2.1. A constructional approach to verb-forming suffixation

The basic assumption of Construction Grammar (CxG) is that a speaker’s knowledge of language consists of established form–meaning associations (constructions and schemas), that are stored in an interrelated network (Lakoff Reference Lakoff1987; Fillmore Reference Fillmore1988; Fillmore, Kay & O’Connor Reference Fillmore, Kay and O’Connor1988; Wierzbicka Reference Wierzbicka1988), known as the ‘constructicon’ (Goldberg Reference Goldberg1995: 5). The notion of construction extends across the lexicon–syntax spectrum, ranging in schematicity from the simplest structures, such as simplex words and phrasal idioms, to the most complex and abstract schemas of morphemic and clausal structures (Goldberg Reference Goldberg2013: 436), the last two of which relate directly to the current article.

The four principal verb-forming suffixes in English are -ize, -ify, -en and -ate; they have Latinate origins, with the exception of -en which derives from Old English. These suffixes form transitive and/or intransitive complex verbs conveying a range of meanings, as shown in (1) and (2), where a contextualized example is presented in (a) and its interpretation in (b).

Examples (1) and (2) demonstrate that the function of verb-forming suffixation is to encode, as an event type, the relationship between the semantics of the base item (in square brackets) and the arguments of the complex verb. Such event types were initially classified into seven semantic categories (Plag Reference Plag1999), the interpretations of which are expressed as glosses or paraphrases, e.g. (1b) and (2b). The complex verbs in (1a) and (2a) correspond respectively to Plag’s Locative and Causative semantic categories. Plag’s (Reference Plag1999) original set of seven semantic categories was extended to twenty by Laws (Reference Laws2023), who conducted an extensive corpus-based study of 429 complex verb types corresponding to 651 verbal derivative senses; the thirteen further semantic categories emerged from the semantic analysis of the corpus data and from additional interpretations proposed by Marchand (Reference Marchand1969) and Dixon (Reference Dixon2014). The twenty semantic categories fall into four argument structure groups (see Appendix).

Of the 651 senses, 576 were assigned to one of the twenty semantic categories, by Laws (Reference Laws2023); these will henceforth be referred to as ‘Categorized’ senses, and the remaining 75 as ‘Uncategorized’.

By adopting a CxG approach, Laws (Reference Laws2023) identified the range of argument structure constructions corresponding to paraphrases associated with each semantic category. For example, the interpretation of the Locative in (1b) is expressed by the form–meaning pair embodied by the Caused-Motion argument structure construction (Goldberg Reference Goldberg1995: 52), represented in (3). The syntactic component on the left-hand side corresponds (↔) to the semantic component on the right. The grammatical roles, Subject, Object and the NP of the Oblique, map onto the argument roles Cause, Theme and Goal, respectively.

The meaning conveyed by the Causative semantic category in (2b) corresponds to the Resultative argument structure construction (Goldberg Reference Goldberg1995: 79), where the Object Complement is an adjective, rather than a noun. Following Laws (Reference Laws2023: 132), the Resultative argument structure construction relating to the Causative semantic category is shown in (4). Here, the grammatical roles, Subject, Object and adjectival Object Complement, map onto the argument roles Cause, Patient and State, respectively.

The theoretical framework of CxG was extended by Booij (Reference Booij2010) to the analysis of inflection, derivation and compounding. With respect to complex words, Construction Morphology (CxM) treats all derivatives, e.g. un-fair and central-ize, as lexical structures that are specified in terms of a hierarchy of schemas and subschemas that represent the shared interpretation of derivatives bearing the same prefix or suffix. Laws & Booij (Reference Laws and Booij2025) present a generalized schema for verb-forming suffixation in English, as shown in (5), based on principles proposed by Booij (Reference Booij2010) for polysemous affixes, such as the nominalizer -er. The left-hand formal component of the schema in (5) corresponds to the right-hand meaning component, co-indexed j.

The left-hand component specifies the morphological composition and syntactic characteristics of the complex verb. Since the derivative base, indexed i, may be a noun (hyphen-ate), an adjective (sweet-en), a truncated stem (synchron-ize), or a bound stem (spec-ify), the grammatical class of the base is unspecified (x). The attachment to the base of any of the four suffixes (indexed q) forms a complex verb V, indexed j. Verbal derivatives can be either transitive or intransitive, therefore optionality of the Object NP_m is indicated with parentheses.

The right-hand meaning component denotes an event type involving the relation R and the meaning of the base, SEM_i. The relation R in generalized affix schemas is defined by Booij (Reference Booij2010: 17) as an unspecified relationship determined by the meaning of the base, the suffix and associated conceptual and real-world knowledge, and that it ‘is filled in by specific subschemas and interpretation mechanisms based on the semantics of the base words’. With respect to verbal derivatives, Laws (Reference Laws2023: 113) proposed that R represents the relationship encoded by the argument structure constructions that express the meanings conveyed by the various semantic categories. For example, derivatives such as centralize in (1) denote events with a Locative relation to the base, the generalized schema of which is shown in (6a); the right-hand semantic component of (6a) is instantiated by the Caused-Motion construction in (6b).Footnote ¹

Similarly, events denoting a transitive Causative relation to the base, such as stabilize in (2), are expressed by the generalized schema in (7a), which is instantiated by the Resultative construction in the semantic component of the schema (7b), and so on for each semantic category. Thus, according to the CxM approach, the interpretation of complex verbs is expressed in terms of argument structure constructions.

It must be noted, of course, from a formal perspective, that the productivity of verb-forming suffixation is affected by constraints that determine whether a particular suffix can attach to a potential base, such as the prosodic characteristics of the base and the nature of the final base phoneme (see Plag Reference Plag1999 for a comprehensive review). Of the four English verb-forming suffixes, -ize is the most productive, accounting for 59.9 per cent of all complex verbs in spoken language, followed by -ify (18.4 per cent), -en (13.3 per cent) and -ate (8.4 per cent), as reported by Laws (Reference Laws2023: 38). Thus, formal constraints can restrict productivity, but it is the potential for both verb-forming suffix schemas and their related argument structure constructions to exhibit productivity and creativity that forms the focus of the present discussion.

2.2. Productivity, creativity and their interrelationship

Morphological schemas such as the form–meaning pairing associated with the nominalizer -ism, are relatively productive in English (Booij Reference Booij2010). The formal component of the schema specifies that the base can be an adjective, noun or proper noun, as in secularism, consumerism and Marxism; the semantic component of the schema conveys the meaning ‘ideas, principles of a doctrine or system denoted by the base’ (Stein Reference Stein2007). An informal representation of a schema for the nominalizer -ism is shown in (8), adapted from Booij (Reference Booij2010), where the left-hand form component corresponds to (↔) the right-hand semantic component.

Language users readily generate new linguistic structures by utilizing these productive schemas to produce neologisms. For example, in (9), taken from the Corpus of Contemporary American English (COCA, Davies Reference Davies2008–) the term Sharptonism refers to the ideologies of Reverend Alfred Sharpton Jr, an American civil rights activist; the definition of this derivative is not listed in the Oxford English Dictionary (OED) or Merriam-Webster (M-W).

The neologism Sharptonism reflects the form–meaning association for -ism presented in (8): the new complex word is modelled on an existing form–meaning morphological schema, where the base is a proper noun. However, language users may also produce expressions that do not adhere to established morphological patterns, as shown in (10).

Here the bases are not adjectives, nouns or proper nouns, as would be expected, but instead involve the verb phrases get-even and get-even-with-them. The existing form–meaning function of the -ism schema has been instantiated in a creative way by accommodating a novel base category, the verb phrase. Nevertheless, language users of English are readily able to infer the intended meaning of this neologism.

A similar phenomenon can be observed at clause level. Argument structure constructions (Goldberg Reference Goldberg2006), e.g. the Caused-Motion construction (3), represent form–meaning associations embodied in expressions such as (11).

Thus, the Caused-Motion construction specifies the interpretation of verbs, such as put, which involve a Direct Object (Obj) in the formal component of (3). This productive argument structure construction is then reflected in novel formulations of Locative complex verbs based on (6), as illustrated in (12) from the British National Corpus (BNC).

Yet, as shown by the classic example documented by Goldberg (Reference Goldberg1995: 54–5) and reproduced in (13), language users may apply this constructional schema to intransitive verbs, such as sneeze, thus illustrating that it is the Caused-Motion construction itself that confers a sense of ‘caused motion’ to the event of sneezing.

Thus, examples (10) and (13) exhibit ‘violations of selectional restrictions’ that are resolved through coercion: a process that renders potentially impossible expressions interpretable, as Bergs (Reference Bergs2018: 283) observes in relation to other novel manifestations of established constructions. Such innovative formulations of existing constructions may then become entrenched in language use, as evidenced by corpus analysis in relation to the Resultative construction (Hoffmann Reference Hoffmann and Dancygier2017: 295). It will be noted that the notion of coercion has also been challenged (Ziegeler Reference Ziegeler2007), but the details of these counter-arguments are beyond the scope of this article.

The productivity–creativity debate has considered the question: to what extent are novel expressions merely additional formulations of an existing pattern, as in (9) and (12), and to what extent can examples (10) and (13) be considered creative? These examples illustrate that a construction can be seen to express two kinds of productive properties:

Sampson (Reference Sampson and Hinton2016) proposed a productivity–creativity dichotomy: the application of an existing productive process, e.g. (9) and (12), is described as F-creativity, or Fixed-creativity. In other words, the meaning of the new form is totally predictable from an existing schema, i.e. (14a). From a constructional perspective, therefore, Sampson’s notion of F-creativity essentially reflects the productivity of the specific construction in question.

In contrast to F-creativity, Sampson (Reference Sampson and Hinton2016) suggests that E-creativity, or Enlarging/Extending creativity is employed when established linguistic patterns are adapted in unconventional ways. Examples in the literature have included the language style of the character Yoda in Star Wars and the use of nonsense words in place of noun, adjective and verb stems in Lewis Carroll’s Jabberwocky (Uhrig Reference Uhrig2018: 298). These are clearly instances of highly creative language, but the question arises as to whether examples (10) and (13) also reflect E-creativity. In proposing a resolution to this question, Uhrig (Reference Uhrig2018) argues that Sampson’s F–E dichotomy is more usefully envisaged as an F–E creativity cline, as illustrated in figure 1, although the boundary between F- and E-creativity is far from uncontentious (Bergs Reference Bergs2018; Flach Reference Flach2025).

Figure 1. F–E creativity cline

F-creativity, then, corresponds to productivity, as reflected in (14a). By contrast, at the other extreme, ‘[t]he products of E-creativity are unpredictable and can break with existing rules in new, innovative ways’ (Bergs Reference Bergs2018: 290), such as Yoda-speak, i.e. extreme coercion. Thus, in the middle of the cline, constructs display more E-creativity than others: examples (10) and (13) can be viewed as less ‘Fixed’ than (9) and (12), and less ‘Enlarging/Extending’ than Yoda-speak. It is proposed here that (10) and (13) exhibit ‘Extended-Fixed’ (henceforth ExFx) creativity, as reflected in (14b), and that the degree of ExFx creativity covaries with the strength of the associated coercive mechanism. Figure 2 illustrates the relative position of the above examples on an elaborated F–E creativity cline, referred to henceforth as the F–ExFx–E creativity cline.

Figure 2. F–ExFx–E creativity cline

As pointed out by Flach (Reference Flach2025), the creativity cline is not linear and the position of linguistic phenomena that fall along it are not evenly distributed.

The application of a creativity cline to verb-forming suffixation is proposed in section 2.4; firstly, the role of linguistic coercion and its relationship to the F–ExFx–E cline is discussed in the next subsection.

2.3. A continuum of coercive mechanisms

It was mentioned in section 2.2 that, when sectional restrictions are violated, coercion plays a critical role in resolving construction-internal incompatibilities, as discussed in relation to (10) and (13). Therefore, the productivity of a construction can be facilitated by licensing novel constructs through coercive mechanisms, which inherently involves increasing degrees of ExFx creativity, as illustrated in figure 2.

Audring & Booij (Reference Audring and Booij2016) examined three types of coercion, Selection, Enrichment and Override, proposing that they form a unified continuum that reflects the relative influence of context on the interpretation of a lexical item. From the weaker to the stronger end of the continuum, these coercive mechanisms increase in the ‘degree to which the lexical meaning contributes to the meaning of the utterance’ (Audring & Booij Reference Audring and Booij2016: 635). Although the authors argue that coercion by Selection, Enrichment and Override lie along a unified continuum, they also provide cases where these categories overlap. The examples presented here have been chosen to reflect the characteristics of the three categories as unambiguously as possible for illustrative purposes.

These three types of coercion correspond to points on the ExFx portion of the creativity cline (figure 2), and relate directly to the semantic analysis of verbal derivatives, as shown in the following subsections. The nature of extreme coercion with respect to the formation of complex verbs is also illustrated.

2.3.1. Selection by coercion

As the weakest form of coercion, Audring & Booij (Reference Audring and Booij2016: 620) argue that Selection involves ‘contextual adjustments’ that resolve ambiguities in the construal of a word in context, as illustrated in (15), adapted from Pustejovsky (Reference Pustejovsky2011: 1403).

The noun book in (15a) refers to the ‘physical object’, whereas in (15b) it refers to the ‘information’ it contains. Thus, the semantic properties, or qualia (Pustejovsky Reference Pustejovsky1995), of book provide the context that ‘selects’ the appropriate construal of the noun in context.

Laws & Booij (Reference Laws and Booij2025) propose that Selection by coercion occurs in the interpretation of complex verbs through the implementation of semantic rules. The precise interpretation of a verbal derivative depends on the meaning of the predicate that is instantiated by the underlying argument structure construction; the properties of the base item and the grammatical object of the complex verb determine the ‘selection’ of that predicate. For instance, the productivity of the Locative semantic category, corresponding to the Caused-Motion argument structure construction (6), is enhanced through the process of contextual adjustment, i.e. the range of predicates that can be instantiated by the construction is extended. For example, in place of move in (1b), the predicates put, copy or direct may be selected, based on the characteristics of the Direct Object, as illustrated in (16) to (18). Laws (Reference Laws2023) relates such predicate groupings to verb-class-specific constructions (Croft Reference Croft, Cuyckens, Berg, Dirven and Panther2003).

Thus, the semantic scope of the Locative is extended by Selection to include nuanced, contextually adjusted interpretations of the Caused-Motion Construction, and hence its productivity is increased through low-level Extended-Fixed creativity. Selection is observed in ten of the twenty semantic categories (Laws Reference Laws2024;Footnote ² Laws & Booij Reference Laws and Booij2025).

2.3.2. Enrichment by coercion

Enrichment by coercion has attracted considerably more attention than Selection, particularly with respect to aspect (Jackendoff Reference Jackendoff1991; Talmy Reference Talmy2000; Michaelis Reference Michaelis2024) and predicate-argument constructions (Pustejovsky Reference Pustejovsky1995, Pustejovsky & Ježek, Reference Pustejovsky and Ježek2008). Enrichment involves the resolution of a semantic conflict through the ‘addition of unexpressed semantics to the utterance’ (Audring & Booij Reference Audring and Booij2016: 626).

A classic example of Enrichment is shown in (19), adapted from Audring & Booij (Reference Audring and Booij2016: 631). The complement X in the construction I’m done with X is an activity predicate, as in I’m done with cleaning the windows / writing the letter. However, in (19), the NP is the prepositional complement, thus, the constructional incompatibility is resolved by an ‘unuttered’ activity predicate (cleaning, writing) appropriate for the semantic properties of the NP.

Laws & Booij (Reference Laws and Booij2025) propose that Enrichment, through the execution of semantic rules, accounts for the elaboration of argument roles in the interpretation of complex verbs, such as Causative legalize and digitize, as illustrated in (20) and (21).

As shown in (7b), the Object of the Resultative construction that underlies the Causative semantic category has a Patient argument role. However, in (20), it is not Sunday trading that is directly affected by the action, but its ‘status’ according to law. Similarly, in (21), it is the ‘format’ of the documents that is digitized, not the documents themselves. The unuttered expressions ‘the status/format of’ thus fuse with the Patient role of the Resultative construction and the roles of Sunday trading and documents are subordinated to Theme modifier status. Examples (20) and (21) are typical cases of ‘reference transfer’ (Jackendoff, Reference Jackendoff, Hoffmann and Trousdale2013: 82–3), where additional implicit information acts as a non-realized semantic operator expressed in the meaning component of the relevant NP construction of the argument structure. Enrichment occurs in nine of the twenty semantic categories; further examples can be found in Laws (Reference Laws2024)Footnote ³ and Laws & Booij (Reference Laws and Booij2025).

2.3.3. Override by coercion

The ‘strongest’ class of coercion on the continuum is Override (Audring & Booij Reference Audring and Booij2016), where the utterance meaning overrides the lexical semantics of an item inserted into a construction. Word class conversion is an example of Override, as shown in (22), adapted from Audring & Booij (Reference Audring and Booij2016: 632).

In (22), the noun 2013 is coerced into functioning as an adjective, the lexical semantics of which are overshadowed by the meaning conveyed by the construction in which it occurs, i.e. 2013 is interpreted as ‘old-fashioned, dated’.

Override by coercion involving word class conversion was demonstrated in (10) for the suffix schema -ism. It is argued here that the fusion of the intransitive verb sneeze with the Caused-Motion construction (13), which requires a transitive predicate, is another example of Override. As mentioned in section 2.2, the event of sneezing is combined with the semantics of the Caused-Motion construction.

With respect to the interpretation of complex verbs, it is proposed here that Override involves some kind of adaptation of an argument role within an argument structure schema relating to a semantic category. A complex verb from the Similative category in which Override does not occur is vandalize, as illustrated in (23). The argument structure construction is presented in (23a), (23b) is a contextualized example, and the ‘expected’ instantiation of the argument structure construction is provided in (23c). The base vandal is the Theme in (23a) and the bus shelter is the Patient in the towards-adverbial expression. In this use of the Similative, the target of vandalization, the bus shelter, is clearly a Patient, since it becomes destroyed during the process.

Item (24), by contrast, presents an example of Override in the argument structure schema underlying the Similative. When the verb evangelize is used in the sense of ‘supporting an idea or project’, the target of the event, the cause in this case, is abstract (24b). Therefore, in contrast to (23), this target is not affected by the evangelization process; instead, the cause also functions as a Theme, as shown in (24a). Thus, the argument role of Patient is overridden by the properties of Theme2.

Laws (Reference Laws2023: 163) refers to the version of the Similative in (23) as ‘strong’ and that in (24) as ‘weak’, since the target encoded in the towards-adverbial expression in the former is affected by the event, whereas in the latter it is not. Thus, the argument role representing the target is overridden by coercion to resolve the semantic incompatibility within the argument structure construction in (24). Examples of the four of the twenty semantic categories that exhibit Override can be found in Laws (Reference Laws2024).Footnote ⁴

2.4. Productivity, creativity and coercion

A fully integrated creativity/coercion continuum is presented in figure 3: degree of coercion, from none to Unruly, is aligned with the F-ExFx-E creativity cline presented in figure 2. The items in brackets refer to examples provided earlier that relate to argument structure constructions associated with the interpretation of complex verbs.

Figure 3. Integrated F-ExFx-E creativity and coercion continuum for complex verbs

It is proposed here that, in accordance with the definitions of creativity discussed earlier, F-creativity corresponds to ‘pure’ productivity, where no coercion is involved, as in (2), stabilize. Selection, Enrichment and Override represent increasing levels of Extended-Fixed creativity, facilitated by increasing levels of coercive force. E-creativity relates to the schemas of Analysable Uncategorized verbs which exhibit unique argument structure patterns, i.e. that reflect Unruly coercion.

3.1. Aims and research questions

Verb-forming suffixation is constrained, on the one hand, by formal characteristics of the four suffix types and the bases to which they may attach, and facilitated, on the other, by coercive mechanisms that increase the scope of the verb-forming suffix schema to accommodate novel complex verbs that are readily understood by users of English. Thus, the current study is concerned with the nature of coercion in conventionalized complex verbs (the exact senses of which are listed in dictionary sources) and newly coined forms (not listed in dictionary sources).

The database of verbal derivatives and their associated constructional analysis originate from Laws (Reference Laws2023, Reference Laws2024). The classification of complex verb senses adopted here is summarized in (25) with the number of senses that occurred under each subclassification.

The first aim of the current study is to identify a creativity/coercion profile of complex verbs with respect to the integrated continuum presented in figure 3. To achieve this, the coercive mechanisms observable in suffix schemas and associated argument structure constructions of complex words were analysed in accordance with the creativity/coercion continuum. This analysis provides a benchmark profile for conventionalized verbal derivatives occurring in contemporary speech, even though the lexeme–sense combination in question may have been attested in documents dating from the Late Old English period (1100) to the present day (2014), as reported in the OED. It is predicted that language users are likely to employ coercive mechanisms to expand the productivity of suffix categories, thus, the second aim of the study is to evaluate whether the proportion of complex verb senses exhibiting creativity/coercion increases diachronically. The third aim is to identify whether novel complex verbs, formulated over the twenty-year period between 1994 and 2014, display higher degrees of creativity/coercion compared with conventionalized verbal derivatives. Finally, the question of how verb-forming suffixation reflects the notion of E-creativity through Unruly coercion is addressed. Thus, the following research questions are posed:

3.2. Data source and analysis procedures

The current study utilized the database of 429 complex verb types reported by Laws (Reference Laws2023) that had been compiled from an exhaustive search of the spoken elements of the BNC1994 and BNC2014 (Love et al. Reference Love, Dembry, Hardie, Brezina and McEnery2017). Many derivatives exhibit more than one sense, e.g. stabilize can be used transitively or intransitively; identify was found to have 14 distinct senses across the corpora. Laws’ semantic analysis revealed that the 429 complex verb types conveyed 651 senses, and since each sense relates to a single semantic category/argument structure construction, the unit of analysis is verb sense.

To address RQ1, the instances of coercion occurring in Categorized complex verb senses overall and across suffixes and semantic categories were calculated. For RQ2, the dates of first attestation of complex verb senses were extracted from quotes recorded in the OED. In each case, this date corresponds to the earliest OED quote where the sense of the derivative is identical to that occurring in the contemporary corpus; since many complex verbs are polysemous, each sense of each verbal derivative was considered. Thus, it is important to note that dates of first attestation refer to derivative lexeme/sense meanings that have remained unchanged over time, based on OED definitions.

To answer RQ3, the 30 neologisms derived from the spoken components of the BNC1994 and BNC2014 reported by Laws (Reference Laws2023) were supplemented by an exhaustive set of 21 complex verbs (not listed in the OED or M-W) that were drawn, for the purposes of this study, from COCA spoken texts for the periods 1990–4 and 2010–14 (henceforth COCA1994 and COCA2014), thus balancing for time frame across sources.

The BNC1994 sample combined the two subcorpora Demographically Sampled (DS1994, 40 per cent) and the Context-Governed (CG1994, 60 per cent), the former of which contains everyday conversation; the latter is composed of more formal spoken language from meetings, speeches and commentaries (Hoffmann et al. Reference Hoffmann, Evert, Smith, Lee and Prytz2008). The BNC2014 corpus also contains everyday conversation and is thus more closely matched in register to the DS1994 (Love et al. Reference Love, Dembry, Hardie, Brezina and McEnery2017). The COCA transcripts are drawn from unscripted TV and radio programs and thus fall in terms of register range of the British corpora between the DS1994 and BNC2014 at the informal end, and CG1994 at the formal end, but are closer to the latter.

It was noted in section 2.2 that coercive mechanisms can apply at the levels of suffix schema ([10] getevenism/geteven-with-themism) and argument structure ([13] He sneezed the napkin off the table). However, since the generalized suffix schema for complex verbs (5) is unspecified for the grammatical category of the base, the range of items that could function as the base is potentially unlimited, although only bases consisting of nouns, adjectives and bound stems were identified in the dataset. Therefore, coercion at the level of affix schema is not relevant to verb-forming suffixation in the way that it is for -ism suffixation, for example, as shown in (10); thus, the current study only focuses on coercion at the level of the argument structure constructions that underlie the meaning of verbal derivative semantic categories. The following paragraphs describe the process of identifying the types of coercion relevant to this study.

Selection by coercion is illustrated for Locative complex verbs in (16), palletize bricks, ‘put’, (17), computerize documents, ‘copy’ and (18), canalize water, ‘direct’: productivity of this semantic category is enhanced by the increased range of contextually adjusted predicates that enter the Caused-Motion construction.

Enrichment by coercion is demonstrated in (20), legalize Sunday trading, ‘make the status of Sunday trading legal’ and (21) digitize the documents, ‘make the format of the documents digital’, in relation to the Causative semantic category. Here, unuttered expressions, such as the ‘status’, ‘format’ or ‘characteristics’ of the Direct Object function as operators that resolve semantic incompatibilities between participant and argument roles in the construction.

Override by coercion, as it applies to complex verbs, is exemplified in (24), evangelize the cause ‘act like an evangelist towards the cause’, in relation to the Similative semantic category. This mechanism entails the replacement of one argument role in the ‘standard’, uncoerced argument structure construction (here, the Patient Direct Object) with another (here, a Theme), without radically altering the meaning conveyed by the construction.

Finally, Unruly coercion, as described in section 2.3.4, occurs with Uncategorized Analysable complex verbs, e.g. unionize, womanize and the neologism sanctionalize. In these cases, the interpretation of argument structure schemas involving the verb base is unique with respect to all other senses in the corpus.

In the following results section, multiple pairwise comparisons of coercion type frequencies are reported, e.g. Selection vs Enrichment. Log Likelihood (LL) analyses are used. In each case, df=1 and Bonferroni adjustments are applied, depending on the number of comparisons. Following Wilson (Reference Wilson, Bieswanger and Koll-Stobbe2013: 6), the Bayesian Information Criterion (BIC) is the effect size recommended for the LL statistic, where evidence against H₀ is indicated as follows: values 2-6 = positive, 6-10 = strong and >10 = very strong.

4.1. RQ1: A creativity/coercion profile of Categorized complex verbs

Table 3 presents the frequency of each coercion type for all Categorized senses, together with related percentages based on the number of Categorized senses in the BNC corpora (547 excluding neologisms). There are 570 instances of coercion, since a complex verb schema may express more than one type of coercion.

Table 3. Distribution of coercive mechanisms in Categorized complex verbs

The first important observation from table 3 is that in 61 per cent of the Categorized verb senses no coercive mechanisms were involved. Comparisons between frequency values revealed significant differences with strong effect sizes between No coercion and Selection (LL=125.57, p<0.0001, BIC=118.57), No coercion and Enrichment (LL=103.72, p<0.0001, BIC=96.72), No coercion and Override (LL=386.42, p<0.0001, BIC=379.42), Selection and Override (LL=91.47, p<0.0001, BIC=84.48) and Enrichment and Override (LL=110.94, p<0.0001, BIC=103.94). No significant difference was found overall between the instances of Selection and Enrichment, and instances of Override appear to be rare.

Figure 4 plots the proportion of coercive types as a function of suffix class. Figure 4 reveals that 90 per cent of -en complex verb schemas exhibit no coercion at all; this relative proportion was significantly greater than that of -ize (LL=10.82, p<0.01, BIC=4.80), although, compared with -ify and -ate, differences failed to meet the 0.017 adjusted α criterion. No other differences reached significance for No coercion.

Figure 4. Percentage of coercive types by suffix class

Regarding Selection, no significant differences were found across the suffix classes. Given that no -en verb schemas exhibit Enrichment, it is unsurprising that significant differences were obtained between this suffix category and -ize (LL=42.00, p<0.0001, BIC=35.97), -ify (LL=28.94, p<0.0001, BIC=23.67) and -ate (LL=23.36, p<0.0001, BIC=23.35). No other differences reached significance for Enrichment. Override only occurred in ten verb schemas (table 3), most of which involved the most productive -ize suffix class; these low numbers did not reveal any significant differences between suffix classes.

Turning now to the twenty semantic categories used by Laws (Reference Laws2023) (see Appendix), figure 5 presents the proportion of complex verb senses in each category that involved one or more coercive mechanism. Semantic categories are rank-ordered so that the degree of coercion increases from left to right along the x axis. Since the size of each category varies considerably (from 3 to 145), it is also important to consider relative category size, thus, category size as a proportion of all Categorized verbs is included in each semantic category label; the abbreviated form ‘Inch-’ refers to ‘Inchoative’. Although Enrichment is slightly more frequent than Selection (table 3), this difference was not significant, therefore, these coercion factors have been combined in figure 5. The graph is represented as a line rather than bar chart to assist the reader in identifying trends.

Figure 5. Percentage of senses exhibiting coercion within each semantic category

Figure 5 indicates that Selection and/or Enrichment are involved in the interpretation of the majority of the larger semantic classes (Causative, Ornative, Representative, Conformative, Imposative and Locative). Interestingly, the schemas underlying these semantic categories express three-place argument structure constructions such as the Caused-Motion, Resultative, with-Applicative and as-Predicative (see Appendix). By contrast, semantic categories that exhibit No coercion or Override, e.g. Determinative, Confirmative, Achievement, Defining and Acknowledging, are expressed by simple Transitive argument structure constructions. Finally, Override is mutually exclusive with Selection and/or Enrichment.

In response to RQ1, the above analyses indicate that coercion is a common feature of Categorized verbal derivatives (39 per cent), in particular Selection and/or Enrichment in relation to the Latinate suffixes and more complex argument schemas; this suggests that with respect to these suffixes, the combination of the suffix and argument schemas is ‘flexible’ and thus conducive to Extended-Fixed creativity. By contrast, coercion seldom occurs in the interpretation of -en verbal derivatives but, when it does, the ‘weakest’ form, Selection, is involved, suggesting that the relationship between this Old English suffix and the argument schemas it is associated with is considerably ‘rigid’. Perhaps this rigidity contributes to the partial productivity of this suffix, a characteristic that is normally attributed to its strict phonological constraints (for details see Plag Reference Plag1999).

The following section addresses RQ2 and throws further light on these findings by examining whether, and to what extent, the role of coercion has changed over time in the interpretation of Categorized complex verbs.

4.2. RQ2: A diachronic view of creativity/coercion in Categorized complex verbs

Table 4 reports the breakdown across coercion types of the number of Categorized verb senses first attested in each century from the 1100s to the present day and beyond. As noted in section 3.2, these dates refer to the earliest attestation of complex verbs that convey the same sense in contemporary language (Laws’ 2023 corpus data) as they did in the original OED quote. Neologisms from 1994–2014 are excluded; these are discussed in section 4.3.

Table 4. Distribution of Categorized complex verbs first attested by century

Since -en is derived from Old English, complex verbs bearing this suffix have been attested in documents dating back to 950; loan verbs ending in -ize, -ify and -ate entered the language in the 1200–1300s, but it was not until the 1500s (1400 for -ify) that verbs were formed in English with these suffixes (Marchand Reference Marchand1969). The steep increase in the number of new complex verbs in the 1800s is due to a surge in the formation of science-related -ize and -ate derivatives in the nineteenth century (Plag Reference Plag1999).

The actual numbers for the period 2000–14 (the end-date of data collection for the BNC2014) are entered in the 2000 column. To predict the number likely to be generated in the whole of the twenty-first century, this value is scaled up (reported in brackets), based on evidence from the first fourteen years, e.g. the total number of Categorized verbs in table 4 is 6 for 2000–2100 and the adjusted value is 43 ((6/14)*100).

The values in table 4 are visualized in figure 6 as percentages. When interpreting the patterns in figure 6, it is useful to consult table 4, to take the size of the percentage denominator into account. Since the values for 2000 onwards are predicted, dashed lines are used on the plot.

Figure 6. Percentage of Categorized verbs involving coercion by first attestation date

The first notable aspect of figure 6 is that, based on the complex verbs occurring in the BNC corpora examined here, none of the three verb senses first attested during the 1100s, or the four attested in the 1200s involve coercion; in fact, it is not until the 1500s that Selection, and to a lesser extent Enrichment are involved in the interpretation of argument structure schemas of newly coined verbal derivatives;Footnote ⁵ after that time, these types of coercion appear to become a relatively stable feature of complex verb formation. As mentioned above, the predicted values for the twenty-first century are based on data for 2000–14; these predicted values suggest that No coercion and Enrichment are involved more frequently than Selection.

It appears that since 1500, when complex verbs with Latinate endings started to be coined in English, formation processes have primarily involved Selection and/or Enrichment. This pattern suggests that increased productivity, or rather Extended-Fixed creativity was facilitated by coercion, initially by the weaker mechanism Selection, followed by Enrichment, particularly from the 1700s. By contrast, Override, appears to be a predictable, but seldom-utilized productive process.

Selection can account for the expansion in scope of a semantic category through contextual adjustment of the predicate/Direct Object combination that instantiates the construction, e.g. (16)–(18). Although Selection and Enrichment both tend to occur predominantly in three-place argument schemas, Enrichment involves a more complex operation than Selection: that of reference transfer, e.g. (20)–(21). This stronger type of coercion requires the speaker/hearer to imply/infer non-realized lexical material in the interpretation of the complex verb. Thus, the increasing frequency of Enrichment over Selection reflects a greater tolerance of more elaborate underlying interpretations of complex verbs as time progresses.

These findings raise the question whether similar patterns occur with recently coined complex verbs, as posed by RQ3. Therefore, the next section focuses on neologisms.

4.3. RQ3: A creativity/coercion profile of recently coined complex verbs

Fifty-one neologisms were analysed: 30 from the BNC (Laws Reference Laws2023) and 21 from COCA. As mentioned in section 2.3.4, the BNC1994 neologism sanctionalize was classified as Uncategorized yet Analysable. Thus, table 5 presents the frequency with which Categorized neologisms displayed the four coercion types None to Override, together with related percentages based on the total of 50. The single Uncategorized Analysable neologism is included in the Unruly column.

Table 5. Distribution of coercive mechanisms in complex verb neologisms

Compared with conventional Categorized verbs (table 3), table 5 indicates a more even distribution of coercive mechanisms across types: the lower proportion of neologisms formed with No coercion results in higher proportions of Selection and Enrichment, although none of the pair-wise comparisons for overall coercion frequency (Coercion Frq) reached statistical significance. By contrast, unsurprisingly, the low incidence of Override (1) was reflected in significant differences compared with No coercion (LL=26.26, p<0.0001, BIC=21.64), Selection (LL=18.50, p<0.0001, BIC=13.88) and Enrichment (LL=23.66, p<0.0001, BIC=19.03).

When coercion frequencies in table 5 are compared with the equivalent values for Categorized verbs (table 3), it turns out that, although no significant differences were obtained for No coercion and Override, the incidence of Selection (LL=4.93, p<0.05, BIC=-1.46) and Enrichment (LL=7.10, p<0.01, BIC=0.71) in neologisms is proportionally greater than it is for Categorized verbs; however, effect sizes are weak. Nevertheless, these findings indicate that recently coined neologisms continue to involve more Selection and/or Enrichment than Categorized verbs have over time (figure 6).

A breakdown by semantic category and coercion type for each neologism is presented in table 6, which reveals that neologisms exhibit a wide range of interpretations; of the twenty semantic categories discussed by Laws (Reference Laws2023), only seven are not represented: the Inchoative-Ornative, Inchoative-Conformative, Imposative, Inchoative-Imposative, Confirmative, Acknowledging and Similative, all of which are relatively small categories. In fact, in general, the distribution of neologisms across categories is similar to that for Categorized verbs (figure 4 labels), with the two following exceptions.

Table 6. Semantic categorization of neologisms and coercion types

Key: B1994/B2014 = BNC1994/BNC2014; C1994/C2014 = COCA1994/COCA2014; N = No coercion; S = Selection; E = Enrichment; O = Override; U = Unruly coercion.

The Ornative sense is expressed considerably more frequently in neologisms (21/50, 42 per cent) than it is in conventional Categorized verbs (16 per cent, figure 4 labels), although in both cases Selection and/or Enrichment are often involved (14/21, 67 per cent and 77/108, 71 per cent, respectively). Thus, the Ornative sense, with or without coercion, is a well-entrenched schema that speakers readily apply to new contexts.

By contrast, one of the largest semantic classes of conventional Categorized verbs, the Causative, accounts for 27 per cent of all senses (figure 4 labels), whereas only 6 per cent (3/50) of neologisms express this meaning, all of which involve Selection and/or Enrichment. Thus, although the Causative sense is employed considerably less frequently in neologisms, coercion has become a regular feature.

Thus, in line with the diachronic path of conventional Categorized verb formation (figure 6), a greater degree of coercion is involved in generating recent neologisms. The coercive mechanisms of Selection and Enrichment, particularly the latter, appear to have become entrenched in speakers’ conceptualization of complex verbs. Section 4.4 addresses the final question relating to the characteristics of E-creativity in verb-forming suffixation.

4.4. RQ4: The nature and development of E-creativity in complex verb formation

So far, the analysis has dealt with Categorized complex verbs exhibiting F-creativity (No coercion) and those involving Selection and/or Enrichment or Override that fall on the Extended-Fixed portion of the creativity/coercion continuum (figure 3). It was proposed in section 2.3.4 that Analysable Uncategorized complex verbs display Unruly coercion, a characteristic that could be considered to reflect E-creativity.

Examples of the 23 Analysable Uncategorized verbs in Laws’ data (2023, 2024) mentioned in section 2.3.4 are unionize (make join a [union]), womanize (to pursue [women] excessively), first attested with these senses in 1874 and 1893, respectively (OED), and the BNC1994 neologism sanctionalize (to approve by means of [sanction]).

It was reported in section 4.2 that Selection and Enrichment became a stable feature of Categorized verbs around 1500 (figure 6). Based on the proposed nature of E-creativity in complex verbs, the question arises as to whether the frequency of Unruly coercion follows a similar pattern to that of Selection and Enrichment as time passes. Figure 7 presents the proportion of Analysable Uncategorized verbs involving Unruly coercion over time; percentage datapoints are annotated with raw frequency ratios, i.e. Analysable/all Uncategorized verbs first attested per 100 years. Predicted values for the twenty-first century were calculated in the same way as that described in section 4.2 for Categorized verbs. Lines joining datapoints for 1100 and 1200 are absent to avoid distorting potential trends in the graph due to low frequencies.

Figure 7. Percentage of Analysable Uncategorized verbs involving Unruly coercion by first attestation date

Interpretations of figure 7 can only be made with extreme caution, given the low type counts involved (just 3.5 per cent of the whole dataset), nevertheless, it is tempting to suggest that Unruly coercion has always been present in verb-forming suffixation, albeit as the exception, and that it appears to be increasing. Furthermore, by comparing figure 7 with figure 6, it could be tentatively inferred that E-creativity was relatively less frequent between 1400 and 1700 when Selection/Enhancement were becoming an established feature of Categorized verbs, but that an increase has occurred since then. The latter suggestion can only be verified by expanding the dataset to include written sources and language samples post-2014.

So, what could be the motivation for these instances of E-creativity? It is conjectured here that Analysable Uncategorized verbs are coined as a kind of ‘short-hand’ to convey an elaborate concept for which no established schema, or coerced extension of one, exists, thus, the suffix creates a verb that captures a unique, ‘unruly’, but succinct relationship between the argument roles and the base item. It is not that these formations ‘break the rules’, rather no rules are applied. It is worth noting that only 5 of the 23 Analysable Uncategorized complex verbs convey a single sense in the dataset (coded (1/1)Footnote ⁶ in Laws Reference Laws2024). Therefore, although verb-forming suffixation expresses a vast array of semantic categories, it appears that the language user requires an even greater set of interpretations; for example, unionize has four senses, three of which are Uncategorized. Thus, the mere attachment of a verb-forming suffix to a base can be sufficient to generate a meaningful complex verb that captures a unique sense specific to the context in which it is coined, and Unruly coercion appears to fulfil this requirement.