2.1. Chinese studies

Most Chinese studies focus on one oppositive connective. Studies on er generally believed that er in Classical Chinese has various uses, such as the contrastive ‘while/and’, the consecutive ‘and then’, the adversative ‘but’, the additive ‘as well as’, the causative ‘and so’, the conditional ‘if’,Footnote ⁵ etc. (Chen Reference Chen1994). Li (Reference Li2011) claims that the primary function of er is to mark coordination and that the causal or adversative relation indicated by er in Modern Chinese can be obscure. Because er and danshi can sometimes be used interchangeably, some scholars contend that er has an adversative function in Modern Chinese (Wang Reference Wang1955, Lü 1990). However, according to Yan (Reference Yan2009), er does not have an adversative use in Modern Chinese; it only indicates that the conjuncts are contrastive. In (5), for example, ‘what she did’ and ‘what she felt’ can be contrastive, and danshi can be changed with er.

Studies on the adversative relation indicate that danshi has the most general meaning among other adversative connectives, as it covers three subtypes of adversative relation: contradiction (as shown in Example (28)), denial of expectation and restriction (Xing Reference Xing1992, Reference Xing2002, Shi & Sun Reference Shi and Huiyan2010, He Reference He2016, Zong Reference Zong2012). From a diachronic perspective, they argue that the adversative connectives, such as danshi, buguo and zhishi, have all developed through three successive steps: ‘restrictive adverb + verb’ (e.g. dan ‘only/just’+shi ‘be’), restrictive adverb (only/just) and adversative connective. However, only danshi develops from the weakest adversativity (namely, restriction) to the strongest adversativity (namely, contradiction) (Shi and Sun Reference Shi and Huiyan2010, He Reference He2016). Regarding the reason, He (Reference He2016: 109–110) suggests that the restrictive p danshi q ‘absorbs’ the context-specific meaning of denial of expectation.Footnote ⁶

Some studies on bushi…ershi indicate that the negative conjunct in this construction involves either a truth-conditional negation, like (6), or a non-truth conditional negation (Shao & Wang Reference Shao and Yiguang2010, Zong Reference Zong2012, Tan 2016b), like (7):

However, some others argue that bushi indicates the negation of an existing utterance or thought rather than describing a state of affairs (Teng 1978, Yeh Reference Yeh1995). In this view, both (6) and (7) are non-truth-conditional (Zuo Reference Zuo2017, Reference Zuo2020). This issue will be discussed in detail in the rest of this paper.

Tan (Reference Tan2021) is the only study indicating that Modern Chinese uses er to mark corrections with post-positioned negation. An example is (8):

The first conjunct of (8) does not contain a wrong or inappropriate part that needs correction. Actually, (8) is more of a contrast between ‘the city that he went’ and ‘the city that he did not go’ than a correction.

In summary, Chinese studies tend to focus on the semantic traits of the discourse relation(s) embodied by each opposition connective, without giving much attention to how they are connected, let alone comparing the lexicalizations of Classical and Modern Chinese. They thus raise the following questions: 1) If the primary function of er is coordination, how do we explain its adversative and corrective use in Classical Chinese? Does it really have an adversative use in Modern Chinese? 2) How do the corrective connectives er in Classical Chinese and ershi in Modern Chinese relate to one another? What function does shi perform? 3) Why can restriction, denial of expectation, and contradiction be expressed by danshi, whereas contrastive and corrective cannot?

Studies in other languages have long been concerned with the connectives representing various opposition relations. In the following paragraphs, I will examine these studies to see whether they provide any ideas for addressing the questions left by Chinese studies.

2.2. Studies in other languages

Semantic studies have examined the distinction of opposition relations. For instance, Izutsu (Reference Izutsu2008) proposes four semantic parameters – the mutual exclusiveness of different compared items, the number and kind of compared items, the involvement of an assumption/assumptions, and the validity of segments combined – and claims that the last three can distinguish contrast, concessive and corrective. Jasinskaja and Zeevat (Reference Jasinskaja and Henk2008) indicate that i ‘and’, a ‘and, but’ and no ‘but’ in Russian impose different kinds of constraints on the ‘questions under discussion’ addressed by their conjuncts. Semantic analysis tends to focus on describing the semantic or syntactic properties of each opposition relation rather than explaining why certain oppositions can share the same linguistic form, and the description of semantic traits appears complicated at times. Moreover, as semantic and syntactic traits are often language-specific, it happens that Chinese connectives cannot be handled by accounts built on the data of other languages. For example, according to Karagjosova (Reference Karagjosova2012), the adverbial doch in German is concessive and anaphoric, whereas the conjunction doch is not anaphoric, allowing it to mark a wider range of contrastive discourse relations besides concession. Such phenomena are not easily paralleled in Chinese.

Some studies transcend language-specific details and identify more universal features. For example, Jasinskaja (Reference Jasinskaja2012: 1990) notes that the common feature of different uses of but is the contrast between ‘something positive’ and ‘something negative’: a positive statement and a negative statement in semantic opposition, an argument and a counterargument in argumentative use of but, a trigger of an expectation and a denial of it in denial of expectation, syntactically positive and negative applying to the logical form in correction. She thus highlights the central role played by negation in the semantics of adversative connective, which has been suggested in earlier studies. For instance, Anscombre and Ducrot (Reference Anscombre and Ducrot1977) argue that in p mais_PA q ( _PA for German aber and Spanish pero – namely, the adversative but), p is an argument for a possible conclusion r, whereas q is an argument against the conclusion r. Compared with p, q is assigned more argumentative force in favor of ¬r. In p mais _SN q (_SN for German sondern and Spanish sino – namely, the corrective but), p must be a sentential negation. By saying p mais _SN q, the speaker presents q as the justification for his refusal of p’s positive counterpart. According to Umbach (Reference Umbach2004, Reference Umbach2005), adversative precludes the possibility that the second alternative is true in addition to the first one, whereas corrective eliminates the possibility that the first option applies in place of the second one. Briefly, previous studies have demonstrated that negation and opposition are tightly related. The investigation into how negation aids in identifying the connection and difference between opposition relations will continue in this study.

Pragmatic analysis was also conducted in previous studies. To analyze the Hebrew aval (adversative) and ela (corrective), Dascal and Katriel (Reference Dascal and Katriel1977) propose the ‘layers of meaning’: from an inner ‘core’ of propositional content to an outer ‘shell’ of conversational implicature, via modality, illocutionary force, felicity conditions, etc. Koenig and Benndorf (Reference Koenig, Benndorf and Koenig1998) distinguish aber from sondern based on the type of inferential processes they trigger, which is related to logical expressions, world knowledge and Horn’s (Reference Horn1989) Q-Maxim and R-Maxim. As per Blackmore’s (Reference Blakemore1989, Reference Blakemore2000) examination within the framework of relevance theory (RT), adversative but indicates that the proposition it introduces is relevant as a denial of an ‘expectation’ created by the utterance of the first conjunct, whereas contrastive but indicates that there is no guarantee that each of the conjuncts is relevant in their own right. Based on the ‘pragmatic ambiguity’ within the three cognitive domains that she proposes, Sweetser (Reference Sweetser1990) suggests that but in denial of expectation is interpreted in the epistemic domain, whereas but in contrasting evaluations is interpreted in the speech-act domain.

Foolen (Reference Foolen1991) emphasizes the pragmatic interpretation of opposition relations as part of a general theory of interpretation, inferencing and reasoning that costs nothing extra. This study concurs with Foolen (Reference Foolen1991) that pragmatic approaches are more inclusive and eliminate the unnecessary complexity associated with some semantic research. However, the classifications of opposition relations remain puzzling despite some insightful pragmatic findings. For instance, some studies adopt the trichotomy of contrast, adversative and corrective (Spooren Reference Spooren1989, Blakemore Reference Blakemore2000, Soffner Reference Soffner2006, Izutsu Reference Izutsu2008), whereas some hold that the contrastive use of but is only a pragmatic realization of the denial of expectation (Lang Reference Lang1984, Foolen Reference Foolen1991, Winter & Rimon Reference Winter and Mori1994). Contrastive use also has multiple interpretations. For instance, it only refers to restriction in Jasinskaja (Reference Jasinskaja2012) but includes and-contrast (e.g. Peter lives in Paris, and Ana lives in Lyon.) in Blakemore (Reference Blakemore1989) and Izutsu (Reference Izutsu2008). The classifications of the adversative relation are also inconsistent. For example, Izutsu (Reference Izutsu2008) uses concessive and adversative interchangeably to refer to ‘denial of expectation’, whereas other studies indicate that adversative and concessive serve different functions (Oversteegen Reference Oversteegen1997, Sæbø Reference Sæbø2003, Malchukov Reference Malchukov2004).Footnote ⁷ Moreover, according to different studies, adversative relation may include any of the following opposition relations: contradicting evaluations, restriction, miratives, concessive, contradiction (Xing Reference Xing1992, Malchukov Reference Malchukov2004, He Reference He2016, Tan Reference Tan2021).

Furthermore, typological research incorporating semantic and pragmatic analysis does not incorporate Chinese data. For instance, Malchukov’s (Reference Malchukov2004) semantic map shows that correction is more closely associated with contrast than adversative without providing any linguistic evidence. Indeed, Modern Chinese’s contrastive connective er and corrective connective ershi, which share a similar form, might provide some support. Mauri (Reference Mauri2008) argues that in a language that uses the same construction for all forms of contrast relations, all the specific relations indicated by the general marker are undercoded and must be inferred from context. This is not the case for Classical Chinese, where the ‘general marker’ er typically codes the adversative relation. In addition, the adversative connective in Modern Chinese, danshi, originates from dan, the restriction marker in Classical Chinese, rather than from the adversative er. Typological investigations could have taken this phenomenon into account.

In this view, this study aims to obtain a general account of opposition relations that incorporates Classical and Modern Chinese data. Some approaches taken by earlier studies – such as the role of negation in the semantics of the polyfunctional adversative connective, the inference of opposition relations within RT, and their interpretation in the ‘three domains’ – are not language-specific and inspired this study. All these techniques, however, are thought to be amenable to further investigation.

3.1. The meta-representation approach and RT

Negation is central in the semantics of the polyfunctional adversative connective (Jasinskaja Reference Jasinskaja2012). The different uses of negation have been discussed with the adversative and corrective connectives. For instance, Anscombre and Ducrot (Reference Anscombre and Ducrot1977) hold that the metalinguistic negation licenses the corrective mais, whereas descriptive negation triggers the adversative mais. A more thorough delimitation of negation is provided by the meta-representation approach in RT, which may enable a further delimitation of the opposition relations.

Sperber and Wilson (Reference Sperber and Wilson1995: 232) propose that a representation can be used in two ways: either descriptively, representing some state of affairs (SOA) in virtue of its propositional form being true of that SOA, or interpretively, representing some other representation, which also has a propositional form. Every utterance is the interpretive expression of the speaker’s thought (Idem: 228). ‘The interpretation of a description’ determines a first-order interpretation, simplified as ‘representation’. ‘The interpretation of an interpretation’ always involves a second-order interpretation – namely, the meta-representation – which relies on the resemblance of the higher-order representation and the lower-order representation. The approach of meta-representation can be illustrated by Figure 1.

Figure 1. Representation and meta-representation.

The meta-representational approach has been applied to explain descriptive negation and meta-representational negation (Horn Reference Horn1985, Carston Reference Carston and Turner1999). The latter includes metaconceptual negation (refuting the content of an existing representation) and the metalinguistic negation (refuting the linguistic form of an existing representation) (Noh Reference Noh1998, Albu Reference Albu2017, Moeschler Reference Moeschler, Scott, Clark and Carston2019, Zuo Reference Zuo2020). The meta-representational negation has no truth-conditional contribution, as shown in the following example:

Only (9a) is truth-conditional, as it describes the objective world. (9b–c) echo another representation and add a refutative attitude toward its propositional content (e.g. 9b) or form (e.g. 9c) at the meta-representation level, thus having no truth-value contribution (Albu Reference Albu2017, Zuo Reference Zuo2020).Footnote ⁸

The ability of meta-representation allows one to infer the communicator’s intention: based on an ostensive stimulus, the addressee will construct a hypothesis about the speaker’s meaning, which satisfies the presumption of relevance conveyed by the utterance (Wilson 2002). An input is relevant to an individual when the processing in a context of available assumptions yields positive cognitive effects, which include ‘strengthening of the existing assumptions’, ‘contradiction and elimination of the existing assumptions’ and ‘derivation of a contextual implication’. The process of producing positive cognitive effects is also involved in the studies of opposition relations. Blakemore (Reference Blakemore1989) suggests that but means ‘and + something else’. This ‘something else’ is non-truth-conditional and related to how the interpretation of the but-conjunct contributes to the relevance, which varies in adversative and contrast. This study proposes that different opposition relations trigger various processes to positive cognitive effects, which will be discussed in Section 4.

3.2. Three conceptual domains

When examining opposition relations, Sweetser’s (Reference Sweetser1990) three domains are also underapplied, despite not being a novel concept.

According to Sweetser (Reference Sweetser1990: 76), a single semantics is pragmatically applied in different ways according to pragmatic context. ‘We model linguistic expression itself not only as a description (a model of the world), but also as action (an act in the world being described), and even as an epistemic or logical entity (a premise or conclusion of our world of reasoning)’ (Idem: 21). In this respect, a connective does not have multiple semantic values, but has meanings so general that they apply equally to the conceptions of content, epistemic and speech act domains, among which only the content domain has a truth-value contribution. The adversative but provides support for Sweetser’s (Reference Sweetser1990) three domains, as shown in the following example:

But in (10a) is interpreted in the epistemic domain because it is derivable from John keeps six boxes of pancake mix on hand that ‘John likes pancakes’, which is contradictory with the subsequent conjunct he never eats pancakes. As for (10b), suggesting eating in King Tsin and suggesting eating in China First are two speech acts that cannot be accepted at the same time, resulting in discordance. It is argued that the adversative but never appears in the content domain because it is hard to figure out how a discordance exists outside of the speaker’s mental concept of harmony or non-contrast (Sweetser Reference Sweetser1990: 104). However, the ‘ambiguity’ of the opposition connective – not only in adversative but also in corrective and contrast – is more intricate than what Sweetser (Reference Sweetser1990) addresses, which will be shown in the rest of this study.

In summary, RT and ‘three domains’ can be further explored to explain opposition relations, and not all opposition relations have been studied in previous research employing these two theoretical frameworks. This study will add a new perspective to investigate the opposition relations as well as the two lexicalizing systems in Classical and Modern Chinese.

4.1. The opposition relations as meta-representation or representation

The dichotomy of representation and meta-representation first allows distinguishing the contrastive relation from the adversative and corrective relations. The contrastive conjuncts describe two SOA that are not contradictory. For instance, born of misery and died of happiness in (1a) are two SOAs that do not contradict one another in the real world. Reversing the two descriptions of SOAs or omitting a connective does not change the meaning of the sentence. Therefore, by describing two coexisting SOAs, the contrastive conjuncts form a lower-order representation, which is truth-conditional. Adversative and corrective, however, indicate the higher-order meta-representation because they represent another representation and express a refutative attitude toward it.

Even though they both exist at the level of meta-representation, the lower-order representations that the adversative and corrective relations reject could be different. The three categories of lower-order representation identified by Wilson (Reference Wilson and Sperber2000: 414) are utterances (public representations), thoughts (mental representations) and sentences or propositions (abstract representations). The adversative conjunct contradicts the implied conclusion of the preceding conjunct, which is typically expressed in an attributed thought, as shown in the following example:

In (11a), fei neng shui ye ‘some people are not proficient swimmers’ implies that ‘they cannot traverse the river’. This implicature is attributed to the speaker as a thought, which is immediately refuted by jue jianghe ‘traverse the Yangtze and Yellow Rivers’. Likewise, from Zhangsan mai le henduo dangao ‘Zhangsan bought many cakes’, the addressees will attribute the thought ‘Zhangsan likes eating cakes’ to the speaker, which is refuted by the adversative conjunct ‘but he never eats cake’. In brief, the adversative conjunct refutes the ‘thought’ (mental representation) implicated by the preceding conjunct.

In contrast, the corrective conjunct can refute the existing representation embodied either in an attributed thought or in an attributed utterance, as shown in the following examples:

In (12a), the refuted representation is embodied in a thought – namely, ‘You set the rulers and subjects against each other’. Zichan attributed this thought to his interlocutor Zidashu because Zidashu’s question ‘What does that make our neighbor countries think of us?’ had a conversational implicature: ‘The neighboring countries would criticize us for dividing the rulers and subjects (by granting a fief to a minister)’. In (12b), what is refuted is an utterance (public representation). This utterance is refuted not because it is false at the level of representation but because the speaker finds it nonassertable at the level of meta-representation. To be more precise, for the speaker, the term xihuan ‘like’ is not appropriate to describe Zhangsan’s fascination with music. In this respect, a corrective conjunct (COR) is provided to indicate the meta-representational nature of the negative conjunct (NEG); otherwise, NEG will be regarded in default as an ordinary negation: Zhangsan does not like music.

According to the semantic study of Ludovico (Reference Ludovico2016: 125), ‘in corrective contexts, what is expressed in the first conjunct is not true, while the second conjunct is true under the same circumstances’. Dissociated from this point of view, this study argues that NEG and COR are both at the level of meta-representation, making no truth-conditional contribution. The meta-representational nature can be revealed by the negative marker bushi in Modern Chinese (Zuo Reference Zuo2019, Reference Zuo2020):

When bushi in (12b) is changed to bu, negative marker for descriptive negation, the two conjuncts are contradictory because bu xihuan (不喜欢) ‘Neg + like’ must be interpreted as dislike, which excludes all the elements higher than xihuan ‘like’ in the quantitative scale < neither dislike nor like, like, adore, addictive…> and blocks the metalinguistic interpretation. In languages like English and French, [NEG + GOR] can focus on the negated element and allow the metalinguistic interpretation. However, in Modern Chinese, ‘negative marker + gradual predicate’ must be interpreted as an integration (Lü Reference Lü1999, Zhao 2007) and cannot allow the metalinguistic interpretation.

Zhao (2007) suggests that the copula shi functions as a focus marker, which restricts the scope of the negation to the element being denied. That is why the metalinguistic negation of a gradual predicate must use bushi as a negative marker rather than bu or mei, as shown in (12b).Footnote ⁹ However, Zuo (Reference Zuo2017, Reference Zuo2020, 2021) contends that the purpose of shi is to denote echoic use rather than to concentrate on a single element. There are three reasons: First, even if shi is present in NEG, COR is still needed to identify the negated element, especially in a verb–object locution. Compare the metalinguistic negation in (12b) and the metaconceptual negation in (14):

Although shi appears in NEG, it is still not sure which element, like or music, is negated and corrected. That is why addressees will always expect additional information when hearing a bushi-negation (Shen Reference Shen1993).

Second, as a focus marker, shi should always precede the focused element. If the negated element had no syntactic position, there would be no location for shi in this sense, but this is not the case, as shown in the following example:

(15) refutes the perspective of the speaker, which has no specific syntactic position, but shi is still present.

Third, it has been argued by previous studies that the negative utterance employing bushi as a negative marker negates an antecedent representation (Teng 1978, Yeh Reference Yeh1995). In fact, in Modern Chinese, different negative markers can distinguish metaconceptual negation from descriptive negation, as shown in the following example:

Different from (14), the negative marker in (16) is bu, indicating that it is a descriptive negation. The two conjuncts in (16) describe two SOAs considered to be contrastive or additive. When the second conjunct is interpreted as an additive conjunct, the connective should be replaced with the anaphoric pronoun (i.e. ta ‘he’ in (16)). Based on the three arguments mentioned above, Zuo (Reference Zuo2019, Reference Zuo2020) proposes that shi marks that the utterance is to be treated at the meta-representational level. In this view, the metaconceptual bushi A ershi B, as in (14) and (15), is non-truth-conditional, although it refutes the content of A instead of its linguistic form.

In conclusion, the adversative connective refutes a representation inferred from an existing conjunct, which is always embodied in thought. The corrective connective refutes the propositional form or the content of an existing representation embodied either in thought or utterance. They are both at the meta-representational level and not truth-conditional. In addition to this difference, the adversative and corrective relations also differ in the process to yield cognitive effects, which will be presented in 4.2.

4.2. Production of cognitive effects

Both the adversative conjunct and the corrective conjunct yield positive cognitive effects by contrasting and eliminating an existing assumption (Blakemore Reference Blakemore2000, Iten Reference Iten2005, Moeschler Reference Moeschler2018, Zuo Reference Zuo2019), but they trigger different processes: the former eliminates the existing assumption by directly providing a contradictory new information, whereas the latter contradicts it through the process [NEG + COR]. Compare B1 and B2 in (17):

A’s utterance implies that Zhangsan has only three sons. B₁ eliminates this implication by directly providing contradictory information (he has four sons) and B2 first refutes A’s utterance by pointing out its inappropriateness and then provides new information. Both the adversative conjunct and the corrective conjunct provide new information that eliminates the existing assumption, but they do not trigger the same inference process.

Restriction and contrasting evaluations also eliminate an existing assumption by directly providing contradictory new information, as shown in the following example:

‘I like her’, derived from she is cute, is restricted by ‘I don’t like her that much’, derived from but a little lazy. (18) can also be understood as two contrasting suggestions: she is cute, so you should marry her, but she is a little lazy, so you should not marry her. The second conjunct eliminates the assumption that the first conjunct activates by providing contradictory new information.

Malchukov (Reference Malchukov2004) proposes that mirative differs from adversative in that the event denoted by the second conjunct is unexpected in its own right, without any relation to the first one. Dissociated from Malchukov (Reference Malchukov2004), this study believes that the initial conjunct still has an impact on the unexpectedness in mirative. According to RT, the interpretation of an utterance depends on the person’s cognitive environment, which includes short-term memory, medium-term memory, long-term memory (logical, encyclopedic and lexical information on concepts) and the physical environment in which communication occurs (Sperber & Wilson Reference Sperber and Wilson1995). In denial of expectation, the assumption eliminated by the adversative conjunct is inferred mainly based on short-term memory activated by the first conjunct, whereas in mirative, the eliminated assumption was primarily activated through long-term memory, as shown in the following example:

In (19a), ‘bump into a rock’ and ‘fall’ are causally related, so ‘he didn’t fall’ is unexpected based on people’s short-term memory created by the first conjunct. In (19b), there is no causal relation between ‘run’ and ‘fall’. The second conjunct is unexpected given the interlocutors’ encyclopedic knowledge that people normally do not fall when they run. However, the first conjunct still activates the information he will not fall, making he fell unexpected. According to Sperber and Wilson (Reference Sperber and Wilson1995), based on an ostensive stimulus, the addressee will follow the path of least effort when assessing cognitive effects and stop when their expectations of relevance are satisfied. The most relevant interpretation of he started to run should be he started to run in an ordinary situation because it requires the least effort. If the first conjunction indicates or implies that the subject runs on a rope or during a storm, it is very possible that the subject will fall, and the second conjunction is not unexpected.

In a nutshell, in denial of expectation, restriction, contrasting evaluation and mirative, the conjunct introduced by danshi creates positive cognitive effects by offering new information that eliminates an existing assumption implied by the first conjunct. This common process allows them to be grouped into the adversative relation.

The analysis within RT also explains the distinction between contrastive, adversative and corrective relations and the connections between adversative and corrective relations. To further explain the connections of all three opposition relations, I will examine them in the ‘three domains’.

5.1. The content domain

Two conjuncts in contrastive relation describe two coexisting and comparative SOAs, as shown in the following example:

In (20a), shesheng ‘give up their lives’ and quyi ‘take righteousness’ are two coexisting SOAs. They are semantically oppositive (give up A vs. take B) but not contradictory. Similarly, the two SOAs in (20b) are also not contradictory, although they are semantically oppositive (affirmative big vs. negative not sweet). Therefore, (20) falls under the content domain with a truth-value contribution. Both in Classical Chinese and in Modern Chinese, er connects two SOAs and presents their contrastive relation in the content domain. The next question is whether er is used as adversative connective in Modern Chinese. What distinguishes two conjuncts joined together by er from the same conjuncts joined together by danshi? For instance, only the connective distinguishes (21) from (20b):

Danshi requires the construction of a contradictory relation between two conjuncts, which moves the propositions from the content domain to the epistemic domain. As in (21), danshi can be considered to connect two contradictory comments: ‘these apples are good’ because they are big and ‘these apples are not that good’ because they are not sweet. Danshi may also connect two suggestions: ‘buy these apples’ because they are big and ‘do not buy the apples’ because they are not sweet. More information about this problem is provided in the section below.

5.2. The epistemic domain

A reasoning process based on epistemic beliefs is necessary in the interpretation of adversative conjuncts, as clarified by Sweetser (Reference Sweetser1990):

In (22a), the causal relationship between awful weather and not having fun exists in the social-physical world. In (22b), a person can be both rich and stupid at the same time in the real world. Therefore, (22a) and (22b) show the use of because and and in the content domain. They are both instances of the description of SOAs, which have a truth-value contribution. However, ‘if two states coexist in the real world, then they cannot be said to clash at a real-world level’ (Sweetser Reference Sweetser1990: 104). In (22c), ‘John is rich’ and ‘John is dumb’ coexist in the social-physical world, so they clash only in the epistemic domain: rich people are generally smart (smartness is often the reason why people become rich), but John is stupid. This ‘denial of expectation’ is realized through reasoning, based on the epistemic experience. Along this line, a process of reasoning is indispensable in adversative. If no adversative relation can be constructed, danshi and er are not interchangeable, as in (23):

The adversative varies in degree. As mentioned in 2.1, some Chinese studies have classified adversative relations (p danshi q) into contradiction, denial of expectation and restriction, ranging from the strongest to the weakest (Xing Reference Xing1992, Reference Xing2002, He Reference He2016). In denial of expectation, it is necessary to draw r from p, but not necessary to draw ¬r from p. As shown in (19a), from p ‘he bumped into a rock’, one can conclude r ‘he fell’, which is directly contracted by q ‘he did not fall’.

The ‘restriction’ in previous Chinese studies, which is considered to represent the weakest adversative relation, includes also contrasting evaluations (Xing Reference Xing1992, Shi & Sun Reference Shi and Huiyan2010, He Reference He2016). This study holds that contrasting evaluations are most likely to represent the weakest adversative relation because it is necessary to draw r from p and ¬r from p, and a special context is often needed to construct the adversative relation, as shown in the following example:

In an ordinary context, (24) appears illogical: wind and rain usually come together. It is thus hard to construct an adversative relation between it is windy and it rains in the epistemic domain. However, imagining that a director wants to film a scene on a sunny windy day, (24) can be interpreted as two contradictory speech acts: let’s begin shooting as it is windy and keep waiting until the rain stops. Given an appropriate context, even (23) can move from the content domain to the epistemic domain:

If (25B) is a response to the question in (25A), the first conjunct implies ‘you are right’ (because the wettest weather has been in Preston), whereas the second conjunct implies ‘you are wrong’ (because the driest weather has been in Ashford); these two answers are contradictory and thus connected by danshi. The adversative between two responses is constructed through reasoning; therefore, (25B) crosses over the epistemic domain and the speech-act domain.

Now, we see how the contrastive relation and the adversative relation are connected: without any reasoning based on the epistemic knowledge, two conjuncts in weak adversative in the epistemic domain can also be interpreted as a description of two comparative SOAs in the content domain. This is why danshi can sometimes be replaced by er in Modern Chinese. If the two conjuncts are connected by danshi, it is indispensable to draw contradictory conclusions from them, and the conclusion drawn from the second conjunct is semantically dominant, as it eliminates the first one. The weakest adversative relation is often interpreted both in the epistemic domain and in the speech-act domain, as shown in (24) and (25). A special context is needed to construct the weakest adversative relation. In fact, the more this context is accessible, the more the adversative relation is evident. In (26), the contrastive er is assumed to be adversative er in Classical Chinese since the context supports the adversative reading.

Big and no edge, go and return are coexisting SOAs in the content domain, but it is easy to put them in conflict since people frequently put positive and negative remarks about a person in opposition: it is good that he can develop and present his opinions, but it is not good that he does not put an end to his talk. As mentioned above, this weak adversative relation often represents contracting evaluations. Facing two coexisting SOAs, it is our epistemic experience that drives us to build an adversative relation in the epistemic domain. Danshi’s diachronic development can support this claim:

As shown in (27a), danshi originally represented ‘only be’, which then developed into the adverb ‘only’ in (27b). As indicated by Wen and Zeng (Reference Wen and Rong2018), the lexicalization of the adverb danshi marginalizes the previously central meaning just/only and highlights the previously marginal ‘slight twist’. This ‘slight twist’ is essentially a contrast. The two conjuncts in (27b) represent two contrastive SOAs; given a context, such as the question ‘Do the Void school and Nature school both explain things through exclusion?’, an adversative relation can be established between the two conjuncts in (27b). Due to its frequent use in contexts where both contrastive and adversative readings are discernible, the adverb danshi has evolved into an adversative connective that denotes varying degrees of adversativity. As shown in (27c), the danshi-conjunct can represent a restriction to the first conjunct: one should be in awe of the honored one, but not always. It can also be interpreted as a denial of expectation: people generally do not point out the flaws of the person they respect, which the second conjunct denies.

It is also notable that the ambiguous er in (26) is different from that in (2). In the latter case, it is just essential to deduce r (indigo blue should be less blue than the indigo plant) from q (indigo blue is extracted from the indigo plant); r and p (but it is bluer than the plant) will immediately contradict each other. Therefore, er in (2) is unambiguously adversative because no contrastive relationship can be built. Likewise, in Modern Chinese, the adversative connective danshi cannot be replaced with the contrastive connective er if there is no contrastive relationship existing between the two conjuncts, as shown in (19).

Whereas contrasting evaluations are the weakest adversative relation, the contradiction is the strongest adversative. According to Xing (Reference Xing2002: 45–57), ‘contradiction expresses two SOA that coexist in the real world and are opposed to each other’, as shown in the following example:

In the contradiction ‘q danshi p’, q is almost equal to ¬p, so the reasoning processes from q to r and from p to ¬r both seem to be dispensable. This situation, in which the two SOAs coexist and clash in the real world, seems to be overlooked by Sweetser (Reference Sweetser1990). However, how can two SOAs coexist and be contractionary at the same time? It is obvious that statements like ‘I saw him, and/but I did not see him’ are unacceptable. According to this study, contradictions always require a reasoning process. For instance, in (28), q (they feel like knowing all the colleagues) implies that ‘they just feel knowing them, but not really know them’, and it is contradictory with r ‘they cannot feel like knowing them’, which is derived from p (they don’t know any of the colleagues). Compare (29a) and (29b):

The coexistence of ‘we know each other’ and ‘we do not know each other’ in (29a) is unacceptable because they directly contradict each other. In contrast, in (29b), ‘we know each other’ is implied rather than explicitly stated, allowing hearers to distinguish between ‘I know her, and she knows me’ and ‘we know each other’. (29b) makes sense when hearers understand that ‘we know each other’ suggests that we have profound communication, as opposed to ‘I know her, and she knows me’, which implies ‘we do not have any profound communication’. In short, contradiction is not a description of two contractionary SOAs in the real world. It is interpreted in the epistemic domain since it still requires a reasoning process.

In brief, the adversative relation is interpreted in the epistemic domain, where a reasoning process is required. When it is necessary to deduce r from p and ¬r from q, the adversative relation is weak. If there is no appropriate context, the two conjuncts may be read as contrastive. In the contrasting evaluations, the utterance is not only interpreted in the epistemic domain but also in the speech-act domain. In the next section, I will provide another instance of the adversative relation that is interpretable in the speech-act domain.

5.3. The speech-act domain

Shen (Reference Shen1993) and Shi and Sun (Reference Shi and Huiyan2010) propose an adversative danshi that could be interpreted in the speech-act domain but has not been addressed by Sweetser (Reference Sweetser1990), as shown in (30):

The most common answer to ‘she succeeded in losing weight’ should be expressing astonishment, congratulations or compliments, but (30) instead asks a question, violating the assumption of what others will say about her success. As a result, (30) is interpreted both in the epistemic domain and speech-act domain.

Corrective relation is different from the cases involving both the epistemic domain and speech-act domain – such as (30) and the contrasting evaluations in (24) – as it does not involve any reasoning process and is interpreted only in the speech-act domain. As stated in previous sections, NEG in correction refutes an existing representation rather than describing SOA; hence, it should be a sentential negation. As indicated by Anscombre and Ducrot (Reference Anscombre and Ducrot1977), the speaker presents COR as a justification, which does not necessarily require that COR logically implies the falsity of the positive counterpart of NEG. For instance, bushi xihuan yinyue ‘NEG + like music’ in (12B) is interpreted either as ‘less than like’ or ‘more than like’ because xihuan yinyue ‘like music’ is refuted as an inappropriate expression, not a false proposition. This is why the negative marker must be the bushi, which marks the echoic use.

Prior research has shown that NEG and the succeeding COR can only appear in a single statement of one speaker, which is supported by the fact that the syntactic component that the negative and corrective conjuncts share must be dropped (Anscombre & Ducrot Reference Anscombre and Ducrot1977; Shen Reference Shen1993; Iten Reference Iten2005). According to this viewpoint, COR is not an independent utterance, but is just a justification of the preceding refutation of an existing representation. Therefore, NEG and COR should be interpreted as an integration in the speech-act domain.

By far, I have investigated how the three opposition relations differ from one another as well as the connections between contrast and adversative relations and adversative and corrective relations, respectively. I will now examine the connections between contrast and correction based on Chinese data. Corrective relation has been considered as ‘contrastive negation’ (Silvennoinen Reference Silvennoinen2019), suggesting that contrast and corrective are interconnected in some way. I contend that the corrective connective connects two meta-representations that coexist in the speech-act domain, just as the contrastive relation describes two SOAs that coexist in the content domain, as shown in the following example:

B₁ and B₂ both refute A’s utterance and point out its flaws: B₁ refutes it explicitly with a negative proposition, and B₂ implies its opposite through a positive proposition. Their coexistence in the speech-act domain and their ability to be reversed without altering the meaning of the utterance serve as the first examples of their contrastive relation, as shown in (32):

Furthermore, the contrastive relation and the corrective relation may both be emphasized or not. In the content domain, if the contrast is not emphasized, the conjuncts can be connected by a comma instead of er. Compare (33) and (4):

(33) places less emphasis on the contrastive relation than (4). Similarly, in the speech-act domain, (32a) and (34) are both acceptable; however, (32a) favors the corrective information, whereas (34) draws attention to the contrast between ‘not come from Beijing’ and ‘come from Tianjin’:

The data of Modern Chinese confirm the metaphorical relationship between contrast and corrective: the connective representing contrast is er, whereas the connective representing corrective is ershi. The latter has just one more character shi than the former. As presented in 4.1, shi serves as the marker of echoic use, indicating that the utterance should be interpreted in the speech-act domain instead of the content domain. In this view, the difference between er and ershi in Modern Chinese is that the latter is used in the meta-representational level and the speech-act domain, implying that correction is the metaphorical mapping of contrast from the content domain to the speech-act domain.

In sum, based on the ‘three domains’, the contrastive relation is between two truth-conditional descriptions of SOAs in the content domain. The adversative relation is interpreted in the epistemic domain or in both the epistemic domain and the speech-act domain. The corrective relation is always interpreted in the speech-act domain. Each of these three opposition relations has a connection to the other two: in a particular context, the contrastive relation can be moved from the content domain to the epistemic domain and understood as a weak adversative relation. The adversative and corrective conjuncts are both exhibited in the cognitive domains which are non-truth-conditional.Footnote ¹⁰ The corrective relation is the mapping of the contrastive relation in the speech-act domain.