Attraction or differentiation: diachronic changes in the causative alternation of Chinese change of state verbs

2.1 A diachronic perspective on the causative alternation

The causative alternation has been extensively examined in previous research. Some studies have concentrated on the derivational relationship between the causative and inchoative variants (Guerssel et al. 1985; Hale and Keyser 1987; Levin 2004, 2015; Song 2023), while others have explored the conditions under which a verb may participate in the causative alternation (Alexiadou et al. 2006; Guo 2011; Haspelmath 1993; Jiang 2017; Levin and Rappaport Hovav 1995). Theoretical studies have offered divergent perspectives on whether the fundamental structure is monadic or dyadic (Hale and Keyser 1987; Levin 2004). Typological studies, on the other hand, have highlighted crosslinguistic variations and complexities related to the morphosyntactic form of alternating pairs (Alexiadou et al. 2006; Haspelmath 1993; Heidinger 2015). Regarding alternating verbs, the variation in transitive sᴜʙᴊᴇᴄᴛ (whether an agent, a cause, or an instrument) has been considered a key condition for the formation of causative alternations (Reinhart 2002; Zúñiga and Kittilä 2019). For example, Heidinger and Huyghe (2024) found that, with French COS verbs, a higher likelihood of the transitive sᴜʙᴊᴇᴄᴛ being interpreted as a cause corresponds to an increased propensity for the verb to be used in the inchoative construction.

Compared to the above synchronic studies, there is a lack of studies that address causative alternations from a diachronic perspective (Kulikov and Lavidas 2013). Among the limited diachronic studies, most are concerned with change in lability, which refers to a verb’s ability to participate in alternation patterns. For instance, evidence from both Old Indo-Aryan and French demonstrates that the decrease in lability is a very slow process (Heidinger 2014; Kulikov 2014). Regarding Chinese, as mentioned above, numerous studies have examined the transitivity of COS verbs (Jiang 2000; Wei 2000; Wu 1999; Xu 2005). However, little attention has been paid to the diachronic changes in the encoding of causative alterations. In two recent studies, Jiang (2017) analyzes the derivational relationship between the causative and inchoative variants with data from Archaic Chinese and reveals that the inchoative pattern is basic for stative verbs, whereas the causative pattern is basic for COS verbs. Wuyun (2024) provides evidence from the Zuo Tradition, an archaic Chinese narrative history from the Pre-Qin period (before the second century) and shows that the causative alternation incorporates a great variety of verbs, including denominal verbs, unergative verbs, and “pure” unaccusative verbs.

Despite these findings, previous studies are either confined to the derivational relationship or the variety of alternating verbs but fall short of addressing the diachronic evolution of causative alternations. To fill this research gap, the present study aims to elucidate the diachronic changes involved in the encoding of Chinese causative alternations. It should be noted that although previous studies mention the features of alternating verbs (Guo 2011; Haspelmath 1993; Jiang 2017), they have not provided criteria for identifying causative alternation pairs. Therefore, to investigate the diachronic evolution of causative alternations, we propose the following four identification criteria.

1. The alternating verb denotes a change of state;

2. The alternating verb exhibits both causative and inchoative uses;

3. The alternating verb usually has the same form in both variants, though it may undergo formal changes due to the systemic evolution of the language;

4. The arguments in both variants should be of an identical type.

Regarding the first and second criterion, the alternating verb denotes both causative and inchoative changes of state. For instance, the two verbs pò ‘break’ and kāi ‘open’ indicate the separation of objects, and they are used in both causative and inchoative patterns. By contrast, agentive transitive verbs (e.g., invite) that do not imply a change of state, and agentive intransitive verbs (e.g., talk) that do not permit spontaneous reading, fall outside the scope of COS verbs (Haspelmath 1993). As for the third criterion, monomorphemic verbs in ancient Chinese have transformed into Verb-Result Constructions (VRC) and parallel Verb Constructions (VVC) through disyllabification (Duanmu 1999; Feng 2000; Shi 2002). Consequently, alternating verbs manifest themselves in three forms: the monomorphemic COS verb, the VRC (V+COS), and the VVC (COS+V). Finally, the type of patient in the causative variant must be congruent with that in the inchoative variant. For example, if the causative variant describes the action of opening a door, the inchoative variant should also describe the spontaneous opening of a door, rather than a book. The patient in both the causative and inchoative patterns should ideally be the same. However, if this is not possible due to constraints, the patients in these patterns should at least belong to the same category or type (see Section 3.4 for the categorization of patient types). Section 4 delineates the causative alternation pairs existing in the evolution of Chinese with specific examples.

2.2 From competition to attraction and differentiation

Analogous to the competition for resources and habitat among species in nature, grammatical structures within a language compete for the expression of similar functions (Berg 2014; Goldberg 2019; Karjus et al. 2020). This competition arises from speakers’ choices among variants based on their expressive needs. For example, in modern Chinese, the expression of affectedness engenders competition among the bǎ-construction, bèi-construction, and SVO construction (Liu and Ambridge 2021). Competition conforms to the principle of isomorphism, which posits that each form corresponds to a distinct meaning (Dik 1988; Geeraerts 1997). However, this perspective overlooks the rise and persistence of functional overlap as well as the reorganization of near-synonyms (De Smet et al. 2018). To address this gap, recent studies propose two analogical processes: attraction and differentiation (De Smet et al. 2018; Traugott 2020). These studies assert that functionally similar expressions converge through attraction, while simultaneously maintaining distinctiveness through differentiation.

Attraction arises from the overlapping features among expressions, leading to increased similarity and functional overlap over time. Differentiation, on the other hand, prevents constructions from becoming near-synonyms by aligning their behavior within their bigger constructional families. Regarding differentiation, De Smet et al. (2018) posit that it is an epiphenomenal by-product of the relationships that exist in a larger constructional network. By contrast, Traugott (2020) argues that differentiation is a default outcome of attraction, contending that a construction’s attraction to the constructional network results in differentiation from its synonyms. Both perspectives are valid, suggesting that whether the attraction arises from a near-synonym or from a higher-level structure, differentiation is ultimately a consequence of attraction.

To uncover attraction or differentiation, researchers typically begin with a construction and examine the collocational preferences of its syntactic slots. For example, by analyzing the predicate slot, filled in by a noun phrase or a non-noun phrase, De Smet et al. (2018) discover the attraction between [consider+zero−SPC] and [consider+as−SPC]. Similarly, Traugott (2020) observes the attraction between the two dative alternation patterns, by focusing on the complement slot filled by definite or indefinite complements. Other evidence includes the functional distribution of connectors (Gillmann 2021), the position of relative clauses (Hilpert 2021), and the semantic types of verb phrases (Huang 2023). Changes in the relative frequencies or proportions of these behavioral profiles, particularly when they become convergent or divergent, serve as indicators of attraction or differentiation.

This study extends the aforementioned research in two ways. First, from a theoretical perspective, we propose that both attraction and differentiation incorporate two subtypes. Attraction can be subdivided into mutual attraction and model attraction. Mutual attraction occurs when expressions at the same level influence each other, resulting in feature exchange and increased similarity. Model attraction involves an expression aligning, either fully or partially, with a pre-existing expression that has developed earlier. Differentiation, which in essence stems from attraction, is driven by higher-level constructional families or by systemic language shifts. Second, methodologically, this study contends that syntactic slots serve as loci for seeking evidence. In this study, the diachronic changes of Chinese causative alternations are attested with syntactic slots encompassing the predicate slot and the ᴏʙᴊᴇᴄᴛ slot, as will be demonstrated in Section 5.

3.1 Periodization and corpus

The corpus used in this study was the CCL corpus (Center for Chinese Linguistics PKU, Zhan et al. 2003). Considering the availability of the data for pò and kāi, four of the seven chronological stages outlined in Traugott and Dasher (2002: xiii) were taken into consideration (Table 1). We also provided the number of characters in each of the four periods. It should be noted that MdMand ranges from 1912 to 1949 for the sake of data controllability. Three earlier periods, including Pre-Archaic Chinese (PAC), Early Archaic Chinese (EAC) and Late Archaic Chinese (LAC) were disregarded because of the sparsity of pò and kāi.

3.2 Selection of COS verbs

Out of the eleven Chinese monomorphemic COS verbs^[3] (Liu 1998; Lv 1999; Shi 2011: 31), pò and kāi were chosen for two reasons. First, their usages extend across both ancient and modern Chinese, allowing a systematic tracing of the causative alternations over diachronic periods. Second, previous studies revealed that both pò and kāi operated as monomorphemic predicates in ancient Chinese but as complements in VRC in modern Chinese (Hu 2005; Xu 2005), promoting further exploration of their evolving argument structure constructions. Only literal usages of COS verbs that denote visible and physical state changes in objects were discussed, such as (4a). Those abstract usages describing abstract state changes, such as (4b), were beyond the range of this study (but see Du and Li 2022).

3.3 Data extraction

To trace the argument structure constructions of COS verbs, we collected usages of the two COS verbs from the CCL corpus following four steps. Initially, in the four periods, usages of COS verbs were extracted randomly with each limited to 4,000 concordances to keep the dataset within a manageable range. Subsequently, repetitive and non-verbal usages of the COS verbs were removed. Table 2 showcases the raw frequencies of the two keywords in each of the four periods.

3.4 Data annotation

To answer the first research question – what causative alternations are present in the diachronic evolution of the argument structure constructions of Chinese COS verbs – we annotated both the argument structures of COS verbs and the causative or inchoative patterns of COS verbs. First, we annotated the argument structures of COS verbs, and found six major syntactic patterns: NP1+COS+NP2, NP1+V+COS+NP2, NP1+COS+V+NP2, NP2+COS, NP2+V+COS, and NP2+COS+V. NP1 stands for the agent, COS for COS verbs, V for action verbs, NP2 for the patient. As exemplified in (5), the agent Hào was marked as NP1, the action verb chuí as V, and the COS verb pò as COS. The pronoun zhi stands for the patient and is annotated as NP2. The usages were numbered sequentially based on the period and the concordance. For instance, EMC173 represents the 173rd line of pò in EMC.

Then, the causative and inchoative patterns of COS verbs were coded based on both their syntactic structures and semantic meanings, as shown in (6a–b). Formally, the causative patterns follow the structures of NP1+COS+NP2, NP1+V+COS+NP2, or NP1+COS+V+NP2, whereas the inchoative patterns follow the structures of NP2+COS, NP2+V+COS,^[4] or NP2+COS+V. Semantically, the causative pattern illustrates how an agent (NP1) causes a patient (NP2) to undergo a change of state (6a). In contrast, the inchoative pattern denotes the patient (NP2) undergoing a change of state spontaneously (6b). Finally, the identification of alternation pairs was based on the four criteria outlined in Section 2.1.

To answer the second research question – how attraction and differentiation influence the alternation between the causative and the inchoative patterns over time – the patients (NP2) and the action verbs were annotated. The patients were annotated for two reasons. First, pò and kāi are COS verbs used to denote separations applicable to the material destruction of various objects (Pye 1996; Majid et al. 2007). Second, while the causative patterns are similar in their agent types,^[5] their patients constitute the core argument and demonstrate diachronic changes in their material composition over time due to social-cultural development (Xu 2005). Based on their material composition, the patients were annotated according to five categories: fleshy objects, fragile objects, flexible objects, stiff objects, and delicate objects. These five categories have proven effective in delineating the functional range of pò and kāi (Du et al. 2020). Fleshy objects typically refer to human organs (7a). Fragile objects, such as glassware represented by ‘oil jar’ in (7b), form a distinct category. Flexible objects include cloth or paper (7c), while objects made of bamboo or wood (7d) are classified as stiff objects. The fifth category, delicate objects, includes items like vegetables or fruits (7e).

The action verb was annotated because it is associated with the COS verb, together constituting the alternating verb complex in causative and inchoative forms. As mentioned earlier, the semantic meaning and syntactic structure of the alternating verb complex determine the possibility of causative alternations. Therefore, annotating the action verb and its collocation with COS verbs provides another means to uncover the diachronic changes of causative alternations. Action verbs associated with COS verbs were annotated as either manner verbs or cause verbs (Talmy 2000). Manner verbs delineate the mode or way in which the patient undergoes a state change. For instance, the pomegranate separates in a mode of cracking (8a). Cause verbs denote the forceful action conducted by the agent and result in the state change of the patient. For instance, in (8b), the disruption of the grass is caused by the horse’s trampling.

To guarantee maximal reliability, two coders were employed to annotate the above variables, yielding an initial interrater reliability (Cohen Kappa) of 0.90. Disagreements on annotation were resolved after further discussion.

In what follows, the argument structure constructions of pò and kāi are extracted and the emergence of causative alternations are investigated in Section 4. In Section 5, we further compare their causative and inchoative patterns across the four diachronic periods and attest how attraction and differentiation interplay in the diachronic evolution of causative alternations.

4.1 CA1: the alternation between NP1+COS+NP2 and NP2+COS

CA1 represents the alternation between NP1+COS+NP2 and NP2+COS. In this pair, the predicate is a monomorphemic synthetic COS verb, which is used to encode the result of the state change and imply the action exerted to cause it (Hu 2005; Li 2023).

CA1 was confirmed to have been established in Archaic Chinese (before the second century). In the literature that covers the second century to the sixth century, existing studies have demonstrated the presence of COS verbs such as bài ‘defeat’, dàng ‘swing’, ruò ‘weak’, luò ‘fall’, as well as pò and kāi in CA1 (Dai 2001; Jiang 2017; Mei 1991; Wuyun 2024). For instance, Jiang (2017) found pò appeared 337 times in NP1+COS+NP2 and 146 times in NP2+COS, and kāi occurred 85 times and 34 times, respectively. These verbs were alternatively used in CA1 to describe the state change of the same patient. As exemplified in (9a–b), both causative and inchoative patterns are used to denote the opening of a door.

Our dataset demonstrates the continued evolution of CA1 in EMC and LMC. As displayed in Figure 1, the causative and inchoative patterns in CA1 accounted for the highest proportions in these two periods, with percentages of 58.6 and 53.5 % respectively for NP1+COS+NP2, and 13.6 and 28.4 % for NP2+COS. Moreover, both patterns were collocated with a wider range of patients, as detailed in Table 5. However, this developing trend gradually faded away in EMand and MdMand. Although, in these two periods, NP1+COS+NP2 continued to be prominent, a closer examination suggested that over half of pò were ancient Chinese relics such as pò-ké-ér-chū ‘break out from the shell’. For kāi, two-thirds of its instances were confined to phrases such as kāi-mén ‘open door’.

This alternation pair further evolved along two main diachronic trajectories. In both trajectories, the causative variant was replaced by a compound structure, while the inchoative variant displayed a somewhat more complicated picture.

4.2 CA2: the alternation between NP1+V+COS+NP2 and NP2+COS

Along the first trajectory, NP1+V+COS+NP2 was gradually replacing NP1+COS+NP2, with the monomorphemic COS verb giving way to the VRC structure. As depicted in Table 3, the normalized frequency of NP1+V+COS+NP2 has grown significantly from 3 in EMC to 474 in MdMand (p < 0.05, using a Fisher’s exact test, φ = 0.25),^[6] while that of NP1+COS+NP2 declined from 263 in LMC to 242 in EMand. As highlighted in Section 4.1, in MdMand, NP1+COS+NP2 became formulaic expressions, suggesting its replacement by NP1+V+COS+NP2. In contrast, NP2+COS has continued to be the primary inchoative pattern, showing an increase in LMC and maintaining a stable presence with a percentage of 14.7 % in EMand and 10.1 % in MdMand.

In EMC, NP1+V+COS+NP2 was established as an optional variant of NP2+COS, together constituting CA2. They are causative and inchoative patterns characterized by distinct fillers in their predicate slots: the former employs a VRC (a manner or cause verb preceding the COS verb), while the latter uses a monomorphemic COS verb. Despite these differences, they form an alternation pair due to their shared COS verb (Chen 2007; Guo 2011:161).

Our dataset recorded five pairs of CA2 in EMC. The alternating verbs included zhì-pò ‘throw break’(10a), bò-pò ‘peel break’(10c), jué-pò ‘cut break’(10e), dǎ-pò ‘hit break’, and niè-pò ‘crush break’. For instance, the causative breaking of the earthen pot was denoted as zhì-pò ‘throw break’, with the first verb zhì ‘throw’ denoting the cause of state change and the COS verb pò denoting the result of state change (10a). The intransitive breaking of the ounce was described solely with the COS verb pò (10b). Meanwhile, the causative and inchoative alternants were used interchangeably to denote the state change of various patients such as fragile (10a–b), delicate (10c–d), and fleshy (10e–f) entities. As illustrated in (10c–d) and (10e–f), CA2 was used to verbalize the causative and inchoative breaking of fruit and body parts respectively.

In the latter three periods, CA2 continued to develop, showcasing a great diversity of V+COS and detailing the state changes in various patient types. Furthermore, the establishment of V+COS (VRC) in the twelfth century (Shi 2002) greatly enhanced the variety of alternating verb types.

4.3 CA3: the alternation between NP1+V+COS+NP2 and NP2+V+COS

In CA3, NP1+V+COS+NP2 alternates with NP2+V+COS because both structures utilize the same V+COS predicate slot and describe the state change of the same patient types. Semantically, this alternation pair can be divided into two types: the alternation between the causative and inchoative variants and the alternation between the causative and the pseudo-passive forms.

In the former type, the verb (V) in the compound is a manner verb that allows for causative/inchoative readings. As exemplified in (11a) and (11b), the NP1+V+COS+NP2 pattern denotes how the agent (e.g., the rhinoceros) causes the state change in the patient, while the NP2+V+COS pattern denotes the inchoative state change of the “water” in a “dividing” manner. The NP2+V+COS pattern is identified as inchoative because it continues to be grammatical when the phrase zì-jǐ ‘by itself’ is added, a well-acknowledged test for inchoative patterns (Beavers and Koontz-Garboden 2013; Cheng and Huang 1995). This suggests that the state change is due to the inherent properties of the patient, rather than an external agent. For example, (11b) can be interpreted as a natural division occurring when the current encounters a change in terrain. Similar verbs include blossom and decay, which describe natural changes of state that are inherent to the patients (Rappaport Hovav 2014). Our findings corroborate the observation that the less likely a cause is agentive, the more likely the event is expressed using the inchoative pattern (Heidinger and Huyghe 2024).

In the latter type, the verb compounds consist of cause verbs and COS verbs. The transitive form indicates a causative reading. As shown in (12a), the transitive form denotes a causative event where the agent causes the state change of the face through a squeezing action. The intransitive form does not allow an inchoative reading and is instead viewed as a pseudo-passive pattern in which the passive marker bèi is omitted (Chen 2007). As exemplified in (12b), the RVC zā-pò specifies both the cause and the state change, making it incompatible with the phrase zì-jǐ ‘by itself’. Consequently, it is interpreted as “the face was squeezed broken”. Alternatively, it can be treated as an intransitive pattern that invites a stative reading, such as “the face is in a squeezed-broken state”.

In EMC, the alternating relation between NP1+V+COS+NP2 and NP2+V+COS has not yet been established due to the absence of identical verbs. By LMC, instances of the same V+COS began to emerge in both transitive and intransitive patterns. The data attested two alternating pairs, zā-pò (12a–b) and liè-pò, which were employed to denote the state change of fleshy objects.

In EMand, CA3 became established, as evidenced by the increased frequencies of alternates, the expansion of available alternating verb complexes, and its co-occurrence with diverse patients. Firstly, there was an increase in the normalized frequencies of NP1+V+COS+NP2 and NP2+V+COS, reaching 243 and 14, respectively, compared to 60 and 6 in LMC. Secondly, alternating verb complexes expanded to include chě-pò ‘tear-broken’, niān-pò ‘wither-broken’, làn-pò ‘rot-broken’ and fēn-kāi ‘divide-open’. Thirdly, these complexes were used interchangeably in both NP1+V+COS+NP2 and NP2+V+COS patterns to denote state changes of various patients, including fleshy, fragile, delicate, and stiff entities. For instance, examples (13a–b) and (13c–d) are two pairs depicting the separation of flexible objects through dividing and tearing, respectively.

4.4 CA4: the alternation between NP1+COS+V+NP2 and NP2+COS+V

Along the second trajectory, the causative pattern NP1+COS+NP2 was substituted by NP1+COS+V+NP2, and the inchoative pattern NP2+COS was superseded by NP2+COS+V. In these two patterns, the predicate is a parallel verb compound constituted by the COS verb and a synonymous verb (Li and Thompson 1989: 68). Both the COS verb and its near-synonymous verb imply the action and specify the result of state change. Inheriting from CA1, the COS verb in CA4 continues to denote an action meaning (Hu 2008; Jiang 2017). As illustrated in examples (14a–b), the verb compound pò-huài ‘break destroy’ occurs in both the causative and inchoative alternates, allowing for the interpretation of both the agent’s destroying of a physical object and the object’s state of being broken.

Figure 1 reveals that from EMC onwards, NP1+COS+V+NP2 and NP2+COS+V have consistently existed with relatively small and comparable percentages: 4.5 and 5.6 % in EMC, and 3.9 and 4.5 % respectively in MdMand. The limited percentages can be attributed to the relatively small inventory of result verbs in Chinese. Chinese has only 24 result verbs (Shi 2011), which consequently restricts the number of possible COS+V compounds.

It was not until EMand that CA4 became established. This was supported by the occurrence of alternating pairs with a variety of alternating verbs and interchangeable patients. Specifically, seven pairs of alternating instances were identified. The alternating verbs included pò-suì ‘break shatter’, pò-kāi ‘break open’, and kāi-fàng ‘open blossom’. These verbs were used interchangeably with various patients, such as flexible objects like clothes, fragile objects such as coal, and delicate entities. Examples (15a–b) illustrate the causative and inchoative breaking of flexible objects like clothes.

Overall, the four alternation pairs exhibit distinct trajectories and became established in different diachronic periods, as demonstrated by the taxonomic structure in Figure 2. The dotted lines represent replacement and the full line signifies development. The periods indicate when the alterations are established. Specifically, CA1 (NP1+COS+NP2 and NP2+COS) became established in Archaic Chinese (AC for short), CA2 (NP1+V+COS+NP2 and NP2+COS) in EMC, CA3 (NP1+V+COS+NP2 and NP2+V+COS) and CA4 (NP1+COS+V+NP2 and NP2+COS+V) in EMand.

Figure 2:

Diachronic trajectories of the argument structure constructions of COS verbs.

Collectively, these four alternation pairs form a constructional family whose members share an overlapping conceptual space while their developmental trajectories intertwine, rather than being independent and isolated. In the following section, we will explore how attraction and differentiation are involved in the diachronic changes of Chinese causative alternations.

5.1 Attraction

The following subsections demonstrate the processes of attraction involved in causative alternations. It is revealed that functionally overlapping expressions do not merely substitute or compete with each other but rather exhibit a tendency to attract each other and become more similar over time. The corpus dataset of pò is employed to document the attraction processes.

5.1.1 Attraction between NP1+V+COS+NP2 and NP1+COS+V+NP2

The attraction between causative variants is attested with the mutual influence between NP1+V+COS+NP2 and NP1+COS+V+NP2. The other causative construction, NP1+COS+NP2, is not considered because it has become obsolete, with most of its remaining instances being relics from ancient Chinese. As noted in Section 3.4, patient types were used as evidence for attraction because they represent the core arguments of COS verbs, reflecting the functional range of these verbs in describing the material destruction of various objects (Majid et al. 2007; Xu 2005). The two stacked bar charts in Figure 3a and 3b demonstrate the collocation of these two causative patterns with fragile (e.g., glass), fleshy (e.g., abdomen), flexible (e.g., book), stiff (e.g., wooden fence) and delicate (e.g., petal) patients across the four diachronic periods. To attest the functional overlap overtime, we conducted multi-factor ANOVA tests to examine the correlation between the two causative constructions based on the proportional distributions of five patient types.

Figure 3a:

The proportional distributions of five patient types in NP1+V+COS+NP2.

Figure 3b:

The proportional distributions of five patient types in NP1+COS+V+NP2.

In the period of EMC, the correlation between the proportional distributions of the five patient types for the two causative constructions was found to be significantly different (p < 0.05, df = 8), indicating the two causative variants were significantly different from each other. However, over time, these two causative variants exhibited a greater functional overlap in terms of their semantic profiles. As demonstrated in Figures 3a and 3b, these two causative patterns gradually converged in their co-occurrence with diverse patient types. By the period of MdMand, confirmed by the multi-factor ANOVA test, these two causative constructions were not significantly different from each other in terms of their correlation with the five patient types (p > 0.05, df = 4).

One of the reviewers insightfully noted the difficulty in distinguishing establishment from attraction. These two processes appear intertwined, often relying on the same evidence for confirmation. However, previous studies on verb compounds have indicated that the two causative constructions were established in LMC (around the twelfth century, Shi 2002). During their further development in EMand and MdMand, these two constructions continued to reorganize the proportional distributions of the five patient types due to attraction. As a result, they were becoming proportionally more similar in their patient types. For instance, these two constructions became similar in their association with fleshy objects as the proportion of fleshy objects in NP1+V+COS+NP2 gradually increased from 25.9 % in LMC to 46.2 and 42.4 % in the latter two periods, while the proportion of fleshy objects in NP1+COS+V+NP2 gradually decreased from 61.5 % in LMC to 38.1 and 46.3 % in the latter two periods. Conversely, due to attraction, the proportion of delicate objects in NP1+V+COS+NP2 dropped from 13.2 % in LMC to 12.0 % in MdMand, while in NP1+COS+V+NP2, it increased from 0.0 % in LMC to 17.1 % in MdMand. In this sense, the two constructions continued to adjust their association with the five patient types due to mutual attraction. It should be pointed out that when only their frequencies or proportions were compared, they demonstrated a relation of substitution. Further speaking, while establishment and attraction are distinct processes, attraction also involves development from a diachronic perspective, necessitating recognition of their intertwinement (De Smet et al. 2018).

5.1.2 Attraction between NP1+V+COS+NP2 and NP2+V+COS

Functional overlap is also found between the causative and inchoative/pseudo-passive variants, with the intransitive variant modeling itself after the causative variant. For reasons of operationalizability, we substantiate this attraction through the modeling effect exerted by the NP1+V+COS+NP2 structure on the NP2+V+COS structure. Figure 4a and 4b illustrate the proportions of “cause verb + COS verb” and “manner verb + COS verb” in the respective patterns. The multi-factor ANOVA analyses demonstrate that the two constructions are significantly different from each other in EMC (p < 0.01, df = 3). The predicate slot of the causative pattern NP1+V+COS+NP2 is primarily composed of a cause verb and a COS verb across all four periods. In contrast, prior to LMC, manner verbs were frequently associated with COS verbs and constituted the predicate slots of the intransitive pattern NP2+V+COS.

Figure 4a:

The proportions of two predicate types in NP1+V+COS+NP2.

Figure 4b:

The proportions of two predicate types in NP2+V+COS.

Manner verbs specify the mode or way in which the patient undergoes a change of state (Talmy 2000: 28–29). In inchoative patterns, manner verbs enrich the inchoative description by adding details to the mode or style of the action. For example, in (16a), the verb kuì ‘ulcerate’ signifies the mode in which the breaking action unfolds. On the other hand, cause verbs denote the actions performed by agents that bring about a state change in patients (Talmy 2000: 28–29). They are typically used in causative constructions. The relationship between cause verbs and causative patterns is rooted in the functional need to indicate who or what initiates the action. For instance, in (16b), huá ‘rip’ is the action performed by the agent to trigger the state change in the patient.

(16)

a.

MdMand944
左足二趾溃破。 (inchoative, NP2+V+COS)
zuǒ	zú	èr	zhǐ	kuì	pò
left	foot	two	toe	ulcerate	break
‘Two toes of the left foot ulcerated and broke.’

b.

EMand488
三寨主…把金头虎衣服划破。 (causative, NP1+V+COS+NP2)
sān	zhài-zhǔ…	bǎ	Jīn-Tóu-hǔ	yī-fu	huá	pò
third	master…	BA	Jin Touhu	clothes	rip	break
‘The third master… ripped Jin Touhu’s clothes broken.’

c.

MdMand802
衣服都划破了。 (inchoative, NP2+V+COS)
yī-fu	dōu	huá	pò	le
clothes	even	rip	break	PFV
‘Even the clothes were ripped broken.’

The bar chart in Figure 4b suggests cause verbs have gradually been replacing manner verbs in the predicate slot of the NP2+V+COS construction since the LMC period. It should be noted that the transition from EMC to LMC is abrupt, with all action verbs changing from manner verbs to include both manner and cause verbs. This could be attributed to the sparsity of the NP2+V+COS construction in EMC, which becomes frequently used in LMC. While a larger dataset might reduce the perceived abruptness of this increase, it will not alter the predominant usage of manner verbs in EMC compared to LMC. The substitution of manner verbs with cause verbs in NP2+V+COS was an attraction process accompanied by replacement. Firstly, attracted by its transitive counterpart where cause verbs were more prevalent (16b), the NP2+V+COS construction also incorporated cause verbs, leading to more collocations of cause verbs and COS verbs, such as huá-pò ‘rip-broken’ in (16c). As noted above, when NP2+V+COS is associated with cause verbs, it evokes a passive or a stative reading. For instance, example (16c) can be interpreted as a pseudo-passive pattern, denoting the clothes being ripped. Alternatively, suppressed by the intransitive pattern, it can be interpreted as depicting a ripping-broken state. Secondly, the collocation of manner verbs with COS verbs in the NP2+V+COS construction was at the same time being replaced by the NP2+COS+V structure. For instance, pò-kuì replaced kuì-pò (17a), and pò-làn took the place of làn-pò (17b). As a result, the collocations between manner verbs and COS verbs appeared more frequently in the NP2+COS+V structure rather than in the NP2+V+COS pattern.

(17)

a.

MdMand565
瘢痕…反复破溃。 (inchoative, NP2+COS+V)
bān-hén…	fǎn-fù	pò	kuì
scar…	repeatedly	break	ulcerate
‘Scars… repeatedly broke and ulcerated.’

b.

MdMand953
皮肤破烂。 (inchoative, NP2+COS+V)
pí-fū	pò	làn
skin	break	rot
‘The skin rotted.’

An exclusive examination of frequency changes in either a single causative or inchoative variant can easily lead to the conclusion that functionally similar constructions are in competition. However, a detailed analysis of their functional changes reveals that functionally similar expressions are more likely to become convergent in their association with different patients or verb types.

5.2 Differentiation

This section aims to show that functionally similar expressions not only align their behavior with each other but also with their bigger constructional network. The division of labor among such expressions is a consequence of constructional inheritance and systemic change.