Language learners, chess champions, and piano prodigies – insights from research on language contact and expert behavior

2.1 Entrenchment and experience

Entrenchment, one of the central mechanisms in Cognitive Linguistics, was first introduced to linguistic research by Langacker (1987, p. 59), who wrote that there is a “continuous scale of entrenchment in cognitive organization. Every use of a structure has a positive impact on its degree of entrenchment, whereas extended periods of disuse have a negative impact. With repeated use, a novel structure becomes progressively entrenched, to the point of becoming a unit; moreover, units are variably entrenched depending on the frequency of their occurrence.” This quote illustrates two important aspects of entrenchment: its close relation to usage and its different effects, such as affecting both strength of representation and holistic processing. The current section mainly focuses on the strength of representation; see section 2.3 about chunking for more information about entrenchment and holistic processing. Since its introduction to linguistic research, there has been a lot of research on the effects of entrenchment, generally showing that every time that speakers hear or use a word or a construction, its representation becomes increasingly entrenched, the word or construction therefore becomes activated more easily, and as a result is also more likely to be selected again (Blumenthal-Dramé, 2016; Bybee, 2011; Schmid, 2016b), resulting in “a feedback loop in which frequency comes to serve as both a cause and an effect of entrenchment” (Schmid, 2016b, p. 10).

The notion of entrenchment also plays an important role in research on language contact and language change (Backus, 2021; Bybee, 2011; Hakimov & Backus, 2021). To illustrate that, I briefly describe one of my research studies on Dutch-German language contact (Barking et al., 2022a). In that particular study, we compiled a corpus consisting of German e-mails written by nine native German speakers living in the Netherlands (total word count: 1,370,708 words; ranging from 13,760 to 513,654 words per speaker). These speakers worked for a holiday home rental company situated in the Netherlands and wrote these e-mails as replies to customer questions. Therefore, the content was mostly about holiday home rental, such as information about houses, booking requests, payment information, etc. The e-mails contained many examples of transfer, sometimes lexical as in (1) and (2) in which an employee told the customer that their payment was ‘outwardly due’ on January 30th or that the holiday home was ‘thicker to’ to the pool. In other cases, the transfer was more grammatical in nature, such as inserting the Dutch complementizer om or taking over the Dutch spelling as in (3).

To analyze the language transfer more systematically, we zoomed in on one specific case of transfer in these e-mails, the placement of prepositional phrases (PPs). PPs are placed more frequently in the so-called postfield position in Dutch than in German (i.e., after the main verb; see example sentence (4) based on Fitch, 2011, p. 372, who conducted a similar analysis in Pennsylvania German) instead of the middlefield position (i.e., before the main verb, see example sentence (5)). This might potentially result in covert transfer (Mougeon et al., 2005), that is, the speakers living in the Netherlands might over-use the postfield position in their German due to transfer from Dutch. Additionally, we collected data about participants’ language use by using the bilingual language profile (Gertken et al., 2014). This questionnaire combines questions about language history, language use, language proficiency, and language attitudes into a continuous measure, referred to as dominance score, which has been shown to be indicative of other measures of bilingual speakers’ language use (but see Titone & Tiv, 2023; Treffers-Daller, 2016, for a discussion of the limitations of these and similar self-reported measures).

The results showed that participants differed in the extent to which they used the postfield position (M = 10.46%, SD = 8.58, minimum: 3.34%, maximum: 26.42%) and that these percentages of postfield use were strongly correlated with their dominance score (r =. 82, p =. 007), indicating that they used the postfield position more frequently with increasing dominance in Dutch. These results illustrate a common finding in research on language contact: on the one hand, entrenchment results in less transfer (based on the conserving effect of frequency; Bybee, 2006, 2011), as evidenced by the speakers who still regularly speak German and who do not experience much language transfer from Dutch; on the other hand, entrenchment can also result in more transfer (based on “interference through entrenchment”, Backus, 2015, p. 30), as evidenced by the speakers who use a lot of Dutch in their daily lives and who, likely due to the increasingly entrenched Dutch words and constructions in their mental language representations, experience more transfer from Dutch. Based on this dual effect, De Smet (2016, p. 95) concludes that “entrenchment crucially underlies not only what is replicated but also what is newly created”, summarizing the complex way in which entrenchment influences our language use.

Shifting now to the research field of expert behavior, here too, the mechanism of entrenchment and its reliance on frequency and repetition plays an important role. In particular, researchers assume expertise to be organized in cognitive schemas, also referred to as expert schemas, which develop “through continual training, practice, and performance” (Dane, 2010, p. 583) and researchers therefore stress the importance of repetition of behavior for expertise to emerge. One of the most influential lines of research on this topic is by Ericsson on the effects of what he refers to as deliberate practice. In a pivotal study, Ericsson et al. (1993) asked professors at the Music Academy in Berlin to nominate violin students who had the potential for careers as international soloists (“best violinists”). They also selected students from the same department who were not nominated (“good violinists”) and students from the music education program that had lower admission standards (“music teachers”), all matched for age and gender to the “best violinists” group. They then asked these musicians to estimate the amount of practice per week for each year they had been playing the instrument. On average, at age 20, the “best violinists” had practiced more than 10,000 hours, the “good violinists” had practiced about 7,800 hours, and the “music teachers” had practiced about 4,600 hours. These and similar findings lead Ericsson and his colleagues to conclude that it is deliberate practice – and deliberate practice only – that results in expert performance. Ericsson has maintained this view over the past two decades (e.g., Ericsson, 2007, p. 4: “the distinctive characteristics of elite performers are adaptations to extended and intense practice activities that selectively activate dormant genes that all healthy children’s DNA contain[s]”) and it has been largely influential in the field. For example, his article from 1993 is one of the most cited articles in the psychological literature and many researchers, inspired by his studies, conducted similar research projects to study the effects of deliberate practice on expert performance.

Despite this significance, Ericsson’s view on deliberate practice has also met with harsh criticism from other researchers, especially in regard to whether practice should be considered the only factor influencing expert performance. For example, Macnamara et al. (2014) conducted a meta-analysis on expert performance in games, music, sports, education, and professions, and showed that, while in many of these domains, practice could explain a substantial part of the variance between participants (i.e., 26% for games, 21% for music, 18% for sports, 4% for education, < 1% for professions), there also was a lot of variance that is left unexplained. Hambrick et al. (2016) take up this and similar findings in their detailed overview paper ‘Beyond Born versus Made: A New Look at Expertise’ and discuss what factors other than deliberate practice might explain individual differences in expert performance. Figure 1 shows the resulting model, also referred to as Multifactorial Gene-Environment Interaction Model (MGIM). In this model, domain-specific performance – such as playing chess, playing the violin, or speaking a language – is directly related to domain-specific knowledge, which conceptually corresponds to what we refer to as speakers’ mental language representations in linguistic research. These representations are in turn influenced by domain-specific experience factors, basic ability factors, and personality factors. Interestingly, research on expert behavior has also extensively focused on how these factors interact with each other, for instance that ability and personality shape experience factors (e.g., you are more likely to do an activity that you are good at and that you enjoy), making it in turn more likely that someone achieves expert performance in the associated behavior. Ultimately, with this model, the researchers argue that it is crucial to focus on various factors like people’s experiences, their cognitive abilities, personalities and attitudes, and the interaction of all these factors to explain expert performance. I will return to this point in the discussion of this paper.

Figure 1:

Multifactorial Gene-Environment Interaction Model, illustrating possible influence of different types of factors on expertise, adapted from Hambrick et al. (2016, p. 43).

Research on expert behavior also focuses on the effects of non-exposure, by testing what happens when people stop practicing a certain behavior. One interesting research line in that regard is conducted by Woollett et al. (2009) on London taxi drivers. To become a taxi driver in London, taxi drivers have to undergo extensive training and testing over a period of 2-4 years in order to obtain an operating license. During that training, they learn the layout of 25 000 streets in London (see Figure 2 for an illustration of what these taxi drivers have to memorize) and thousands of places of interest. Additionally, the taxi drivers continuously practice how to navigate around central London, both during the training and also when they start working as actual drivers. In a series of experiments, Woollett et al. (2009) showed that this knowledge and practice results in some exceptional behavior, for instance in regard to their memory and even neurological changes that can be linked to their navigational practice and skills. In one particular study, the researchers wanted to test what happens when the taxi drivers stop all that practice, for instance because they retire. To do that, they invited a group of taxi drivers and retired taxi drivers (matched for age; years since retirement: M = 3.6 years) to their lab where they performed the so-called VR London navigation test. This test is based on the videogame ‘The Getaway’ and allows players to drive around a faithful representation of central London in VR (see Figure 2 for an example screenshot). Both the practicing and the retired taxi drivers were asked to navigate between different points in London. For each of the routes, the researchers determined the minimum length of that route and then calculated the difference between that length and the length of the routes that the taxi drivers drove during the VR experiment. The results showed that the routes of the retired taxi drivers were significantly longer than those of the practicing taxi drivers. Additionally, they performed worse on tests about their London layout knowledge, such as making proximity judgment between landmarks. Finally, the researchers even demonstrated that the previously observed neurological changes had largely disappeared for the retired drivers. Taken together, these results demonstrate the profound effects of the taxi drivers stopping their navigational practice due to retirement.

Already now, there are some clear links emerging between linguistic research and research on expert behavior, with both fields stressing the importance of usage/experience/practice for cognitive (expert) schemas to emerge and to become entrenched. In particular, frequent exposure to a particular behavior – for example frequent use of Dutch, frequent practice of the violin, frequently navigating around central London – strengthens representations associated with that behavior. Infrequent or potentially even non-existent exposure to the behavior results in a decrease of entrenchment (see again Langacker, 1987, p. 59: “extended periods of disuse have a negative impact [on entrenchment]”). The effects of this lack of exposure can in fact be quite far-reaching. For example, in the case of Dutch-German language contact, we see that the more participants become dominant in Dutch –and thus the more the Dutch words and constructions become entrenched for them relative to the entrenchment of their German counterparts – the more they experience transfer from Dutch. Building on this finding, in a follow-up experimental study, we even found that some speakers would not only use the transferred constructions such as *Wecker setzen, but that they would also accept them in a judgment task and sometimes, likely due to extended non-use, would even reject the conventional German counterparts such as Wecker stellen (Barking et al., 2022b; see also Verschik, 2006, and Bahtina et al., 2021, for similar findings in acceptability tasks of Russian-Estonian and English-Estonian language transfer). Similarly in the case of the London taxi drivers, their retirement resulted in a drastic decrease in performance on London navigation and knowledge tests and even in a reversal of the changes in their brain structures. Of course, more research is needed to test whether the concept of usage as it is used in linguistic research (i.e., exposure and own usage of words and constructions) and the concept of experience/practice as it is used in research on expert performance (e.g., Ericsson’s concept of deliberate practice; knowledge and skills due to daily taxi driving) do indeed trigger the exact same cognitive mechanisms. The links across the two fields as described here provide a good starting point for further exploration.

Figure 2:

Central London, UK, 2009 (left; Woollett et al., 2009, p. 1408) and a screenshot of the VR London navigation test (right; Woollett et al., 2009, p. 1408).

2.2 Entrenchment and interference

Other research, both in the field of linguistics and in the field of expert performance, has focused on the effects of entrenchment, particularly in terms of memory interference.

In research on language contact, including my own research on Dutch-German language transfer, memory interference plays a crucial role. In particular, research has shown that language transfer often happens unconsciously and automatically (Barking, 2024; Davidson, 2019; Nycz, 2016; Rodríguez-Ordóñez, 2019), as words and constructions from the contact language, likely due to their high entrenchment, become activated automatically and subsequently interfere with the selection of any alternatives in the speakers’ other language. For example, native German speakers living in the Netherlands might be so used to some Dutch words and constructions – such as achternaam ‘last name’, wekker zetten ‘to set an alarm’, or the Dutch postfield position – that they experience interference when speaking German and transfer these words and constructions, resulting in unconventional German expressions such as *Hintername, *Wecker setzen, or a marked frequency increase of the postfield position. This automatic nature of language transfer becomes especially evident in the case of transfer by speakers who actually want to avoid it. To study this, we conducted a number of focusgroup discussions among native German speakers in the Netherlands (Barking, 2024) and showed that many of these speakers hold rather negative attitudes towards language transfer. They think that they should be able to speak their native language German without any mistakes and they see Dutch transfer as a deviation from standard German and thus as something that should be avoided. However, zooming in on their actual language use during the focusgroup discussions, we observed many cases of language transfer, often because participants did not notice the transfer while speaking or because they no longer recognized a construction as transfer (see again also Barking et al., 2022b, and Barking et al., 2024, who showed that German speakers in the Netherlands often accept transferred constructions in an acceptability judgment task).

Research in the field of expert behavior also focuses on the effects of entrenchment. On the one hand, researchers have found advantages of expertise: due to the repeated activation of the expert schemas, activation of these schemas and their associated behaviors becomes automatic, allowing experts to multi-task, to quickly assess situations, and to take effective decisions (see Ericsson, 2006, for a detailed overview of the advantages of expertise). On the other hand, researchers argue that there might be a trade-off between expertise and flexibility. They argue that repeated activation of the expert schemas might lead to what they refer to as cognitive entrenchment (Dane, 2010), leading experts to focus on only one solution instead of considering potential alternatives, making them less flexible and less able to adapt to new situations, and overall resulting in less creative thinking (Dane, 2010; see also Sternberg, 1996, p. 347: ”One such cost is increased rigidity: The expert can become so entrenched in a point of view or a way of doing things that it becomes hard to see things differently.”). To test this experimentally, Woollett and Maguire (2010) conducted another experiment using London taxi drivers as a case study. This time, the researchers invited a group of practicing taxi drivers to the lab and again presented them with a VR environment of London (see Figure 2). However, the researchers had made some changes to the layout (see Figure 3), and they then asked the participants to drive around that altered layout and to commit it to memory. After the training phase, the taxi drivers were tested on their abilities to navigate around the altered layout, and they had to draw a sketch map. Comparing their performance to their performance on a similar task, in which they had to learn the layout of an area that was completely new to them, the researchers showed that the taxi drivers performed significantly worse on this task. Based on these findings, Woollett and Maguire (2010) hypothesize that the drivers experienced difficulties inhibiting their strong expert schemas of the actual London layout and that these expert schemas competed and interfered with the new representations that the taxi drivers tried to memorize during the experiment. As such, this study provides experimental evidence that in some cases, expertise might indeed hinder performance.

Figure 3:

Map of London. Left: London as it is normally. Right: A map showing existing London integrated with ‘new’ London, where modifications are depicted in red. Note that participants never saw these maps, Woollett and Maguire, 2010, p. 569.

Again, these findings are similar to linguistic research, particularly research on language contact. For example, my own research on Dutch-German language contact showed how Dutch words and constructions can become so entrenched for the native German speakers living in the Netherlands that they become activated automatically, similarly to how the highly entrenched representations of the layout of central London become activated automatically for the London taxi drivers during the VR experiment. This activation in turn can result in memory interference, as evidenced both by language transfer (i.e., speakers transferring the activated Dutch words and constructions to their German, resulting for example in an over-use of the postfield position or in innovative constructions such as *Hintername or *Wecker setzen) and by the taxi drivers’ difficulties learning the altered layout of central London. Looking at these similarities, especially also in combination with the similarities already discussed in section 2.1 that both research fields stress the importance of usage/experience/practice, it indeed seems that the cognitive mechanism of entrenchment plays a central, often similar role in both linguistic research and research on expert behavior. These results therefore echo Langacker’s (2016, p. 39) claim that entrenchment is “not specific to language but a general phenomenon observable in any kind of learned human activity” and that it might therefore provide “the opportunity to establish a new meeting ground for psychology and linguistics” (Schmid, 2016a, p. 3).

2.3 Chunking

Similar to entrenchment, the cognitive mechanism of chunking is also assumed to play an important role in human cognition. In fact, in their overview paper ‘Chunking mechanisms in human learning’, Gobet et al. (2001, p. 236) call it “one of the key mechanisms of human cognition […] where each chunk collects a number of pieces of information from the environment into a single unit” and they argue that in doing so, it forms an important link between “the external environment and internal cognitive processes”. When introducing the concept of chunking in their overview paper, the researchers refer to a wide range of studies, including research on chess (see below for more information about chunking in chess research) and interestingly also on language acquisition. Similarly, linguists sometimes refer to other areas of human cognition when introducing the concept of chunking in their theories (see for example Langacker, 2013, p. 16 who argues that similar processes to language learning are observed when “learning to tie a shoe or recite the alphabet: through repetition or rehearsal, […] a structure undergoes progressive entrenchment and eventually becomes established as a unit”). Such close links across research fields can also be observed when looking at linguistic research and research on expert behavior. To illustrate that, have a look at the quotes in (6) and (7).

The first quote is about chunking in regard to our language use, making a distinction between holistic processing on the one hand (i.e., a frequently used collocation like good morning is likely retrieved as a unit) and an online, computational procedure on the other hand (i.e., combining the words on the spot, for example, good + morning, which happens when a given expression is not (yet) stored as a chunk). The second quote, coming from the field of expert behavior, describes the so-called Instance Theory and it is applied by Bilalić et al. (2008a) to chess players, who they argue to first rely on “crude, slow and effortful strategies” to process a given chess position, but who over time store “specific instances” of various chess positions including what moves to do next. Eventually, these “instances start winning the race against general algorithms” of calculating every possible move, that is, the chess players start to rely on them to make quicker – and often, although not always – better decisions (see below for a more detailed description about chunking in chess). Both quotes, despite coming from two different fields, thus stress the tension between “an online, computational procedure”/ “a general algorithm” and processing based on specific instances in a holistic manner,^[7] already illustrating the close links between the research fields when it comes to the cognitive mechanism of chunking.

Further zooming in on linguistic research, there is ample evidence that chunking greatly influences our language processing and production. For instance, research has shown that processing of a chunk proceeds automatically once started (e.g., people activate the next words in a chunk before they have been uttered, helping them to efficiently process their speech input; Schmid, 2016b), that chunks are more likely to be affected by processes of language change (e.g., frequent chunks such as going to become processed as one unit potentially resulting in phonetic and morphological reduction as in gonna; Bybee, 2011), and that a lot of language contact effects seem to rely on transfer of entrenched chunks rather than the general pattern (e.g., Doğruöz, 2014; Doğruöz & Backus, 2009). I observed similar effects in my own research on Dutch-German language contact. For example, analyzing the corpus consisting of German e-mails written by native German speakers living in the Netherlands for their placement of prepositional phrases (Barking et al., 2022a; see explanation in section 2.1), we were also interested in the specific lexical instances that the speakers would use in combination with the postfield position. We therefore noted the verb, the preposition, and the verb-preposition-combination for each PP use and ran collostructional analyses (Gries, 2007; Stefanowitsch & Gries, 2003) to test which lexical elements were most attracted to the postfield position (Barking et al., 2022c). Zooming in on some of the highest attraction values, we found that many of them were the result of a single fixed expression. For example, sentences (8; prepositional phrases in the postfield position marked in bold) to (10), which were written by the same speaker, all make use of the same expression to start the sentence. This suggests that this speaker does not experience interference from the Dutch postfield position every time that she uses this expression, but rather that this has become a fixed multiword unit for her.

Chunking also plays an important part in research on expert behavior. In fact, researchers argue that this process is what solves one of the main contradictions in the research field: that, on the one hand, research has shown people to have a limited working memory capacity, but that, on the other hand, experts are able to quickly process highly complex situations (Van Merriënboer & Sweller, 2005). For example, some of the first research in the field focused on chess players (De Groot, 1965; Chase & Simon, 1973) and presented them with different chess positions (see Figure 4 for an example). Some positions were taken from tournament games, others were just random positions. Participants saw the positions for around ten seconds and were then asked to recall them. The results showed that chess players at and above master level recalled the positions almost perfectly when presented with positions from tournament games, but they performed only slightly better than novice players when presented with random positions (Chase & Simon, 1973). To explain these and similar findings across a variety of expert performances (Larrick & Feiler, 2015; Nee & Ward, 2015), researchers make use of the mechanism of chunking: they argue that people identify patterns in their experiences, group them together in their memory, and start to process them as holistic units, even though they might internally consist of many different elements such as many different chess pieces on a chess board. As such, this chunking process can explain why experts are able to quickly process and respond to complex situations despite their limited working memory capacity (Van Merriënboer & Sweller, 2005).

Figure 4:

Types of positions typically used in chess memory research. A game position taken from a masters’ game (left) and a random position obtained by shuffling pieces of a game position (right).

Interestingly, research on expert behavior often discusses the process of chunking in close relation to experience (see also the quote by Langacker, 1987, at the beginning of section 2.1 for a similarly close reasoning between chunking and entrenchment in linguistic research), both in relation to the emergence and the effects of chunks. For the emergence of chunks, Gobet (2005, p. 192) argues that “a direct consequence of chunk-based theories is that the time at task is essential to gain knowledge, as each process in the growth of the discrimination network [referring here specifically to a network of chunks] […] is subject to a time cost. Thus, in learning algebra or becoming a violin virtuoso, time must be invested.” To test this empirically, research has for example investigated the chunk inventory of chess players, showing that experienced chess players have stored both more and larger (i.e., containing more chess pieces) chess positions compared to novice chess players. In relation to the effects of chunking, researchers argue that chunks allow people to make quicker and better decisions. This is because people also include information about how to respond in their chunk inventory, resulting in expert schemas to become “abstract, prototypical maps or mini-recipes regarding how to respond” in a given situation (Nee & Ward, 2015, p. 2). For example, asking their participants to think aloud, Chase and Simon (1973) showed that experienced chess players often relied on heuristics to evaluate a chess position and to decide on their next move, whereas less experienced players were hindered by a step-by-step approach.

At the same time, just as was the case for the cognitive mechanism of entrenchment (see section 2.2), researchers have also observed negative consequences of chunking, often referring to the phenomenon of Einstellung (set) effect (Luchins, 1942), which can be explained with another example from chess research. In this experiment, Bilalić et al. (2008a) presented chess players at different levels (Grand Masters: N = 6, International Masters: N = 6, Masters: N = 11, Candidate Masters: N = 11) with chess problems such as the ones presented in Figure 5 and asked them to find the shortest move sequence resulting in a check mate. The first chess problem offered two solutions: the well-known theme of so-called smothered mate that results in check mate in five moves and a less familiar solution that results in check mate in three moves (see Figure 5). Most pieces in the second chess problem are placed in the same way, but with the black bishop on a different square, smothered mate is no longer possible, and the players needed to find the other solution instead. The only difference between the two problems was thus whether or not a familiar solution was available, yet the results across the two problems were very different: all Grandmasters, but only 50% of the International Masters, 18% of the Masters and none of the Candidate Masters found the shorter solution in the first chess problem, even though they were all able to find it in the second one. Based on the players’ think-aloud protocols, the researchers conclude that the smothered mate solution was quickly spotted by all players and that they were then “trapped by the immediate appeal of the familiar solution” (p. 85), failing to find the more optimal one. In a follow-up study, Bilalić et al. (2008b) also added eye-tracking data, showing that, even when players were trying to search for a better solution, they remained fixated on the squares that were critical for the familiar solution, illustrating the powerful blocking effect of entrenched chunks, in this case, the entrenched smothered mate chunk.

Figure 5:

The 2-solution (critical) and 1-solution (extinction) problem (White to play). In the 2-solution problem the well-known smothered mate solution is possible: 1. Qe6+ Kh8 2. Nf7+ Kg8 3. Nh6++ Kh8 4. Qg8+ Rxg8 5. Nf7#. The shorter solution is: 1. Qe6+ Kh8 (if 1... Kf8 2. Nxh7#) 2. Qh6 Rd7 3. Qxh7#. In the 1-solution problem the black bishop has moved from c6 to h5. This disables the smothered mate solution but still allows the shorter solution (Bilalić et al., 2008a, p. 83).

Overall, when it comes to the cognitive mechanism of chunking, there are some clear links between linguistic research and research on expert behavior, with both fields assuming a similar chunking process: people first notice patterns in their experiences, for example frequently co-occurring words or chess pieces that are placed similarly in relation to each other across various chess positions, and they then store these patterns as chunks. As a consequence, these chunks are processed as holistic units, allowing both fast and effortless production (e.g., in the case of one of the native German speakers living in the Netherlands who likely produced the expression “Wie wir entnehmen können aus unserem System…” in a relatively automatic manner) and processing (e.g., in the case of the experienced chess players who were able to process and remember complex chess positions in just a number of seconds). At the same time, this rather automatic production and processing has also been shown to come at a cost, specifically when the activated chunks are not the most appropriate solution in a given situation. For example, for the native German speakers living in the Netherlands, transferred chunks might become activated and used, even when those speakers want to avoid language transfer. Similarly, chess players might not find the optimal move in a chess position because they focus too much on a familiar solution instead.

3.1 Potentials for theory building

Overall, the discussion of research on language contact and on expert behavior showed that there are a lot of similar patterns and processes, such as (cognitive) entrenchment and its relation to frequency and repetition of behavior, people observing patterns in their experiences and chunking them together, behavior becoming automatic due to repeated exposure, or different schemas competing and interfering with each other. At the same time, the description of the two research fields also revealed some slight, but important differences, which I believe to be areas in which the research fields can benefit most from each other.

3.2 Potentials for theory testing

We can also think about the shared links across research fields as a way to test whether the cognitive mechanisms underlying our language use are domain-general and thus apply to behavior beyond language. As argued in the introduction, this is one of the most important assumptions in Cognitive Linguistics, but ultimately, it remains an assumption and we need to formulate falsifiable hypotheses to test it (see also Divjak and Milin, 2023, p. 10: “Another thing we need to get better at is generating testable predictions and then actually testing them.”).

This is actually more difficult than it sounds, which also became evident in my own research on Dutch-German language contact. As described in the previous sections, one of my main focus points was to investigate the cognitive mechanisms driving speakers’ language transfer. To do that, I used these mechanisms to formulate several hypotheses and I designed different experimental studies to test them. In most cases, however, my hypotheses were actually not confirmed. For example, based on the cognitive mechanism of entrenchment, I hypothesized that native German speakers living in the Netherlands would over-use the German complementizer um due to transfer from Dutch, in which the complementizer is used much more frequently. I expected this to be the case for both constructions in which the use is already possible in German and in new, innovative constructions (Barking et al., 2024; see also (3) for an example). While this was indeed the case for the innovative constructions, my participants were in fact less likely to use the complementizer in existing German constructions compared to a control group of German speakers not in contact with Dutch. Clearly, my hypothesis was thus rejected. However, even though I had based this hypothesis on entrenchment, one of the core mechanisms within Cognitive Linguistics, this rejection did not necessarily challenge the theory itself. Instead, I showed that social factors, which are explicitly included in many usage-based models (e.g., Backus, 2021; Dabrowska, 2020), could explain this finding, as speakers with negative attitudes towards transfer engaged in hypercorrection and intentionally overruled entrenchment-related processes. As such, this example illustrates how flexible Cognitive Linguistics can be and the range of linguistic phenomena that it is therefore able to explain. At the same time, this flexibility also means that it is really difficult to actually test the theory, that is, to find falsifiable hypotheses that – if rejected – would actually challenge Cognitive Linguistics itself.

Importantly, the shared links across research fields, for example the shared links between research on language contact and expert behavior, might present the perfect opportunity to solve this problem and provide starting points for hypotheses that directly test some of the core assumptions of Cognitive Linguistics, specifically the assumption that the cognitive mechanisms driving our language use are domain-general. As already argued in the introduction, the reasoning here is simple: if these mechanisms are indeed domain-general, then findings regarding these mechanisms in other areas of human cognition, for example findings in the field of expert behavior, should hold in similar ways for our language use. One way of testing this is to take prominent studies in the field of cognitive psychology and to ‘translate’ them to linguistic research, for example as is done in the research study by Ibbotson et al. (2012) which conceptually replicated Franks and Bransford’s (1971) classic study on categorization (see section 1.0). Similarly, we can think about the research field of expert behavior and its studies on the underlying cognitive mechanisms of entrenchment and chunking. For example, we could test whether speakers are more likely to remember existing versus randomized chunks (conceptually replicating the study by Chase and Simon, 1973, on chess players) or whether the Einstellung (set) effect (Luchins, 1942) also occurs in regard to our language use (conceptually replicating the study by Bilalić et al., 2008a). In doing so, we can test whether findings regarding cognitive mechanisms – such as entrenchment and chunking – hold in similar ways across different domains of human cognition and as such test the assumption whether the cognitive mechanisms driving our language use are indeed domain-general.

Please note here that there is an important caveat to this set-up. As argued above, language – developed as a system for social interaction – is inherently social, which means that it is impossible to ever truly study (domain-general) cognitive mechanisms in isolation (Backus, 2021; Dąbrowska, 2016, 2020; Schmid, 2016b). This in turn raises the question as to how similar we should actually expect cognitive effects in language and other cognitive domains to be – even when the underlying mechanisms are domain-general –, precisely because language is so social and thereby potentially creates a very different ‘working environment’ for the cognitive mechanisms to operate in. A similar argument is raised by Titone and Tiv (2023) regarding the much-contested bilingual advantage, that is, the idea that bilingual speakers constantly have to suppress one of their languages and through that practice achieve advanced executive control in general. Research on this advantage has produced very mixed results, which Titone and Tiv (2023) argue should not surprise us. In particular, they reason that these findings might reflect that “language use really is socially-rooted, as certain cognitive strategies (e.g., proactive control) may be effective in responding to certain social-environmental demands, but certainly not all” (p. 12). Applying this reasoning to the comparison of cognitive effects across different domains, we too have to take into account the different social dynamics in these domains and how they affect what any cognitive effects might look like. Note for instance how the results in the study by Ibbotson et al. (2012) slightly differed from the conceptually replicated study by Franks and Bransford (1971). In particular, participants in the original shape study were more likely to believe that they had seen prototypical shape combinations, which they had not seen previously, than less prototypical combinations that they had in fact seen during the training phase. Similarly, the participants by Ibbotson et al. (2012) often believed to have seen prototypical transitive sentences during the test phase, but not more so than less prototypical sentences that they had seen. Based on these findings, Ibbotson et al. (2012) conclude that the prototype effect might in fact be more pronounced for shapes than for language, illustrating the complexities involved in comparing cognitive effects across different domains. Clearly, more conceptual as well as empirical research is needed to further explore how to best test the core assumption that the cognitive mechanisms driving our language use are indeed domain-general, as such targeting the heart of Cognitive Linguistics and allowing us to directly test the theory itself.