2.1. Matrix verbs

Previous studies on matrix verbs in the CPC have largely revolved around a specific subset of verbs (Visser Reference Visser1963–73; Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985; Huddleston & Pullum et al. Reference Huddleston and Pullum2002; Toyota Reference Toyota2006; Puckica Reference Puckica2009; Kim Reference Kim2018). For instance, Visser (Reference Visser1963–73: 1888) identified an initial set of 22 such verbs, including abide, avoid, await, bear, continue, deserve, desire, escape, fear, hate, lack, merit, miss, mot (obsolete), need, prefer, prevent, repay, require, stand, suffer and want, which have been used historically in the construction. On the other hand, Quirk et al. (Reference Quirk, Greenbaum, Leech and Svartvik1985: 1189–90) and Huddleston & Pullum et al. (Reference Huddleston and Pullum2002: 1199–1200) only included the matrix verbs that generally take a to-infinitive complement such as deserve, need, require and want.

Subsequent studies have explored the diachronic distribution changes of these matrix verbs, utilizing corpus data (Toyota Reference Toyota2006, Reference Toyota2009; Kim Reference Kim2018). According to Toyota (Reference Toyota2009), throughout the history of English, matrix verbs in the CPC can be categorized into three types in terms of chronological order: Type I includes abide, avoid, continue, escape, lack and suffer, used around the seventeenth–eighteenth century. Type II, including await, miss, prevent, repay and stand, emerged in the nineteenth–twentieth century. Type III includes bear, deserve, merit, need, require and want, used from 1400 to the present day. Kim (Reference Kim2018) further demonstrated that the most frequently used matrix verb in the construction has changed over time, based on raw frequencies of matrix verbs extracted from two diachronic corpora: CLMET 3.0 and BYU-BNC fiction register. Specifically, he showed that during the Late Modern English period (1710–1920), want was the most frequently used matrix verb, while in Present-day English (1980s–1993), need took that role.

2.2. -ing complements

Next, previous studies have observed that verbs in this slot exhibit various syntactic types, as illustrated in (3) (Puckica Reference Puckica2009; Kim Reference Kim2018):

Building on this finding, Puckica (Reference Puckica2009) compared these complement types with those of the standard be-passive construction and argued that the verbs in the VP[-en] complement of the be-passive construction are indistinguishable from those in the -ing complement of the CPC. From a diachronic perspective, Kim (Reference Kim2018) noted that in the early stages of the late modern English period, the CPC predominantly featured monotransitive verbs in the -ing complement. Over time, however, this usage expanded to a wider variety of verb types.

Another important aspect of the -ing complement is its syntactic nature; specifically, whether it is as a VP or an NP (Huddleston & Pullum et al. Reference Huddleston and Pullum2002: 1200; Toyota Reference Toyota2006; Puckica Reference Puckica2009; Strelluf Reference Strelluf2022).

In (4a), the presence of an agent by-phrase renders coaxing verbal while in (4b), the presence of some that modifies coaxing shows its nominal property. The more complex case is (4c), where coaxing appears on its own, leaving the interpretation ambiguous; it could be verbal, meaning ‘to be coaxed’, or nominal, meaning ‘some coaxing’. Huddleston & Pullum et al. (Reference Huddleston and Pullum2002) argued that the -ing complement here should be analyzed as a VP, given that ‘coaxing’ is directed toward the children, suggesting a passive construction. Strelluf (Reference Strelluf2022) expanded on this idea by indirectly supporting the VP analysis, comparing the -ing complement in the CPC to the Alternative Embedded Passive (AEP; see also Edelstein Reference Edelstein, Zanuttini and Horn2014). He compared three structural variants: the infinitival passive (e.g. needs to be fixed), the AEP (e.g. needs fixed) and the CPC (e.g. needs fixing), as demonstrated in (5).Footnote ¹

Although these structures are truth-conditionally equivalent, the nature of their complements varies. According to Tenny (Reference Tenny1998) and Edelstein (Reference Edelstein, Zanuttini and Horn2014), verbs with reversative un- (e.g. unwrap) are compatible with both the infinitival passive structure and the AEP, as in The package needs {to be unwrapped/unwrapped}. However, verbs with non-reversative, negative un- (e.g. unwash) are incompatible with the AEP but remain acceptable in the infinitival passive structure, as in The car needs {to be unwashed/*unwashed}. They explain this discrepancy by arguing that reversative un- can function verbally, while negative un- cannot, serving only an adjectival function (e.g. the unwashed car). In Strelluf’s corpus study, AEP and CPC examples only appeared with reversative un-, such as Just like Pereira would unlock Pogba, now Lingard needs unlocked too (AEP) and Looks like Klopp needs unlocking (CPC) (Strelluf Reference Strelluf2022: 58); however, no instances were found with non-reversative/negative un- items, which are typically adjectival (e.g. unwashed). Based on these findings, Strelluf concluded that both the AEP and CPC involve VP structures.

2.3. Agent by-phrases

Another notable grammatical property of the CPC is the use of an agent by-phrase. Researchers have debated over this property with differing grammatical judgments. Huddleston & Pullum et al. (Reference Huddleston and Pullum2002: 1200) argued that the CPC can optionally include an agent by-phrase, similar to the canonical be-passive construction, by providing examples like The article needs checking (by the editor) and The article needs to be checked (by the editor). On the other hand, Toyota (Reference Toyota2006) claimed that the CPC differs from the canonical be-passive in its inability to co-occur with an agent by-phrase, considering sentences like This TV needs fixing by the electrician to be ungrammatical, unlike This TV needs to be fixed by the electrician.

Puckica (Reference Puckica2009) contributed to this discussion by providing corpus examples, such as Cataclysmic first-quarter figures from MIPS Computer Systems Inc. … underline how much the company needs rescuing by Silicon Graphics Inc. This indicates that the construction can, in practice, include an agent by-phrase. Furthermore, Kim (Reference Kim2018) showed that out of his 609 CPC examples only one example contained an agent by-phrase, suggesting its rare usage throughout the history of English.

2.4. Subject (in)animacy

The final characteristic of the CPC concerns the animacy of the subject within the construction. Research, including Toyota (Reference Toyota2006) and Kim (Reference Kim2018), has shown that approximately 40 percent of CPC examples feature an animate subject. This trend has remained relatively consistent since the late modern English period. However, Kim (Reference Kim2018) further observed a notable change in Present-day English, where the frequency of CPC examples with want as the matrix verb and an animate subject has risen significantly to 85.7 percent. This represents a substantial increase from 30–35 percent during the Late Modern English period. This trend suggests a diachronic change in the usage of the CPC when want serves as the matrix verb, in particular in terms of the preference for the animate subject.

2.5. Research questions

We have thus far reviewed several key grammatical properties of the CPC discussed in previous studies. However, many of these studies were based on the researchers’ own intuitions, leading to varied grammatical judgments (Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985: 1189–90; Huddleston & Pullum et al. Reference Huddleston and Pullum2002: 1199–1200; Albrespit Reference Albrespit2007; Puckica Reference Puckica2009). Additionally, while some studies incorporated corpus examples, they often provided limited frequency data from relatively small samples (Visser Reference Visser1963–73: 1886–8; Toyota Reference Toyota2006). Moreover, a significant gap in the literature is related to the lack of exploration into strongly associated matrix verbs (or verb classes) and possible interrelations between matrix verbs (or verb types) and other constructional elements. Lastly, previous literature lacks an account of the passive interpretation of the CPC induced without any overt passive morphology.

Considering all these together, our present study seeks to address the following research questions:

1. 1. What is the distribution of verbs in Visser’s list within the CPC’s matrix verb slot in Contemporary American English, and how strong are their statistical associations? Should these verbs be grouped together or divided into subtypes?

2. 2. Among the various semantic types of -ing complements, are there any correlations between matrix verbs (or verb types) and -ing complement types?

3. 3. How common is the use of the agent by-phrase in the CPC, and are there any specific verbs (or verb types) that are strongly associated with its presence or absence?

4. 4. What are the distribution patterns of animate and inanimate subjects in the CPC, and which matrix verbs (or verb types) show strong correlations with the (in)animacy of the subject?

5. 5. What are the mechanisms that can account for its passive interpretation induced without any overt passive morphology?

To answer these questions, we conducted a comprehensive analysis of the CPC using real-life data from a large and balanced corpus, adopting both quantitative and qualitative methods.

3.1. Data

For this study, we utilized data from the Corpus of Contemporary American English (COCA) (Davies Reference Davies2008). The corpus is composed of more than one billion words, with an annual contribution of 24–25 million words from 1990 to 2019. This extensive collection is fairly equally distributed into eight distinct registers: five fundamental registers (i.e. spoken, fiction, popular magazines, newspapers and academic texts) and three newly added informal registers (TV and movie subtitles, blogs and webpages).

For our current study, we focused on the five primary registers. The initial step in our methodology involved extracting all combinations of the 22 matrix verbs identified by Visser (Reference Visser1963–73: 1888) and -ing verb forms. To achieve this, we employed specific search strings in COCA, as exemplified in the following:

In this query, [need]_vv represents all variants of the verb need (i.e. need, needs, needing and needed), and v?g corresponds to an -ing verb. This search string helped us to gather all instances where any form of need as a lexical verb was followed by an -ing verb. We replicated this method for each of the 22 matrix verbs listed in Visser’s work, and the initial raw frequencies of these combinations, before manual filtering, are presented in table 1.

Table 1. Raw frequencies of combinations of 22 matrix verbs listed in Visser (Reference Visser1963–73) and -ing verb forms in COCAFootnote ²

Note: Frequencies in table 1 represent raw instances retrieved from COCA prior to manual exclusion of irrelevant tokens (see (7)).

Then, we manually filtered out irrelevant examples as in (7):

In (7a), the matrix verb hated is followed by an -ing complement having, but it does not yield a passive interpretation. In (7b), needed is not directly complemented by hiding; instead, hiding modifies places and, thus, is not a complement of the verb needed. After this data-cleaning process, we identified a total of 2,541 CPC examples for analysis. To ensure reliability in this filtering and classification step, all 39,130 instances initially retrieved from the corpus were checked manually by both authors and subsequently reviewed by an independent non-author. Clear non-CPC cases were excluded. Borderline cases, such as suffer bruising and bear repeating, were retained when a passive reading remained contextually available, even if a non-passive interpretation was also plausible. We acknowledge that this conservative inclusion strategy may allow some non-passive tokens into the dataset, but we consider it appropriate because such borderline cases reflect how English speakers actually use the CPC in practice.

3.2. Methodology

Collostructional analysis is a family of quantitative methods in corpus linguistics used to examine the interaction between words and their grammatical constructions. In this study, we applied three types of collostructional analysis to statistically investigate the real-life grammatical properties of the CPC: collexeme, covarying collexeme and distinctive collexeme analyses.

Firstly, we employed the collexeme analysis to scrutinize verbs within the CPC’s matrix verb slot. This method reveals how specific semantic classes of words are naturally attracted to, or repelled from, particular construction slots (Stefanowitsch & Gries Reference Stefanowitsch and Gries2003; Gries & Stefanowitsch Reference Gries, Stefanowitsch, Achard and Kemmer2004a; Wulff Reference Wulff, Gries and Stefanowitsch2006; Hilpert Reference Hilpert, Allan and Robynson2012; Stefanowitsch Reference Stefanowitsch, Hoffmann and Trousdale2013, Reference Stefanowitsch, Herbst, Schmid and Faulhaber2014; Perek Reference Perek, Glynn and Robinson2014). A notable application of this method can be seen in Stefanowitsch & Gries’ study, where they discovered that in the British National Corpus (BNC), the V1 slot of the transitive into -ing construction is often filled with trickery verbs like trick and mislead or force verbs like coerce and force (e.g. They {tricked/forced} him into believing them).

Secondly, we used the covarying collexeme analysis to identify patterns between verbs (or verb types) in the matrix verb slot and those in the -ing slot of the CPC (Gries & Stefanowitsch Reference Gries, Stefanowitsch, Achard and Kemmer2004a, Reference Gries, Stefanowitsch, Newman and Rice2010; Stefanowitsch & Gries Reference Stefanowitsch and Gries2005; Stefanowitsch Reference Stefanowitsch, Hoffmann and Trousdale2013; Hilpert Reference Hilpert, Glynn and Robinson2014). This method is particularly useful for discovering combinations of two lexemes that co-occur more frequently than expected by chance within two slots of a construction. For example, Stefanowitsch & Gries (Reference Stefanowitsch and Gries2005), using BNC data, found a strong correlation between a trickery frame and a belief frame in the transitive into -ing construction, as evidenced by the four strongest covarying collexeme pairs such as fool–thinking and mislead–believing (e.g. They {fooled/misled} him into thinking he would never find out the truth).

Thirdly, we also applied the distinctive collexeme analysis to examine whether specific verbs (or verb types) in the matrix verb slot tend to be more strongly associated with certain types of subjects (animate or inanimate) and the presence/absence of an agent by-phrase (Gries & Stefanowitsch Reference Gries and Stefanowitsch2004b; Gilquin Reference Gilquin2006, Reference Gilquin, Leino and von Waldenfels2012, Reference Gilquin2015; Hilpert Reference Hilpert2006; Stefanowitsch Reference Stefanowitsch2006, Reference Stefanowitsch2018; Wulff Reference Wulff, Gries and Stefanowitsch2006; Wulff et al. Reference Wulff, Stefanowitsch, Gries, Radden, Köpcke, Berg and Siemund2007; Gries & Wulff Reference Gries and Wulff2009). This method is typically used to analyze one slot in two or more related constructions. For instance, Gries & Stefanowitsch (Reference Gries and Stefanowitsch2004b) used it to identify distinctive verbs in the ditransitive versus the to-dative constructions. They showed that verbs like give, tell, show and offer, which deliver a sense of direct or metaphorical contact between the agent and recipient, are more strongly associated with the ditransitive construction, while verbs like bring, take and pass, conveying a sense of accompanied motion or distance between the agent and recipient, are more preferably used with the to-dative construction.

At this juncture, it is important to note that raw frequency counts alone cannot reliably capture the strength of association between a verb and a constructional slot. A verb may occur frequently in the corpus simply because it is common overall, without being especially characteristic of the construction. Collostructional analysis, therefore, employs statistical association measures that compare observed frequencies against expected distributions, thereby identifying genuine attraction or repulsion (Gries & Stefanowitsch Reference Gries, Stefanowitsch, Achard and Kemmer2004a; Stefanowitsch Reference Stefanowitsch, Hoffmann and Trousdale2013; Desagulier Reference Desagulier2017).

However, the ‘canonical’ version of collostructional analysis has been the subject of sustained debate. Bybee (Reference Bybee2010), from a usage-based perspective, questioned the normalization procedures and, more broadly, the cognitive plausibility of downweighting frequent lexemes. In her view, frequency is not statistical ‘noise’ but a key factor in entrenchment and category formation. For instance, highly frequent verbs such as go or know often appear across many constructions, and thus may receive a low collostructional score. Yet, as Bybee shows with be going to or the discourse marker I don’t know, these high-frequency items are cognitively central: they anchor constructions in speakers’ mental grammar, undergo reduction and semantic bleaching, and drive processes of grammaticalization. From this viewpoint, procedures that minimize the role of frequent items risk overlooking precisely the elements that most strongly shape grammatical knowledge.

Schmid & Küchenhoff (Reference Schmid and Küchenhoff2013) and Küchenhoff & Schmid (Reference Küchenhoff and Schmid2015) added further methodological critiques, particularly of the Fisher-Yates exact test (FYE), one of the most widely used statistics in early collostructional studies. They argue that FYE tends to conflate effect size with sample size, rests on questionable assumptions of randomness and complicates direct comparisons with alternative descriptive measures such as attraction and reliance.

In response, Gries (Reference Gries2012, Reference Gries2015) has emphasized that collostructional analysis is not bound to FYE and that multiple association measures can be applied depending on the research question. Moreover, he has shown that results obtained via collostructional analysis consistently align with experimental findings, often more reliably than approaches based on raw frequency alone.

In line with Gries’ argument that association measures should be chosen to match the research question (rather than relying on raw frequencies alone), we adopt collostructional analysis in this study and implement it in the coll.analysis 4.0 package in R Studio, with the log-likelihood ratio (LLR) as the association measure.Footnote ³ LLR has become one of the most widely used statistics in collostructional studies (Stefanowitsch & Gries Reference Stefanowitsch and Gries2003; Hilpert Reference Hilpert, Allan and Robynson2012) and ensures comparability with prior work. It is sensitive to both frequency and association strength, and G²/LLR is among the most widely used association measures in collostructional analysis (Gries Reference Gries2019).Footnote ⁴

5.1. Syntactic issues

The most intriguing syntactic and semantic characteristic of the CPC is its passive interpretation without any overt passive verb form. Such a mismatch between an active form and a passive meaning has been observed in other constructions like the middle and passival constructions as in (15) (O’Grady Reference O’Grady1980; Ackema & Schoorlemmer Reference Ackema and Schoorlemmer1994; Warner Reference Warner1995; Hundt Reference Hundt, Mair and Lindquist2004; Smitterberg Reference Smitterberg2016):

As with the CPC in (15a), the middle construction as in (15b) and the passival construction as in (15c) do not involve any overt passive form, but they still induce passive interpretations, such that their subject referent should be understood as an undergoer of the event denoted by the active verb form sell in (15b) or the -ing complement building in (15c).

An important question is how a passive interpretation of the CPC is achieved. To address this, one may consider the structure developed by Strelluf (Reference Strelluf2022), which was originally proposed for similar constructions by Tenny (Reference Tenny1998) for the AEP need washed and by Huddleston & Pullum et al. (Reference Huddleston and Pullum2002) for the CPC need washing.

First, as discussed in section 2.2, AEP complements, unlike to-infinitival passive complements, do not allow the adjectival prefixes like un- ‘not’ as in (16a). Furthermore, they also do not allow negation within the complement, whereas to-infinitival passive complements do, as illustrated in (16b).

Edelstein attributes this contrast to syntax, which distinguishes AEP from the to-infinitival passive. Edelstein provides structures as in (17). The to-infinitival passive (17a) involves need selecting a full Tense Phrase (TP), which contains a Passive Phrase (PassP). This allows elements such as negation to appear within the complement (e.g. The car needs not to be fixed). By contrast, (17b), the structure for the AEP lacks those additional projections and involves only Aspectual Phrase (AspP).

Because the AEP lacks the projections needed to host negation, forms such as The car needs not fixed are ungrammatical.

Strelluf (Reference Strelluf2022) provides empirical support for this analysis using Twitter corpus data, where he examines three structures: the to-infinitival passive, AEP and CPC. His findings confirm that, like AEP, CPC constructions do not appear with the adjectival prefix un-, as seen in (16a), and do not include negation within the complement, as seen in (16b). These patterns suggest that CPC shares the same structural properties as AEP, aligning with (17b), whereas the to-infinitival passive follows (17a). Considering all these together, Strelluf concluded that they are all raising constructions, where the argument (e.g. the car) of the verb fix within the embedded verb phrase is ‘raised’ to the matrix subject position.

Strelluf’s (Reference Strelluf2022) so-called ‘raising’ analysis primarily focuses on the verb need, but this analysis can also be extended to other verbs that appear in the CPC. To understand why the CPC with need involves raising, consider the examples in (18):

In (18a), this construction is not a matrix argument. Semantically, (18a) can be paraphrased either as (18b) or as (18c). This demonstrates that the verb need in the CPC does not directly assign a thematic role to the subject but instead raises it to the subject position.

Similarly, other matrix verbs in the CPC, such as want, require, deserve, bear and merit, can be taken to involve this type of ‘raising’ structure, as in (19):Footnote ⁸

This parallel suggests that a similar ‘raising’ analysis can be applied to these verbs. Like need, these verbs in the CPC do not directly assign a thematic role to the subject but raise it to the matrix subject position. Therefore, want, require, deserve, bear and merit can be analyzed similarly, supporting a unified syntactic analysis for this class of verbs in the CPC.

However, this generalization does not apply to all matrix verbs found in the CPC. For instance, verbs such as await, escape and prevent exhibit different syntactic behavior, as shown in (20):

The example in (20a) shows that these verbs can be used in the CPC, but they cannot take an extraposed clausal complement along with a dummy it subject as in (20b) or a clausal subject as in (20c).

The discrepancy between (19) and (20) indicates that different syntactic structures are required to account for the CPC with certain verbs. In other words, while the ‘raising’ analysis works well for verbs such as need, want, require, deserve, bear and merit, a different account is necessary for verbs like await, escape and prevent. Footnote ⁹

Before concluding this section, it is important to emphasize a key distinction between the examples in (19) and (20). In (19), the matrix verb does not assign a thematic role to its subject directly, whereas in (20), it does. This distinction suggests that the CPC consists of two constructions, CPC-raising and CPC-control. We adopt the terms raising and control and the associated diagnostics in the standard sense (Davies & Dubinsky Reference Davies and Dubinsky2004; see also references therein). In CPC-raising, the subject’s thematic role is assigned by the embedded verb phrase; in CPC-control, it is assigned by the matrix verb. Additionally, in CPC-control, the matrix subject controls the understood subject of the embedded -ing verb phrase. Recognizing this distinction is crucial for understanding the syntactic and semantic behavior of CPC matrix verbs and necessitates distinct structural analyses for these subtypes.

5.2. A construction-based analysis

We now provide an account of the CPC from a Construction Grammar (CxG) perspective. The key assumptions of CxG, as summarized from Goldberg (Reference Goldberg1995, Reference Goldberg2006, Reference Goldberg, Hoffmann and Trousdale2013), Sag (Reference Sag, Boas and Sag2012), Kim & Davies (Reference Kim and Davies2016), Kim & Michaelis (Reference Kim and Michaelis2020), are as follows:

• • All levels of linguistic description (including morpheme, word, phrase and clause) involve pairings of form with semantic or discourse functions.

• • Constructions vary in size and complexity, with form and function specified if not immediately transparent.

• • Constructions are learned based on linguistic input and general cognitive mechanisms.

According to these assumptions, constructions in CxG involve pairings of form with semantic or discourse functions. Importantly, any linguistic pattern is considered a construction. This includes patterns whose form–meaning relations are not entirely identifiable from their component parts or other related constructions (e.g. idioms like jog (someone’s) memory) and those whose form–meaning relations are fully identifiable (e.g. words like avocado and anaconda). Moreover, even rule-like patterns such as argument structures (e.g. transitive, ditransitive, etc.) are treated as constructions within CxG.

Within the CxG framework, argument structures are also considered constructions, and a verb’s inherent core meaning is distinct from the semantics linked to these argument structure constructions. This means that in CxG, a verb with its own core meaning can occur in different argument structure constructions, as illustrated in the following examples:

In these examples, the verb kick has a core meaning of striking forcibly with the foot. However, a range of distinct interpretations arises from the various argument structure constructions. When each of these constructional meanings in table 6 is linked to the matrix verb kick, the proper interpretations are evoked accordingly. In (21a), the transitive construction involves a direct action affecting an object. In (21b), the ditransitive construction suggests giving something to someone through the action. In (21c), the caused-motion construction indicates causing an object to move in a certain direction. In (21d), the resultative construction implies causing a change of state in an object. In (21e), the way-construction suggests movement along a path by means of the action.

Table 6. Related construction and semantic properties

This demonstrates how different argument structure constructions modify the core meaning of kick to produce specific interpretations based on the syntactic context. The flexibility of verbs to participate in various argument structure constructions is a fundamental aspect of the CxG framework, showcasing how a single verb can generate multiple nuanced meanings.

For a formal account of the CPC, we propose that it consists of two distinct argument structure constructions: CPC-raising and CPC-control. First, consider the constructional constraints of CPC-raising below:

The constructional constraint in (22) indicates that syntactically CPC-raising takes two arguments (a nominal subject NP/CP and an -ing complement) and that semantically the construction denotes the proposition that it is necessary or worthy (relating to V) that the subject (x) undergoes the event depicted by the -ing complement (y). This constructional constraint ensures that the subject is understood only as the undergoer of the event described by the -ing complement. That being the case, the sentence the house needs painting can be paraphrased as it is necessary that the house be painted and the sentence the situation deserves watching can be paraphrased as it is worthy that the situation is watched.

Next, let us consider the CPC-control type, as illustrated below:

The constructional constraint in (23) states that the CPC-control construction takes two syntactic arguments: a subject NP and an -ing complement. Semantically, it expresses the proposition that the subject NP (x) either performs the event V on the event (y) that is acted upon the subject NP (x), or the subject NP (x) experiences the event V from the event (y) that is acted upon it. This constraint then ensures that the subject can be interpreted as either the actor or experiencer of the matrix verb, and, crucially, as the undergoer of the event described by the -ing complement, at the same time. For example, in the actor case, Payne awaits sentencing can be paraphrased as Payne performs the event of awaiting, which is directed toward the event of sentencing that affects him. In the experiencer case, Almost one trainee doctor in eight has suffered bullying at work can be paraphrased as the trainee doctor has experienced the event of suffering from the event of bullying acted upon him/her.

A significant advantage of distinguishing between the two CPC subconstructions, CPC-raising and CPC-control, is that it helps explain discrepancies in the syntactic realization of the subject. The prediction is that in CPC-raising the subject can be either an NP or a CP but in CPC-control, it can only be an NP. This prediction is indeed borne out in our data. Observe the examples in (24).

In (24a) the matrix subject is a CP while in (24b) it is the dummy pronoun it and it is linked to the extraposed CP at the end. In our data, examples like these are only found with the verbs that license CPC-raising, but not with those that license CPC-control. Therefore, the examples above provide an additional justification for the distinction between CPC-raising and CPC-control.

6.1 Co-occurrence with an agent by-phrase

The distinctive collexeme analysis results revealed that matrix verbs like escape, await, avoid and prevent are strongly associated with the presence of an agent by-phrase; conversely, need and bear are strongly associated with its absence. Notably, most of the former fall under the CPC-control category (e.g. escape, await, avoid and prevent), whereas the latter belong to the CPC-raising category. This pattern suggests that although the agent by-phrase is generally not frequent in the CPC, its presence statistically prefers CPC-control to CPC-raising.

6.2. (Dis)preference for animate subjects

The distinctive collexeme analysis results also showed clear (dis)preference patterns regarding subject animacy. Matrix verbs strongly associated with animate subjects include await, avoid, deserve, want, suffer, escape and need, whereas bear and require strongly favor inanimate subjects. Again, most of the former belong to CPC-control (e.g. await, avoid, suffer and escape), while the latter fall under CPC-raising. This indicates that while animate subjects are generally not frequent in the CPC, statistically they favor CPC-control over CPC-raising.

6.3. Patterns in -ing complements

The covarying collexeme analysis results revealed that certain matrix verb (or verb types) and -ing complement pairs occur more frequently, forming distinct semantic frames in the CPC. CPC-control and CPC-raising structures exhibit different -ing complement patterns based on subject animacy, intentionality and event interpretation. CPC-control verbs, with animate subjects, select action-oriented complements linked to negative outcomes or obligations, such as escape–bombing, avoid–questioning or suffer–bruising. These emphasize external obligations requiring avoidance, anticipation or endurance. In contrast, CPC-raising verbs, typically with inanimate subjects, favor complements connoting passive transformations, evaluations or requirements, as seen in bear–repeating, require–priming or need–replacing. Since these subjects lack volitional agency, the complements highlight necessary changes rather than deliberate actions. The findings highlight how CPC-control verbs align with agentive, negative-event management, whereas CPC-raising verbs favor passive necessity. This distinction thus suggests the interplay between the matrix verb and subject type and event structure as key factors in the -ing complement selection.