2.1 A lexicalist approach: Li (Reference Li1990, Reference Li1998)

Li (Reference Li1990, Reference Li1998) offers a lexicalist account of V₁-V₂ RCs. In this account, V₁-V₂ predicates are a combination of their component verbs. Li attributes the complex theta-role assignments in these constructions to rules that derive the V₁-V₂ predicates’ theta-grids from the theta grids of the individual verbs. In the formation of a resultative complex V₁-V₂, shared theta-roles of V₁ and V₂ are first merged and assigned to a single argument. This reduces the number of arguments required by the V₁-V₂ to a maximum of two, and circumvents the violation of the Theta Criterion. Then, head-feature percolation ensures the theta-roles of a resultative complex follow the thematic hierarchy of the head verb of the complex, that is to say, V₁. In terms of zhui-lei in (5), its component verbs have the theta-grids in (6), with V₁ assigning two theta-roles and V₂ assigning one.

In Li (Reference Li1990), the three readings of sentence (5) are explained as different identification possibilities, as presented in Table 3, with “ = ” symbolizing identification of theta-roles.

Table 3. Varied theta-identifications for different readings of sentence (5)

Assuming that V₁ is the head, the theta-grid of zhui-lei ‘chase-tired’ needs to follow zhui's thematic hierarchy. Li's analysis can easily accommodate Readings A and C but not Reading B, because, in Readings A and C, the agent of the head verb V₁ appears in the external argument position of sentence (5). As for Reading B, it involves reversed theta-role assignments, violating the thematic hierarchy percolated from V₁ zhui ‘chase’. To solve this, Li introduces causative structures wherein the causer is held responsible for putting the causee in a specific state. It is stipulated that the causative hierarchy < Causer < Causee >> is able to override the thematic hierarchy percolated from a head verb. Following this, the patient of V₁ and the agent of V₁ under Reading B are assigned the new roles of causer and causee respectively, and are thus not subject to V₁'s (zhui) thematic hierarchy. The introduction of causative structures does provide an extended explanatory scope, but it also renders Li's analysis less integrated. A complex predicate's theta-grid is derived from several incongruent rules, and the analysis involves stipulations and lacks independent theoretical motivations. Also, Li's approach raises but does not answer the question of how patients can be understood to be causers.

2.2 Conceptualizing RC sentences as events

Hoekstra (Reference Hoekstra and Roca1992: 160) proposes that an RC sentence encodes an event with a result state, and the predicate representing the result state functions to “turn a non-telic predication into a telic one, by specifying the state which terminates the event”. Sybesma (Reference Sybesma1999) follows this conceptual basis and analyzes Mandarin RCs on a par with their English counterparts. In Sybesma's treatment of V₁-V₂ RC sentences, V₂ and its argument (also the internal argument of the complex) are included in a small clause. For example, the underlying structure of (1b) contains a small clause [_SC table clean], as seen in (7), whose structure is based on Sybesma's analysis.

1. (7) Partial derivation of sentence (1b) following Sybesma (1999):

   

Immediately above the small clause is the projection of Extent Phrase, which serves as the complement of V₁. V₂ within the small clause needs to undergo head movement and combine with V₁ to derive the surface form. Unfortunately, Sybesma's analysis can account for sentence (5) only under Reading A, which is, in essence, a Type I RC sentence; it fails to derive Type II RC sentences such as (2). In this sentence, a bottle of wine got the drinker drunk through the drinker's drinking; a drinking event is terminated by the drinker's insobriety. Following this reasoning, the result state (i.e., V₂ ‘drunk’) and the result state's holder (i.e., the drinker Lisi) should originate in the small clause, and the event (i.e., V₁ ‘drink’) should be merged as a V head above ExtP. A problem arises when V selects Lisi as its specifier, because then there are two copies of Lisi (the drinker); one in the small clause and the other in the specifier of VP. Meanwhile, the sentential subject ‘wine’ is not represented at all.

Lin (Reference Lin2004), Ramchand (Reference Ramchand2008) and Liu (Reference Liu2019) demonstrate approaches that appeal to interactions at the syntax-semantics interface. They agree that predicates should be analyzed semantically as events, but they go further and decompose predicates into semantic primitives of events. These semantic primitives are mapped to syntactic argument structures. In Ramchand (Reference Ramchand2008), predicates are decomposed temporally into the subevents of initiation, process and result. The three subevents are mapped onto syntactic structures as the functional phrases, initP (initiating/causing projection), procP (process projection) and resP (result projection). In terms of an RC sentence, there is an initiation phase with the necessary conditions leading to the result; after experiencing the process during which a certain action happens, the affected entity obtains a result state. Following this analysis, sentence (1b) should have the derivation as in (8). The external argument ‘waiter’ is introduced as an initiator and merged as the specifier of initP. The internal argument ‘table’ is not only the undergoer of change/process, but the resultee (i.e., entity holding the result state), so that ‘table’ is introduced twice as the arguments in both procP and resP. Though Ramchand does not go into the details of the derivation of RC sentences, it is conceivable that some mechanism could be adopted to reduce one copy of ‘table’ and ensure a surface form that does not crash.

1. (8) Partial derivation of sentence (1b) following Ramchand (Reference Ramchand2008):

   

When it comes to deriving Type II RC sentences, it is unclear how roles are assigned, especially those of Initiator and Undergoer. In interpreting and deriving sentence (2), for instance, the undergoer can either be the drinker or alcohol, as both of these entities undergo a change or a process. The change involved in (2) is getting drunk, but the process involved is drinking, indicating inconsistent undergoers. Besides, the referent of Initiator changes depending on what is initiated, the process of drinking or the change. In short, Ramchand's account with Initiator and Undergoer faces problems when an agent acts on a patient/theme but at the same time the agent undergoes a change.

Lin (Reference Lin2004) also analyzes predicates from the perspective of event structures, but unlike Ramchand (Reference Ramchand2008), Lin makes use of a different set of semantic primitives (including do, become and be) to decompose a V₁-V₂ resultative complex. These primitives are realized as verbalizing heads little v, which select for verbal roots and license the roots as verbs. Semantically, the three little vs represent three subevents; v _DO realizes the causing event and v _BE realizes the result state, while v _BECOME represents the cause-and-effect connection between the causing event and the result state. Structurally, a v _DO head takes a v _BECOMEP as complement, which in turn has v _BEP as complement. In (9), I present how Lin would derive sentence (1b). V₂ combines with v _BE and V₁ combines with v _DO as modifiers to produce little vs with enriched meanings. Little vs at lower levels then undergo head movements to the highest little v and merge into the complex predicate surfacing as V₁-V₂ (not pictured in (9). As zhuozi ‘table’ is both the entity with a state of change and the entity subject to the action ‘wipe’, it is represented at two places in (9), with an overt form represented as v _DO's argument and an empty form PRO encoded as v _BECOME's argument. The introduction of PRO saves the Theta Criterion. As v _DO's argument is the closest c-commanding DP for PRO, PRO is necessarily controlled by v _DO's argument, ensuring the co-indexation between the two.

1. (9) Partial derivation of sentence (1b) following Lin (Reference Lin2004):

   

In line with Kratzer (Reference Kratzer1996) and Pylkkänen (Reference Pylkkänen2002), a Voice head is merged to introduce the external argument waiter. Lin recognizes that non-agentive entities can also trigger changes of states, for example, ‘wine’ in the Type II RC sentence (2). In this case, to use v _DO to represent the effect of ‘wine’ is inappropriate. It is then proposed that causers like wine have special properties/states capable of producing certain effects that require another projection of v _BE to be inserted between VoiceP and v _DOP. This signifies a specific property/state and mediates between the non-agentive causer and its relevant result state. Despite Lin's efforts to accommodate Type II RC sentences, I believe it is counter-intuitive to claim that an activity-denoting v _DO is always included in the derivation, even when alcohol, which does not ‘do’ anything, has an effect on its drinker.

Concerning the syntactic structure in Lin (Reference Lin2004), Liu (Reference Liu2019) argues that v _BE should be removed, as its presence should allow adverbs that modify states to appear and modify V₁-V₂ resultative predicates, which turns out to be false. Consider the contrast in (10).

Moreover, the contrast in (11) demonstrates that adverbs in V₁-V₂ RC sentences modify the causing events but not the caused events (i.e., result states), as is the case in English direct causative sentences. For example, grumpily in the sentence “John awoke Bill grumpily” cannot be used to modify the state of awake and mean that John is/was awake grumpily.

Following Pylkkänen's (Reference Pylkkänen2002) analysis of direct causatives like awake, Liu proposes that V₁-V₂ resultative predicates involve v _CAUSE, rather than v _DO. The replacement of v _DO with v _CAUSE also helps account for why adverbs modifying activity predicates (i.e., those with v _DO) are not necessarily compatible with V₁-V₂ resultative predicates, as seen in (12). Besides, v _CAUSE more appropriately represents the effect that alcohol has on its drinker than v _DO does.

Following the above amendments concerning little vs, Liu's analysis suggests a revised structure in (13) for sentence (1b). With v _DO's replacement by v _CAUSE, ‘table’ is no longer subject to the action ‘wipe’, so it is represented only once as the entity with a change of state.

1. (13) Partial derivation of sentence (1b) following Liu (Reference Liu2019):

   

Note that the introduction of v _CAUSE is not without its problems. Recall that ‘depressed feeling’ is demonstrated to be a possible cause for one's getting drunk through drinking. If v _CAUSE does exist, there is no straightforward reason why the RC sentence in (4) is not acceptable with ‘depressed feeling’ as the external argument. In the analysis conducted in this paper, v _CAUSE is excluded from the derivation of Mandarin RC sentences and the roles played by v _CAUSE will be undertaken by other existing functional heads.

2.3 An overlooked phenomenon: multiple possible effects following from a single event

In an attempt to explain the fact that the same individual/entity can appear in different argument positions as in (14), Croft (Reference Croft2012: 233) points out that in mental events of emotion, cognition and perception, there can be two directions of transmission of force: “the experiencer directs their attention to the stimulus and the stimulus causes a change of (mental) state in the experiencer”. The two sentences in (14) thus embody forces in different directions, though the force from the music to the listener caused a change, and the force from exerted from the listener to the music did not.

1. (14)

   1. a. I listened to the loud music.

   2. b. The loud music frightened me.

The bidirectionality of force transmission is not unique to mental events. Returning to the sentences in (1), a waiter performed a “wiping” action, which made the wiped entity clean. For the event with the physical action “wiping”, the waiter's force directed toward the wiped entity (i.e., table) has an effect on it; it gets the wiped table clean. It is easily conceivable that the same event of “wiping” can also result in an effect on the wiping agent (i.e., waiter), for example, the waiter's tiredness. The latter effect reflects the force transmission from the wiped entity to the wiping agent, and is characterizable by the V₁-V₂ RC sentence in (15). Actually, the effects embodied in all Type II RC sentences apply to agents of the relevant causing actions, and do not apply to the patients/themes.

The effect of a causing event can apply to entities completely external to the causing event as well. RC sentences with this type of effect are represented by Type III. In the Type III sentence (16) (repeated from (3)), Zhangsan's crying got a handkerchief wet; the handkerchief's (the result state holder) presence is not a necessary condition for Zhangsan's shedding tears. However, the handkerchief did get wet because of Zhangsan's crying, demonstrating an outward effect from a causing event. Therefore, there are multiple potential effects arising from different entities involved in a causing event that can be directed toward different entities internal or external to the causing event.

Many studies of Mandarin RCs, especially those analyzing RCs from the perspective of event structures, generally overlook the multiple potential effects following from a single causing event. As seen in the previous section, those event-based analyses realize that a Mandarin RC sentence involves an event followed by a result state, with the former characterized as the causing event, and the latter as the caused event/effect. As causing events usually include sentences with of action verbs and their agents and patients/themes, it is easy to dwell on the potential effect that an agent has on a patient/theme and ignore other potential effects following from the same causing event, for instance, the effect a patient has on an agent. Therefore, previous event-based analyses fail to distinguish the possible different directions of effects within a causing event, which makes these analyses inaccurate with regards to what roles the external arguments and the component verbs (especially V₁) in a V₁-V₂ RC sentence play, and how they are licensed.

For instance, the external arguments in Mandarin RC sentences are considered sometimes to be agents but sometimes to be causers; even when they are unified into initiator or causer roles, it is still unclear which individuals/entities can be licensed as such. A V₁ in a V₁-V₂ RC is often characterized as the manner in which a result state is caused, for example, Liu (Reference Liu2019). While a waiter can make a table clean in the manner of wiping, it is inappropriate to make the claim for sentence (15) that the table got the waiter tired in a wiping manner. As will be seen in section 3, a recognition of the multiple effects following from a single causing event and a distinction between these effects are critical for both the interpretation and the production of Mandarin RC sentences. To characterize the multiple potential effects, Croft has used the notion of force, which captures the causal power from any entity in an event, without necessarily entailing an actual effect.

The notion of force is not new; it has been adopted to characterize dynamic verbal meanings in the cognitive linguistic tradition in Talmy (Reference Talmy1988, Reference Talmy2000), Croft (Reference Croft1991, Reference Croft1998), Langacker (Reference Langacker2003) and others. Formal literature (Asher Reference Asher1992; Portner Reference Portner1998) also mentions the notion of force in an intuitive sense in order to characterize some external or unexpected effects. Moreover, the notion of force is explicitly exploited in Shen and Sybesma's (Reference Shen and Sybesma2010) analysis of the Mandarin ba-construction, which is closely associated with Mandarin resultative constructions. Nearly all the Mandarin RC examples introduced in this paper can be easily rephrased with the addition of ba, forming ba-construction sentences. For example, the Mandarin RC sentence (1b) with ‘wipe-clean’ has been modified into a ba-construction sentence as in (17). According to Shen and Sybesma, object DPs in ba-construction sentences are affected by external forces, eminating from the subject DPs preceding ba. In terms of sentence (17), it should encode the meaning that the waiter exerts a force, causing the table to experience a change of state, that is, become clean. Mandarin RC sentences are almost semantically identical to their ba-construction counterparts (e.g., sentence (1b) and sentence (17)) and only differ in word orders. Therefore, the notion of force can possibly be extended to the interpretation of Mandarin RC sentences as well.

With the introduction of “force”, the two different directions of the effects encoded respectively in sentences (1b) and sentence (15) actually reflect the bidirectionality of forces (or causal influences) between the two participating parties (i.e., agent and patient/theme) in the causing event “the waiter's wiping the table”. Moreover, the handkerchief's getting wet in (3)/(16) should arise from the force directed from Zhangsan to the handkerchief. Following this line of thought, the result state encoded in an RC sentence can be attributed to a specific entity involved in the relevant causing event, rather than the event as a whole. This will allow conceptual structures of RC sentences to match their corresponding linguistic representations, in which the subject/object arguments refer to specific entities, rather than events. Therefore, this article undertakes a reconceptualization of Mandarin RC sentences using forces. It maps force structures to syntactic structures and argument realizations, leading to a more precise analysis of V₁-V₂ RCs and more plausible answers to the problems concerning V₁-V₂ RC sentences. To accomplish this, I will first show how force structures are represented linguistically.

2.4 Introducing the force-theoretic framework for event structure

Copley and Harley's (Reference Copley2015) force-theoretic framework originates as an alternative account of accomplishment verbs, such as open in John opened the door. Traditional approaches consider accomplishment verbs to be composed of two sub-events chained together in a causal relationship, ich is to say the causing sub-event e₁ “John's opening” and the result sub-event e₂ “the door's being open”. This chain is represented as $\exists $e₁$\exists $e₂: e₁ CAUSE e₂. When it comes to a sentence like (18), the two sub-event analysis runs into problems, as no result sub-event e₂ has occurred.

1. (18) Mary was painting the dresser black, but she didn't finish.

In order to account for the non-culmination of the event in (18), Copley and Harley (Reference Copley2015) develop a syntax-semantics interface theory of accomplishment verbs which draws on the notion of force. According to these authors, an accomplishment verb encodes a force, which represents the causal influence from a specific source. This force is inherently defeasible and thus entails no necessary effect. Semantically, forces are realized as a new semantic type, f(orce): <s,s>, denoting the function from an initial situation (s₀) to a final situation (s₁) that occurs if nothing external intervenes. In the sentence “John opened the door”, s₁ is encoded in the small clause [_SC the door open], and s₀ is the situation immediately before s₁ with the door closed. This sentence has the basic structure as in (19), with v _BECOMEP representing the force leading to the situation change from “door closed” to “door open”.

1. (19) Partial derivation of “John opened the door”:

   

The external argument John is “introduced by a Voice head, which takes a predicate of forces as its complement and returns a function from individuals to forces” (Copley and Harley Reference Copley2015: 125).

The force-theoretic framework is also applied to compositions of other Vendlerian eventuality types, based on the underlying conception that dynamic predicates are forces while stative ones are situations. Note that ‘dynamic’ here is defined based on whether an input of energy is involved, rather than whether any motion or change occurs. Such a treatment successfully captures non-culminated cases of accomplishment verbs, and sheds light onto the question of why maintaining verbs (e.g., keep and stay) behave like change-indicating predicates (e.g., become, cry), but do not encode any kind of change. Imagine a door installed with a mechanism such that its resting state is to always be closed. A sentence like “The door stays open” does not encode any change of state, but it can take progressive aspect and become “The door is staying open”, which demonstrates a dynamic quality. Following a force-theoretic treatment, the dynamic quality of maintaining verbs is ascribed to energy input rather than actual changes of state. Although no change of state happens to the door, an external force is needed to counteract the door's desire to be closed. The intervention by external forces is more noticeable in the sentence “Mary is keeping the door open”, though including the force's source Mary is not necessary to license the dynamic quality and the progressive staying open. Therefore, this force-theoretic framework not only accommodates sources and recipients of forces in syntactic representations, but tackles problems confronting event-based analyses, for example, non-culminating cases of accomplishment verbs.

3.1 Adapting the force-theoretic approach to Mandarin V₁-V₂ RCs

A force-theoretic approach to Mandarin V₁-V₂ RCs begins with a conceptual reinterpretation of these RCs with forces and situations. Resultative constructions are traditionally characterized by causing events and result states. From the perspective of Copley and Harley (Reference Copley2015), the result states represent new situations; to bring in new situations, forces must be involved. In a V₁-V₂ RC sentence, it is V₂ that encodes the resultant property in the new situation, while the internal argument of this sentence encodes the holder of the resultant property. This is demonstrated by the tests below, where sentence (20a) (replicated from (1b)) and sentence (20b) only differ in V₂, and sentence (20a) and sentence (20c) only differ in their internal arguments. A change of V₂ results in a completely different resultant property for ‘table’; a replacement of the internal argument results in a different entity being subject to the resultant property.

Turning back to (1b)/(20a), the table in s₁ is clean, and in s₀ the table was not as clean. Forces are defined as the causal functions from initial situations to new situations; in (1b)/(20a), the force is what changes the table from “not being as clean” to “being clean”, and the table undergoing this change is the force recipient. The subject waiter is this force's source, because it is the waiter that makes the change happen. In terms of V₁ ‘wipe’, it seems to represent the manner of the relevant force, as a replacement of V₁ in (1b)/(20a) with shua ‘scrub’ (cf. sentence (21)) only suggests a difference on how the change happening to ‘table’ is realized. Based on the arguments above, a preliminary mapping between linguistic elements and conceptual interpretations are provided in (22). Each linguistic element corresponds to a conceptual primitive. A force structure involves four conceptual primitives, which are all relevant to and indispensable to the relevant force.

The mapping relations in (22) are derived based on Type I RC sentences. When similar reasoning is applied to Type II and Type III RC sentences, the mapping relations still hold except in the case of V₁. For instance, the Type II RC sentence in (23a) (copied from (2)) involves the causal influence from ‘wine’ to the drinker Lisi, leading to the drinker's intoxication; however, it is less appropriate to claim that ‘wine’ got Lisi drunk in the manner of ‘drinking’ (the relevant V₁ here). Moreover, the Type III RC sentence in (23b) (copied from (3)) contains a new situation with “handkerchief being wet” resulting from a force directed from Zhangsan to the handkerchief, but it is far-fetched to make the interpretation that Zhangsan got the handkerchief wet in the manner of ‘crying’.

In this article, I propose that V₁ represents a property of the relevant force, indicating the causing action's input of energy, which either directly or indirectly leads to the situation change.Footnote ⁵ For instance, sentence (20a) and sentence (21) both encode the waiter's force (or causal influence) applied to the table, which made the table clean, but the two sentences differ in the use of V₁, suggesting that different causing actions were responsible for the table's getting clean, that is to say, wiping and scrubbing. As for the Type II RC sentence in (23a), V₁ ‘drink’ is a property denoting the relevant causing action, which serves as a prerequisite to the causal influence from ‘wine’ to Lisi. Though the influence from ‘wine’ to Lisi is passive, the presence of V₁ ‘drink’ here still encodes the input of energy that makes wine's influence on a drinker possible. Similarly, ‘cry’ in (23b) encodes the causing action of ‘crying’ that helps realize Zhangsan's influence on the handkerchief. With this changed interpretation of V₁, a revised mapping from linguistic elements to conceptual interpretations is presented in (24).

1. (24) Linguistic elements⇐⇒Conceptual interpretations/primitives (revised)

   V₂: relevant property in the new situation s₁

   IA (internal argument): holder of the relevant property (force recipient)

   EA (external argument): source of the force that leads to the change from s₀ to s₁

   V₁: property suggesting the causing action with an input of energy

In terms of the derivation of a V₁- V₂ RC sentence, Copley and Harley's treatment of “John opened the door” can be used for reference. The accomplishment verb open has a V₁- V₂ counterpart, da-kai, in Mandarin, and da-kai literally means ‘hit-open’. The morpheme da ‘hit’ here has lost its semantic meaning and only functions to make the form disyllabic, which is much preferred over monosyllabic forms in modern Chinese (Duanmu, Reference Duanmu2007). In the same way that open is decomposed in (19), complexes like da-kai ‘hit-open’ can be derived as well, with da represented as v _become. Here, however, I will not use ‘open’ and will instead use sentence (1b)/(20a) as an example derivation, so that comparisons can be easily made with previous analyses.

Integrating all the conceptual primitives, sentence (1b)/(20a) depicts a force that makes the table clean. The force is asserted from the waiter to the table, and is realized through the causing action of wiping. Following the example derivation of “John opened the door” in (19), I propose that sentence (1b)/(20a) is formally represented as in (25) with one more layer of intermediate projection, vP. The small clause in (25) represents the result situation of the table's being clean’; v _BECOMEP represents the force that evokes the situation change.

1. (25) Partial derivation of sentence (1b)/(20a) following Copley and Harley's (Reference Copley2015) analysis:

   

Note that V₁-V₂ resultative complexes are interpreted within a lexical-decomposition syntax, so that V₁ and V₂ are understood not as independent verbs, but as verb roots. Therefore, V₁'s root $\sqrt {\mkern 1mu} $wipe denotes a property about the force (i.e., suggesting the causing action), modifying the force that leads to the situation change denoted by v _BECOMEP.Footnote ⁶ Structurally, the property-denoting $\sqrt {\mkern 1mu} $wipe adjoins to the force predicate vP, while compositionally, $\sqrt {\mkern 1mu} $wipe and v _BECOMEP are combined by Predicate Modification. After that, an empty voice head is merged and introduces the force's source ‘waiter’. Following Copley and Harley's denotations for v _BECOME and Voice, I provide the parallel semantics involved in the derivation of sentence (1b)/(20a) below.

Copley and Harley (Reference Copley2015) also hint at the morphological derivation of complex predicates. In the case of ca-ganjing ‘wipe-clean’, V₂'s root $\sqrt {\mkern 1mu} $clean first undergoes head-movement to v _BECOME; the intermediate outcome [$\sqrt {\mkern 1mu} $v ^o]_v^o then combines with $\sqrt {\mkern 1mu} $wipe via m-merger (Matushansky Reference Matushansky2006), deriving the surface form ca-ganjing.

3.2 Pragmatic enrichment in comprehending Mandarin RC sentences

As seen in (24) and (26), the linguistic elements in a Mandarin V₁-V₂ RC sentence can be mapped to their respective conceptual primitives, which are then integrated to derive this sentence's foundational interpretation: because of a causing action, represented with V₁, EA has a causal influence on IA, leading to a change of state for IA. This foundational interpretation can be represented semantically, as in (27).

1. (27) Foundational interpretation of a V₁-V₂ RC sentence:

   [[VoiceP]] = λf. fin(f) = V₂ holds of IA in fin(f) & V₁(f) & source(f) = EA

For instance, sentence (1b) or (20a) has the foundational interpretation that because of the ‘wiping’ action, the waiter has a causal influence on the table, leading to the table's being clean. However, this foundational interpretation does not specify the agent/theme of wiping, which is different from the common sense understanding that the waiter wiped the table, making the table clean. There is thus a gap between a sentence's foundational interpretation and its common sense understanding. As a matter of fact, previous event-based analyses differ from the current force-theoretic analysis in that they rely on the common sense understanding with two clear-cut sub-events, that is to say the causing event “the waiter's wiping the table” and the result state “the table's being clean”.

This article acknowledges that people's common sense understanding of an RC sentence tends to come with a causing event and a result state, but the thematic relations indicated in a causing event (between external/internal arguments and V₁ in a V₁-V₂ RC sentence) are proposed to be derived from pragmatic inference, rather than formal structures. In the following, I will first argue that there is no uniform pattern specifying how each argument is thematically related to V₁, which suggests that it is implausible to encode these thematic relations in formal structures. Then I will demonstrate that the thematic relations manifested in causing events can be cancelled, confirming that pragmatic enrichment plays a role in building the thematic relations. Therefore, the current force-theoretic analysis allows flexibility in interpreting a Mandarin RC sentence, and this flexibility is not seen in previous event-based analyses. As will be shown, the flexibility based on various possibilities of pragmatic enrichment can account for why a Mandarin RC sentence may have multiple interpretations. Moreover, the interpretation with reversed theta-role assignments will no longer be problematic, because arguments in Mandarin RC sentences are interpreted in relation to the relevant forces, rather than relative to V₁. The seemingly reversed theta-role assignments actually embody the causal influence from the patient/theme of a causing action to the agent of that action.

For the Type I RC sentence in (1b)/(20a), the external argument ‘waiter’ and the internal argument ‘table’ are associated with the property-denoting V₁ ‘wipe’ as its agent and its theme respectively. However, there can be great variability concerning how arguments in Mandarin RC sentences are thematically related to V₁. Sentence (28a) is copied from (2), which involves a reversed arrangement of theta-roles in relation to the property-denoting V₁ ‘drink’, with the external argument serving as the theme of V₁ ‘drink’ and the internal argument taking the role of agent.

There are also times when the internal argument of an RC sentence is irrelevant to V₁ in that sentence, that is, not a necessary part to make the causing event happen, as in Type III RC sentences. For instance, in sentence (28b) (copied from (3)), the internal argument ‘handkerchief’ does not receive any theta-role from its relevant V₁ ‘cry’. As the causing event involved in an RC sentence needs to be reconstructed by relating this sentence's arguments to V₁ thematically, the great uncertainty and variability in building the thematic relations makes it very difficult to propose a uniform pattern that maps each argument to a specific theta-role in relation to the property-denoting V₁. Unfortunately, previous event-based analyses generally take it for granted that the interpretation that contains a causing event and a result state is readily available, and rarely address the issue how people, in the face of an RC sentence, come to derive this interpretation, especially the relevant causing event.

The gap between the foundational interpretation and the common sense understanding of an RC sentence must be bridged. Huang et al. (Reference Huang, Audrey Li and Li2009) propose that this can be done by inference based on common sense. This paper also holds pragmatic inference responsible for the uncertain and variable theta role assignments in relation to V₁. For instance, common sense requires the external argument ‘waiter’ and the internal argument ‘table’ in (1b)/(20a) to be the agent and the theme of to the causing action of ‘wiping’ so that ‘waiter’ can get ‘table’ clean. However, these thematic relations indicated in a common sense understanding can be cancelled in a hypothetical world where one can get a table clean by simply wiping something on a screen, and this action performed on the screen will trigger a mechanism that makes the table clean. So in this hypothetical world, the internal argument ‘table’ is not necessarily the theme of ‘wipe’, as evidenced in (29).

As for sentence (28a) with the reversed theta-role assignments, the external argument ‘wine’ and the internal argument Lisi serve as the theme and the agent in relation to the causing action of ‘drinking’ because, based on common sense, only in this way can the wine get Lisi drunk. Therefore, when people perceive an RC sentence, the foundational interpretation tends to be enriched with pragmatic inference based on common sense. Pragmatic enrichment centers around how arguments are thematically related to the property-denoting V₁, which helps reconstruct what happened before the resultant situation s₁, and provides a more concrete idea of the causing action.

Pragmatic enrichment can account for the existence of multiple readings of a single sentence. Recall sentence (5) (replicated as (30) below) with three possible readings. The foundational interpretation of sentence (5)/(30) only presents the idea that through the causing action of chasing, Zhangsan got Lisi tired.

Both Reading A and Reading B conform to the foundational interpretation, though they differ in the causing events reconstructed, specifically, how each argument is associated with V₁ zhui ‘chase’. This is where pragmatic enrichment comes in. The differences between Reading A and Reading B should derive from different enrichment possibilities. According to common sense, Zhangsan can get Lisi tired either when Zhangsan chased Lisi, or when Zhangsan was chased by Lisi. In the former scenario, Zhangsan's chasing caused Lisi to run, which tired Lisi; in the latter scenario, Zhangsan's running required Lisi to chase hard, which got Lisi tired. The two scenarios involve quite different causing events, but both allow the causal influence from Zhangsan to Lisi, with Lisi becoming tired. The two enrichment possibilities then lead to two readings (i.e., A and B) with different causing events. As to Reading C, it is derived differently, and will be discussed in a later section.

Another instance involving multiple readings is provided in (32). Its foundational interpretation only suggests that through the action of singing, a song brought a large audience to tears. However, there is no definite answer to what happened before the audience's shedding of tears, and it is unclear who did the singing.

It is possible that the audience themselves sang and were moved to tears, or that some other person (e.g., a famous singer) sang and moved them to tears. In the second case, it is difficult to reconstruct the event of singing with a definite singer if no other contextual information is provided. This demonstrates that pragmatic enrichment is constrained and that it is not always possible to reconstruct a complete picture of a causing event, that is, with all necessary entities to realize the causing action. This poses a great problem for those analyses that take causing events for granted or assume a fixed pattern in reconstructing causing events from Mandarin RC sentences.

3.3 Settling force structures in the production of Mandarin RC sentences

The previous subsection has shown that in interpretation, people can reconstruct from a Mandarin RC sentence a scenario with a specific causing event and its result state by integrating the conceptual interpretation of each linguistic element in that sentence and then enriching the integrated foundational interpretation with pragmatic inference. When it comes to the production of a Mandarin RC sentence, the derivation in (25) illustrates how a V₁-V₂ RC sentence can be derived step by step, with the necessary building blocks. But if one intends to frame RC sentences within a real-life scenario where an event causes multiple resultant situations, it is necessary to first settle the force structure to be represented, which includes four conceptual primitives: the source of the force, the recipient of the force, the property of the force (characterizing the causing action) and the resultant property in the new situation. As these conceptual primitives all have their corresponding linguistic elements, a Mandarin RC sentence cannot be derived without determining the relevant force structure. Specifically, the DP representing the source appears as an external argument; the DP encoding the recipient appears as an internal argument; the verb representing the causing action serves as V₁ in a V₁-V₂ complex, and the verb representing the resultant property is V₂. This ensures that only possible linguistic elements (external argument, internal argument, V₁ and V₂) are mapped to form RC sentences. In this way, a Mandarin RC sentence can be derived by mapping the conceptual primitives of a force structure to corresponding linguistic representations. Previous event-based analyses only distinguish causing events and result states, so result states are ascribed to causing events, without specifying the entity that causes a result state. It is therefore no wonder that they tend to face problems in deriving RC sentences, especially in determining their external arguments, whose referents are the specific entities to which result states should be ascribed.

For instance, the event “Zhangsan's beating a big drum” can result in any of the three resultant situations in (33). These situations affect different entities, that is, the theme of beating, the agent of beating, and an entity irrelevant to the causing event. Therefore, if encoded by RC sentences, these situations will correspond to three different types. Since force is defined as the causal influence leading to a resultant situation (f: <s₀, s₁>), the three resultant situations (i.e., s$_1^a $, s$_1^b $ and s$_1^c $) in (33) represent three different forces (i.e., f^a, f^b and f^c), though they share a common s₀. Note that the three resultant situations can co-exist; if so, these situations should belong to one general situation. But since a Mandarin RC sentence only allows the expression of one entity's resultant property at a time (i.e., only one force is encoded each time), it is necessary to first decide which resultant situation or which entity's resultant property is to be expressed. In the following, I will assume that (33) presents three separate resultative situations, and illustrate how the force structure relating to each situation is determined and mapped to linguistic elements to derive the corresponding RC sentences.

It is relatively straightforward to articulate the scenario in (33a) with the resultant situation s$_1^a $. For the scenario with a broken drum, the drum is the force recipient with the property ‘broken’; the force that got the drum broken comes from the drummer Zhangsan through the causing action of beating. In this way, the four conceptual primitives of the involved force structure have the linguistic representations as seen in (34). An RC sentence corresponding to this scenario is then derived, as presented in (34a).

When it comes to the resultant situation s$_1^b $ in (33b), the net force involved made the drummer Zhangsan tired. While s$_1^b $ and its corresponding force f^b (including the force recipient and the property relating to this force) are determined immediately, this force's source requires further discussion. It seems that both Zhangsan and ‘big-drum’ can serve as the source in relation to f^b, because conceptually, Zhangsan and the drum jointly contributed to Zhangsan's tiredness. When a drummer beats a drum, the drum has a causal influence on the drummer, tiring the drummer; but since the drum's influence to the drummer requires the input of energy from the drummer (i.e., the drummer's beating), the drummer also has a causal influence on themself, though the influence from ‘big-drum’ is non-energetic whereas the influence form the drummer is energetic. Considering the two sources, two RC sentences encoding different sources are derived, as presented in (35a) and (35b). Note that when the force's source coincides with its recipient, the reflexive ziji ‘self’ is used to encode the force recipient (cf. sentence (35b)), in contrast to the use of Zhangsan in (35a).

With regard to the resultant situation s$_1^c $ in (33c), the resultant property holder “passers-by”, though external to the causing event, can still be affected. As seen in (36), the passers-by received the causal influence and became impatient. Like in (35), both entities involved in the causing event, that is, ‘big-drum’ and ‘Zhangsan’, contributed to the impatience of the passers-by via the production of drumming sounds; correspondingly, two sentences encoding distinct sources of the relevant force, that is, (36a) and (36b), are presented for s$_1^c $.

The RC sentences in (34), (35) and (36) all encompass different types of resultant situations following from a single causing event. The production of these sentences demonstrates that determining a force structure requires selecting the resultant property to be expressed and its holder, and identifying the underlying force and its source. In actual usage, people are free to express any of the resultant properties and compatible holders that they intend to focus on, determining V₂ and IA. V₁ can be fixed via the causing event's action. When it comes to EA, previous event-based analyses are not clear on what is mapped to an EA. Following the force-theoretic framework and the mapping relations in (24), for a DP to appear in the EA position of an RC sentence, it needs to encode the source of the force encoded in the RC sentence. As forces in RC sentences are defined by the relevant situation changes and modified by properties characterizing the relevant causing actions, what appears in the EA position of an RC sentence should represent the entities that cause the relevant situation change through the relevant causing action. Consider a scenario where a singer sang a song so that a large audience were brought to tears; since both ‘singer’ and ‘song’ are responsible for the audience's shedding tears due to the causing action of ‘singing’, either can appear in EA position, as seen in (37).Footnote ⁷ But when the scenario changes with a different causing event, such as a large audience listened to a song so they were in tears, it is no longer acceptable for ‘singer’ to appear in EA position, as demonstrated in (38a).Footnote ⁸ This is because the causal influence from ‘singer’ is not realized through the causing action relevant in (38), that is to say, listening.

Turning back to the question why the causer “depression” cannot replace the external argument “wine” in (2) (reproduced below in (39)), there is a straightforward explanation: V₁ indicates a causing action of ‘drink’; the sources relevant to this causing action only include the drinker Zhangsan and ‘wine’; ‘depression’ is not a possible source.

This section has detailed the complexity in determining the relevant force and its structure in the production of a Mandarin RC sentence. The determination of a causing event and of a resultative property to be represented involves all four force primitives in a force structure. After these primitives are mapped to their respective linguistic representations, the linguistic building blocks can then be merged into the derivation of a Mandarin RC sentence, regardless of its RC type.

The current force-theoretic analysis has demonstrated success in the interpretation and production of Mandarin RC sentences, but it may not be readily applicable to resultative constructions in other languages. Cross-linguistically, resultative constructions share the conception that a causing event results in a state of change to an entity, but different linguistic means can be used to encode this conception. For instance, all three types of RC sentences in Mandarin are represented with V₁-V₂ complexes, with sources of forces in EA position; however, a special sentence pattern (e.g., wipe…clean) is used for the English counterparts of Mandarin Type I RC sentences and event-denoting complex DPs are utilized for Type II and Type III RC sentences, as seen in the contrasts between each Mandarin sentence and its English gloss in (40).

In a strict sense, an English RC sentence with the pattern like wipe…clean does not correspond to a Mandarin RC sentence with a complex V₁-V₂ (e.g., ca-ganjing ‘wipe-clean’). As demonstrated earlier, the thematic relation indicated in the common sense understanding of the relationship between an EA entity and V₁ (e.g., the relation between the table and the action of wiping in (40a)) is cancellable in a Mandarin RC sentence, but not so in its English gloss. It seems that English resultative constructions stress a complete representation of causing events so that all necessary theta-roles (mostly agent, patient/theme) of a causing action verb need to be assigned. But in Mandarin, V₁ simply modifies a relevant force by providing information about the causing action, without presenting a complete picture of the causing event. The stress on event representation in English is also manifested in the glosses of the other two types of Mandarin RC sentences. For instance, in (40b) and (40c), result states are directly attributed to events, that is, drinking that bottle of wine and Zhangsan's crying. This event representation of causes in English allows adverbials to adjoin and modify the causing action verbs drink and cry, but modification intended for V₁ (in a V₁-V₂ complex) is not always acceptable in Mandarin Type II and Type III RC sentences, for instance the use of buduandi ‘continuously’ in (40b). Therefore, Mandarin and English are different in how resultative constructions are represented; while English stresses representation of causing events, Mandarin focuses on representation of the exact source that brings about a result state. The difference between English and Mandarin RC sentences can help set up a typology of resultative constructions across languages, though a force-theoretic analysis of English-type resultative constructions requires further research.

3.4 Explaining Reading C of sentence (5)/(30)

In the exploration of the interpretation and production of Mandarin RC sentences, the current force-theoretic analysis has been able to address most of the problems put forward in the introduction, including the reversed theta-role assignments, the constraints on external argument (corresponding to source of the force) selections, and two of the readings of sentence (5)/(30). Despite this, Reading C of sentence (5)/(30), simplified as sentence (41a), is yet to be resolved, as it involves a resultant property for the subject Zhangsan. In other words, the entity in EA position seems to be interpreted as a force recipient, contradicting the mapping relations in (22), which require external arguments in Mandarin RC sentences to be interpreted as sources of forces.

Before attempting to account for Reading C, it is necessary to discuss a different type of RC sentence where entities in subject position are force recipients (also resultant property holders). This type of RC sentence contains a V₁-V₂ predicate, but only selects a subject argument, as exemplified in sentence (41b), which differs from (41a) only in the presence/absence of an apparent object. For a sentence of this type, it is the entity represented by the subject that undergoes a situation change, and serves as the rheme as defined in Ramchand (Reference Ramchand2008). Considering this, some previous studies (e.g., Lin (Reference Lin2004), Liu (Reference Liu2019)) have described this RC type as subject-affected, as opposed to object-affected RC sentences that select two arguments. Just as a causative sentence like “John opened the door” can be expressed with an inchoative counterpart “the door opened”, this study agrees with Lin and Liu regarding the proposal that object-affected RC sentences and subject-affected RC sentences represent the alternation between causative and inchoative uses of the same V₁-V₂ resultative complexes. As seen in (42), sentence (41b) has a causative counterpart, with Zhangsan serving as an object. The addition of Lisi in subject position specifies the source of the force and all the other conceptual primitives in (41) still apply to (42).

Following the current force-theoretic analysis, resultative complexes in inchoative RC sentences can be decomposed and interpreted in the same way as they are decomposed and interpreted in causative RC sentences. The subject arguments in inchoative RC sentences (e.g., (41b)) are resultant property holders and take the role of force recipient; there is no representation of the relevant force's source. Every step in the derivation of a causative RC sentence (cf. (25)) is also involved in the derivation of an inchoative RC sentence, except that VoiceP is not projected, and therefore the force's source is not introduced. Subsequently, the DP encoding force recipient is raised to the subject position, explaining why subjects in inchoative sentences appear to represent force recipients with changes of states.

Turning back to Reading C presented in (41a), it involves a force that rendered Zhangsan tired, which is identical to the force encoded in the inchoative sentence (41b). Formally, sentence (41a) seems to have an extra object Lisi in comparison to sentence (41b). As claimed in Huang (Reference Huang2006) and Lin (Reference Lin2004), this apparent object is not a true argument, for the following reasons. First, Lisi in (41a) under Reading C does not allow passivization or topicalization, as evidenced in (43). Note that Lisi under Reading A or B can be passivized or topicalized, so sentence (43a) and sentence (43b) can be grammatical, but not with the intended readings.

Second, only a limited number of V₁-V₂ resultative complexes allow apparent objects, including chi-bao ‘eat-full’, he-zui ‘drink-drunk’, and those in the form of V-lei ‘tired’, etc. Even for these complexes, only bare nouns are allowed in the apparent object position, as seen in the contrast in (44).Footnote ⁹ This strongly suggests the idiosyncrasy of these resultative complexes and the relevant sentences. According to Cheng (Reference Cheng1997), these bare nouns are mostly non-referential, as is advantage in the idiom “take advantage of”. Following Cheng's analysis, the apparent object Lisi in (41a) is originally selected as a patient of V₁ zhui ‘chase’, but later gets incorporated with V₂ and V₁, forming the complex verb zhui-lei-Lisi ‘chase-tired-Lisi’.

This article agrees with Cheng's analysis of sentences like (41a) under Reading C. That is, the apparent objects are incorporated to the resultative complexes to form even more complex predicates. This incorporation serves to provide supplementary information which constrains possible interpretations. For instance, the extra apparent object in (41a), in comparison to its inchoative counterpart in (41b), provides more information about who Zhangsan chases. Just like in the derivation of an inchoative RC sentence, the force recipient Zhangsan is then raised to subject position to derive sentence (41a), making this sentence subject-affected. Though the details of how a bare noun (e.g., fan ‘meal’ in (44a)) is incorporated still await further research, Cheng's analysis successfully captures the fact that the apparent objects cannot be interpreted as arguments, hence disallowing passivization and topicalization. Therefore, sentence (41a) under Reading C is explained as involving the incorporation of a bare noun (i.e., the apparent object) into an originally inchoative RC sentence. This study also successfully predicts the absence of a fourth reading of sentence (5)/(30). Consider the various readings in (5)(30). It is easily seen that Reading A and Reading C involve the same causing event “Zhangsan's chasing Lisi” but differ in the force recipient, that is, Lisi for Reading A and Zhangsan for Reading C. One may wonder whether there is Reading D, which shares the same causing event as Reading B, but has a different force recipient, that is to say, “Lisi's chasing Zhangsan” → “Zhangsan's tiredness”. No previous literature has reported Reading D. Informal consultation with native Mandarin speakers does not support Reading D, either. The unavailability of Reading D can be explained from two perspectives. First, Reading D would violate the mapping relation that requires that force recipients with changes of states should be the entities in IA position. Considering that Lisi appears in IA position, Zhangsan cannot be the one becoming tired. Second, Reading D cannot be derived in the same way that Reading C is derived. V₁ zhui ‘chase’ should first license the role of patient; thus, what is adjoined to V₁ ‘chase’ through nominal incorporation should be a nominal with the role of patient, that is, Zhangsan as the patient of ‘chase’. Consequently, Reading D does not allow Lisi to undergo nominal incorporation, making the sentence form in (5)/(30) incompatible with Reading D.