Are We Indeed So Illuded? Recency and Frequency Illusions in Dutch Prescriptivism

Abstract

In 2005, Arnold Zwicky posited two misapprehensions about language: the Recency Illusion, or the false idea that certain language variation is new, and the Frequency Illusion, the erroneous belief that a particular word or phrase occurs often. Since their conception, these concepts have received widespread attention in popular scientific linguistics, but quantitative research investigating their application is scarce. The purpose of this paper is to provide an empirical investigation of Zwicky’s proposed illusions. It does so by collecting statements about recency (‘this word is new’) and frequency (‘this construction occurs often’) from a database of Dutch prescriptive publications (1900–2018). I assessed their accuracy by comparing them to linguistic sources, including dictionaries, and usage corpora and other data. Our research showed that recency statements were rare, but that frequency statements, especially using high frequency terms such as vaak (‘often’), were commonplace. Compared to usage, most prescriptive recency and frequency statements for both lexis and grammar indeed constituted Zwickian illusions. This seems partly due to genuine erroneous or unsupported beliefs by authors, but also partly to prescriptive genre conventions and rhetorical choices. Our explorative research highlights the complex usage–prescriptivism interface, and argues for more research into this aspect of language perceptions.

1. Introduction

In 2005, American linguist Arnold Zwicky published a post on the well-known blog LanguageLog entitled ‘Just Between Dr. Language and I’ (Zwicky 2005). In what would later be called a “classic post” (Zimmer 2019), he wrote about a language column which claimed that the construction between you and I had only started to appear over the last few decades. Zwicky invalidated this claim by showing that its usage dated back at least 150 years, and possibly as much as 400 years. He also speculated more generally about such recency statements, claiming that the claim in this column was not an isolated occurrence, but rather an example of several “systematic dogged misapprehensions” (Zwicky 2006) about language that people entertained. Zwicky described two of these closely related misapprehensions as follows:

• Recency Illusion: If you’ve noticed something only recently, you believe that it originated recently.
• Frequency Illusion: Once you notice a phenomenon, you believe it happens a whole lot. (Zwicky 2005)

These illusions manifest in metalinguistic statements of various sorts. In the case of the aforementioned between you and I, the columnist wrote that “about 20 years ago U.S. English-speakers began switching me and X to X and I everywhere the phrase occurs” (quoted in Zwicky 2005). Here, the temporal marking is a form of recency statement. Other statements include the word recent or variations thereof, such as this word has recently come into use, or statements such as this has lately come into fashion. As for the Frequency Illusion, Zwicky reports in his original blog post how members of a certain research group expressed a belief about the common use of quotative all, which some even perceived to be used “all the time” (2005). All the time is a clear statement about the frequency with which a word occurs; other statements include other high frequency terms such as (very) often or repeatedly, but also frequency terms denoting presumably lower degrees of frequency, such as sometimes, rarely or hardly ever. Even a seemingly neutral description this word occurs can implicitly refer to frequency.

The crucial part about these illusions is that the perceived recency and frequency may not actually be based in fact, as Zwicky’s example shows. As such, these illusions are clearly related to the large and well-researched group of cognitive biases, or the “systematic errors in judgment and decision-making common to all human beings which can be due to cognitive limitations” (Wilke and Mata 2012, p. 531). The existence of such ‘errors’ does not mean that people are always wrong in assuming recency and frequency: there are well-known cases in which people make decisions based on demonstrably accurate information (cf. Arnold et al. 2000; Brysbaert et al. 2017). However, for some reason, sometimes this impression is wrong. One of the main reasons why people exhibit cognitive errors is selective attention. More specifically for language, Zwicky claims, people are influenced by the fact that “hardly anyone has a panoptic view of language variation” (2005). The illusions, then, are not only an effect of selective attention to language variation, but also of the (almost inevitable) restricted access that language users have to a language or language variety as a whole. Consciously or subconsciously, language users take their personal experience as representative for a larger variety, lect, or language, resulting in inaccurate estimations of the recency or frequency of certain words or phrases.

The above explains why these illusions occur; Zwicky also hypothesizes about who are likely to suffer from them, and what language is prone to being the subject of these illusions. According to him, education plays a key role in determining which language users are susceptible to false claims about recency and frequency (similarly, Cameron notes that knowledge of prescriptive discourse is ‘common’ for educated language-users (Cameron 1995, p. viii). Especially those with higher levels of education are vulnerable, as they have more “faith in the engines of correctness” (Zwicky 2008). Language professionals, linguists, and really anyone who is “reflective about language” are also susceptible (Zwicky 2005). As for which language is targeted, Zwicky claims that language users are particularly disposed towards certain types of language, which he states are “non-standard, informal, and spoken variants” (Zwicky 2008). One thing that variants with these parameters have in common is that they are often the subject of normative attention. For example, the whole point of the aforementioned column about between you and I was to eradicate mistakes in the use of this phrase. It seems, then, that recency and frequency statements are probably made in relation to usage items, or particular cases of disputed variation.

This combination of who (reflective language users) and what (disputed language variation) means that prescriptive publications, such as usage guides and style guides, are particularly likely to contain Recency and Frequency Illusions. After all, when we talk about these so-called ‘engines of correctness’, then surely prescriptive publications, which contain nothing but usage items, and aim at fixing language mistakes and eradicating unwanted variation, are the vehicles that run on these engines. Earlier research has shown that such works do indeed contain statements about usage, particularly relating to frequency (Sundby et al. 1991, p. 38; Kostadinova 2018, pp. 154–66; Ayres-Bennet 2020). However, this research is limited in scope, and recency statements do not seem to have been studied at all. Whether there is any systematicity to the use of frequency and recency statements, for example depending on the variable, author, or time period, is as of yet unclear.

Equally unclear is to what extent prescriptive statements about usage are an accurate reflection of actual language use. Recent work has shown that the position of prescriptivism being “often in defiance of normal usage” (Trask 1999, p. 246) may not be as categorically true as was long assumed to be the case. For example, van der Meulen and Rutten (2022) show that in a work by the Dutch linguist Matthijs Siegenbeek (1774–1854) there is a relation between the degree of frequency term used to describe a disputed variant and the actual frequency with which this variant occurs in usage (i.e., higher terms, such as vaak (‘often’) show higher occurrences of usage). Similarly, in their study of the one of the (few) X who Y-construction, Hogeweg et al. claim that a frequency statement in a particular style guide is correct in asserting that this construction occurs vaak (‘often’) with singular agreement (Hogeweg et al. 2018, p. 339).1 By contrast, in the one example in which an alleged Recency Illusion is explicitly tested against usage data, Liberman (2010) comes to an uncertain conclusion about whether the claim of recency for the sentence-initial so is correct. However, he admits that his analysis is “crude”, as he only looks at frequencies per million words in the Corpus of Historical American, without checking whether instances are relevant or not, without an operational definition of recency, and without taking potential genre differences into account.

In summary, although we have some evidence for recency and frequency statements being used in prescriptive publications, we know little about the extent and manner of their use; whether or not these statements are accurate is similarly uncertain. The present paper addresses these two issues, by first mapping recency and frequency statements in Dutch prescriptive publications from 1900 onwards. Second, I assess whether these statements are illusions in the Zwickian sense, or whether they conform to actual usage patterns, by comparing the contents of these statements against actual language usage. In order to investigate whether certain types of linguistic variable are more likely to be correctly spotted than others, I look at both lexical and grammatical usage items. In the absence of any quantitative work exploring these matters, the purpose of the present paper is explicitly explorative: I want to see whether this type of methodology works, and what issues we encounter. The methodology and the materials are discussed in Section 2. Next, I look at the results in the prescriptive publications for recency (Section 3.1) and frequency (Section 3.2), after which I turn to an assessment of the accuracy of recency statements (Section 3.3) and frequency statement (Section 3.4). I end the paper with a discussion of all these findings (Section 4).

2. Materials and Methods

In this section, I first describe our prescriptive material, how I extracted recency and frequency statements from this material, and the tags I gave to these statements (Section 2.1). Following that, I explain how I tested the veracity of these statements. For recency (Section 2.2) and frequency (Section 2.3), I detail how each concept is operationalized, which statements I used for our evaluation of accuracy, and what usage data I base my evaluation on.

2.1. Prescriptive Data

To study prescriptivist statements about usage, I searched the Normative Database of Dutch. This database contains 5678 entries from prescriptivist publications aimed at mother tongue speakers of Dutch in the Netherlands for the period 1910–2018 (see van der Meulen 2020 for more details). The database focuses primarily on morphosyntactic usage items, but includes a substantial number of entries on lexis, especially for the earlier decades, as well as some entries for other linguistic levels, such as spelling or pronunciation. Entries in the database are quite heterogeneous in terms of length and contents, because their structure follows the varying build-up of the publication they originate from. Consequently, while many entries pertain to a single usage item, some contain more than one. Any usage item can attract more than one recency or frequency statement.2 I approached the data using #Lancsbox (Brezina et al. 2020).

To investigate the Recency Illusion, I used a set of queries based on words denoting recency, manually cleaning up the results.3 Over the course of the investigation, it became apparent that there are in fact two types of recency statements. Next to the ‘narrow’ Zwickian type, which is about recent origins (Example 1), there are also statements that comment on a recent development in frequency, often by using the word tegenwoordig (‘nowadays’, Example 2). Although this latter type of pronouncements is not strictly Zwickian, it does constitute a recency claim, and thus I included them in my initial survey of statements.

• In dit geval gebruikt de jongere taal niet zelden het woord ontbranden
  (‘In this case, recent language not uncommonly uses the word ontbranden’) (Moortgat 1925, p. 50)
• Per. Tegenwoordig een veel gebruikt voorzetsel: per 1 januari
  (‘Per. Nowadays an often-used preposition: per January 1st’) (Meijers 1959, p. 103)

For frequency statements, I extracted utterances from the database that referenced any kind of frequency of use.4 Again, I cleaned up the results, including only explicit statements about actual usage (Example 3), rather than opinions about usage, as this latter group of statements does not have to be grounded in the reality of usage (cf. Theissen 1978, p. 7). This meant that I excluded statements about how usage ought to be (Example 4), as well as statements about attitudes (Example 5).

3.

Driedubbel: vaak gebruikt waar drievoudig wordt bedoeld. Driedubbel is zes. (‘Driedubbel: often used where threefold is meant. Driedubbel is six’) (de Raat 2012, p. 29)

4.

Waar kan nooit redengevend zijn.

(‘Waar should never indicate a reason’) (Algemeen Nederlandsch Verbond 1925, p. 4)

5.

Ondertussen raakt dit gebruik van betreffend echter steeds meer geaccepteerd.

(‘Meanwhile this usage of betreffend is becoming more and more accepted’) (Tiggeler 2001)

As I was interested in delving deeper into the patterns of recency and frequency statements, I tagged all of them for linguistic level (i.e., grammar, lexis, and spelling, etc.) and usage item (for example, comparative conjunction als/dan ‘as/than’). In certain cases, I grouped certain usage items together. This applied, for example, to several closely related syntactic usage items about the correct combination of dependent and independent clauses. Next, as I was also interested in how prescriptivists applied recency statements according to whether they targeted recent origins or recent increased frequency, I tagged this dimension as well. I also labelled the frequency statements as absolute when they denotated a fixed frequency (e.g., vaak ‘often’, soms ‘sometimes’), or diachronic when they referred to a development over time (e.g., meer en meer ‘more and more’, steeds vaker ‘increasingly often’). Finally, I classified all frequency statements according to their grade. Terms could be either low, such as soms (‘sometimes’); middle, such as gebruikt (‘in use’); or high, such as vaak (‘often’). For both absolute and diachronic, as well as grade, I started from the classification scheme in van der Meulen and Rutten (2022), but as the present research included many more terms, I expanded this framework where necessary.

2.2. Establishing Recency

To evaluate recency statements, I had to date the origins of the word or construction under scrutiny. Although very common, especially in lexicography, this practice nevertheless comes with a host of issues (for an overview see van der Sijs 2001, pp. 36–43), some of which complicated the current research. For example, when Moortgat (1925, p. 33fn2) questioned the recent use of halfmens (‘halfman’), he objected not to the wordform as such, which had existed since at least the 16th century (Woordenboek der Nederlandsche Taal, WNT, s.v. half^I), but to a new meaning. Such a semantic shift was hard to distinguish from the already existing meaning of a word. A second problem was that a word could be established in specialized domains or genres, such as scientific publications or medical jargon, before it percolated to general usage. However, an early occurrence of a word in a specialized context does not prove general usage, nor does it seem feasible to expect prescriptivists to come into contact with, base themselves on, or police such specialized occurrences (unless specifically mentioned).

Because of these issues, I limited my investigation to recency statements denoting new wordforms, rather than new meanings. I dated these wordforms based on general usage only, which we can reasonably expect prescriptivists to be aware of, or have come into contact with. It is very likely that certain authors also came into contact with more specialized genres, but to take these into account would require both a more fine-grained approach than this paper aims at, and specialized data that we do not have. I only looked at recency statements concerning recent origins (Example 1 above), rather than statements denoting an increase in frequency (Example 2 above). Although the latter type of statement does concern recency, and is thus included in the first half of this paper, the survey of statements does not fall within the strict Zwickian interpretation of this concept. Moreover, as I also note below, Dutch lacks the proper data to investigate recent changes in frequency for usage items. From the statements that were left after these selection criteria, I randomly selected a maximum of two usage items per decade, for both lexical items and grammatical ones. In taking this coarse-grained approach, I avoided basing our conclusions on the perceptive ability of particular authors. Moreover, this approach enabled me to see whether the accuracy of recency statements showed a development over time with regard to prescriptivists’ perceptions.

The data I used to pinpoint the origins of our usage items came both from primary and secondary sources. Firstly, I checked whether the earliest known example of a usage item was mentioned in one of four secondary sources: the historical Woordenboek Nederlandsche Taal (WNT),5 the Chronologisch Woordenboek (‘Chronological Dictionary’, van der Sijs 2001), the Etymologiebank (‘Etymology bank’, van der Sijs 2010), or the Geschiedenis van het Nederlands in de twintigste eeuw (‘History of Dutch in the twentieth century’, van der Horst and van der Horst 1999). Secondly, I used primary usage sources. Recent years have seen a great increase in the availability of Dutch language materials, which makes antedatings possible and even likely as compared to the linguistic sources mentioned above. For that reason, I checked results from our metalinguistic investigation in Delpher, an online repository containing over 120 million pages from newspapers, books, and magazines.6

After dating the words, I had to evaluate for each usage item whether the date of its origins can indeed be called recent, as compared to the date of the prescriptive recency statement. Clearly, the scope of what can be considered ‘recent’ depends very much on context. Set against the development of the faculty of language in humans, for example, any change that happened since the standardisation of Dutch in the 16th century is extremely recent. However, this is presumably not what language users mean when they say that a form is ‘new’ or ‘recent’. Zwicky gave the following example of what is not recent for him:

Charles Hockett wrote in 1958 (A Course in Modern Linguistics, p. 428) about “the recent colloquial pattern I’m going home and eat” (…). But Hockett’s belief that the construction was recent in 1958 is just wrong; David Denison, at Manchester, has collected examples from roughly 30 years before that.

(Zwicky 2005)

For Zwicky, then, recent had to be well within 30 years, although he gives neither an explanation for this judgement nor a more exact timeframe. The fact that he stated Hockett’s belief is ‘just wrong’ implies that it was even substantially less than 30 years. However, in the absence of any operationalisations of recency within linguistics that I was aware of, and in light of the fact that the rest of our research is also built upon Zwicky’s definitions, I used this threshold value. Moreover, when we take into account that all prescriptive authors are over 30 years old, this meant that recency implies a word or construction originated within the lifetime of an author. If my data, then, showed the origins of a word or phrase to be more than 30 years before the prescriptive utterance, I considered this to be a case of the Recency Illusion.

2.3. Establishing Frequency

The results for the frequency statements presented us with similar issues as the recency statements. These included the aforementioned usage items that discussed new meanings for existing wordforms. Additionally, several syntactic types of ellipsis, such as the omission of the complementizer om (‘for to’) proved hard to study quantitatively. An additional selection criterion resulted from Zwicky’s formulation of the Frequency Illusion as the belief that certain linguistic variants occur often. In other words, although low and middle frequency statements, such as sometimes or rarely, are statements about frequency, they do not qualify as possible Frequency Illusions in the Zwickian sense, and thus were left out of the subsequent analysis.

Within these parameters, for lexis I again randomly extracted two usage items per decade. I matched the variant targeted by the frequency statement with its proposed counterpart or counterparts, mapping out the raw and relative frequency of both variants in a usage corpus (with relative frequency being the proportion with which a variant occurs as compared to all variants combined for a variable, expressed as a percentage). I used different corpora for different periods, depending on availability and corpus quality. For lexical usage items, I initially searched C-CLAMP, a corpus containing ±200 million words from various magazines and newspapers published between 1837 and 1999 (Piersoul et al. n.d.). When this corpus yielded fewer than 100 combined hits for both variants for any usage item, I again used Delpher. It is important to note that this dataset unfortunately has some serious limitations with regard to quantitative research, as its OCR quality is quite low for certain time periods, and its representativeness is questionable (see van der Sijs 2019). Still, as I base my conclusions on relative frequency, this becomes less of an issue, although it has to be taken into account when interpreting the results. As with the recency statements, I again use a time window of 30 years prior to the statement. So, when Moortgat calls duistering (‘darkness’) the “usual” variant as opposed to duisternis in 1925 (Moortgat 1925, p. 48), I look at occurrences of both variants in the 1896–1925 segment of C-CLAMP.

For grammatical usage items, establishing the frequency of two variants in a certain time period of corpus data presented us with problems. The polysemous nature of many grammatical words, coupled with their high occurrence, would require more cleaning-up effort than I could achieve given the overall scope of the present paper, even after sampling. To still be able to test some grammatical usage frequency statements, I based my study on data as reported in (Van der Meulen n.d.). For this paper, spoken and written usage data were collected and cleaned-up for nine morphosyntactic variables, in order to test the distribution of variants against prescriptive evaluations (i.e., whether a variant was acceptable) and attitudinal data. I used either written, spoken, or all data from this dataset, depending on the scope of the statement. The usage data comprise the period 1995–2004, and thus I looked at frequency statements from roughly that period (i.e., 1990–2015).

To evaluate whether any result constituted a Frequency Illusion, I turned to a survey of Dutch probability and frequency terms and phrases, as reported in Willems et al. (2020). Conveniently, they asked participants to give numerical interpretations of their target phrases in percentages. Thus, for example, vaak (‘often’) has a mean numerical interpretation of 73%, although there is considerable variability in the answers (Willems et al. 2020, p. 9). However, as very few of the high frequency terms found in our data were included by Willems, Albers, and Smeets, I only used the result for vaak, arguing that higher frequency terms such as heel vaak (‘very often’) can be expected to at least entail this meaning. In other words, taking this relatively low threshold, I am being lenient towards prescriptivists’ judgments. In terms of the Frequency Illusion, this approach meant that if less than 73% of the combined variants for a given usage item constitute the one that the prescriptive author targets, this constitutes a case of the Frequency Illusion. Taking our example from above, the disapproved variant duistering occurs 8 times in our data, the approved variant duisternis 909 times. Relatively speaking, then, duistering only occurred in 0.9% of all 917 combined variants. This is clearly far below our threshold of 73%, and we can thus conclude that this is an example of the Frequency Illusion.

3. Results

First, I discuss the results of our prescriptive survey for recency terms (Section 3.1) and frequency terms (Section 3.2). After that, I zoom in on the comparison with usage, looking at lexis and grammar separately, again first for recency statements (Section 3.3) and then frequency statements (Section 3.4).

3.1. Prescriptive Statements about Recency

Our data showed 238 entries that contain at least one recency statement. Most entries (214, 89.9%) were aimed at a single usage item; 24 entries discussed more than one. For example, Charivarius listed eight bruikbare nieuwe Nederlandse woorden (‘useable new Dutch words’, Charivarius 1940, p. 73). By far the most extreme cases were two series of modewoorden (‘fashionable words’), both from the same work by Haje (1932), which contained 40 and 24 words, respectively. In total then, this leads to 341 usage items being targeted. The majority of these pertained to lexis (230 or 67.6%). Grammar comprises a much smaller proportion (98 or 28.8%); the final twelve items concerned pronunciation, stylistics, or spelling. This distribution was particularly striking when we take into account that the database as a whole focused more on morphosyntax. It seems then, that lexical items are more likely to attract recency statements.

Looking at which particular usage items were mentioned, we saw that the 341 usage items belonged to 307 types (TTR = 0.9). A total of 283 items were mentioned only once (83.3%); the remaining 58 belonged to 22 types (see

Table 1. Usage items which are mentioned more than once with recency statements.

Usage Item	Translation	Level	No. of Recency Statements	Year(s) of Recency Statements
aantal N is/zijn	‘a number of N is/are’	grammar	5	1979, 1994, 2000, 2001, 2002
N dat/wat	‘N that/which’	grammar	5	1932, 1963, 1990, 1996, 1996
als/dan	‘as/than’	grammar	4	1984, 1994, 1994, 2006
trappen van vergelijking	comparative/superlative formation	grammar	3	1940, 1946, 1994
omdat/doordat	causal conjunctions	grammar	3	1962, 1994, 1996
nieuwbouw	‘new construction’	lexis	3	1932, 1942, 1964
grootstad	‘big city’	lexis	3	1932, 1942, 1964
belevenis	‘experience’	lexis	3	1932, 1964, 1978
woordgeslacht	noun gender	grammar	3	1993, 1993, 1994
11 items	N/A	lexis	2	N/A
3 items	N/A	grammar	2	N/A

). The two types mentioned most often (five occurrences each) were both morphosyntactic: the relative pronouns dat/wat (‘that/which’) and number agreement een aantal N is/zijn (‘a number of N is/are’). The usage items that were mentioned more than once usually stemmed from different years. The largest time difference, 64 years, was found with regard to N dat/wat (‘N that/which’), for which the first recency statement stemmed from 1932, and the last from 1996. This difference in itself makes Recency Illusions inevitable: if particular variation already existed in 1932, it cannot be a recent phenomenon in the 1990s.

To see whether there was any development in the use of recency statements, I looked at the distribution of recency judgements over time (see

Table 2. Recency statements per decade as raw frequency and percentage of total amount of entries.

Decade	1910s	1920s	1930s	1940s	1950s	1960s	1970s	1980s	1990s	2000s	2010s
No. of usage items with recency statements (n = 341)	0	18	97	42	29	35	44	8	50	15	4
No. of entries with recency statements (n = 238)	0	16	38	32	26	22	38	6	43	13	4
Total no. of entries in database (n = 5678)	216	193	563	396	531	544	336	524	1082	887	406
% of entries that contain recency statements compared to total no. of entries in database	0.0	8.3	6.7	8.1	4.9	4.0	11.3	1.1	4.0	1.6	1.0

). Looking first at usage items (n = 341), the highest number of statements was found in the 1930s, which was largely caused by the two wordlists mentioned before. The 1990s was the second-highest scoring decade, as it had 50 usage items attracting recency statements. However, as the number of prescriptive publications varied greatly over time in our database, we cannot draw conclusions based on these raw usage items frequencies. Instead, we should look at relative occurrences per decade as compared to the total database. Unfortunately, as we do not know the exact number of usage items for every entry in the whole database, I had to base our study here on the number of entries, which we do know. Looking at entries containing recency statements (n = 238) gives a somewhat different view. Now, the 1970s show the highest proportion of entries containing at least one recency statement (11.3%). The lowest percentage is in the 2010s (1%). Furthermore, there seems to be a declining tendency to use recency statements, as earlier decades show higher proportions than later. This tendency is not linear, but the largest outlier, the 1970s, is due to the predilection of one particular author to use such statements.7 The same goes for the 1990s, albeit to a lesser extent. Although the preferences of particular authors influence the presence of statements, we can conclude that the use of recency statements does proportionally decline over time.

Lastly, the division of recency statements into origins and recent development showed that the former, strictly Zwickian remarks comprise the majority of statements (80.9%, 276). Almost all statements that comment on a development in usage signal an increase; only in a few isolated cases do writers indicate that a variant is losing ground. This preference points towards the rhetorical nature of such statements: after all, no-one needs to be ‘warned’ about a disapproved variant disappearing.

3.2. Prescriptive Statements about Frequency

Our material showed that 1249 entries contain at least one mention to frequency. Frequency statements are thus far more prevalent in Dutch prescriptive publications than recency statements. In total, these 1249 entries contained 1786 frequency statements. A difference with the recency statements was that the majority of frequency statements pertained to grammar (1064, 59.5%), while 36.8% of statements was found in relation to lexis. Again, a small number of 67 cases referred to other levels, such as spelling and pronunciation. This distribution seems to be more in line with the general contents of the database. Another notable difference between the recency and frequency utterances was that the latter were distributed over a proportionally much smaller number of different usage types, namely 601 (TTR = 0.33). Moreover, for the frequency items, just 400 (66.6%) of the types were found only once.

Just as with the recency statements, those particular items that attract the most frequency statements are grammatical in nature (see

Table 3. Linguistic items mentioned more than once with recency statements.

	Usage Item	English Translation	Level	No. of Times Mentioned
1.	woordgeslacht	noun gender	grammar	88
2.	hen/hun	3rd person personal pronoun direct and indirect object	grammar	77
3.	als/dan	comparative conjunctions	grammar	66
4.	dat/wat	relative pronouns	grammar	61
5.	zinsbouw	sentence structure	grammar	39
6.	tante betje	word order in subordinate clauses	grammar	37
7.	congruentie	agreement	grammar	36
8.	beknopte bijzin	participle clauses	grammar	30
9.	trappen van vergelijking	degrees of comparison	grammar	27
10.	omdat/doordat	causal conjunctions	grammar	26

). However, for the frequency statements this effect is much stronger. In fact, when we look at all 33 types that occur more than ten times, only one type, verbal formation, refers to lexis, and then only partly. Another noteworthy property of the items in

Table 4. Frequency statements per decade as raw frequency and percentage of total amount of entries.

Decade	1910s	1920s	1930s	1940s	1950s	1960s	1970s	1980s	1990s	2000s	2010s
No. of usage items with recency statements	0	66	74	211	153	283	155	121	340	277	107
No. of entries with recency statements	0	53	59	132	82	195	108	94	233	214	79
Total no. of entries in database (n = 5678)	216	193	563	396	531	544	336	524	1082	887	406
% of entries that contain recency statements compared to total no. of entries in database	0.0	27.5	10.5	33.3	15.4	35.8	32.1	17.9	21.5	24.1	19.5

is the fact that the top five most frequently found usage items in the 20th and 21st century (hen/hun, als/dan, dat/wat, tante betje and congruence, see van der Meulen 2021, p. 170) are also among those items that attract the most frequency statements. This may show that frequency statements are at least partially related to the prescriptive canon. By contrast, however, two of the other types in this top-10, namely zinsbouw and beknopte bijzin, relate to quite complex syntactic phenomena, which are generally rare in prescriptive publications (cf. Chapman 2021).

The patterning for the development of frequency statements over time, again looking at entries rather than specific items (n = 1249), is more haphazard than for recency statements (see Table 4). There is another cut-off point between the 1970s and 1980s, but the decline in the last decades is less pronounced, and the presence of frequency terms remains higher than for recency terms. Finally, the influence from specific authors is more marginal. One exception is B. Cees Damsteegt, who uses frequency statements in 102 entries in his In de Doolhof van Het Nederlands (Damsteegt 1964). These comprise 52.3% of all mentions in this decade, but this percentage is actually in line with Damsteegt’s overall presence in that time period: the entries from his detailed guide comprise 51.5% of all the 544 entries for the 1960s.

As I was interested in the way prescriptivists evaluate usage, I next looked at the specific frequency terms, and the distribution among categories of frequency term and degree (

Table 5. Top-10 most frequently used frequency terms.

	Frequency Term	English Translation	Category	Degree	No. of Times Mentioned
1.	vaak	‘often’	absolute	high	317
2.	soms	‘sometimes’	absolute	middle	107
3.	steeds vaker	‘increasingly more often’	diachronic	high	51
4.	weleens	‘occasionally’	absolute	low	51
5.	veel	‘much/many’	absolute	high	47
6.	gebruikt	‘used’	absolute	middle	36
7.	vooral	‘mostly’	absolute	high	36
8.	steeds meer	‘increasingly’	diachronic	high	35
9.	vaker	‘more often’	diachronic	high	33
10.	dikwijls	‘often’	absolute	high	32

). As for type, the top-10 contained seven absolute frequency terms and three diachronic ones. This was an underrepresentation of absolute frequency terms, as they comprise 85.3% (1524 tokens) of all terms. I also found seven high degree frequency terms, which is an overrepresentation as compared to the presence of such terms in total, as high degree frequency terms comprise 59% (1054 tokens), as opposed to middle (331, 18.5%), and low terms (402, 22.5%). Still, it was clear that prescriptive authors favour high frequency terms. This may be another sign of the rhetorical nature of such terms: the sense of urgency is lower for low frequency terms, which defies the purpose of prescriptive publications.

3.3. Recency and Usage

Of the 23 recency statements I investigated, 17 lexical usage items are instances of the Recency Illusion, which is by far the majority of statements (20, 87%). Just three recency statements are not Illusions (see

Table 6. Lexical usage items with year(s) of prescriptive utterance, earliest occurrence in usage, and evaluation with regard to Recency Illusion.

Usage Item	English Translation	Year(s) of Recency Statement	Source8	Illusion?
toonkunstenaar	‘musician’	1925	1756	yes
mentaliteit	‘mentality’	1932	1902	no
vandaag de dag	‘today’	1940	1895	yes
nieuwbouw	‘new housing estate’	1932, 1942, 1964	1859	1932: yes; 1942: yes; 1964: yes;
halfbroeder	‘half-brother’	1932, 1964	1748	1932: yes; 1964: yes
grootmacht	‘superpower’	1932, 1964	1871	1932: yes 1964: yes
hoogspanning	‘high-voltage’	1932, 1978	1907	1932: no 1978: yes
belevenis	‘experience’	1932, 1964, 1978	1898	1932: yes; 1964: yes; 1978: yes
witlof	‘chicory’	1959	1854	yes
festival	‘festival’	1959	1872	yes
emballage	‘packaging’	1980	1745	yes
uitonderhandelen	‘negotiate’	1989	1955	yes
mond-aan-mondreclame	‘word-of-mouth advertising’	1995	1989	no
met behulp van + person	‘with the help of’ + person	2001	1857	yes
in de loop der tijd	‘over time’	2008	1946	yes
uitboeën	‘booing’	2012	1966	yes

). These exceptions are mond-aan-mondreclame (‘mouth-on-mouth marketing’), for which there was only six years between the recency statement and the first observed occurrence in usage; hoogspanning (‘high voltage’), for which the time gap was 25 years (a later statement claiming the recency of this word from 1978 is a Recency Illusion according to our definition), and mentaliteit (‘mentality’), for which the time gap is exactly 30 years. However, these results highlight the arbitrariness of our definition: mentaliteit just makes the cut, while belevenis (‘experience’), for which the time gap is 34 years, is only an illusion by a small margin. However, such occurrences are actually quite rare: for 17 of the statements, the earliest occurrence of a word or phrase predates the recency statement by more than 50 years.

Within our selection criteria, there were few recency statements that commented on the origins of a particular grammatical variant, and that could be checked. Still, for those 13 that did occur for eight usage items, the picture is much the same as for lexical items: they are almost all Recency Illusions (see

Table 7. Grammatical usage items with year(s) of prescriptive utterance, earliest occurrence in usage, and evaluation with regard to Recency Illusion.

Usage Item	Recency Statement	Source10	Illusion?
relative pronouns dat/wat	1932, 1990	1930s	1932: maybe; 1990: yes
periphrastic comparative/superlative	1940, 1994	1890	1940: yes; 1994: yes
complementizer om	1948	1898	yes
article ellips before certain nouns (e.g., ondergetekende ‘undersigned’)	1959	1900	yes
causal conjunctions omdat/doordat	1962, 1994, 1996	1860	1962: yes; 1994: yes; 1996: yes
comparative conjunctions als/dan	1984	17th century	yes
noun gender	1993, 1994	1700	1993: yes; 1994: yes
hun-subject	2000	1911	yes

). The only possible exception in our data is relative pronoun wat, which, according to Haje, is tegenwoordig (‘at present’) frequently used in written language following neuter nouns (Haje 1932, p. 118). Although the general development of d-forms to w-forms is well-established (see, for example, Schoonenboom 2000), relatively little is known about the particular progression of this phenomenon with regard to neuter nouns (cf. Rutten 2020). However, as van der Horst and van der Horst posit that the use of the disapproved variant wat must have been limited to spoken language until approximately 1900 (van der Horst and van der Horst 1999, p. 170), I evaluate this case then as maybe constituting a Recency Illusion.

An interesting difference with the lexical items is that the time gaps are much larger for grammatical items. Several of the usage items go back to the early days of Dutch normative writing. For example, when Heldring complained in 1993 that many neuter proper nouns for countries and cities were increasingly being referred to by using feminine possessive pronouns (for example, Amsterdam en haar grachten, ‘Amsterdam and her canals’), this echoes a very similar comment made by Van Hoogstraten in 1700 (quoted in van der Sijs 2021, p. 402). On the one hand, it seems unlikely that Heldring, a journalist, would have read a grammar book from 300 years ago. On the other hand, the referential practice of using feminine possessive pronouns is a staple of prescriptive publications, dating back to at least the 1930s.9 With regard to these grammatical cases, then, Zwicky’s complaint that “if only they’d thought to consult some standard sources or look at some facts” seems warranted (Zwicky 2005).

3.4. Frequency and Usage

Within our parameters, I found two checkable lexical examples in all decades except the 1920s and 1950s, for a total of 18 (see

Table 8. Lexical usage items with year(s) of prescriptive utterance, absolute and relative frequency counts of condemned variant in usage, and evaluation with regard to Frequency Illusion.

Condemned Variant	Proposed Alternative(s)	English Translation	Year of Meta Utterance	Absolute Frequency of Condemned Variant11	Relative Frequency in % of Condemned Variant	<73%
duistering	duisternis	‘darkness’	1925	8	0.9	yes
(on)toelaatbaar	(on)geoorloofd	‘impermissible’	1932	73	17.9	yes
jongeman	jonkman	‘young man’	1935	81	38.8	yes
onmeedogenloos	meedogenloos	‘merciless’	1940	49	3.5	yes
zodoende	dus, daarom	‘thusly’	1941	20	0.1	yes
-toename	-toeneming	‘increase’	1957	50	47.2	yes
vanwege	wegens	‘because of’	1962	186	34.0	yes
zwempoel	zwembad	‘swimming pool’	1964	69	0.1	yes
ons inziens	onzes inziens	‘in our opinion’	1975	7791	65.7	yes
begeesterd	verrukt, geestdriftig	‘enraptured’	1978	4	1.2	yes
verplichtend	verplicht	‘obliged’	1980	16	2.5	yes
ik mankeer	mij mankeert	‘I lack’	1986	205	58.6	yes
middels	door middel van	‘by means of’	1991	328	25.5	yes
behartenswaardig	behartigenswaardig	‘worthy of consideration’	1997	146	56.6	yes
overnieuw	opnieuw	‘all over again’	2000	552	0.1	yes
meer of mindere mate	meerdere of mindere mate	‘a greater or lesser degree’	2005	1457	76.7	no
scherpst van de snede	scherp van de snede	‘sharpest part of the knife’ (idiom)	2013	297	47.9	yes
zei af	zegde af	‘cancelled’	2013	25	27.5	yes

). Our data from the lexical variants showed that all but one of our examples constituted Frequency Illusions. The only exception was meer of mindere mate (‘a greater or lesser extent’), for which the condemned variant comprised 76.7% of all variants. In seven of the cases, the proportion of targeted variants comprises less than 5% of all variants. Noteworthy is the case of compounds ending in -toename (increase), which is targeted by Grauls (1957, p. 197) in favour of the proposed alternatives ending in -toeneming. In our data, the disapproved variants occur in almost half of all cases (47.2%), making this a Frequency Illusion. However, data from the present day showed that the disapproved variant has all but disappeared, as I found 64 hits for -toeneming, against 8767 for -toename. So, while the statement was not correct at the time, Grauls was perhaps aware of a change in progress.

Of the nine grammatical usage items under investigation, seven occurred in our data with high frequency statements, sometimes multiple times (see

Table 9. Grammatical usage items with year(s) of prescriptive utterance, absolute and relative frequency counts of condemned variant in usage, and evaluation with regard to Frequency Illusion.

Condemned Variant	Proposed Alternative(s)	Year of Meta Utterance	Absolute Frequency of Condemned Variant12	Relative Frequency in % of Condemned Variant	<73%
person waarvan	person van wie	1999, 2000, 2001, 2001	120	18.6	yes
Neuter noun wat	Neuter noun dat	1996, 1998, 2001, 2007, 2009, 2013	339	7.4	yes
subject-hun	subject-zij	1994, 2000, 2008, 2011	197	2.7	yes
een aantal mensen zijn	een aantal mensen is	1994, 1996, 2000, 2005, 2007, 2011	465	60.7	yes
je kan	je kunt	1993	1792	28.6	yes
u heeft	u hebt	2009	827	74.1	no
hele mooie auto	heel mooie auto	1993, 2011	2482	79.7	no

). For two of these usage items, u heeft (‘you-form have) and hele mooie auto (‘very-infl nice car’), the proportion of condemned variants is above our threshold of 73%. Thus, the three frequency statements I find for these usage items are not Frequency Illusions. However, these correct assessments are the minority, as the other 21 statements, distributed over five variants, show relative frequencies below the threshold. Thus, as with the lexical variants, by far the majority of statements can be classified as Frequency Illusions.

Of particular interest was one of the frequency statements made for relative pronoun wat. Houët (2000, p. 249) claimed that in spoken language this variant is used vrijwel uitsluitend (‘almost exclusively’). When I extracted the spoken part out of the data used, I observed that, although with 35.8% it was much higher than the combined written and spoken data reported above, it was far from used exclusively. In fact, it was not even the dominant variant. This label uitsluitend (‘exclusively’) was quite rare in our data, and I found only two other instances. In 1962, Heidbuchel claimed that omdat (‘because’) is used vrijwel uitsluitend, to the detriment of alternative doordat (Heidbuchel 1962, p. 144). However, in C-CLAMP (1932–1962) I found 832 examples of doordat (8.8% of all variants). Similarly, Weverink stated that vrijwel iedereen (‘almost everybody’) used ze (‘them’) in spoken language to indicate direct or indirect object, rather than hen or hun (Weverink 2012, p. 52). In this case, I found over 2300 instances of hen and hun in these functions in the Corpus Gesproken Nederlands (‘Corpus of Spoken Dutch’, see Oostdijk 2000). Both statements were thus clearly illusions. Arguably, these examples embodied a ‘worse’ Frequency Illusion than the use of vaak, as uitsluitend is much more categorical. This was also shown in the research by Willems et al.: the interpretation of more categorical words, such as always, had a much smaller range of answers than that of high frequency words such as often (Willems et al. 2020, p. 10). These examples highlight the rhetorical dimension of the use of such extreme frequency statements, but also the risk of their use.

4. Discussion

Our research shows that recency and frequency statements are part of the modern Dutch prescriptive tradition, but in different ways, and to different degrees. Recency statements occur mostly in relation to lexis, and seem to be falling out of fashion over the course of the twentieth century. Although, as I mentioned, this may partly be due to the design of our database, it may also be an effect of the solidification of the prescriptive canon. Tieken-Boon van Ostade notes this development for the English usage guide tradition when she says that “many usage problems came to have a remarkably stable presence” (Tieken-Boon van Ostade 2020); the same has been shown for Dutch (van der Meulen 2021). This lack of innovation in the genre presumably diminishes the need and even the possibility for variation to be perceived as recent. Thus, when recency statements are used, they are often illusions, but the fact that prescriptivists use fewer of them over time points to some awareness on their part with regard to ‘canonised’ usage items.

Frequency statements differ in three ways from recency statements in our data: they are far more widespread, are mostly used for grammatical usage items, and do not show a decline over time. It is noteworthy that frequency statements regularly occurred with the ‘old chestnuts’ of Dutch prescriptivism, those cases of (grammatical) variation that have been targeted for decades or even centuries (van der Meulen 2021). This implies that we are seeing a rhetorical dimension of prescriptivism: frequency statements belong to the prescriptive genre, partly because of tradition, partly to keep on creating a sense of urgency. The fact that there is a tendency to use high frequency terms further supports this hypothesis. More in-depth qualitative research should investigate to what extent prescriptive authors genuinely believe their frequency statements, and to what extent they use them, consciously or not, as rhetorical devices.

For the recency statements, the fact that most of them target unique linguistic variants initially pointed towards these statements possibly correctly identifying new variation. However, as our investigation showed, this was not the case: most of them constituted Zwickian illusions. Why the type:token ratio was so high remains unclear. One explanation for the statements that did occur more than once could again be intertextuality. The example of grootmacht, which was correctly identified as recent in 1932, according to our definition of recency, was inaccurately stated in 1964. Perhaps this later claim was not a genuine observation, but simply copied from the earlier work. As for the scope of our investigation, our results could be skewed by the fact that I did not look at recency statements targeting new meanings. However, it seems unlikely that prescriptivists would be able to detect sometimes very subtle shifts more accurately than more straightforward wordform occurrences. The same applies to statements observing recent developments in frequency: there is no reason to assume that this arguably more difficult type of observation would yield a higher degree of accuracy. Still, such observations would be worthwhile exploring further, given the availability of adequate usage corpora.

Almost all frequency statements, both for lexis and grammar, were also illusions. Although for grammatical usage items, our research was admittedly limited in scope, both for type and temporal coverage, the results were similar across the board. Moreover, while we may have expected earlier prescriptivists to be inaccurate in their assumptions because they lacked available data, this becomes less of an excuse for later periods, when increasing numbers of data sources become available. Nevertheless, the absence of a panoptic view, as Zwicky calls it, or more specifically the extrapolation of the personal linguistic experience to a language as a whole, could be the main factor explaining these results. One way of further testing this is by attempting to better match the linguistic reality of specific prescriptivists to the statement. An example of this can be found in the aforementioned van der Meulen and Rutten (2022), who, in their investigation of frequency terms used in a Dutch normative work from the nineteenth century, are able to approximate the type of language targeted by a particular prescriptivist, Matthijs Siegenbeek. Their results deviate from the present conclusions, in that the prescriptivist work under scrutiny does show broadly correct assumptions about frequency. Whether such an approach would yield different insights for the prescriptivists under investigation here remains the question. The difference in results may also be an effect of the time period: it seems possible that, with linguistic communities being smaller in the nineteenth century, the personal exposure to language of Siegenbeek was more representative of the language as a whole.

The present research provides solid empirical evidence for the existence of the Recency and Frequency Illusions. Of course, this by no means should be taken as evidence that all prescriptivists, let alone all language users, are always ‘wrong’ in their assessments of recency and frequency. Much more research should and can be conducted to further our understanding of the circumstances in which people make such evaluations. From a methodological standpoint, we consider our approach to be viable, and well worth extending further. Care should be taken with using the present research design however, as the interpretation of seemingly similar frequency terms may differ between different languages (Willems et al. 2020, p. 11). More generally, as much depends on the definitions of recency and frequency, it is important to further delve into what language users actually mean when they use these terms. We saw that sometimes a recency statement explicitly noted a recent increase, rather than a recent origin. Perhaps this is implicitly meant by other recency statements as well. Similarly, both types of statements may accurately observe the emergence or occurrence of a variant in a new genre, even when this is not mentioned explicitly. Perhaps more fine-grained qualitative work would show that prescriptivists do have better antennas for evaluating usage, given different parameters. However, when we take prescriptive statements at face-value, as I did in the present paper, and as language users perusing the prescriptive publications cannot help but do, we have to conclude that Zwicky was right: we are indeed illuded.