1. Introduction
In bilingual brains, lexical items of different languages are stored in the mental lexicon [
1]. When late sequential bilinguals, also multilinguals who have acquired their L1 language system from birth on and their L2 language system during or after adolescence, learn a second language, their brain has to be aware of the fact that there is already a language stored in that brain. Therefore, the bilingual brain needs a certain control mechanism, not only to prevent it from between-language interference, but also to provide access to the right language during two-language processing [
2]. This controlled language processing, which in the literature is often referred to as language control, takes place in a neural language control network, involving the prefrontal cortex (PFC), the anterior cingulate cortex (ACC), the inferior parietal cortex, and the caudate nuclei in the basal ganglia [
3,
4,
5]. The function of this neural network not only involves language control, but also implicates cognitive control in other domains: the primary processes of this network are (response) inhibition, updating information in working memory, and shifting of mental sets [
5,
6].
Indeed, only when two or more languages are simultaneously accessible and activated is cognitive language control needed. The Language Mode Hypothesis [
7], e.g., tries to point out when both languages are activated in the bilingual brain, and thus when language control is actually needed. The Language Mode Hypothesis assumes a continuum ranging from a purely monolingual context to a purely bilingual context. Any language context or communicative context, named ‘language mode’, can be seen as a point on this continuum. According to this theory, the L1 of a multilingual is always fully activated, but the more an actual language context approaches the bilingual endpoint, the more the L2 is activated and the more cognitive control the multilingual needs to avoid between-language interference. However, if the actual language context coincides with the monolingual endpoint, the L2 will not be activated at all and the multilingual will not need any cognitive control [
7].
The Language Mode Hypothesis, and especially the deactivation of the L2 in the monolingual mode, has received much criticism from a connectionist distributed learning perspective. Further research has revealed that multilinguals cannot turn off their L2 in a monolingual context [
1]. As a consequence, in bilingual brains, lexical items of all known languages are accessible and active, and are interacting with each other, even if only one language is being used at a given time point [
8,
9,
10,
11,
12,
13], both in language reception [
14] and in language production [
15]. That is why the bilingual brain constantly has to deal with a conflict situation: for each lexical item, it has to choose the right form, belonging to the right target language [
8]. Bilingual language reception and production therefore requires the constant involvement of the cognitive control system.
Information about the way in which a bilingual brain saves and processes information is essential for understanding the question of whether and how multilinguals inhibit irrelevant information. More specifically, there needs to be clarity on the way that bilingual brains save and process two languages in one and the same brain. Following a first account of the connectionist Bilingual Interactive Activation (BIA) model [
16], lexical items of the L1 and L2 are stored together in the bilingual brain. This would mean that every time the bilingual brain wants access to a word, all the words in the brain are activated, both the L1 words and the L2 words alike [
16]. The BIA model consists of four levels: the lowest level is the feature level, followed by the letter level, the word level, and finally the language level or node. In order to be able to recognize a word, the bilingual brain should go through these four levels bottom-up (that is, from the lowest level to the highest level). Within this model, however, there are also top-down processes; the recognition of words is not only a bottom-up process, according to the BIA model, but also an interactive process in which both bottom-up and top-down processes take place. When a multilingual sees a word, such as the word ‘table’, the feature level gets activated: all features that match the letters of the word that needs to be recognized. The letter T, for example, consists of the features | and ‾. The activated features then form letters at the letter level. Letters that do not match the activated features are now suppressed. When all letters are formed, the word level forms words that match those letters. The word order of the letters is always respected at the word level: words that contain the same letters as the target word, but in a different order, are suppressed. For the example ‘table’, the letters T + A + B + L + E have a fixed order. In a bilingual brain, the word level contains two lexica: one lexicon for the L1 and one lexicon for the L2. However, those two lexica are stored together in the bilingual brain, so that every word that contains the right letter in the right order gets activated, independently of the language system the word belongs to. When all matching words have been activated, the language level activates the right language tag of the target word. This language tag or node then suppresses all activated words that do not match the target language [
16].
The suppression takes place in the post-lexical phase: all words that match the activated letters, including the words of the non-target language, have already been activated in the bilingual brain. Only after this activation does the right language tag get activated, and only from then on, can the language tag suppress the previously activated words from the non-target language. However, a language tag cannot prevent words of the non-target language from being activated at the word level [
16].
Evidence for the BIA comes from the Neighborhood Density effect (ND). This effect is based on orthographic neighbor words, words that only differ from each other in only one feature. Two visual neighbor words thus differ from each other in one letter, like ‘bee’ and ‘see’. According to the ND effect, the more orthographic neighbor words there are for a word, the longer it takes to recognize the word [
17]. This effect also appears across different languages. The German word ‘Tee’ is also considered as a neighbor word of the English words ‘bee’ and ‘see’ [
18]. “It has been shown that the recognition of a word belonging to L1, the active language, can be significantly affected by a large orthographic neighborhood in L2, the non-active language” [
19] (p. 203). However, it is not clear whether this hypothesis still holds if there are two orthographically similar word forms with the same meaning.
The original BIA model only applies to the form recognition of individual words. That is why the BIA model was revised after a few years [
20]. The BIA + model considers both the linguistic and the non-linguistic (i.e., semantic) context. The BIA model did not refer to the semantic context, whereas the BIA + model does. According to this model, the semantics of a word can thus give feedback to the orthographic word forms [
21]. Words that agree both orthographically and semantically, such as cognates, would then, according to this model, be activated, and the semantic level would give semantic feedback to both orthographic word forms. Thus, if a cognate (e.g., the Dutch and German nouns
brief/
Brief) appears, two orthographic word forms (the Dutch and the German) are activated, and the semantic level gives feedback to both word forms, making the orthographic activation faster. According to the BIA+ model, cognates are then activated faster than non-cognates [
22]. But what would this mean for the activation of orthographic neighbors with the same meaning (i.e., cognates)?
In order to answer these and similar questions, Jacquet and French (2002) introduced a further adaptation of the BIA+ model, which they refer to as BIA++. In this model, they suggest (according to the idea of a connectionist distributional learning network) the existence of unified multilingual lexicons, for which the existence of a language node is not needed [
19]. We would like to hypothesize that according to such a model, orthographic neighborhood could have an effect, even when L2 learners have no knowledge of the target language in question, which would become apparent as a function of response inhibition control in a bilingual context, since this model encompasses a learning mechanism.
Controlled language processing in multilinguals could give them several cognitive advantages [
23]. In the literature, these advantages are often referred to as the bilingual advantage. The bilingual advantage can be explained by the fact that multilinguals constantly need cognitive control in order to prevent the bilingual brain from between-language interference [
8]. Several studies have shown that the bilingual brain is better trained in inhibiting irrelevant information compared to the monolingual brain [
24] or less proficient bilingual brains [
25]. More recent studies refer to the difference in the level of bilingualism to the bilingual advantage: the higher the level of bilingualism and the lower the age of acquisition of the L2, the higher the bilingual advantage [
26,
27]. However, some other studies did not find such a ‘bilingual advantage’ at all [
28,
29]. When looking at neuroimaging studies about this topic, it seems that bilingual brains are more efficient in dealing with interference: the brain network used by the monolingual brain is much bigger than the brain network used in the bilingual brain when dealing with interference, even if there is no difference in the behavioral level [
30].
A good way of testing the efficacy of the bilingual cognitive control system is by running a Stroop task [
31]. The Stroop task is a linguistic task measuring response inhibition control. In an original Stroop task, words are shown in a particular color. The words themselves can be color words or other nouns. Participants are asked to name the color in which the words are written. This task is easy when ‘neutral’ nouns are presented (‘control trials’, e.g., the word TABLE written in blue), and even easier when the color and the meaning of the word match (‘congruent trials’, e.g., the word BLUE written in blue). The task is much more difficult, however, when the color and the meaning of the presented word do not match (‘incongruent trials’, e.g., the word BLUE written in green).
Within response inhibition control, the Stroop task entails three effects: a facilitation effect, an interference effect, and a general Stroop effect. A facilitation effect occurs when the participant has to deal with a congruent trial: the time needed to name the color of the congruent trials is lower than the time needed to name the color of the control trials. The time needed to name the color of the incongruent trials, however, is longer than the time needed to name the color of the control trials. This effect is called the interference effect. An interference effect occurs because the automatic reading process and the color naming process are in conflict [
32]. Finally, the overall Stroop effect is the sum of the facilitation effect and the interference effect, which is the time needed to name the color of the incongruent trials minus the time needed to name the color of the congruent trials. The overall Stroop effect is mostly used when a Stroop task only contains congruent and incongruent trials, but no control trials.
Considering both behavioral and neuroimaging studies about the bilingual advantage, it is clear that there is a difference in executive functioning between the monolingual and the bilingual brain, at least in tasks that involve interference suppression. Possible causal factors leading to that difference, however, remain much more speculative. In the current study, we investigated two factors that can modulate the effect of bilingualism on cognitive control: cognateness on the one hand, and orthographic neighborhood on the other hand.
Previous research has shown that cognates, which are defined as identical words with the same meaning in different languages [
10], are processed faster by the bilingual brain than by the monolingual brain [
33]. Cognates not only have an (almost) identical spelling in different languages, but are also identical on the phonological level, having the same meaning in those languages [
10]. An example of Dutch-German cognates would be
nacht/
Nacht (night) or
vragen/
fragen (to ask). Because of the homologous meaning in different languages, cognates are processed significantly faster by bilingual brains: both items are supposed to be linked to the same semantic cue at the word level [
33]. In fact, interlingual homographs, defined as words with an identical spelling and an identical phonology but with a different meaning in different languages, are processed significantly slower by the bilingual brain compared to the monolingual brain (i.e., the brain that does not know the target language in question). Because of the different meanings in the different languages, according to the BIA+ model, both items would be linked to different semantic cues in the mental lexicon of the bilingual brain [
10]. Those differences between the monolingual brain and the bilingual brain would occur in all contexts, bilingual and monolingual language contexts alike [
1]. Within bilingual brains, cognates are processed faster than non-cognates, both in a bilingual context and in a monolingual context [
23]. Thus, the orthographic and semantic similarities are believed to have a facilitation effect on the bilingual brain [
10,
23]. In real life, however, most cognates have an almost identical, but no complete identical, orthography. The color words used in the present study also slightly differ in orthography. However, previous research affirms that those cognates follow the same tendency as completely identical cognates: the more identical the cognates, the bigger the facilitation effect [
23].
What is less clear is the effect that cognates might have on multilinguals who only speak one of the cognate languages. In this case, the words cannot really be considered cognates. Instead, we speak of orthographic neighbors. The question then is, does the similarity between two different languages have a bilingual advantage in terms of lower interference and higher facilitation effects, even if multilinguals only speak one of the two similar languages? Such an advantage could only be explained from a BIA++ perspective, since this model incorporates the possibility of cognates and orthographic neighbor words being part of the same unified multilingual lexicon, resulting in “a distributed (i.e., non-localist) encoding” for the words in each (new) language [
19] (p. 203). The advantage of this model is that it also includes a learning mechanism which is linked to this idea of distributed encoding. Word frequency is an important variable to be considered in distributed learning mechanisms, however. In this way, the BIA++ model is compatible with the Temporal Delay Hypothesis [
20], which states that the more frequently a certain word is used, the faster it is believed to be activated. Therefore, in general, the activation of a word in the L1 would be faster than the activation of a word in the L2, because the L1 word is used more frequently than the (in the case of foreign language learners, sometimes yet to be learned) L2 word. As a consequence, cognates and orthographic neighbors with the same meaning could have an effect in an L2 Stroop task, because the L1 version of the cognate resp. orthographic neighbor is activated faster than the L2 version thereof. Previous research with primary school children with Dutch as L1 and English as L2 also found a beneficial cognate effect in the L2, but not in the L1 [
34].
These issues could be dealt with by running a Stroop task in each of the cognate languages. On the phonological level, the similarity between two different languages has already been proven to have an effect on the bilingual brain: in a bilingual Stroop task (English and Japanese), the Stroop effects (i.e., the interference effect plus the facilitation effect) were bigger when the color words of both languages were phonologically similar [
13]. However, English and Japanese have a different orthographic system. In another study, Dutch-English cognates were used in a Stroop task to test the possible effects that language similarity can have on cognitive control (i.e., response inhibition) [
35]. The results of this study indeed showed a facilitation effect for the Dutch-English bilinguals, but the results were not compared to results of multilinguals who only spoke Dutch or English, combined with another language.
The question remains whether the orthographic similarity of two languages has an effect on multilinguals who do and do not speak both cognate languages and whether such an effect can be explained from the BIA++ model perspective of unified multilingual lexicons in foreign language learning in a distributed connectionist setting [
19] (p. 203). In the current study, we investigated the results of L1 Dutch-L2 German multilinguals, L1 French-L2 German multilinguals, and the multilingual group L1 Dutch without knowledge of German, using a bilingual Stroop task with color words that are cognates in Dutch and German, but that are not cognates in French and Dutch. All participants are said to be multilinguals, because complete monolinguals hardly exist. The aim of the current study was to investigate whether the similarity between the Dutch and the German color words only had an effect on the Dutch-German multilinguals, or also on the multilingual group L1 Dutch without knowledge of German. Against the background described above, the current study will address the following research questions:
What influence, if any, do cognates have on the cognitive control in multilinguals?
What influence, if any, do orthographic neighbors have on the cognitive control in multilinguals?
These research questions can be supplemented with the following sub-question:
Do and to what extent do Dutch speaking learners of German experience a cognitive advantage (in terms of response inhibition) compared to
As for the Dutch-German multilinguals, we predict that the Dutch-German cognates will have an influence on the Stroop effects. The interference effect, on the one hand, would be bigger with cognates than with non-cognates, because both meanings of the cognates would not correspond with the color of the word. This double contradiction would lead to slower reaction times. The facilitation effect, on the other hand, would be bigger with cognates than with non-cognates, because both meanings of the cognates would correspond with the color of the word. This double confirmation would lead to faster reaction times. Taken together, because of the similarity between the color words in German and Dutch, the Stroop effects would be bigger in Dutch learners of German compared to the Stroop effects in French learners of German.
When comparing the general Stroop effects within the different groups, the Stroop effects should be bigger in the L1-Stroop task than in the L2-Stroop task, because of a higher interference effect and a higher facilitation effect in the L1 than in the L2. A Stroop effect will only occur if a participant understands the language the color words are written in. Dutch speaking participants who do not speak German and for whom the words are only orthographic neighbors to be learned, would therefore experience no Stroop effects in a German Stroop task. Any Stroop effects in the German Stroop task for these participants could only be explained through an orthographic neighborhood effect.