping position that John likes few music styles. With the operators on top, however, the sentence implies that he likes many music styles. That is a contradiction. (8b), on the contrary, negates the possibility that John likes as little as one music style, and that doesn't result in a contradiction.
More evidence to the effect that cai defines a scope position of its own comes from (9), a sentence contributed by an anonymous reviewer.
(9) Zhiyou zhe-zhong qufeng Yuehan cai bukeneng jieshou. only this-kind music.style John CAI impossibly accept 'Only for as little as this music style is it impossible that John accepts it.' With the scalarity operator scoping above the negated epistemic modal, the deviance of (8a) with cai turns into impeccability. This reading is predicted by Hole's (2017) proposal for cai's function. 8 Interestingly, Sun (2021: 342-343) herself adduces evidence to the effect that cai contributes this kind of scalarity. Consider (10) and (11), where Yangmi represents an actress of average quality, Tangwei one of higher quality, and Zhangziyi one of excellent quality.  (2017), though, is that cai can take scope above epistemic modals at all. Hole (2017) assumes that scalar cai sits in a fixed position right above T, but below epistemic modality. That parametric words may scope in different positions which must be at least above T has also been shown for the parametric-scalar use of ye 'also' by Yang (2020). I have adjusted the statements about cai's syntactic position in the main text accordingly.
These dialogues only differ with regard to the presence or absence of cai in B's turn. While the presence of cai makes it possible to include higher-ranking options, B's turn in (11) excludes all other alternatives, no matter what their rank is. Sun (2021) uses this kind of data to steer in the direction of her emphatic-cai proposal. I think I have shown above that this is not viable. What this kind of data shows, however, is that cai may very plausibly be assumed to induce an order over alternatives, and that this order associates the asserted option with a low scalar value. It is important to note at this point that this scalar lowness holds even in cases like (10). The levels of excellence that will do in (10) are stated to be few, with higher levels of excellence than that of Tangwei trivially making B's turn true. In (11), on the contrary, no scalarity is invoked, and all alternatives are excluded independently of any concomitant ranking.
To sum up, Hole (2017) is probably wrong in claiming that each zhiyou focus involves cai scalarity. He is right, I think, in claiming that the evidence points towards assuming that zhiyou foci, if cai is present, interact with cai in a way that may well be analyzed as a spec-head relationship, if only as one step in a longer movement chain. 9 2.3 Sun's (2021) architecture is more similar to Hole's (2015Hole's ( , 2017 than might first appear Sun (2021) proposes that there is an exclusive focus head where needed above vP which indicates the scope position of 'only' operators and that zhiyou phrases obligatorily move through or land in that projection. While Hole (2015Hole ( , 2017 in his work on German, Dutch, Vietnamese and Chinese doesn't consider the possibility that zhiyou phrases must value features in that projection, he proposes precisely this overall architecture for sentences that have a scalar and exclusive meaning. This overall architecture is provided in (12). (Hole, 2017: 404 treats 'even' foci on a par with 'only' foci, that's why the scalarity head LITTLE 0 in (12) has a MUCH 0 variant. AdFoc stands for "ad-focus particle" like Chinese zhiyou 'only'. FocAdv stands for "focus adverb" like Chinese zhi 'only'.) In (13) the topicalized quantifier must reconstruct to the c-command domain of the subject quantifier kein Mann 'no man'. Otherwise the possessive pronoun could not be bound. However, the scope of the exclusive operator must not be below the subject quantifier. Büring and Hartmann (2001) use this kind of example to argue against the constituenthood of nur ein Bild von seiner Frau in (13), something that is really hard to accept if one knows that the German topicalization position always hosts exactly one constituent. Now, Hole (2015) proposes that the semantically active focus operator is not pronounced and sits at the edge of vP. The ad-focus particle in (13) is not interpreted. This will yield the PF and LF in (14).
Whether it is the exclusive operator or the ad-focus particle that is pronounced, is regulated by the principle in (15), which holds for German.
(15) First Come, First Spell-Out (cf. Hole 2015: 59) From among EXCL and AD-FOC in a single focus-background structure, the linearly first operator will be pronounced (where EXCL is the exclusive operator and AD-FOC the ad-focus particle/operator).
(16) Exactly one head in the bipartite structure of 'only' must be phonologically overt.
In sum, Sun (2021) proposes an architecture that, among other inspirations, goes back to Hole (2015Hole ( , 2017, but with the important addition that zhiyou phrases are seen as partaking in feature valuation not in scalarity phrases, but in the projection of the exclusive operator. I would like to propose that the zhiyou phrase might well value features both in the exclusive position and in the scalarity position.

Sun's reconstruction evidence is inconclusive
The present and the following subsections are different from the preceding ones in that they critically review data that don't really affect the core of Sun's proposal or its compatibility with the proposal to be made in Sect. 3. These subsections aim at pointing out that some of Sun's (2021) data should only be used with caution. Sun (2021: 326) uses the data in (17) to argue in favor of the reconstruction of zhiyou phrases. This is done to corroborate the A′-movement nature of the movement involved. (17)  The reflexive-like element in (17a) is expected if the zhiyou phrase reconstructs. After reconstruction, Lisi in (17b) will be c-commanded by ta, hence the reported Condition C effect. In (17c), finally, the availability of the idiomatic reading of chui zhe-ge niu 'boast' is available, which should mean that the direct object undergoes reconstruction.
The problem with (17a) is that taziji is not the short-distance reflexive taziji here, but a sequence of the ordinary Condition B pronoun ta and an intensifier 'he himself' (Hole 1998(Hole , 2008. This can be seen clearly in (18), a sentence which is as good as (17a).
I only this-CL about him DE book give-EXP Lisi '[Only this F book about him i ] did I give Lisi i t 1 .' (18) could either mean that the zhiyou-phrase does not reconstruct, or that it reconstructs, but that ta 'him' is in an exempt position. Note the impeccability of (19), where no movement has taken place and the ordinary pronoun is fine under c-command from its antecedent.
I give-EXP Lisi this-CL About him DE book 'I gave Lisi i this book about himself i .' (19) renders it highly plausible that ta's position in (18) is exempt. Hence this sentence cannot be used to argue in favor of reconstruction.
What is more, as Sun (2021: 334) notes herself, Chinese is known to have reconstruction for (reflexive) binding only, and not for scope. It is, hence, a bit unclear what the claim that zhiyou-phrases reconstruct really means.
Let us now turn to Sun's other two examples from (17)  As concerns (20a), I would like to deny its ungrammaticality. Condition C violations are known to be a bit tricky, and I would claim that the oddness of (20b) is of a much milder kind than that of, say, *Ta i dai-lai-le Zhangsan i -de shu '*He i brought along Zhangsan i 's book.' Turning to (20b) now, I would like to point out that the canonical form of the idiom involves a bare object: chui niu 'boast'. The version in (20b) with the definite preposed object is interpreted in an ad-hoc compositional fashion which doesn't require reconstruction. The English translation gives a hint in that direction in that zhe-ge niu can be paired with the DP this boast in the translation. (Note that according to Diesing 1992, Carlson 2003 and many others definites obligatorily move out of VP and don't reconstruct.) To sum up, Sun (2021) does not show beyond doubt that zhiyou phrases reconstruct.

The purported WCO effect is something else
The data in (21) is used to argue for a Weak Crossover Effect (WCO) with preposed zhiyou foci (Sun 2021: 327). This, if such an effect can be shown to exist, would strengthen the idea that zhiyou foci are not interpreted in their surface positions. To be sure, the zhiyou phrases have moved across the co-indexed pronoun, which makes these cases look like Weak Crossover violations. However, Yuehan doesn't c-command the co-indexed pronouns in (21), only zhiyou Yuehan does. This makes Sun's footnote 7 appear in a new light. There she reports marginal acceptance of the sentences in (21) by some of her consultants. If no WCO violations existed in the first place, this would be expected. The footnote goes on to insist on the absence of the binding readings in (21) even if these sentences are deemed acceptable, those where other people are not introduced to their respective bosses, and not to John's boss. Again, the absence of this reading is expected, because Yuehan in its surface position does not c-command tade. 10 2.6 Taking stock I have presented strong evidence to the effect that the cai frequently co-occurring with zhiyou is not emphatic cai, as proposed by Sun (2021). Instead, I reinstantiated Hole's (2004Hole's ( , 2017 view that there is morphosyntactic interaction between cai and zhiyou-foci. I went on to show that cai defines a scope position for 'only'-foci of its own which does not coincide with the position of Sun's empty exclusive operator. Furthermore, I pointed out the large similarities in the overall architectures of Sun's and Hole's analyses. In the last portions of this section. I warned against using some of Sun's data without caution. It is probably fair to say that, at the present point, we don't know whether zhiyou-phrases really reconstruct.

The new proposal
The proposal that I wish to propose is quite obvious. It includes everything that Sun (2021) proposes for the interaction of zhiyou-foci with the phonetically empty exclusive operator. It adds to this the scalarity head instantiated by cai. Zhiyou-foci without a scalar component (such as the one in (11), repeated here as (22)), require no modifications of the theory. Zhiyou-foci co-occurring with cai (such as the one in (10), repeated here as (23)), require the assumption of an enriched feature structure of zhiyou-foci with a matching feature structure on cai. Again, zhiyou-foci have to move to, or through, Spec,Scal to value the scalarity feature on the Scal head cai. This means, then, that zhiyou-foci come in two flavors, either as the QP type from (24), or as the QP type from (26) depending on whether cai co-occurs or not. This proposal maintains all the advantages of Sun's analysis while integrating the-empirically needed-morphosyntactic treatment of cai into the picture. Let me point out one important difference between Foc 0 and Scal 0 . The syntactic position of Foc 0 , the empty exclusive operator, varies above vP according to scopal needs. As shown in Sect. 2.2 and argued for by Hole (2015Hole ( , 2017, the syntactic position of Scal 0 is fixed, only allowing for a scoping position right above TP/AspP or epistemic modality (the choice between TP and AspP depending on whether one assumes a tense node for Chinese or not). 11 Some readers may be uneasy about treating zhiyou as a constituent (Liu Mingming, p.c.). The you in zhiyou has uses elsewhere (jin-you 'only', hai-you 'furthermore', the you with indefinites in subject position), and its 'if'-like semantics appears to be stable across contexts. One might be tempted to assume, then, that you combines with a sister to its right first, to be merged with zhi only after that (researchers have entertained parallel proposals for English only if or even if; Vidal 2017: 259-260). Note, however, that a compositional account that treats zhi as taking you as its first argument (or that combines with it by function composition) can easily derive this stable behavior of you (Wimmer 2022).
The price to pay for this overall solution is that we are forced to assume a kind of morphosyntactic polysemy for zhiyou. The two variants differ in whether an instance of zhiyou carries the [uSCAL(+)] feature, or not. 12 If all of this is on the right track and, specifically, if the presence of cai invariably indicates scalarity, one loose end still remains. Why is cai truly necessary in biclausal zhiyou-structures ('only-if'-structures) as in (27)  Even the most advanced treatment of such structures to date (Wimmer 2022) assumes no obligatory scalarity in this case, whereas such scalarity is assumed by Wimmer (2022) for zhiyou's recalcitrant counterpart zhiyao 'if'. It would certainly increase the symmetry of Wimmer's proposal if arguments could be found to the effect that zhiyou-conditionals always involve scalarity, too. I'll leave this for another occasion.

Conclusions
In this short article, I have critically reviewed some conclusions arrived at by Sun (2021) in her article on bipartite monoclausal zhiyou structures. I hope to have convinced the reader that, despite my criticism, Hole's (2015Hole's ( , 2017 and Sun's (2021) theories are probably both on the right track and are, in fact, rather similar. Sun delivers the bipartite syntax for the interaction of zhiyou-phrases with the scope-taking exclusive operator of variable syntactic height, while Hole delivers the syntax of zhiyou-phrases with the scope-taking scalarity operator. For concreteness's sake, I developed an explicit featural architecture strictly following Sun (2021) and Quek & Hirsch (2017) to derive the behavior of zhiyou-phrases and cai in interaction.
An interesting avenue for future research would be to clarify whether bi-clausal zhiyou structures ('only-if'-conditionals) should all be analyzed as structures involving scalarity. The morphosyntactic evidence renders this highly likely. the past about the lasting achievements of Viviane Alleton (especially Alleton, 1972). I feel greatly indebted to her pioneering work.
Funding Open Access funding enabled and organized by Projekt DEAL.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creativecommons.org/licenses/by/4.0/.