Lexical projections other than verbs can have specifiers that function as their subjects. This constituent, lacking any functional projection [_SC Subject XP], was called a small clause (SC) because unlike matrix clauses (cf. (2a) above), SCs in English allow for NPs, Adjectival Phrases (AdjP) and PPs as predicates without the copula:2

An analysis involving SCs was likewise proposed for matrix sentences with copular verbs such as be, become, seem in combination with NPs, AdjPs or PPs as predicates, by among others Partee (1986), Heycock (1992), Moro (1997), Bowers (2001), den Dikken 2006 (cf. Citko 2011 and Balazs 2012 for detailed discussion and references). This led to a structure where the matrix copular verb selects an SC complement:

Returning to the main focus of this article, i.e. SCs in non-root contexts, examples (5)–(10) provide the complete array of non-root SCs postulated for English, in general subdivided into the following types:

While included here for the sake of exhaustiveness, constructions with bare VPs (cf. (10)) are in general not subsumed under SC, because they show important differences with respect to the other types of SCs (cf. Matushansky 2019, section 2).3 First, bare VPs in English are restricted to complements of let, have, help, make, and of modals as well as perception verbs, thus contrasting with the many environments allowed for non-verbal SCs. Second, bare VPs are unacceptable with copular verbs, and third, the external argument of a verb can be absent in passives or middles. Nothing similar is observed for the subject of non-verbal predicates. These properties of bare VPs and the properties of the external argument of verbal predicates are captured by the introduction of v/Voice. The head v/Voice is to be distinguished from the functional head Pred° postulated to accommodate the non-verbal predicates in SC (cf. Baker 2003: 37–39 for further arguments against unifying v/Voice and Pred°).

Stowell’s (1981) initial hypothesis that an SC corresponds to the projection of the lexical category serving as its predicate, with the subject in its specifier, was inter alia challenged by data involving movement of an apparent non-maximal projection (cf. (11a) and (12a) from Svenonius 1994) and an already occupied specifier position for non-bare SC predicates, given the prohibition of multiple specifiers (cf. (13a) from Williams 1983):

This led to proposals postulating a functional head in SC such as Pred° mediating the predicational relation (cf. Bowers 1993). As a result, small clauses no longer involved different lexical projections, but a unique functional projection (headed by Pred°).

As demonstrated by Matushansky (2019), current minimalist theory can account for the cases in (11)–(13) without postulating any functional category in SCs such as Pred° resulting in the same categorial identity for all SCs.

(11–12) can be analysed as raising-to-object (Johnson 1991; Lasnik & Saito 1991 a.o.): the subject in the SC complement of a verb moves to SpecVP, and the verb raises to a higher head. The sentence can then be derived via moving the entire AdjP SC, a maximal projection, containing the subject trace:

Concerning (13), the prohibition of multiple specifiers has been abandoned in the meantime for theoretical and empirical reasons (cf. Matushansky 2019, section 2.3) and can thus no longer serve as evidence for the PredP-analysis of SCs.

In general, as argued for extensively by Matushansky (2019), the theoretical and semantic considerations adduced as arguments in favour of the obligatory presence of Pred° (or other functional head) in SCs do not bear further scrutiny or are amenable to alternative analyses (cf. section 4 below for Marelj & Matushansky’s (2015) analysis of ‘take NP₁ for NP₂’ in English, Serbo-Croatian and Russian as ditransitives). The identical distribution predicted by a uniform PredP analysis of SC is not observed, either (cf. Matushansky 2019: 67–73). As a consequence, Matushansky (2019) rejects the idea that SCs are headed by the same functional head and returns to Stowell’s original analysis of SCs as lexical projections.

Recast into the Minimalist Program, Matushansky (2019) proposes to treat predication as the last thematic merge to an extended projection of a lexical head:

Given the now heterogeneous syntactic status of SCs (DP, NP, AdjP, DegP etc.), it is the semantics which acts as common denominator, i.e. the predication relation with its resulting propositional denotation.

Taking a different angle, Bruening (2018) likewise challenges an SC analysis of resultative constructions (She hammered the metal flat), caused motion constructions (Jerome waltzed Matilda across the room), particle verb constructions (They sponged the water up), and double object constructions (Melinda wrapped her friend a present). His main argument is that an SC analysis predicts wrong semantics, because it uncouples the object DP from the verb event by representing it exclusively as the subject of the SC: hammer [_SC the metal flat]. Bruening (2018) thus confirms Marelj & Matushanky’s (2015) approach and removes further constructions from the domain originally claimed to involve SCs.4

While for Marelj & Matushanky (2015), Matushansky (2019) and Bruening (2018) there still remain genuine SCs in English and other languages, Chinese will be shown to lack SCs altogether, i.e. non-root clauses licensing the predicative function of an XP in a form unacceptable for the same XP when in a matrix context.5 This is independent of the approach chosen, SCs as a unique functional projection (e.g. PredP) or as extended projections of a lexical head.

Given the lack of SCs in Chinese, the reader might ask how the English sentences with non-root SCs in (5–9) are translated into Chinese. The more so as proponents of SCs in Chinese focus on adjectival SCs as in They consider [_SC John intelligent] (cf. (17a) below), for the simple reason that the lexical items in the Chinese translation show the same linear order as in English. However, when going beyond the surface, we see that a different construction must be used for translating each type in (5–9), none of which involves an SC.

The Chinese equivalents of the English sentence (16a) with a denominative SC are given in (16b) and (16c):

(16b) shows the verb rènwéi ‘think, assume’ that selects a clausal complement, with the obligatory copula shì ‘be’ and the nominal predicate. Unlike English John/him a genius, Zhāngsān shì tiāncái is a well-formed independent sentence: ‘Zhangsan is a genius’, hence not an SC. (Cf. Section 6 below for discussion of matrix sentences with copular verbs in Chinese). (16c) features the ditransitive verb dāng ‘take sb. for, treat as, consider as’, which requires the bǎ construction (also cf. Sections 5.2 and 5.3 below).

The Chinese equivalent (17b) of an adjectival SC in English (17a) does not feature an SC, either. For in Chinese, adjectives such as cōngmíng ‘be intelligent’ function as autonomous predicates, both in matrix clauses (cf. (17c) and in complement clauses (cf. (17b)) (cf. Section 4.2 below for further discussion):

Concerning the English absolute construction [_SC With Peter sick/out of office], we’ll never get the job done in time, it has has no Chinese equivalent.

For the English resultative SC in (18a), the Chinese equivalent (18b) once again features the bǎ construction with a ditransitive compound verb, selecting two nominal complements:

Turning to subject and object depictives, the adjectival SC predicate in English subject depictives (19a) must be encoded in Chinese as a preverbal adverb or a preverbal adjunct clause (whose null subject is co-indexed with the matrix subject).

By contrast, there is no systematic correspondance in Chinese for English object depictive SCs as in (20a):

(20b) features a verb compound shēng-chī ‘raw-eat’ = ‘eat sth. raw’, while in (21a) the object depicting ‘hot’ is rendered as an adverb, hence in preverbal position. The fact that this possibility is not available in (22) shows the idiosyncratic nature of this case; instead, nóngnóngde ‘strong’ must be encoded as a modifier of the object DP.7

The so-called “mad magazine” sentences [_SC Me mad]?! Ridiculous are irrelevant for Chinese, given that adjectives can in any case function as predicates without the copula in both root and non-root contexts (cf. (17b–c) above).

To summarize, the Chinese translations of English sentences with non-root SCs display very different constructions, none of which involves an SC. While it is in principle possible that languages might have some cases of SCs without showing the entire array known from English, this is not the case in Chinese, which lacks SCs altogether.

Concerning constructions with bare VPs as complements of a small set of verbs (modals, perception verbs and causative verbs) such as Paul saw [_SC Peter cross the street], they are in general not subsumed under SCs in English, because they show important differences, linked to v and the dichotomy external vs internal argument (cf. Section 2.1 above), with respect to the other types of SCs.

Notwithstanding this important insight, we encounter claims to the contrary in Chinese linguistics, a.o. in Yang Daran (2003) and Shen & Sybesma (2006), discussed below as per the request by an anonymous reviewer.

There are two classes of intersective adjectives in Chinese. (40a)–(40c) illustrate the adjectives functioning as autonomous predicates (labeled predicative adjectives by Chinese scholars) and corresponding to scalar adjectives:

(41a–c) feature the so-called non-predicative adjectives, corresponding to absolute adjectives, which occur in a “nominalized form” (cf. Paris 1979) and therefore require the copula shì. Recast in an analysis where the subordinator de instantiates different heads on the D-spine, among them Det°, this nominalized form is tentatively analysed as a DP with a covert NP complement: [_DP adj [_De’ de [_NP Ø]]. (Cf. Paul 2012, 2017a for discussion of the subordinator de and the structure of DP.)12

Note that so far the literature has not provided a more precise analysis than Paris’ (1979) nominalization approach; shì ‘be’ …. de is often presented as a “discontinuous” constituent, given that both are obligatory (cf. (41b–c)).

Returning to predicative adjectives, since, as their name suggests, they are autonomous predicates, they combine directly with negation and adverbs (cf. (40a)–(40c) above) and are incompatible with the copula shì ‘be’:13

By contrast, with non-predicative adjectives, it is the copula shì as verbal head that is negated or modified by adverbs (cf. (43), (44)). Requiring a nominal projection as complement, the copula is incompatible with an adjectival complement (cf. (45)):

When functioning as secondary predicates for object DPs, the same facts hold: predicative adjectives are acceptable on their own and exclude the copula shì ‘be’ (cf. (46)), whereas non-predicative adjectives appear in the nominalized form and require the copula shì ‘be’ (cf. (47)), exactly as in matrix contexts:

Incidentally, the obligatoriness of shì ‘be’ in (47)–(48) substantiates Matushansky’s (2019: 84) stand against the presence of a null Pred° in SCs. If indeed there were a null Pred°, it should be capable of turning any intersective adjective into a predicate, which is visibly not the case. Instead, an overt copula is required for the class of non-predicative intersective adjectives.15

The situation for adjectives thus confirms our observations for nominal projections, viz. that there are no differences between root and non-root contexts for their predicative function.

However, an anonymous reviewer points to a possible semantic asymmetry between root and non-root contexts and suggests this as potential evidence in favour of adjectival SCs. The starting point is the well-known fact that in Chinese, a bare scalar adjective as predicate in general conveys the comparative degree:

Accordingly, the same bare form occurs with an explicit standard of comparison:

When modified by degree adverbs (hěn ‘very’, tèbié ‘particularly’, tài ‘too’), adjectives only give rise to the positive degree interpretation. Furthermore, when hěn ‘very’ is not stressed, it does not add any lexical meaning and therefore remains untranslated, in contrast to the other degree adverbs (cf. Paul 2015: 151–156 for discussion and references; also cf. C.-S. Luther Liu 2018).

Interestingly, in some non-root contexts such as complement clauses and conditional clauses, a bare adjective can be interpreted in the positive degree:

However, bare adjectives in matrix clauses can likewise indicate the positive rather than the comparative degree. This is the case in coordinations (53), in yes/no questions (54a) and under negation (54b):

Accordingly, the occasional possibility of a positive degree interpretation for bare adjectives in some non-root contexts can not be (mis)taken as evidence in favour of a systematic root vs non-root asymmetry and hence as evidence for the existence of adjectival SCs.

There is ample evidence for the existence of both prepositions and postpositions in Chinese (pace a.o. Huang, Li & Li 2009; Cheng & Sybesma 2015). For detailed discussion and references, cf. Ernst (1988); Djamouri & Paul (1997; 2009); Djamouri, Paul & Whitman (2013); Paul (2015, chapters 3 and 4).

We have seen that like verbs, scalar adjectives function as predicates, exclude the copula and are directly negated with bù. This holds both for root and non-root contexts (e.g. sentential subjects, complement clauses, secondary predicates).

Among the nominal projections, NumPs are licit predicates in root and non-root contexts. When negated, however, NumPs either require the negated copula bù shì ‘not be’ (bù shì NumP) or the negated form of the verb yǒu ‘have’, i.e. méi yǒu ‘not have’ (méi yǒu NumP). NPs, but not DPs, may sometimes function as predicates on their own, but this constrained possibility is limited to root contexts. In non-root contexts, the copula shì ‘be’ is obligatory.20 The copula is also required for negation, irrespective of root or non-root context.

The predicative function is in general excluded for Adpositional Phrases (AdpPs), notwithstanding the presence of the copula shì ‘be’. More precisely, there are only very few cases of well-formed PP-predicates involving the copula shì ‘be’, and the conditions at work here are unclear. Concerning PostPs, they are incompatible with the copula shì. Some native speakers accept PostPs with a NumP complement as autonomous predicates in both root and non-root contexts, treating them on a par with NumPs.

Crucially, the constraints holding for a given XP as predicate in a non-root context are the same as those holding in root contexts, i.e. an XP unacceptable as predicate in root contexts is likewise unacceptable as predicate in non-root contexts. NP predicates are an exception here, insofar as the copula is required in non-root contexts, while sometimes optional in root contexts. Importantly, this asymmetry is exactly the opposite of what is observed for SC predicates in e.g. English, where an NP is licit without a copula only in the non-root SC context, but requires the copula in root contexts.

So far, secondary predicates have served as the non-root context par exellence when checking the possible predicative function of lexical XPs in root and non-root contexts. The analysis of the constitutent XP following the matrix object in structures such as (68a)–(68d) as a secondary predicate goes back to C.-T. James Huang (1984; 1987). He provides three arguments against the analysis of this XP as a DP-internal modifier; while some of his observations have been questioned in the meantime, his overall analysis still remains unchallenged.

First, Huang (1987: 231–232) observes that the acceptability of a secondary predicate depends on the properties of the matrix VP. Besides yǒu ‘have’, other transitive verbs are allowed when suffixed with the perfective aspect -le or the experiential aspect -guo. Verbs with the progressive aspect zài in general exclude a secondary predicate on the matrix object (cf. (68d–e)):21

Huang (1987: 231) tries to capture the common factor of these different VP formats by stating that “the verbs used all have to do with ‘existence’ of some sort”, while being very well aware of existing counterexamples.22 For the purpose of this article, his approximation is sufficient, because the aim is to show that the format of the predicate is the same in both matrix and secondary predicates, independently of the conditions under which a secondary predicate is possible. Note that no more accurate analyses have been proposed since Huang (1987).

Second, there is also a constraint on the object DP itself in order to be compatible with a secondary predicate: it must be specific indefinite, to the exclusion of definite DPs and non-referential bare nouns (cf. Huang 1987: 249):23

Third, a secondary predicate is excluded from wh-questions (Huang 1987: 249, (86)):

Note, finally, that only postverbal DPs allow for a secondary predicate.

None of these constraints hold for DP-internal modifiers followed by de; a DP-internal modifier XP does not depend on the aspectual nature of the VP nor on the specificity of the DP and the position (pre- vs postverbal) it occurs in, and a modified DP is naturally acceptable in a wh-question:

For all these reasons Huang (1984; 1987) adopts an analysis of the XP following the matrix object DP as a secondary predicate, which does not form a constituent with the object DP. Instead, the matrix verb, the matrix object DP and the secondary predicate are “dominated by VP”, hence are “sisters to the verb” (Huang 1987: 232–233).24 This also allows him to maintain the robust generalization that in Chinese, the modifier always precedes the modifiee in the nominal projection.

Taking up Chomsky’s (1980) observation that a purposive clause is a predication on a main clause NP, Huang (1984: 569; 1987: 252: fn 5) proposes an analysis of the secondary predicates in Chinese on a par with purposive clauses and analyses them as CPs. The empty category in object position is analysed as a variable bound by an abstract operator in the embedded C, which in turn is coindexed with the matrix object by the Generalized Control Rule (GCR).25

Huang’s (1984; 1987) structure in (72) has basically remained unchallenged up to today, as witnessed by a.o. Lin Jo-wang & Tsai Wei-tian (2015: 117) who maintain his operator analysis and treat secondary predicates as “integrated non-restrictives of some sort”. Note that they do not address at all the internal structure and label of the projection including the verb, the matrix object DP and the secondary predicate, although it is evident that Huang’s analysis from back in the 1980’s with a ternary-branching VP can no longer hold. Before addressing this latter issue, let us first examine the size/projection of the secondary predicate.

The CP analysis of purposive clauses raises a number of problems. It is unclear what the status of abstract operators is within the Minimalist Program (Chomsky 1995 and his subsequent works). On the empirical side, as argued for by Wei Haley Wei & Y.-H. Audrey Li (2018), the size of purposive clauses with a null subject in Chinese cannot be CP, given that e.g. topicalization is only possible to a position in the matrix clause, not within the purposive clause:

In addition, while manner adverbs and adjunct PPs and NPs are allowed, higher adverbs and auxiliaries are excluded from purposive clauses:26

Wei & Li (2018) show that these constraints hold for purposives irrespective of the presence or absence of lái ‘in order to’. While a purposive headed by lái projects a PredP (with lái as Pred°), a “bare” purposive without lái ‘in order to’ projects a vP. Importantly, both types of purposives are first merged with the verb, just like control complements. The resulting projection merges with the matrix object DP:

Purposives thus correspond to infinitival complements of control verbs such as kāishǐ ‘begin’, jìxù ‘continue’, which have the size of vP according to Huang (2017), not to the larger projection selected as complement of control verbs such as jìhuà ‘plan’, zhǔnbèi ‘prepare’, which allow topicalization within their control clause (Wei & Li 2018: 317, (41)):

While topicalization in (76) is subject to variation among native speakers, the contrast with (73b) above nevertheless demonstrates that topicalization can serve as a diagnostic for different types of complements. Also note that according to Huang (2017), the acceptability of topicalization inter alia patterns with the presence of auxiliaries such as yào ‘will’.

As demonstrated below, while disallowing topicalization (cf. (77b)), secondary predicates can feature auxiliaries (cf. (77a)), aspectual suffixes (cf. (78)) and negation (cf. (79)) and thus differ both from purposives (excluding auxiliaries) and the WollP control complements (allowing for topicalization):28

Allowing for auxiliaries and aspectual suffixes, a secondary predicate is thus clearly larger than purposives (vP or PrdP according to Wei & Li 2018). At the same time, it lacks a projection able to host a topicalized DP, unlike the WollP complement. The secondary predicate can therefore be plausibly assumed to project a TP with an always covert subject.

Concerning the hierarchical position of the secondary predicate in the sentence, it must be merged with either VP or vP. Merging with a higher projection in TP is excluded by the overall syntax of Chinese, where due to the systematic head-initiality of the extended verbal projection (including TP), postverbal material must be merged within the VP, as emphasized by Huang (1987: 232) himself: “That the XP [i.e. the secondary predicate; WP] when it appears, is under VP, but not immediately under S, is assumed in all discussions.”29

Two scenarios are possible: either the secondary predicate is merged with the verb, as proposed for purposive clauses by Wei & Li 2018 (cf. (80a)), or it is merged with the VP (cf. (80b)). Merging with vP (or AspP) is not an option, because it would make it difficult to rule out adjunction of adverbial XPs, which are completely excluded from the postverbal position, a very robust fact of Chinese syntax known since Huang (1982) (cf. Paul 2017b for discussion and references).

Concerning (80a), it is not clear whether the secondary predicate should be likened to a purposive control clause, i.e. to a complement of the verb (cf. (75) above), complementation and predication not being the same.

This potential problem does not arise for (80b), where the secondary predicate TP is merged with VP (also cf. Irimia 2005). PRO within the secondary predicate TP is coindexed with the matrix object via “cyclic” or “weak” c-command and a predicative relation is established. (For the relevance of “weak” c-command in Chinese, cf. Huang, Li & Li 2009: 335). For the resulting projection, the label TP seems plausible, thus leading to a configuration where the matrix verb (raised to v) now combines with a clausal complement.

It is correct that Huang (1987) rejected such an analysis on the grounds of problems with subcategorization, because the open class of transitive verbs would have to be stipulated to select a clausal complement instead of a DP, provided the latter is followed by a secondary predicate. However, he also noted that the various constraints holding for the aspect type in the VP allowing for a secondary predicate on its object DP (cf. (68) above) as well as the unacceptability of secondary predicates in wh-questions (cf. (70) above) can in any case not be captured by the selectional restrictions imposed by a transitive verb. Last, but not least, for sentences with the existential verb yǒu ‘exist’ as matrix verb, Huang (1987: 236; 1988: 57) himself tentatively suggested an analysis of the secondary predicate and the matrix object as constituting a clause:30

Further research must decide between the configurations (80a) and (80b), because the few studies on secondary predicates after Huang (1987) (cf. a.o. Tsai 1994; Lin & Tsai 2015) never address the precise hierarchical position of secondary predicates on the clausal spine.

Looking at (81) one might question the necessity of the covert subject PRO in the secondary predicate as assumed so far. However, as shown in (83), a secondary predicate requires an overt matrix object in postverbal position, not a silent copy thereof.

This is different from complement clauses:

The contrast between (83) and (84b) indicates that the secondary predicate indeed features a PRO. Since PRO is subject to the GCR (cf. footnote (25)), this leads to the undesired co-indexation with the nearest DP, i.e. the matrix subject wǒ ‘I’ in (83), given the lack of an overt matrix object.31

Finally, as already demonstrated by Huang (1987), the projection consisting of the secondary predicate TP and the matrix object can not be a DP, because this would wrongly predict the acceptability of the resulting DP elsewhere than in the postverbal object position, i.e. in all the other positions available for DPs, such as the subject position, complement of adposition position or as object in the bǎ construction.

To conclude, while secondary predicates clearly involve a non-root context, they cannot be analysed as SCs, because they contain material (auxiliaries, aspect, negation, adverbs) normally said to be excluded from (genuine) SCs and thus project a TP with an always covert subject. Furthermore, the XPs allowed as predicates here are identical with those in matrix clauses.32 As a consequence, secondary predicate TPs are likewise acceptable as root clauses on their own, Chinese being a pro-drop language. In Chinese, there are thus no non-root contexts where otherwise non-predicative XPs are licit as predicates, a situation which was the very reason to postulate SCs to begin with. The non-existence of ECM verbs in Chinese, discussed in the next section, further supports the lack of SCs.

However, the Chinese translations corresponding to (85a–b), often cited as illustrating ECM verbs with an SC complement, turn out to simply involve verbs selecting a clausal complement (for rènwéi ‘think’, cf. already Huang 1987: 235):

This error is partly due to the fact that mainly English adjectival SCs were chosen and translated into Chinese (c. a.o. Tang Sze-Wing 1998), despite the well-known caveat by Y.-H. Audrey Li (1985: 270–72, note 8; 1990: 130–134). She argued against ECM verbs in Chinese, emphasizing the well-formedness as an independent clause of tā hěn bèn ‘He is stupid.’ in sentences such as (86), given the predicative nature of scalar adjectives in Chinese. This contrasts with English him foolish in I consider [him foolish] which is not an independent clause (also cf. Tang Ting-chi 2000: 209, fn 34).33

As soon as nominal predicates are included, this is obvious, because the copula shì ‘be’ is obligatory here, as is the case in matrix contexts:

The verbs rènwéi ‘think, assume’, yǐwéi ‘believe’ etc., often presented as ECM verbs on a par with English consider, believe etc., in fact select a full-fledged clausal complement, perfectly acceptable as an independent root clause, featuring inter alia negation (cf. (86) above) and auxiliaries (cf. (88) below). Accordingly, this clause is not as “bare” as expected for a SC, characterized by the absence of tense, aspect, modality.

Importantly, the subject (tā ‘s/he’ in (86–88)) is case-licensed within the embedded clause, not via the matrix verb rènwéi, hence the unacceptability of the bǎ construction in (89), which presents tā ‘he’ as object of the matrix verb rènwéi ‘think’:34

Furthermore, if rènwéi and other alleged ECM verbs such as xiāngxìn ‘believe’, xián, tǎoyàn ‘dislike, mind’ etc. indeed selected SC complements (as claimed by a.o. Niina Zhang 2016), these alleged SCs should also feature non-predicative intersective adjectives (cf. (90)), PostPs (cf. (91)) and PPs (cf. (92)), in addition to predicative AdjPs, licit as autonomous predicates. As shown below, this prediction is not borne out by the data: