The sociopragmatic parameters steering the reported selection of Anglicisms or their Dutch alternatives

1 Introduction

Language variation and change induced by contact with English has become a prominent research topic within linguistics over the past decades. This holds for both intense contact situations, where English is a dominant language because of immigration or colonization (cf. Wolf and Polzenhagen 2009), and for weak contact situations, where English does not have official status, and contact with the language occurs more indirectly, predominantly through the media (cf. Androutsopoulos 2012). In the latter case, scholars have mainly focused on lexical borrowing, as that is the most common and apparent result of language contact (Thomason 2001: 69).

Anglicism research, which foregrounds lexical borrowing from English, has historically typically been conducted from a systemic and structural perspective. Most attention so far has been paid to the position of the loanwords in the structure of the recipient language (see Zenner and Kristiansen 2014: 1), resulting in descriptions of the different morphophonological and orthographic adaptation processes (Núñez Nogueroles 2017; Saugera 2017) or the establishment of a borrowability hierarchy comparing different word classes (Czech-Rogoyska and Krawiec 2018). More recently, however, two paradigm shifts in Anglicism research can be remarked: from a systemic to a more sociopragmatic and usage-based perspective (see, among others, Andersen 2014; Ilić 2017; Onysko and Winter-Froemel 2011; Onysko et al. 2013; Peterson 2017; Peterson and Vaattovaara 2014; Winter-Froemel 2011: 295–319, 2017), and, as a result, from a semasiological to an onomasiological point of view (Soares da Silva 2014; Zenner and Kristiansen 2014; Zenner et al. 2012).

The first shift “implies a reorientation of its locus from the borrowed lexemes per se, to how the use of borrowed items is constrained by cultural, social or cognitive factors” (Andersen et al. 2017: 71). This means that researchers now look at lexical borrowing as an instrument of local identity construction, speaker evaluation communication, and as manifesting sociocultural norms and values, or in short: as socially meaningful (Andersen 2014; Babel 2016; Peterson and Beers Fägersten 2018; Zenner et al. 2019). The second shift refers to taking the concept expressed by the Anglicism, rather than the lexical borrowing itself, as the point of departure, suggesting that attention is paid to the naming instead of the meaning, and as such to the selection made by language users between the loanword and alternative lexicalizations (see for instance Winter-Froemel [2008] who gives an overview of criteria for judging between alternative lexical strategies in contact situations, and Serigos [2017] and Winter-Froemel [2018] who show that Anglicisms are often semantically more specific than their Spanish and Italian alternatives using a concept-based approach).

In adopting these sociopragmatic perspectives on lexical variation induced by contact with English, most researchers have concentrated on the production of Anglicisms in various contexts and genres, while fairly exclusively relying on corpus-based methods (Balteiro 2011; Burmasova 2010; Yang 1990). Experimental studies, targeting language users’ perception and (social) evaluation of Anglicisms, are as yet far more scarce, with studies combining information from corpus data and experimental data virtually absent altogether (though see Sandøy [2014] as an exception). Nevertheless, a proper sociopragmatic account of lexical borrowing that enhances our understanding of the lexical choices language users make between loanwords and heritage alternatives, is contingent on a combination of insights from language production, perception and (social) evaluation (Backus 2020).

Against this background, the objectives of the present study are twofold. First, we aim to combine both the sociopragmatic and the onomasiological shift by investigating the under-explored issue of lexical choice between loanwords and heritage alternatives. Second, in selecting the appropriate methodology to obtain the first objective, this study targets to complement the strongly corpus-based orientation of Anglicism research by relying on an experimental design, though not without bearing the fruit of existing corpus-based Anglicism studies. Specifically, we investigate the sociopragmatic parameters steering the lexical preference of Belgian Dutch and Netherlandic Dutch speakers for English lexemes or Dutch alternatives through a forced-choice experiment that is designed based on the corpus study of Zenner et al. (2012).

By focusing on English loanwords in the Low Countries, this article presents Anglicism research in a weak contact setting. While most speakers of Dutch claim to be fairly bilingual according to the Eurobarometer of 2012 (European Commission 2012), English does not have official status in either country. Consequently, contact with English happens predominantly indirectly via marketing, television, movies, music, and other mass and social media (De Decker and Vandekerckhove 2013; Zenner et al. 2015). The Low Countries are additionally an interesting case study given the pluricentric nature of the Dutch language (Clyne 1992). It is sometimes argued that the Flemish have a different relationship with English than the Dutch because of Flanders’ history and sociolinguistic background (see e.g., Geeraerts and Grondelaers 2000; van der Sijs 1996; Zenner et al. 2012, 2015). Whereas the Netherlands developed a Dutch standard language from the seventeenth century onwards, the social elite and the ruling class in Flanders preferred French as their overt language of prestige. As a consequence, Geeraerts and Grondelaers (2000: 58) argue that “the struggle for recognition of Dutch as the official language in Belgium, often materialized as a competition with the French standard”, culminated in the rejection of French interference, and, even possibly foreign interference altogether (Vandenbussche et al. [2005: 46], and see also Cajot [2004]). While this has not yet been corroborated empirically to our knowledge, this study might shed light on the possible impact of the different linguistic histories of the Low Countries on self-reported preference toward Anglicisms.

The paper is structured as follows: first, we provide an overview of existing works on the perception and (social) evaluation of Anglicisms in Section 2. In the third section, we introduce the research questions and hypotheses. Section 4 proceeds to the design of our study, where the materials and stimuli of our forced-choice experiment will be presented and the sample will be described. Next, the results are explained in Section 5, followed by a discussion and conclusion in Section 6.

2 The perception and (social) evaluation of Anglicisms

As mentioned above, lexical borrowing from English has mainly been documented through corpus-based studies, with a clear focus on production (see e.g., Andersen and Graedler [2020] for English borrowings in Norwegian; Balteiro [2011] for sports Anglicisms in Spanish; De Decker and Vandekerckhove [2013] for the use of English in Flemish youth speak; Garcia-Yeste [2013] for English loans in Swedish advertising; Martin [2008] for the presence of English in French advertising; Schaefer [2019] for English in German, and many others). Experimental designs targeting perception and (social) evaluation of Anglicisms are less frequent, but not entirely absent.

Van Meurs et al. (2014), for instance, designed an association task to study to what extent speakers perceive Anglicisms to have different meanings from their Dutch alternatives. To study the social meaning of English swearwords in Finnish, Vaattovaara and Peterson (2019) modified the classic matched guise technique from Lambert et al. (1960)’s Speaker Evaluation Paradigm by including availability and acceptability ratings. The study of Sandøy (2014) forms a notable exception as well, as it investigates the (social) evaluation of Anglicisms in the Nordic languages by combining newspaper data, interviews, and a matched/verbal guise test.

Whether a speaker has a positive or negative reaction to language variation induced by contact with English depends on various factors, social, pragmatic and linguistic. When investigating the attitudes of Nordic speakers toward English influence, for instance, Thøgersen (2004) discovered that positive attitudes toward Anglicisms correlate with social class. The higher the degree of formal education, the more positive Nordic speakers are oriented toward English influx (Thøgersen 2004: 35).^[1] Devos (2018) on the other hand, found age to have a significant impact: Flemish participants between 15 and 18 years old accepted the use of English in newscasts more than participants between 60 and 75 years old. A Finnish study then, conducted by Sánchez and Tuomainen (2014), showed Finnish women to be more appreciative of Anglicisms than Finnish men, conforming to the “gender paradox”, which states that women often initiate linguistic change yet are more sensitive to linguistic norms (Labov 2001). Besides typical social predictors, context can play a crucial role in shaping attitudes as well (see Garrett [2010] for a general appraisal). An investigation on the acquisition of loanword prestige – via a matched guise method – discovered that Flemish primary school children prefer a cartoon with English loanwords slightly more than a cartoon with only Dutch words (Zenner et al. 2020). While this does not mean that children find English prestigious overall, it does show that in the specific framework of a cartoon, the use of English words is evaluated positively.

Not only in cartoons, but also in sports (cf. Balteiro 2011), job advertisements and commercials (cf. van Meurs et al. 2007), the copious use of English is often explained by referring to its prestige value. It is argued that the English language is associated with modernity and globalization and as such offers a certain appeal for a product (Cheshire and Moser 1994; Martin 2007; Piller 2001). This proposition appears to be confirmed by the research of Ager (1999), who analyzed the attitudes toward English of French native speakers using the results of the SOFRES poll of 1994.^[2] 41% of the respondents described Anglicisms as “modern,” 30% as “useful,” and 19% as “amusing.” A follow-up study by Walsh (2015), however, indicated that, although the participants displayed a generally positive stance toward English lexemes in French, over two-thirds agreed that the use of Anglicisms feels superfluous in product packaging (68%), advertisements (69%) and television/radio (61%). Similar results were found by van Meurs et al. (2017), who studied how English loanwords were perceived by Dutch and Flemish students in job titles. They concluded that English borrowings were largely reacted positively to, but both nationalities preferred the Dutch job titles over the English alternatives nonetheless. These findings support the more general conclusion that English loanwords are appreciated on the one hand because they seem to carry prestige, especially when there is no native alternative, but are rejected on the other because they feel like a threat to the native language, precisely because there is a native equivalent (Sánchez and Tuomainen 2014).

When looking at the existing work on the perception and (social) evaluation of Anglicisms, three methodological oppositions can be made. First, where some of these existing studies rely on direct attitude measurement such as surveys (e.g., Thøgersen 2004), interviews (e.g., Sánchez and Tuomainen 2014), and forced-choice tests (e.g., Walsh 2015), others rely on indirect techniques, typically the matched-guise technique from Lambert et al. (1960)’s Speaker Evaluation Paradigm (e.g., Vaattovaara and Peterson [2019]; Zenner et al. [2020]; or see Sandøy [2014] who relies on the verbal guise technique). These indirect procedures are often preferred over direct methods as they allow an assessment of covert, unconscious attitudes: the participants do not know that their language evaluation is being investigated. While the matched guise technique is indeed a very popular tool for language attitude research, it has been a target of critique as well. Besides the fact that it is hard to find a speaker who can speak multiple varieties naturally, one could also question whether the goal is to keep the informants unaware of one speaker using different language varieties or to keep the informants unaware of evaluating language variation (Pharao and Kristiansen 2019). In that regard, direct acceptability judgments can be rather useful as participants are consciously offering their language evaluations, although one has to keep in mind that these might be steered by a need for being socially or politically correct and that they might not match the actual usage of the participants.

Second, from a usage-based perspective, we can discern between “artificial” and more “naturalistic” research on the perception and (social) evaluation of lexical borrowing. Studies where the participant has to read or listen to a fragment (see e.g., Gerritsen et al. 2000) are generally considered to have a more natural feel because a “real-life” sociopragmatic context is available. This is in opposition to trial experiments, such as a forced-choice task, where stimuli are presented in isolation most of the time, giving the study a more artificial feel (see e.g., Rothe 2014). Nevertheless, as Rosseel et al. (2019) rightfully argue, contextual features need to be included in experimental designs either way, as context plays a crucial role in shaping language evaluations and perceptions as we have discussed above. Hence, this study will attempt to incorporate the context of use into its experimental design.

Third, we find that where some scholars focus on the perception and (social) evaluation of particular Anglicisms, others adopt a rather broad approach aggregating over different Anglicisms. For instance, to study the social meaning of English swear words, Vaattovaara and Peterson (2019) solely focused on the perception of the English loanword shit and the Finnish heritage form paska. They designed a matched-guise technique with six utterances containing either shit or paska, all with the same semantic meaning, but spoken in different varieties. Van Meurs et al. (2014) on the other hand, presented Dutch native speakers with 30 English – Dutch word pairs used in job advertisements to investigate whether loanwords receive different meanings than their native equivalents. If we want to gain more insight into the lexical choices of speakers, however, it is necessary to find a golden mean between specificity and aggregation: on the one hand, enough stimuli should be included to allow conclusions transcending the individual lexeme, but on the other hand, stimuli have to be carefully chosen to unveil the factors impacting lexical choice. For this study specifically, this resulted in 12 Anglicisms that are equally spread over different categories according to the different linguistic variables that are under scrutiny (see Section 4.2.1).

Thus, to enhance our understanding of language users’ in situ selection of loanwords over native alternatives, we propose a forced-choice experiment based on and linked up with corpus data, in which Dutch-speaking respondents are asked to select the most appropriate word from several word pairs consisting of an English loanword and a Dutch alternative, according to a specified context. This way, we can pay attention to both the social and the pragmatic effects steering that choice. The specific research questions and hypotheses underlying the experiment are presented in the next section.

3 Research questions and hypotheses

The main research question underlying this study is: “Which sociopragmatic factors help explain language users’ reported preference for English loanwords or their Dutch alternatives?”. To tackle this question, we presented a sample of over 1,000 Belgian Dutch and Netherlandic Dutch respondents with several experimental trials in which they were asked to select which of two words they find most appropriate in one of two speech contexts (a quality newspaper article or a conversation with friends): an English loanword or a Dutch alternative. As will be explained in Section 4, the experimental design is set up in such a way that the impact of three sets of sociopragmatic factors on the lexical choice made by the respondents can be determined, viz. (i) the community-based usage patterns for the borrowed and heritage form under scrutiny as established in previous corpus-based research (see Zenner et al. 2012); (ii) the socio-demographic characteristics of the respondents participating in the experiment; (iii) the speech context offered as part of the experimental trials. As such, our main research question can be subdivided into three subquestions.

For the first subquestion, we aim to inquire whether the proportions of use of the English loanwords vis-à-vis the offered Dutch alternatives as established in previous corpus-based research relying on mass media sources help explain the lexical choices made by the respondents in our study. Following the usage-based hypothesis (Goldberg 2006; Tomasello 2005) which states that our mental lexicon is shaped by individual usage events, we can expect the corpus-based proportions to effectively correlate moderately to strongly with the choices respondents make in our forced-choice experiment. Although the link between frequency of use and entrenchment should not be overstated (Schmid 2010), we do anticipate language users to find the lexeme that occurs most frequently to be the most appropriate in use. The association between corpus frequencies and self-reported preference in the experiment might however not be absolute, given that purist reflexes might steer the respondents away from borrowed lexical items, even when these are the most frequent option (see e.g., Schmidt and Diemer 2015; Spitzmüller 2007; Walsh 2014).

Four socio-demographic characteristics are considered in this study, viz. the respondents’ age, gender, region, and self-reported attitude toward the influence of the English language on Dutch. As concerns Age (RQ 2.1), we anticipate that younger respondents will be most likely to select the English loanword over the Dutch alternative and that older respondents will be least likely to select the English loanwords over the Dutch alternatives. We base this hypothesis on claims in previous research that the use of English words and phrases is a youth language phenomenon (see De Decker and Vandekerckhove [2013] and Devos [2018] for Dutch; Garley [2010] and Pennycook [2003] on English as the language of hip hop, and also more broadly Piller [2001]) and on research revealing more purist reflexes in older respondents (e.g., Walsh [2015] for French). As concerns Gender (RQ 2.2), previous research shows conflicting results. Where González Cruz and Rodriguez Medina (2011) do not find any notable gender differences, Zenner et al. (2015) reveal higher use of English by male participants on a Dutch reality TV show (later shown as indexing a strong heterosexual male identity, see Zenner et al. [2015]). Sánchez and Tuomainen (2014) on the other hand show women as more appreciative of Anglicisms than men. This is further supported by Balteiro (2014), who highlights the important role of English lexemes in fashion discourse, which is still typically female oriented. Hence, no clear hypothesis for gender effects can be put forward, but the results from previous studies do point out that more research on the potentially gendered nature of the use of English loanwords is needed. To investigate Region effects (RQ 2.3), we contrast Belgian Dutch and Netherlandic Dutch respondents, expecting a stronger purist anti-English reflex for the Belgian Dutch participants. Although to date no synchronic empirical support has been found to support this hypothesis (despite attempts in, among others, Zenner et al. [2012, 2015]), we anticipate that the standardization history of Dutch in Flanders (the Northern part of Belgium) as described in Section 1, will leave its trails in the language attitudes of the Belgian Dutch respondents (see Cajot [2004] on purist reflexes in Flanders, which in his view materialize especially in complaints about French loanwords in Netherlandic Dutch, rather than in the actual language usage in Flanders). It is, however, also possible that the described puristic reflexes mainly pertain to French, given Flanders’ linguistic history, and less so to English. Finally, we include the respondents’ Self-reported attitude toward the influence of English on Dutch (RQ 2.4), verifying to what extent a negative attitude toward English leads to a stronger penalization of English loanwords.

Previous research has highlighted that the evaluation of particular Anglicisms may vary according to the contexts in which the items are used, that is, speakers may rate the adequacy of an Anglicism compared to its heritage alternative differently depending on contextual features (Zenner et al. 2020). This observation can be linked to the variation linguistics framework developed by Koch and Oesterreicher (2001, 2012, where communicative immediacy and communicative distance are introduced as a basic dimension of linguistic variation. Scenarios of immediacy and distance are characterized by parameters such as the dialogic versus monologic character of communication, the familiarity versus unfamiliarity of the communication partners, face-to-face interaction versus communication across temporal and spatial distance, private versus public communication, spontaneous communication marked by expressivity and affectivity versus planned communication on a previously fixed subject, etc. Moreover, communicative immediacy is typically realized in oral settings, whereas distance communication is traditionally linked to written realization. A general crosslinguistic observation is that for scenarios of communicative distance, there is a stronger orientation to the standard language. In our view, this also implies a more conservative attitude and an avoidance of Anglicisms, as purism is a very important part of standardization (cf. Langer and Nesse 2012: 612; Van der Sijs 1999: 111).

In addition to that, immediacy and distance are linked to other dimensions of lectal variation, e.g., informal peer group conversations are perceived as belonging to communicative immediacy in a broad sense, whereas formal style, found in newspaper articles, for instance, is linked to communicative distance. To investigate the potential impact of communicative immediacy and distance on the evaluation of loanwords or their Dutch alternatives, we, therefore, operationalize this distinction by presenting the experimental trials (see Section 4) in two different usage contexts: a quality newspaper article or a conversation between friends. Whereas the first context exhibits typical features of communicative distance such as the monological, premeditated and public character of the utterance, etc., the informal conversation with friends represents one of the most typical scenarios of communicative immediacy (private, dialogic, spontaneous, face-to-face communication, etc.). Given the associations of English with modernity discussed above (see Piller 2001) and the general perceptions of communicative immediacy and distance, we anticipate that respondents will be more likely to select the English loanword over the Dutch alternative when asked to imagine a conversation between friends than when asked to imagine a newspaper article.

4 Methodology

To tackle the research questions, we use data from a forced-choice experiment that was embedded in an online survey. In this section, we outline the design of our questionnaire (Section 4.1), the stimuli and materials used (Section 4.2), the procedure of the study (Section 4.3), the participants (Section 4.4), and the analyses (Section 4.5).

5 Results

Before addressing the results of the mixed-effect regression model, we report the results of the two sanity checks in Section 5.1, complemented with a check of the overall variation for the target items. Based on these sanity checks, two of the 12 target stimuli will be discarded from further analyses that will be discussed in Section 5.2.

5.1 Verifying the guiding assumptions in stimuli selection

Two core assumptions guided our selection of target stimuli (the alternation pairs of English lexemes and Dutch alternatives), viz. (1) that the two words were considered semantic equivalents; (2) that the English words were perceived as English loanwords, or at least as not equally “Dutch” as the heritage terms. The second part of our survey verified to what extent these assumptions hold for our respondent sample (see Section 4.2.2 for more details).

Figure 1 reports the results of the first sanity check, where respondents were asked to rate the semantic equivalence of word pairs on a five-point Likert scale (1 = no similar meaning/not felt to be synonymous, 5 = similar meaning), including not only the 12 target word pairs but also six fillers pairs that were meant to benchmark the rating scale (architect_bouwvakker “architect–construction worker,” basketballer_voetballer “basketball player–football player,” professor_onderwijzer “professor–teacher,” landbouwer_boer “farmer–peasant,” arts_dokter “physician–doctor,” auteur_schrijver “author–writer,” see Section 4.2.2). First, the position of the filler items in Figure 1 (white boxplots) reveals a well-benchmarked scale. Second, the overall results for the target items (grey boxplots) are reassuring. Only three word pairs have a mean synonymy rating under 4 on the 5-point scale, viz. the word pairs covergirl – hoezenpoes (M 3.20, SD 1.34), jockey – rijknecht (M 3.73, SD 1.25), and marketeer – marktstrateeg (M 3.77, SD 1.11). Particularly the boxplot for covergirl – hoezenpoes reveals a problematic sanity check, forcing the removal of the item from further scrutiny.

Figure 1:

Sanity check 1: loanword status score (1 = not perceived as similar in meaning, 5 = perceived as similar in meaning); target items in grey, filler items in white.

The results for the second sanity check, which gauges the perceived loanword status of the English items (1 = felt to be an English loanword, 5 = felt to be part of the Dutch lexicon), are reported in Figure 2. The filler items that are meant to help benchmark the scale are found in white. The boxplots for the target items are found in grey. Again, the filler items reveal an intuitive scale, with foreign words such as police officer or secretary of state receiving very low scores, as opposed to entrenched loanwords included in Dutch dictionaries such as ballerina or architect, and Dutch items such as leerkracht “teacher” or verkoopster “seller,” which receive high scores. The results for the 12 English target loanwords are in this case also reassuring: none of the 12 items has a mean foreignness score under 2, and each of the 12 items has a median foreignness score of 1.

Figure 2:

Sanity check 2: loanword status score (1 = perceived as foreign, 5 = perceived as native); target items in grey, filler items in white.

As a final check, we had a look at the answer patterns for each of the 12 stimuli. Before addressing the core research question on the variation in lexical selection of Anglicisms and heritage alternatives, we wanted to make sure that there is, as we assumed, effectively notable variation in the answer patterns for the loanwords.

As discussed in Section 4.2.1, we deliberately included some alternation pairs of which the corpus-based success of the Anglicism was very high (e.g., covergirl – hoezenpoes). We included these items to gauge whether respondents’ purist reflex might override the usage-based nature of lexical choice, yet we herewith introduce the risk that a sizeable proportion of respondents might not know the Dutch alternative and decide to refrain from lexical choice by selecting the wildcard “I don’t know (one of) these words”. Table 4 reveals the number of times respondents selected this wildcard option for each of the words, complemented with information on the number of respondents who selected the English or Dutch word per trial. Finally and crucially, the table also provides the proportion of respondents who selected the English item when excluding respondents who withdrew from selection through the wildcard, revealing the amount of variation attested in the answer patterns. The table reveals that two items are not fit for further analyses: both jockey_rijknecht en covergirl_hoezenpoes are the only items (1) where more than 10% of respondents selected the wildcard option “I don’t know (one of) these words”; (2) where barely any variation in answer patterns is left when excluding these wildcard answers, with more than 95% of respondents selecting the English term for both alternation pairs. For covergirl – hoezenpoes, this adds to the already problematic results from the first sanity check.

Table 4:

Variation in answer structure per trial.

Target trial	N wildcard	N English	N Dutch	(N English)/(N English + N Dutch)
jockey – rijknecht	147	1,075	6	0.994
covergirl – hoezenpoes	123	1,088	17	0.985
quizmaster – quizleider	1	1,164	63	0.949
ghostwriter – spookschrijver	57	1,109	62	0.947
songwriter – liedschrijver	2	1,149	77	0.937
workaholic – werkverslaafde	0	1,104	124	0.899
marketeer – marktstrateeg	86	910	232	0.797
pitcher – werper	121	853	254	0.771
keeper – doelman	0	921	307	0.750
catcher – vanger	121	761	346	0.687
babysitter – kinderopvang	0	829	399	0.675
businessman – zakenman	0	197	1,031	0.160

To conclude our verification of the guiding assumptions underlying the stimuli selection, two-word pairs are excluded from further analyses: jockey – rijknecht and covergirl – hoezenpoes. This leaves us with 11,892 forced choices for 10 target alternation pairs obtained from 1,228 respondents. In total, the English term was selected in 8,997 trials (75.7%), the Dutch term was selected in 2,895 trials (24.3%). In the next section, we attempt to increase our understanding of what guided respondents to make their selection.

5.2 Addressing the research questions

As explained in Section 4.5 above, we analyze the data using mixed-effects logistic regression analyses. As concerns the response variable, we estimate the chance of selecting the Dutch item (1) over the English item (0). Although this might seem conceptually counterintuitive, it is technically the most robust way to fit the model, considering the reasonable skew in the response (75.7% English, 24.3% Dutch). In terms of the model-building strategy, we started with a null model including only random intercepts for the variables “Respondent ID” and “Stimulus (word pair)”. The fixed effects were included through a manual forward stepwise selection relying on AIC, confirmed through Chi-square tests that compare the null model and the fitted model. The best-fitting model includes four of our six variables, viz. “Corpus success of the Anglicism” (see Table 1),^[5] “Context” (newspaper article or conversation with friends), “Age of the respondent” (group 1: 18–29; group 2: 30–50; group 3: 51–70), and “Self-reported attitude respondent” (bothered vs. not bothered/neutral).

In terms of fit, the model reported in Table 5 (random effects) and Table 6 (fixed effects) significantly reduces the null model’s AIC (from 9,958.580 to 9,749.5, for 11,884 residual degrees of freedom in the fitted model). The relative contribution of the different fixed effects to explaining the variation in the response variable is presented in Table 7, with higher Chi-square values indicative of a stronger contribution to explaining the variation in the response. In terms of power, we find a C-measure of 0.878. This measure has values between 0 and 1, with values over 0.8 seen as indicative of models with predictive power. The conditional R² of the model is 0.463. This value between 0 and 1 indicates the amount of variation in the response variable explained by the model. Where both the C-measure and the conditional R² are calculated taking both fixed and random effects into account, the marginal R² of the model, a value between 0 and 1, indicates how much of the variation in the response variable is captured exclusively by the fixed effects in the model (see Winter 2019: 264). This model achieves a reasonable marginal R² of 0.22.

Table 5:

Information on the random effects in the mixed-effects model.

		Variance	Standard deviation
Respondent ID (N = 1,228)	(Intercept)	0.530	0.728
Stimulus (Word pair) (N = 10)	(Intercept)	0.965	0.9821

Table 6:

Information on the fixed effects in the mixed-effects model (estimating the chance for a respondent preferring the Dutch lexicalization).

	Estimate	Std.Error	Z-value	Pr(>|z|)
(Intercept)	−0.944	0.339	−2.782	<0.01	**
Corpus success of the Anglicism	−0.339	0.010	−3.447	<0.001	***
Context (“Friends” vs. “Newspaper”)	−0.664	0.054	−12.255	<0.0001	***
Age respondent (“group 2: 30 to 50” vs. “group 1: 18 to 29”)	−0.195	0.079	−2.478	0.01	*
(“group 3: 51 to 70” vs. “group 1: 18 to 29”)	0.263	0.091	2.901	<0.01	**
Self-reported attitude respondent (bothered vs. not bothered/neutral)	0.505	0.078	6.437	<0.0001	***

Table 7:

Analysis of deviance table (Type II Wald chi-square tests), revealing the relative contribution of the predictors to explaining variation in the model.

	Chisq	df	Pr(>Chisq)
Corpus success Anglicism	11.88	1	<0.01	**
Context	150.18	1	<0.0001	***
Age respondent	20.75	2	<0.0001	***
Self-reported attitude respondent	41.44	1	<0.0001	***

For our interpretation of the model, we focus on the fixed effects (Table 6); the random effect structure will not be discussed further. Overall, the most relevant information for addressing the research questions is captured in the second and the final column of Table 6. The final column of the table provides information on the significance of the effects, with p < 0.05 indicating significance, and more stars indicating lower p-values. For significant effects, the next step is to scrutinize the second column. This column includes the model’s estimates for the effects, with positive estimates indicating a higher chance of respondents selecting the Dutch item and negative estimates indicating a higher chance of respondents selecting the English item. Below, we make this more concrete for each of the parameters, addressing our research questions one by one.

To answer the first research question, we need to verify whether the community-based usage statistics for the word pairs under scrutiny help explain the respondents’ self-reported preference for English loanwords or their Dutch alternatives in the experiment. Table 6 reveals a significant effect for the predictor “Corpus success Anglicism,” although Table 7 indicates that the parameter has the weakest contribution of the four selected predictors to the model: we see a significant effect with a smaller effect size. For the interpretation of the effect, we turn to Figure 3, which presents us with the numeric predictor “Corpus success Anglicism” (expressed as log(odds)) on the x-axis and the model’s fitted probability of respondents selecting the Dutch term in the trials on the y-axis: the higher the value on the x-axis, the more likely respondents are to select the Dutch term over the English term. We can clearly see the anticipated negative trend: as the corpus success of the Anglicism goes up, the probability of the respondent selecting the Dutch term goes down. We can hence favorably answer RQ1: the community-based usage statistics for the word pairs under scrutiny help explain the respondents’ self-reported preference for English loanwords or their Dutch alternatives in the experiment, with reported preference resonating usage patterns.

Figure 3:

Effect plot corpus success English term.

For the second research question, viz. “Do the socio-demographic characteristics of the respondent help explain the respondents’ self-reported preference for English loanwords or their Dutch alternatives in the experiment?”, the effect of four parameters was tested in the model, viz. “Age,” “Gender,” and “Region” of the respondents and their “Self-reported attitude” toward English influence. The two parameters for which less straightforward hypotheses were presented – “Region” and “Gender” – do not reach significance in the model. The age of the respondent and their self-reported attitude do play a role. For “Age,” the regression model captures the behavior of the youngest respondents (group 1: 18–29) in the intercept, then compares the behavior of the two other age groups to this intercept. Both comparisons yield significant differences: we see a significantly lower estimate for the odds of selecting the Dutch term for “group 2: 30–50,” than for the youngest participants (group 1: 18–29) and a significantly higher estimate for the odds of selecting the Dutch term for “group 3: 51–70,” than for the youngest participants (group 1: 18–29). Put differently: it is not as anticipated the group of the youngest respondents, but it is the active population (group 2: 30–50) who is most likely to select the English lexeme. Congruent with our hypothesis, the oldest age group (group 3: 51–70) is least likely to select the English lexeme. The results for “Self-reported attitude” toward English are highly intuitive: respondents who are allegedly bothered by the use of English have a higher chance of selecting the Dutch term in our experiment than respondents who claim not to be bothered by English influence or who state not to have an opinion (captured in the intercept). We can thus favorably answer RQ2: although not all of the variables included have an effect and although not all effects confirm the specific hypotheses introduced, we do see that some of the socio-demographic characteristics of the respondents help explain their lexical choices for English loanwords or their Dutch alternatives in the experiment.

Finally, we turn to RQ3: “Do speech contexts of communicative immediacy versus distance (conversation with friends or quality newspaper article) help explain the respondents’ self-reported preference for English loanwords or their Dutch alternatives in the experiment?”. Firstly, this parameter has a notably larger impact on explaining the variation than the other three significant predictors (see Table 7). Next, studying the effect, we see that the probability of respondents selecting the Dutch term in trials is lower in the context of “conversation with friends” than in the context of “newspaper article.” English terms are allegedly considered more appropriate in the immediacy context of casual conversation than in the distance context of the newspaper article. Hence, we can favorably answer RQ3: the speech context introduced to the respondents in the experimental trial helps explain the respondents’ lexical preference for English loanwords or their Dutch alternatives in the experiment, and the parameter of communicative immediacy versus distance plays an important role for the lexical choices made. In the next section, we will unpack these results with some more attention to their theoretical implications, meanwhile attenuating the analyses with some methodological reflections where needed.

6 Discussion

From a theoretical perspective, it is interesting to dig deeper into four results introduced in Section 5: (1) the correlation between community-based usage statistics and the reported lexical preferences for the studied word pairs, (2) the non-linear effect of age on the studied lexical preferences, (3) the lack of a region effect on the reported preferences and (4) the overriding effect of speech context as a predictor in respondents’ lexical choices.

While it might not come as a surprise that loanwords that occur more frequently in a newspaper corpus are also more likely to be selected as “most appropriate” in a lexical-preference experiment, the correlation is interesting to highlight, as it aligns nicely with a usage-based perspective on language variation. Usage-based approaches to language assume that (1) linguistic knowledge is an inventory of cognitive “units” – form-meaning combinations of varying schematicity – in an individual’s memory and that (2) this inventory results from and is constantly updated based on the usage events an individual engages in. Hence, a usage-based approach expects a clear link between language usage on the one hand and the mental representation an individual has of language usage on the other. In the case of Anglicisms, the frequency of occurrence can be expected to correlate with the loanword’s “entrenchment” in cognition, which in turn determines how appropriate the Anglicism is deemed, and also influences new usage events. Indeed, our results show a significant, proportional effect of an Anglicism’s frequency in a newspaper corpus on lexical preferences. However, the correlation between corpus success and the probability that a respondent selected an English loanword is not absolute. The regression model shows a significant, but rather small effect of corpus frequency. In our view, two factors seem to be at play here.

Firstly, attitudinal factors can explain why the reported preferences do not perfectly reflect corpus success frequencies. Linguistic purism, for instance, i.e., the desire “to preserve a language from, or rid it of, putative foreign elements, or other elements held to be undesirable” (Thomas 1991: 12), might influence reported preferences more than actual language usage. This was explicitly suggested by respondent 413 in the open comments section of our survey:

Puristic reasoning also clearly emerges in other comments, mainly “metalinguistically” and “ideologically” motivated (cf. Langer and Nesse 2012). Langer and Nesse (2012: 610–611) identify – following Gardt (2001) – four types of discourse underlying puristic reasoning: (1) structural discourse, which “assumes that for each language there is a state of purity at which the linguistic system is perfectly balanced”, (2) ideological discourse, which “emphasizes the superior quality of one’s language based on its genealogical purity and great age” and which views foreign elements “as a corruption of cultural purity”, (3) pedagogical discourse, which argues “that the use of foreign words leads to social division within a society, since the less-educated and the elderly might not understand new borrowings introduced into specific domains such as youth language or technical registers” and (4) metalinguistic discourse, which “more openly acknowledges that purism has to do with taste and aesthetics, rather than a general ability to communicate” and scorns the use of foreign words as “chasing fashionable trends”, “giving the impression of being intellectual and modern” and hence also as “superficial and pretentious behavior”. In our survey, 87 respondents (7.1%) made use of the open comments section, of which 4 respondents criticized English loanwords based on ideological arguments (cf. Examples (2) and (3)) and 5 based on metalinguistic ones (cf. Example 4). Structural or pedagogical arguments were not given explicitly, though reading between the lines, e.g., in Examples (3) or (4), structural or pedagogical reasoning can also be found. The other comments mainly pertained to the personal linguistic background of the respondents (n = 12), the research design (n = 9), indicated an open attitude toward English (n = 19), or commented on the difference between “necessary” and “luxury” loans (n = 15). We refrained from verifying statistically whether there is a correlation between the answers in the comments section and the answers in the forced-choice experiment, i.e., whether participants who commented negatively on the use of English in the open answer field choose fewer Anglicisms and vice versa. We did so because, on the one hand, not all the answers allow for a crude categorization, and on the other, because not enough participants made use of the open answer field. Nevertheless, it would be interesting to investigate this in further research.

Attitudes and beliefs influence the cognitive representations of linguistic items and can explain why reported preferences do not perfectly reflect corpus success frequencies. It has been shown that there is a general tendency for highly salient innovations to frequently give rise to strong judgmental reactions by the speakers (Winter-Froemel 2020), and it could thus be expected that these factors will have an impact on the reported lexical preferences for highly frequent and highly salient Anglicisms. On the other hand, it has often been observed (see e.g., Kabatek 1996) that the speakers’ actual linguistic behavior does not necessarily match and may even contradict their self-reported linguistic attitudes. A complex interaction of factors thus needs to be assumed, and this should of course be accounted for in an adequate usage-based model on language variation (see e.g., Campbell-Kibler’s 2016 proposal of a “cognitively realistic model of sociolinguistic variation,” in which she assumes the existence of a self-regulation system, which allows language users to control their speech production, perception, and attitudes, but would be limited in the number and complexity of constructs it can monitor).

Another factor potentially explaining the weak effect size of corpus success is that the frequency data from Zenner et al. (2012) do not perfectly mirror the frequencies with which the respondents encounter the studied Anglicisms in real life. The entrenchment of linguistic units varies among and within individuals (cf. Verhagen 2019), and hence, the entrenchment of the loanwords for the individual respondents does not completely match the “community entrenchment” measured using frequencies in a newspaper corpus.

Variation in entrenchment might not only account for the weak correlation between the corpus success measured in Zenner et al. (2012) and the reported lexical preferences, but also for the age effects in our data (see also Walsh 2015). Recall from Section 5 that the active population (group 2: 30–50) in our study was most likely to select English lexemes – more than the youngest generation (group 1: 18–29) – and that the oldest age group (group 3: 51–70) was least likely to do so. Our assumption here is that these differences correlate with differences in the degree to which members of the studied age groups encounter the presented Anglicisms. Research by Zenner et al. (2013) has shown an increase in English loanwords in Dutch job advertisements between 1995 and 2008, and hence, interpreted in line with the apparent-time hypothesis, which argues that ongoing language change is reflected in differences among age groups in the present (cf. Labov 1963), the entrenchment of English loanwords can be assumed to be weaker among older language users. Given that all word pairs presented to our test subjects included Anglicisms borrowed in Dutch before 1989, this reasoning seems not to hold quite well though. The relatively old Anglicisms included here could rather help explain why, contrary to our expectations, the youngest generation (group 1: 18–29) was not more prone to select the Anglicisms than the “active population.”

Similar familiarity-based reasoning can be applied to explain the lack of region effects in our data. Our study hypothesized that Belgian Dutch participants would react with stronger purist anti-English reflexes to the stimuli than Netherlandic Dutch respondents, given the standardization history of Dutch in Flanders (cf. Section 1). This idea was also voiced by some of our respondents. One respondent for instance remarked the following in the open comments section:

However, in the lexical preferences themselves, there was no difference between Netherlandic Dutch respondents and Flemish respondents. This might again have to do with the selected Anglicisms. The Anglicisms were selected from the corpus study of Zenner et al. (2012), which also did not report significant differences between the Flemish and Netherlandic subcorpora. At the community level, the presented Anglicisms seem to be entrenched equally in both the Netherlands and Flanders, and it is hence not surprising that the individual appropriateness judgments do not differ significantly. Yet, as we highlighted at the beginning of this section, the correlation between newspaper data and the individual choices reported in this study is far from absolute, so there might be other factors involved. It is also possible that the region effect we hypothesized does not exist in production, but only in attitudes. In this scenario, the reported attitudes of the Belgian Dutch and Netherlandic Dutch participants would be different. This proves to be the case, though contrary to expectations we find a higher proportion of Netherlandic Dutch participants who claim to be bothered by the use of English (117 out of 396, viz. 29.5%) than in the Belgian Dutch sample (164 out of 832, viz. 19.7%). The association between region and self-reported attitude is significant (X2 (1, N = 1,228) = 14.15, p < 0.0001), but fairly weak (Cramer’s V 0.1); more research is needed to grasp this pattern.^[6]

A last and interpretatively more straightforward result to highlight is the importance of context in our experiment. Context was the most important predictor of the success of an Anglicism in our study: the probability of respondents selecting an English term was significantly higher in the context of “conversation with friends” than in the context of “quality newspaper article.” English loanwords are thus considered more appropriate in the immediacy context than in the distance context. The importance of context was also stressed by 15 respondents in the open comments (cf. Examples (6) and (7), with the respondent in Example (7) explicitly pointing to the importance of the written realization, which is typical for communicative distance).

The results suggest that contexts of communicative immediacy or distance are of key importance for the selection of Anglicisms or their Dutch alternatives. This observation potentially has far-reaching theoretical implications, as it means that speakers do not have a uniform evaluation of the Anglicisms, but strongly take into account the specific context of the use of the items, which may lead to different evaluations of the adequacy of the same Anglicism by the same speaker. From a methodological perspective, the results stress the importance of incorporating context into attitudinal experiments (cf. Nejjari 2020; Rosseel et al. 2019). At the same time, it might be the case that the effect of context interacts with the level of entrenchment of the Anglicisms. In this study, we only included well-entrenched loanwords that were borrowed before 1989. However, it is possible that more recent Anglicisms, as well as Anglicisms belonging to other registers and semantic domains, might be evaluated in another way. Therefore, follow-up studies should include a variety of Anglicisms borrowed from different periods covering other semantic domains and registers to obtain a more fine-grained picture of the pragmatic factors steering the lexical choice of speakers.

7 Conclusion

The present study aimed at contributing to Anglicism research by studying the evaluation of Anglicisms experimentally and from an onomasiological perspective. By focusing on lexical preferences of Anglicisms compared to heritage alternatives, the study was able to show that lexical preference is steered by a combination of sociopragmatic parameters, with the context of usage playing a crucial role in the speakers’ self-reported usage. Of course, there are a few methodological concerns to bear in mind. First, replication of this study with more alternation pairs would be a valuable addition to the present study, preferably also integrating loanwords that are less entrenched (e.g., incipient loanwords or single-word switches). Secondly, it might be interesting to complement the current design with an experiment that is more indirect about its purpose. The here presented study did integrate fillers to divert the attention somewhat from the Anglicisms, but of course, the goal of the research was still very clear. To conclude, the “sanity checks” of our design indicated heterogeneity in the degree to which the presented Anglicisms were perceived as “foreign.” This issue was only touched upon briefly in our report – we mainly attempted to keep the degree of synonymy and perceived exogeneity constant – but it would be interesting to investigate why some words are more easily experienced as “foreign” than others. There are in other words still a lot of intriguing questions to tackle, but we do hope to have demonstrated the usefulness of an experimental, attitude-oriented approach toward lexical choice and to have paved the way for further research expanding on the theoretical and methodological implications of the findings obtained.