Researchers interested in lexical borrowing have long mainly restricted their attention to isolated loanwords, perhaps as an epiphenomenon of the traditional quest for sharp dividing lines between lexical borrowing and codeswitching and the ensuing debate on the status of single word switches (Poplack, 1980; Myers-Scotton, 2002; and see Zenner and Geeraerts, 2015). After all, the most prototypical instances of lexical borrowing involve exactly such loanwords, according to Matras (2009, p. 113): “the regular occurrence of a structurally integrated, single lexical item that is used as a default expression, often a designation for a unique referent or a grammatical marker, in a monolingual context.” However, when embracing the idea that there is a continuum from borrowing to codeswitching rather than a sharp dichotomy between both (see Matras, 2009; Zenner and Geeraerts, 2015; Zenner et al., 2019), understudied contact phenomena that are part of the linguistic reality positioned between the outer poles of the prototypical codeswitch and the prototypical loanword instead come to the fore. For instance, attention has been awarded recently to the way language users adapt and integrate larger semi-fixed chunks of source language material into their own language (Andersen, 2020a).

We consider both phraseology and construction grammar¹ to be useful theoretical frameworks for this endeavor. Phraseology research aims to study the form-meaning characteristics of word combinations (Cowie, 2001; Wray, 2002; Feyaerts, 2006), viz. of all structures including minimally two words (amongst others collocations, idiomatic expressions, phrasal verbs, slogans). Construction grammar, in turn, aims to identify and compare recurring linguistic form/meaning combinations on different levels of schematicity across the lexical, phonetic and grammatical domains of language use (Fillmore, 1988; Goldberg, 1995; Croft and Cruse, 2004; Steels, 2011; Boas, 2013). Both frameworks have been applied broadly and have hence each fragmented into separate subfields, making it fairly challenging if not impossible to arrive at a fit-for-all list of necessary and sufficient criteria to define and delineate the paradigms. What is clear, however, is that although both phraseology and construction grammar have largely developed independently of each other (see Gries, 2008; Ziem, 2018), they terminologically and conceptually share a number of properties. This is for instance true when taking the subframework of Cognitive Construction Grammar (Goldberg, 1995, p. 4; Boas, 2013) as point of departure, as this paper aims to do.

Where phraseology research aims to uncover a language's phrasicon, viz. the inventory of phrasemes, Cognitive Construction Grammar is concerned with the constructicon, viz. the structured inventory of constructs or phrase types that are captured by linguists descriptively as constructions. Both constructions and phrasemes are said to be (i) fairly fixed in terms of form and meaning, though allowing for variability in some of their elements (“open slots”, e.g., Oh my X)²; (ii) entrenched as units in the language users' minds; (iii) with degrees of entrenchment depending on frequency of exposure to the unit in language use; (iv) semantically non-decomposable, in the sense that a complex expression can have a meaning that is not attributable to the meanings that the subparts would have independently (Kay and Michaelis, 2011) (the phraseme kick the bucket is not about kicking a bucket, nor does they worked their butts off as an instantiation of the body part off-construction necessarily concern the agent's behind (Goldberg, 1995; Sawada, 2000; Granger, 2009); another example is the fully schematic double object or ditransitive construction V NP NP, which is associated with the meaning “transfer of possession”, despite it containing no lexical elements whatsoever; Goldberg, 1999, p. 199). Not surprisingly then, linguists have also relied on both paradigms to study the borrowability of semi-fixed chunks as grey zones in between prototypical lexical borrowing and prototypical code-switching. In this application, some pivotal differences between the paradigms can, however, be uncovered.

Pointing out phraseological borrowing as a largely unexplored area, Fiedler (2017) aims for an inventory of English-based phrasemes in German, classified according to formal parameters of the borrowed phrasemes, mainly contrasting phraseological patterns that are directly borrowed (nice try! in German), indirectly borrowed (e.g., der Elephant in Raum based on the elephant in the room) or hybrids characterized by partial substitution of English lexemes by German alternatives (e.g., den Turnaround schaffen “to manage the turnaround”) (see Andersen, 2020a,b for further cross-linguistic support). This distinction mirrors the opposition made by Matras (2009) between matter and pattern replication. Yet, the construction grammar emphasis on the various degrees of schematicity of constructions becomes crucial when aiming to study not just which phrases are borrowed form a given source language in what form, but also to describe how language users gradually adapt and change these semi-fixed source language chunks in the receptor language. It enables us to integrate, rather than separate, direct, and indirect borrowing, to simultaneously analyze pattern and matter replication, and to arrive at a more profound understanding of the way in which language change takes shape through individual usage occurrences in which language material from source and receptor language are integrated (Boas and Höder, 2018, 2021, and see Traugott and Trousdale, 2013).

Höder (2012) describes just how language users go about such integration (and see Dogruöz and Backus, 2009 for similar ideas earlier). Through interlingual identification, language users conflate similar constructions in two languages as belonging to a common “diaconstruction”. When instantiating the diaconstruction in specific usage events, users are left with a choice to lexicalize (parts of) the construction (the pattern) with linguistics elements (the matter) from either language. Whether source or receptor language material is used, may vary or change over time, with language users for instance gradually lexicalizing more slots in the receptor language. This contact-induced change can reflect or support more generally attested types of constructional change (Traugott and Trousdale, 2013; Coussé and Von Mengden, 2014), such as the occurrence of open slots in a previously fully fixed expression, semantic specialization or generalization of the construction as a whole or of its constituent parts.

The benefits of a (Diasystematic) Construction Grammar approach for the description of contact-induced variation and change have been illustrated by the papers in (Boas and Höder, 2018, 2021; and further see for instance Noël and Colleman, 2018). Meanwhile, as concerns change in (semi-)fixed chunks borrowed from English, Zenner et al. (2018) may serve as a first study. In this work, Zenner et al. analyzed semantic and formal similarities between the diaconstructional variants [(DET) ADJ_superlative N ever (Ptcp)] (e.g., beste zomer ever “best summer ever”) and [(DET) ADJ_superlative N ooit (Ptcp)] (e.g., beste zomer ooit “best summer ever”). Particularly, Zenner et al. (2018) aimed to link ongoing change in the existing Dutch ooit-construction to the incorporation of the English ever-counterpart in Dutch. The current paper focuses on a different question. It investigates the construction Pimp My Ride, which was borrowed from English into Dutch, and analyzes how its constituent elements have evolved in Dutch both within and outside of the target construction.

In a pilot study, Van de Velde and Zenner (2010) revealed the rapid emergence of the verb pimpen “to pimp, fancify” in Dutch following the introduction of Pimp My Ride as the name of a popular MTV series in 2005. Pimp, part of English vocabulary since the 1600s³, (1) is a noun referring to a person who controls prostitutes; (2) is a verb derived from the noun in the meanings “to act as a pimp” and “to prostitute someone”; (3) exhibits figurative meanings based on (1) and (2) such as he pimped himself out to the media. MTV (Music Television), an American cable channel that spawns numerous affiliated channels across the globe, introduced a then still innovative additional verbalization of the noun when launching the TV show Pimp My Ride in 2004⁴. In essence, the show involves the cosmetic makeover of a shabby car provided to the show by a participant. The title of the show refers to the request of the participant (imperative pimp) to, simplistically put, fancify their car (my ride) to the point where it looks like a car one would stereotypically link to a pimp. Although the show emerged in the socio-cultural context of the US, connecting to cultural ideologies surrounding pimps and ghetto-style, the series, both in the original and in various adapted versions, quickly found an international audience, carrying with it the new meaning of “to pimp”.

The choice made by Van de Velde and Zenner (2010) for this particular construction can be justified by pinpointing several reasons why it is actually fairly unexpected that the English phrase Pimp My Ride would instigate the introduction of a new verb, pimpen “to pimp, fancify” in the Dutch lexicon. First, the verb is introduced in Dutch via an international TV show, and so far the role of media in language change has generally been contested (Tagliamonte, 2014) and, in the context of anglicisms, understudied (Andersen, 2020a, p. 2). Second, its original fairly taboo meaning loaded with social stigma could hinder the spread of the more innocent meaning “to fancify”. Third, the imperative use of the noun pimp concerns a denominal verbalization that is not straightforward in Dutch. Finally, the verb pimpen is borrowed as part of a fixed expression functioning as a proper name. Hence, a process of constructional change could be evoked to explain the process of how the verb has become detached from the original proper name.

This latter point is precisely what Van de Velde and Zenner (2010) set out to study. Mining a self-collected fit-for-purpose Dutch corpus of newspaper articles published between 1998 (well before the first episode of Pimp My Ride in 2004) and 2009 (the time of analysis) for instances of pimp, the authors corroborated the fact that the “fancify” meaning of verbal pimp did not occur prior to the introduction of the MTV show Pimp My Ride in the Low Countries, in this way pinpointing T0 of the constructional change and sidestepping the actuation problem typical of variational studies (Weinreich et al., 1968). Second, the authors show how, even in the quality newspaper corpus, the new verb pimpen with its new “fancify” meaning spread quickly after the introduction of the show Pimp My Ride. Making a case for a construction grammatical approach to the change, they show how the verb pimpen was distilled from Pimp My Ride through a fast process of semantic and formal “deconstructionalization” from the underlying constructional template [pimp POSS N]. By “deconstructionalization” we mean the process of a gradual or stepwise blurring of the formal and semantic contours of the constructional template, leading to a less crisply delineated, recognizable chunk. Deconstructionalization is in that sense used here merely as a descriptive label, not as a bid to engage in theoretical and technical terminological debates. We prefer the term “deconstructionalization” over “schematization”, as the latter is mainly used to refer to the shift from lexically fixed slots to open slots, whereas we also intend to look at insertion at the syntagmatic level, and expansion of morphological productivity.

Semantically, the verb pimpen was increasingly used for other entities than rides and vehicles, even including animates [see (1)]⁵. Formally, a transition was noted from the fixed proper name Pimp My Ride via the semi-fixed target construction [pimp POSS N] [see (2)] to finite uses of the Dutch verb pimpen detached from the construction [see (1)].

In conclusion, by the end point of the measurements of the pilot study, viz. a mere 5 years after the first broadcast of Pimp My Ride in the Low Countries, the (in Dutch) new verb pimpen seemed to have established itself in Dutch, being used with a range of pimpable entities, both in the constructional template [pimp POSS N] derived from the target expression pimp My ride and more freely as a finite verb. This finding is further supported by the inclusion of the verb in dictionaries as of 2006 and the lack of objection to its use in normative reference works⁶.

This impressive trajectory in the Low Countries from English-origin TV show to canonized dictionary entry in less than 3 years might hold implications for our understanding of the role of globalized media for local language change. However, as the presumed lack of longevity is a traditional argument used against media-induced variation and change (Labov, 2001, p. 228), a follow-up study is required. Particularly, we aim to understand what has happened with pimp in Dutch following the 2004 introduction of Pimp My Ride, verifying whether the change is long-lived, and to what extent differences can be found in the trajectories of free use of the new verb pimpen “to pimp” and the [pimp POSS N]-construction that is closer to and hence in part still resonates the verb's globalized media origin.

This study aims to uncover deconstructionalization patterns in the use of the verb pimp in Dutch following the first wave of deconstructionalization as described in Van de Velde and Zenner (2010). A contrast is made between the trajectory of pimp-cases in the constructional template [pimp POSS N] and of occurrences of verbal pimp outside of this original template:

For our study, we made use of a Twitter corpus. Following Androutsopoulos (2010, p. 204), we consider computer-mediated communication (CMC) as optimally suitable when aiming to study how “globalization is not a unidirectional process by which linguistic or cultural elements are diffused and uncritically adopted” but rather a process of local integration. Computer-mediated communication will be a primary source for uncovering the detachment of pimp from its original template [pimp POSS N]. Practical considerations further support the specific choice for Twitter: it allows us to arrive at a diachronically sliceable corpus of sufficient size. Needless to say, caution is needed when interpreting our findings, in the sense that we cannot simply extrapolate the trajectories of use of pimp to other genres or media.

We gathered a dataset of pimp-tweets automatically identified as written in Dutch for the period from 2007 to 2020, viz. from the launch of the platform in the Low Countries to the time of data collection. This means that we started our retrieval after the initial wave of deconstructionalization of Pimp My Ride in 2004 (Van de Velde and Zenner, 2010). Tweets containing conjugated and derivative forms of pimp were gathered through Python, also querying spelling deviations expected to occur in the conjugation of borrowed verb forms in Dutch or in the general CMC context (e.g., gepimpt vs. gepimped)⁷. All 163,046 collected tweets⁸ were lemmatized and POS-tagged with Frog (Van den Bosch et al., 2007). As the quality of POS-tagging is hampered by the multilingual contexts in which many of the pimp-forms occur and by the graphemic instability typical of the informality of CMC, we proceeded to manual coding of a sample of tweets. Specifically, a random selection of 500 pimp-tweets (or less, if no 500 were available) was selected for each of the 13 years under scrutiny. As such, a total of 6,381 pimp-examples was manually coded for their semantic and formal properties.

A first step in the coding procedure consisted of excluding noise from the dataset. A total of 1,820 tweets (28.52%) were excluded from further scrutiny for one of the following reasons: (1) the matrix language of the tweet was not Dutch, but rather English, Afrikaans, German, … (N = 754 of N = 6,381); (2) the target semantics of pimp were not “to fancify”, but rather pertained to the original prostitute controlling meaning, or the tweet was too short to establish the meaning of pimp (N = 1,031 of the remaining N = 5,627); (4) the verb pimp was conjugated following English derivational rules (e.g., past participle pimped rather than gepimpt) (N = 35 from the remaining N = 4,596)⁹.

For the N = 4,561 observations left after noise removal, we first identified the observation type, contrasting instances within the construction, which include both tokens of the original fixed expression Pimp My Ride [N = 254; 5.57%, see (3)] and tokens of the derived constructional template [pimp POSS N] that are not Pimp My Ride [N = 711; 15.59%, see (4)], with instances outside of the construction, viz. free occurrences of the verb pimp [N = 3,596; 78.84%, see (5)]. Table 1 and Figure 1 summarize the tokens per observation type per year, supporting the finding that the first wave of deconstructionalization happened soon after the first introduction of Pimp My Ride in 2004: the “free” use of pimp is the most frequent throughout the period under investigation. Additionally, we see a revival of original Pimp My Ride cases in 2010–2013. This probably relates to a rerun of the show on TV.

Next, we proceeded to coding the formal and semantic characteristics for both groups further. Table 2 summarizes the tokens per year of the construction [pimp POSS N], classified according to the parameters of RQ1. This table was composed as follows. Following Table 1 above, we first split off occurrences of the original fixed proper name Pimp My Ride, which always have a 1SG possessive pronoun, ride as pimpable entity and English as language for the three slots. For the other [pimp POSS N] cases, four parameters are included. First, we indicate whether or not an English first person singular form is used as in the original construction [see my in (3) vs. your in (6) or je “your” in (8)]. For the N-slot, we keep track of the semantics by contrasting pimped vehicles (cars, caravans, motorcycles) with other types of entities [see car in (7) vs. kussensloop “pillowcase” in (6)]. Further, we keep track of the language used to instantiate N, contrasting English slots with Dutch slots [see coffee in (4) vs. autoband “car tyre” in (8)]¹¹. Hybrid forms such as feestoutfit “party outfit” are considered English. Unclear cases and proper nouns are awarded NA (N = 21, marked red in Table 2). Finally, we keep track of lexical elements intruding in the [pimp POSS N]-construction [see nu “now” in (8)].

The “free” pimp tokens do not follow the [pimp POSS N]-template, calling for another set of formal parameters that indicate (even) further deconstructionalization. The variables and token counts can be found in Table 3. For the semantics, we now resort to a ternary classification, contrasting pimped vehicles, other non-animate pimped entities and animate pimped entities [see (9), (11), and (10)]. NAs are awarded to instances where no pimped entity is specified (N = 106).

For the formal classification of tokens, we adopt a verbal and a derivational morphological perspective. In terms of verbal morphology, we contrast imperatives [see (11)], infinitives [see (5)], (adjectival use of) participles [see (9)] and finite uses [see (10)]. For derivational morphology, we keep track of pimp's productivity by checking for derivational morphemes or phrasal extension [see op “up” in (9)].

To arrive at a “holistic” picture of the deconstructionalization process of the lexical expression pimp my ride both within and outside the template [pimp POSS N], we follow a quantitative procedure applied earlier in studies by Van de Velde (2009, p. 334–339), De Smet and Van de Velde (2013), and Petré and Van de Velde (2018). The idea is that an observation collects “points” for each dimension of constructional change that plays a role in the departure of the original construction. To give an example: a point is awarded if we observe pimp your ride instead of pimp my ride, and yet another if the pimped entity is not ride but e.g., pimp your laptop. In a sense, the number of points collected can be seen as a distance value from the original construction¹².

The total score is then used as the response variable in a negative binomial regression, with the year of attestation as the predictor. If the predictor is significant, we can assume there to be a robust trend over time. Before we have a look at the results, we will first detail the scoring procedure, which takes the information from the manual coding as input. We made a distinction between the original construction [pimp POSS N] and the subsequent “free” use of the verb pimp.

For the original construction (RQ1), the following system has been applied:

As such, the maximal number of points an observation of the form [pimp POSS N] can gather is 4.

For the verb pimp in its “free” use, outside of the constructional template (RQ2), the following scoring system has been applied:

The maximal number of points for the “free” construction is 6 points.

For the use of the [pimp POSS N] construction, the negative binomial model (with the natural logarithm as the link function) indicated a trend over time. Diachronically, the distance from the original construction as measured by the four-point scoring system increases significantly [β = 0.07 (on the log scale), p < 0.001]. This is visually represented in the effect plot in Figure 2. Next, Figure 3 verifies to what extent we still find a significant deconstructionalization trend when taking the original lexical construction, viz. all occurrences of the lexically fixed Pimp My Ride, out of the equation. As can be seen in Figure 3, the upward trend remains, but loses some of its oomph and its significance [β = 0.01 (log scale), p = 0.156].

“Free-roaming” pimp, i.e., the occurrence of pimp outside of the [pimp POSS N] template, does not show a trend over time. The effect of the year of attestation is not significant [β = −0.004 (log scale), p = 0.111]. Indeed, the effect plots in Figure 4 shows that the line is flat. Upon closer inspection, however, it seems that we would be remiss to assume that everything remains the same. Over time, the average score does remain more or less stable, but the range widens. Indeed, there is an increase over time in the standard deviation (Pearson's correlation 0.83, p < 0.001), see Figure 5.

What we have, then, is an increase in the higher regions of the scores, but this increase is counterbalanced by a simultaneous upsurge in the more basic use of the construction. This echoes a finding from Zenner et al. (2018), who noted that the use of a new construction can boost the frequency of an older cognate construction, which they call the “a rising tide lifts all the boats” phenomenon.