Language and Cognition in Bilinguals and Multilinguals_ An Introduction

386 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS bilingualism in children. A study by Peal and that the monolinguals should perform better than Lambert (1962) on the relation between child the bilinguals on the tests that measured verbal bilingualism and intelligence marks the turning intelligence, and that monolingual and bilingual point from regarding bilingualism a lamentable children should not show any difference on the human condition to avoid whenever possible tests measuring nonverbal intelligence. (which it generally cannot be) to a state that, under specific conditions, is beneficial for language As observed by these and many later authors competence and cognitive functioning in general. (e.g., Hakuta & Diaz, 1985; Reynolds, 1991), a plausible reason for the deviant results is The participants in the Peal and Lambert that many of the earlier studies lacked experi- study were 10-year-old children from middle-class mental rigor, for instance by not matching the French schools in Canada’s Montreal region monolingual and bilingual groups on age, socio- (where at the time about 80% of the population economic status, or amount of education. Other was francophone) who had been selected on the infelicities occurring in the earlier studies were basis of careful screening. The screening involved that the tests were not always appropriately a number of tests to determine whether a child standardized or that the bilingual participants should be considered French monolingual or fully were not balanced and were tested in their weaker balanced French–English bilingual. Children who language. If bilingual test-takers experience dif- could not unambiguously be classified as mono- ficulties with one of their languages and verbal lingual or balanced bilingual were excluded as intelligence is measured through this language, participants. (Incidentally, “monolingual” in this what else could be expected than lower scores on study did not mean the children in question had verbal intelligence? Similarly, it has been argued no knowledge whatsoever of English, and given (Cummins, 1976, p. 23) that some minimal the fact they were residents of a country where threshold of L2 proficiency should exist for English is the national language it is likely that bilingualism to become advantageous for cogni- they had picked up quite of bit of English in the tive functioning: course of their lives.) The selected children were administered a number of tests that measured . . . there may be a threshold level of L2 their verbal and non-verbal intelligence. Informa- competence which pupils must attain both tion on a number of background variables such as in order to avoid cognitive disadvantages the children’s and their parents’ attitudes towards and allow the potentially beneficial aspects English and French, how well they did in school, of becoming bilingual to influence their their age, sex, and socioeconomic class was also cognitive functioning. Those aspects of collected. bilingualism which might accelerate cogni- tive growth seem unlikely to come into effect Contrary to the authors’ expectations, the until the child has attained a certain bilingual children performed significantly better minimum or threshold level of competence than the monolingual children on the vast in his second language. Similarly, if a child majority of all tests and subtests, both those that [. . .] attains only a very low level of com- measured verbal intelligence and the ones meas- petence in his second language, his inter- uring non-verbal intelligence. The bilingual chil- action with the environment through the dren outperformed the monolinguals in concept medium of that language [. . .] is likely to be formation and in tasks that required mental flexi- impoverished. bility, and bilinguals showed a more diversified set of mental abilities than the monolingual controls. It may have been the case that the earlier studies These results also held when differences between have generally tested children whose linguistic the two groups in socioeconomic class, age, and competence in L2 happened to be below this sex were controlled. These results came as a big threshold level. Finally, the participants’ surprise to the investigators because a thorough bilingualism had occasionally been determined in review of the literature had led them to predict

7. COGNITIVE CONSEQUENCES OF BILINGUALISM 387 rather unorthodox ways: In some studies it had a limbo where neither language is useful as a tool been determined on the basis of where the parti- of thought and expression—a type of “semilin- cipants’ parents were born and in one case it was gualism” (Lambert, 1981, p. 12). In conclusion, even done by simply looking at the participants’ it appears that the bilingual French–English names (Pintner, 1932; in Peal & Lambert, 1962). children in Peal and Lambert’s (1962) study out- Even though some methodological weaknesses performed their monolingual French-speaking remained (see Hakuta & Diaz, 1985, for a discus- peers in intelligence because they had exploited sion), in all these respects Peal and Lambert’s the opportunity, provided by their environment, investigation compared favorably with the earlier to become additive bilinguals. The reason why ones. they, and not their monolingual peers, had exploited this opportunity presumably relates Looking for the reason why for the first time to the more favorable attitudes towards English the evidence clearly pointed to a beneficial effect and the English-Canadian community of these of bilingualism on intelligence, later publications children and their parents (see Peal & Lambert, (Lambert, 1977, 1981) highlighted one potentially 1962, for a substantiation of this claim). crucial feature of Lambert and Peal’s study: In the Montreal region both English and French are However, an alternative explanation of the socially valued and respected languages, English results of Peal and Lambert (1962), of which being the national language and French the the authors were well aware themselves, must be home language of the majority of the population. considered: that the relation between bilingualism Therefore the acquisition of English by franco- and intelligence had worked the other way phone children in this region does not involve around. The more intelligent children may have the risk of the home language French getting been the ones to become bilingual, and because corrupted as a consequence of a social pressure they were more intelligent they obviously per- not to use it. In other words, the learning of formed better on the intelligence tests. Generally, English by French-Canadians does not involve a suitable way to determine the direction of the risk that their L1 French competence is causality between two variables is to design a lowered but truly enriches their linguistic reper- longitudinal study in which one of two matched toire with a new language; it adds a language groups receives some treatment over an extended onto one that does not suffer a cost (although, period of time and the other does not receive this as we have seen on pp. 361–371, it may change treatment. To assess the effect of treatment, through its contact with the new language). after some time performance of both groups on Lambert called this form of bilingualism additive some task or set of tasks is compared. If the two bilingualism, and it is this form of bilingualism groups perform differently on this task, the spe- that he argued to be advantageous for cognitive cific treatment is likely to have caused this functioning. difference. The counterpart of additive bilingualism is A small set of longitudinal studies suggests subtractive bilingualism, a form of bilingualism that bilingualism can indeed improve cognitive “experienced by ethnolinguistic minority groups functioning. One of these studies (Scott, 1973, in who, because of national and educational policies Lambert, 1981) concerned two groups of English- and social pressures of various sorts, feel forced Canadian children. At the onset of the study the to put aside or subtract out their ethnic languages children in one group (the treatment group) had for a more necessary and prestigious national been given the opportunity to become function- language” (Lambert, 1981, p. 12). In this form of ally bilingual through immersion schooling in bilingualism the L2 gradually replaces the L1. French over a couple of years, whereas the Because language serves to buttress thought, this children in the other group (the control group) form of bilingualism, weakening the native lan- had not been given this opportunity but received guage, is detrimental for cognitive functioning their schooling exclusively in English, the for reasons that “it usually places youngsters in conventional school language in Canada. The

388 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS children in both groups were equated on IQ and public schools in Connecticut. The children were social class background at the onset of the study. all tested twice (at Time 1 and Time 2), with an The second study concerned an experiment in interval of 6 months in between the two test ses- which a random selection of schools in an area in sions. The tests included measures of their relative New England, the home of many francophone linguistic ability in Spanish and English and a Americans, were permitted to offer part of their measure of non-verbal cognitive ability. The elementary curriculum in French, the children’s measures of relative ability in the two languages home language. A second group of schools, provided an estimate of degree of bilingualism; with children of similar intelligence and socio- that is, whether or not the children were balanced economic backgrounds and, again, French as bilinguals and, if not, how large the proficiency their home language, offered an all-English difference between their two languages was. The curriculum. When tested a number of years later, investigators did not include a monolingual con- in both studies the group of children who had trol group but were interested to find out whether received (part of) their schooling in French, and degree of bilingualism at Time 1 predicted per- had thus become functionally bilingual, out- formance on the cognitive task at Time 2 or performed the other group on tests that index whether, instead, performance on the cognitive some aspect of cognitive functioning, such as a task at Time 1 predicted degree of bilingualism at math test or a test of divergent thinking. Interest- Time 2. The data suggested that the relation ingly, the children schooled in French also between bilingualism and cognitive functioning showed superior English language skills. works both ways, but “the model claiming degree of bilingualism to be the causal link was more In a third longitudinal study (Kessler & Quinn, consistent with our obtained data than the 1980, in Lambert, 1981), a group of Hispanic- model claiming cognitive ability to be the causal American children were given the opportunity variable” (Hakuta & Diaz, 1985, p. 340). In other to learn subject matter in their home language words, the chances that a child becomes bilingual Spanish and their performance on a number of (in a bilingual environment) are greater if the tasks was compared to that of a socioeconomic- child has strong cognitive skills than when it is ally more privileged group of middle-class, endowed with lesser cognitive skills but, more so, white, monolingual English-speaking children of bilingualism is the cause of enhanced cognitive the same age. Both groups received an extensive ability. training program in science inquiry through dis- cussions, films, and hypothesis testing. Despite An important conclusion to draw from all being less privileged, the pupils in the Hispanic- these studies is that allowing ethnic minority American group outperformed those in the children to nurture their home language (instead English-American group in problem-solving cap- of gradually having it replaced by the L2) and acity, generating hypotheses of a much higher thus to “transform their subtractive experiences complexity and quality. Furthermore, while with bilingualism and biculturalism into additive generating the hypotheses, the Spanish–English ones” (Lambert, 1981, p. 13) improves their per- bilinguals used more complex linguistic structures formance on a varied set of cognitive tasks. That than the English monolinguals. Thus, once again minority children in circumstances of additive it appeared that bilingualism is the cause of bilingualism even outperform monolingual enhanced cognitive functioning. children who are, in socioeconomic sense, more privileged, is an especially appealing result. Finally, Hakuta and Diaz (1985) tackled the Furthermore, all these studies suggest that the question of whether bilingualism improves cogni- children’s bilingualism was the cause of their tive functioning or whether superior cognition relatively high levels of cognitive performance. causes bilingualism in a slightly different way. Finally, there is some evidence to suggest that They tested over 100 Spanish-dominant children intelligence/high cognitive ability helps to become aged between about 4 and 8 years and all enrolled bilingual. in a bilingual education program of a number of

7. COGNITIVE CONSEQUENCES OF BILINGUALISM 389 The beneficial effect of bilingualism on the “repetitiveness and redundancy in grammar cognition: Further evidence and compensate for inexperience” (Ben-Zeev, 1977a, limitations p. 35). Similarly, she assumed that the vocabulary deficit will also diminish or disappear when, over Since Peal and Lambert’s (1962) seminal investi- the years, language experience increases, but gation, in many studies detailed knowledge because of the large size of a language’s vocabu- has been gathered about the type of tasks on lary one might expect the vocabulary deficit to which bilinguals excel, and a number of factors last longer. have been identified that mediate and restrict the positive effect of bilingualism. But prior Other effects of bilingualism that from the to elaborating on some of these it should be fractional view of bilingualism may be considered mentioned that bilingualism, even bilingualism of detrimental have been presented in the earlier the additive form, brings about some minor sections of this chapter. An example is the evi- drawbacks as well. This is especially the case dence of the ubiquitous accents (in phonology, when the fractional view of bilingualism is taken grammar, and semantics) in the language use of (see p. 340), which departs from the (flawed) bilinguals in both their L1 and L2, which result assumption that a bilingual is two monolinguals from interference due to co-activation of the in one person. For instance, several authors non-targeted language and/or from the merging have shown that, as compared to the monolingual of L1 and L2 representations (p. 364). But from norm, bilingual children have relatively small the holistic view of bilingualism (see p. 340), which vocabularies in their separate languages (see, acknowledges that the use of two languages pro- e.g., Ben-Zeev, 1977a, 1977b; Bialystok, 1988; duces a specific speaker-hearer in his or her own Eviatar & Ibrahim, 2000; Martin-Rhee & right (e.g., Cook, 1991; Grosjean, 1989), such Bialystok, 2008; Rosenblum & Pinker, 1983; results are non-normatively regarded as reflecting but see Pearson et al., 1993). The reason presum- just differences between monolinguals and ably is that having two words for one and the bilinguals; differences that obviously result from same concept and using both implies that the bilingual experience. What is more, from the bilinguals use each single word less often than holistic point of view, what at first sight seems monolinguals do. As a consequence it has been to be a drawback of bilingualism may be recog- less well learned. This may also be the reason why nized as an advantage by a slight change in word retrieval in adult bilinguals is slower and perspective. For instance, one may become more effortful than in monolingual controls, as aware of the fact that the two relatively small suggested by longer picture-naming latencies vocabularies bilingual children possess in each (e.g., Gollan et al., 2005; Ivanova & Costa, 2008; of their languages add up to the mastery of a Mägiste, 1979), lesser verbal fluency (Gollan, range of concepts that outnumbers the concepts Montoya, & Werner, 2002), and more “tip-of-the- known by age-matched monolingual children. tongue” states (Gollan & Silverberg, 2001) in This is because a bilingual child’s lexical- bilinguals (but these effects may also result from conceptual knowledge store includes both con- greater lexical competition during word produc- cepts shared between the languages as well as tion in bilinguals than in monolinguals; see concepts specific to each separate language (cf. p. 362). Umbel, Pearson, Fernández, & Oller, 1992). Fur- thermore, from a holistic point of view one might Ben-Zeev (1977a) hypothesized that the not only have an eye for the potentially harmful relative lack of experience with each separate effects of having two words for one concept in language may also be reflected in relatively terms of retrieval speed, but also for its poten- limited knowledge of grammatical rules, in each tially beneficial effects. It is to these, and other, language separately, in bilingual children, but that advantageous effects of bilingualism that I will this effect might be very short-lived because now turn.

390 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS Bilingualism and metalinguistic awareness Other aspects of cognitive performance may then benefit from this increased level of meta- So what are the tasks that have revealed salutary linguistic awareness in bilinguals, either indirectly effects of bilingualism and what aspects of through superior linguistic abilities, or directly, superior cognitive functioning are tapped and for instance because attention to structure is indexed by these tasks? Many researchers in this beneficial for cognitive functioning in general. field of study, and especially those who studied According to Bialystok (e.g., 1988, 1992, 2001a, childhood bilingualism, have used tasks that are 2001b, 2004) metalinguistic tasks appeal thought to index some form of metalinguistic to two cognitive processes that she calls the awareness; that is, the ability “to reflect on and “analysis of representational structures” and manipulate the structural features of language “control of attentional processing” (“analysis” independent of meaning” (Bruck & Genesee, and “control”, in short) and it is in the latter 1995, p. 307), “to look at language rather than that she believes bilinguals excel (see below for through it to the intended meaning” (Cummins, details). 1978, p. 127), or to think about language rather than to think through language (Bialystok, 1992). Word awareness in bilingual and monolingual Some metalinguistic tasks probe the test-taker’s children phonological awareness; that is, “the ability to reflect on and manipulate sublexical phono- The tasks that have been used to measure aspects logical units such as syllables, onsets, rimes and of metalinguistic awareness are too numerous phonemes” (Bruck & Genesee, 1995, p. 308). To and varied to cover all of them here. Instead, do so requires an understanding of the sound by way of illustration I will confine myself to units that make up a word. Other meta- presenting those that were used in two influential linguistic tasks tap syntactic awareness; that is, early experimental studies (Ben-Zeev, 1977b; the language user’s ability to reflect on and Ianco-Worrall, 1972), which were among the manipulate syntactic structures. A further group first to reveal beneficial effects of bilingualism. of metalinguistic tasks probe word awareness; Specifically, both of them demonstrated that that is, the language user’s understanding of bilingualism boosts word awareness and, par- the relation between words and their meanings ticularly, the awareness that a word’s form and its and, especially, the arbitrary nature of this meaning are not inseparable entities but have relation. become associated merely through convention. This effect of bilingualism was first noticed by Some authors (e.g., Ben-Zeev, 1977a; Bruck & Leopold, who dedicated four volumes to the Genesee, 1995) have suggested that metalinguistic speech development of his daughter Hildegard awareness is increased in bilingual children who was raised according to the “one-person because their dual linguistic environment forces one-language principle” (Leopold, 1939, 1947, them to pay special attention to the structural 1949a, 1949b). As summarized by Ianco-Worrall, aspects of language. Such a focus on structure from a very early age Hildegard “readily accepted would help them to separate the two systems and, new names for objects already denoted in one thus, to minimize interlingual interference. As language and asked to be given the name in the Bruck and Genesee (1995, p. 308) put it: second, or even a third, unfamiliar language” (Ianco-Worrall, 1972, p. 1391). This behavior The hypothesis that bilingualism or second suggests the awareness that the relation between language acquisition promotes meta- a word’s sound and meaning is arbitrary, that a linguistic awareness is based on the view that particular thing (or living being or abstract con- bilingualism provides a form of contrastive cept) remains this very same thing if its name linguistics instruction which leads bilingual were to be changed into a completely different children to compare and analyse the struc- one. tural aspects of language in more advanced ways than monolingual children.

7. COGNITIVE CONSEQUENCES OF BILINGUALISM 391 A miscellaneous collection of advantageous words, each accompanied by two choice words. effects has been attributed to this one insight. One of the latter two words was similar in sound For instance, it is thought to be beneficial for to the standard (e.g., standard word: cap; choice abstract thinking (see Hakuta & Diaz, 1985, for a word: can); the other was semantically related to discussion), which is an ability that is likely to the standard word (hat). The participants were promote many facets of cognitive functioning. It asked to select the choice word most similar to is also thought to promote analytical thinking the standard (e.g., “I have three words: cap, can, (e.g., Ben-Zeev, 1977b), to speed up semantic and hat. Which is more like cap, can or hat?). All development (Ianco-Worrall, 1972), and others children were from two age groups: 4 to 6 years, went so far as to regard the separation of word and 7 to 9 years. The bilingual children were sound and word meaning as a liberating, emanci- tested in both languages, their performance in pating attainment: “it frees the mind from the Afrikaans and English being compared with that tyranny of words. It is extremely difficult for a of the Afrikaans and English monolingual monoglot to dissociate thought from words, but groups, respectively. he who can express his ideas in two languages is emancipated” (Evans, 1953, in Peal & Lambert, The investigator predicted that the older chil- 1962, pp. 19–20). dren would select the semantically related alterna- tive more often than the younger children and Ianco-Worrall (1972), seeking empirical that, within each age group, the bilingual children support for Leopold’s observations, hypothesized would select the semantic alternative more often that the early separation of word sound from than the monolingual controls. In agreement with word meaning might have the effect that semantic these predictions, semantic preference indeed development in bilingual children progresses at increased with age, but only in the monolinguals. a faster rate than in monolingual children. She The bilingual children did not show an age effect furthermore hypothesized that, if such were the but the semantic preference was the dominant case, bilingual children more often than mono- pattern in both age groups. From these data the lingual children should perceive the relation- author concluded that “bilinguals, brought up in ship between words in terms of their semantic a one-person, one-language home environment, similarities rather than in terms of their acoustic reach a stage in semantic development, as attributes. The underlying assumption here measured by our test, some 2–3 years earlier than appears to be that attention to form aspects of their unilingual peers” (Ianco-Worrall, 1972, words is a developmentally earlier stage than p. 1398). attention to meaning aspects. In Chapter 3 (pp. 126–128) some evidence from multiple- A second experiment provided a hint that this choice recognition tasks was presented that sup- accelerated semantic development was related to ports this idea. It is furthermore supported by (and possibly caused by) the bilingual children’s early word association studies, which have shown relatively early awareness that the relation that younger children produce relatively many between word form and word reference is arbi- associations sharing a sound relation with the trary. In this experiment all children were asked a stimulus word, whereas in older children the set of questions, of three types: Questions that number of associations sharing a semantic rela- called for an explanation of word names (e.g., tion with the stimulus word increases (Entwisle, “Why is a dog called dog?”); questions that called Forsyth, & Muus, 1964; Ervin, 1961a). for a judgment of whether or not word names could be interchanged (e.g., “Could you call a dog Ianco-Worrall tested these views in a semantic- cow and a cow dog?”); and questions that called phonetic preference test in which Afrikaans– for an interchange of word names in play (e.g., English children (as Leopold’s daughter, raised “Let us play a game. Let us call a dog cow. Does according to the one-person, one-language prin- this cow have horns? And does this cow give ciple) and monolingual English and Afrikaans milk?”). The bilingual children outperformed controls were presented with a set of standard their monolingual controls on the second type of

392 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS questions, whereas performance of both groups geous effect of bilingualism on some or other on the remaining question types was similar. On manifestation of linguistic awareness. But null the basis of the differential results regarding the effects of bilingualism have also been reported, second type of question the author concluded even within the subset of studies that focused on that bilingual children are aware of the arbitrary word awareness, as the above studies did. Very relation between a word and its meaning at a occasionally a reversed effect has even been younger age than monolingual children. found, the monolinguals in fact performing better. Finally, in one study that exploited In a similar study Ben-Zeev (1977b) provided Ben-Zeev’s (1977b) symbol substitution test converging evidence. Her participants were 5- to (Rosenblum & Pinker, 1983), clear differences 8-year-old Hebrew–English bilingual children between monolinguals and bilinguals were and matched Hebrew and English controls. One obtained but these differences could not be quali- of the tests she administered was a so-called fied in terms of an advantage or disadvantage for “symbol-substitution test”, which consisted of either the bilinguals or the monolinguals. In two seven items similar to the questions that Ianco- comprehensive reviews (Bialystok, 2001b, 2004), Worrall (1972) posed to her participants. All and building on her own experimental work items were meant to determine the participants’ (Bialystok, 1988, 1992), Bialystok managed to ability to disconnect the conventional name of a distill a regular pattern out of the ostensibly referent and assign it another name; in other somewhat whimsical results and identified a words, to find out to what extent they were aware couple of factors that seem to constrain the that a name is not an intrinsic property of its bilingual advantage. referent. The items were of increasing difficulty (and increasingly bizarre). In the first two items, A starting point in Bialystok’s analysis is her as in Ianco-Worrall’s third type of questions, the observation—based on her review of the litera- name of an object/entity was substituted by the ture—that metalinguistic ability is not a “unitary name of another object/entity. For instance, the achievement” (Bialystok, 2001b), because if it experimenter showed a toy airplane to the child were the bilinguals might have been expected to and said: “You know that in English this is named excel on all tasks that putatively measure some airplane. In this game it is named turtle. Can a aspect of metalinguistic awareness. They did not, turtle fly?” [Correct answer: yes]; “How does however, not even on all tasks designed to index the turtle fly?” [Correct answer: with its wings]. one subdomain of metalinguistic awareness: word The remaining items asked for a substitution that awareness, syntactic awareness, or phonological violated selection rules, a word from one gram- awareness. Next, scrutinizing the processing matical class to be substituted by one of another requirements of all tasks dubbed “metalinguistic” grammatical class. For instance, the experimenter she identified two types of processing involved in said: “For this game the way we say I is to successful task performance: “analysis of repre- say macaroni. So how do we say ‘I am warm’?” sentational structures” and “control of selective [Correct answer: Macaroni am warm]. On both attention”—in short, “analysis” and “control”. types of items the bilinguals performed better Bialystok regards any task that demands the than the monolinguals, suggesting that it was eas- involvement of one or both of these processes as ier for them to ignore the conventional meanings a metalinguistic task. The more each of these of words. Similar results have since been obtained types of processing must be deployed to perform by Bialystok (1988), testing French–English the task properly, the more difficult the task. bilingual children. Analysis of representational structures con- Analysis vs. control cerns children’s ability to build increasingly detailed and explicit representational structures The above two studies exemplify a much larger of linguistic information that initially is only number of studies that have shown an advanta- stored in implicit representations. Implicit repre- sentations are thought to suffice for ordinary

7. COGNITIVE CONSEQUENCES OF BILINGUALISM 393 conversational uses of language but not for estimated analysis and control demands posed by literate language use. For the literate use of lan- each of them (from Bialystok, 2001b) guage explicit representations are needed that can be consciously accessed and manipulated. The conclusion that the bilingual advantage is Control of selective attention is the ability to the ability to control attention when there is mis- direct attention to specific aspects of either an leading information in language tasks gives rise to external stimulus or a mental representation dur- the question of whether this advantage extends ing task performance. The control demands are beyond the domain of language to other areas of especially high when the task environment con- cognitive functioning. The impact of an increased tains a conflict or ambiguity between two (or ability to control attention in non-linguistic tasks more) possible mental representations of the task, is potentially huge because attention control is and correct performance requires that one of required in all behavior that is not fully auto- them is attended to and the other ignored or sup- mated and thus, arguably, in the far majority of pressed. The higher the level of conflict between our everyday pursuits. the competing representations, the more attention control must be exerted to maintain satisfactory The above studies on the effects of bilingual- levels of performance. ism on metalinguistic task performance and the conclusions drawn from them have thus paved the Scrutinizing the metalinguistic tasks on which way towards a new field of research that connects monolinguals and bilinguals have been compared the study of bilingualism with the vast research (tasks, thus, that require the involvement of at field examining how humans exert attention con- least one of these two processing components to trol while performing cognitively demanding some minimal degree), Bialystok (2001b, p. 178) tasks and, especially, tasks that present some type discovered the following pattern: of conflict. Ellen Bialystok was the first to see this possible connection and to start examining it. Simply put, tasks that are high in their The general hypothesis underlying the ensuing demands for control of attention are solved experiments is that while controlling their lan- better by bilinguals than monolinguals; guages bilinguals exploit more general processes tasks that are high in their demands for and mechanisms of cognitive control (see also analysis of representations are not necessar- pp. 306–310) and that, as a consequence, they ily solved better by either group. The have become experts in cognitive control in bilingual advantage, therefore, is in the general. If so, they should generally excel on tasks ability to control attention when there is mis- requiring cognitive control, even if the tasks in leading information. question are completely non-linguistic. A further hypothesis is that the beneficial effect of bilingual- This analysis explains the bilingual children’s ism on cognitive control in general extends superior performance on the symbol manipula- beyond childhood into adulthood. In the next tion task presented above. In that task the mislead- section I will present some of this work. ing information to be inhibited is the conventional meaning of the substitute word (e.g., turtle in can a Bilinguals as experts in cognitive turtle fly?), where turtle is the substitute for air- control plane). For similar reasons, bilingual children are better than monolingual controls in accepting sen- Evidence from the Simon task tences such as “Apples grow on noses” as gram- matically correct. To provide the correct response To examine whether language control in (“yes”) the meaning of the sentence must be bilinguals exploits more general processes and ignored and attention must be focused on the sen- mechanisms of cognitive control, Ellen Bialystok tence’s structure. Figure 7.10 organizes a larger and her colleagues compared the performance set of metalinguistic tasks on the basis of the of bilinguals and monolingual controls of

394 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS A number of metalinguistic tasks organized according to their estimated demands for analysis and control. From Bialystok (2001). Reproduced with permission from Cambridge University Press. various ages on, among others, the “Simon task” the position of the key to be pressed (e.g., red (Bialystok, 2006; Bialystok, Craik, Klein, & squares require pressing the right-hand key and a Viswanathan, 2004; Bialystok et al., 2005a; red square has just appeared on the right side of Bialystok, Martin, & Viswanathan, 2005b; the screen); in an incongruent condition a stimu- Martin-Rhee & Bialystok, 2008). This perceptual- lus’s screen position and the position of the key motor task assesses the participants’ skill in to be pressed mismatch (e.g., the red square is inhibiting or ignoring irrelevant spatial informa- presented on the left-hand side of the screen). tion, a skill that requires executive control. The Congruent and incongruent trials are presented participants are seated in front of a computer in a mixed, unpredictable order. Typically, incon- screen where stimuli (e.g., squares) in different gruent trials lead to longer response times (and colors are presented in different positions. In the more errors) than congruent trials. This effect is standard set-up stimuli of two different colors are called the “Simon effect”. used, say blue and red, and the stimuli appear, in a randomized order, either on the right or the left As mentioned, the starting point of Bialystok side of the screen. The participants are instructed and her colleagues was the hypothesis that to press a key on the left-hand side of the key- (active) bilingualism boosts cognitive control, and board if the presented square’s color is blue and especially inhibitory control, outside the language to press a key on the right side when the square’s domain, for reasons that bilingualism involves color is red (or vice versa). In a congruent condi- the incessant need to exert this type of control, tion the stimulus’s position on the screen matches inhibiting or ignoring one language to be able to speak the other. As we have already seen, earlier

7. COGNITIVE CONSEQUENCES OF BILINGUALISM 395 studies had provided evidence that balanced increasing working memory demands in the bilingual children are better than matched mono- Simon task, adding a condition with four rather lingual controls at metalinguistic language tasks than two colors and instructing the participants that require the control of attention. Two of the to press the left-hand key in reply to two of these questions the authors posed in the present set of colors and the right-hand key in reply to the new studies were whether this same advantage remaining two colors. This manipulation might also hold when a non-linguistic task has to increases working memory load because now the be performed and if this would also be the case participants have to keep four instead of two for balanced bilinguals of other age groups; associations between specific colors and keys in that is, whether the bilingual advantage would memory during task performance. Second, the sustain into adulthood. A further question built researchers included a control condition in which on the well-known fact that normal cognitive the color stimulus to be responded to always aging involves a decline of the mechanisms appeared in the center of the screen (that is, that enable cognitive control (e.g., Christ, White, without task-irrelevant spatial information). This Mandernach, & Keys, 2001). The authors manipulation was meant to rule out the possi- wondered whether bilingualism might provide a bility that a bilingual advantage to be obtained “defense” against this decline of controlled pro- was merely due to overall faster responding of cessing with aging. Finally, they attempted to bilinguals rather than to increased efficiency of determine the neural correlates of a bilingual inhibitory control (as demonstrated by a smaller advantage in inhibiting irrelevant information. Simon effect and/or faster responding to congru- ent and incongruent trials). Third, the authors Across these studies, a smaller Simon effect in looked at the effect of task practice: If excessive bilinguals than in monolinguals was regarded practice in inhibitory control underlies the as the main marker of more efficient inhibitory relatively small Simon effects in bilinguals (if cognitive control in bilinguals. In addition, over- these were to emerge at all), with increased all task performance was looked at as well practice on the Simon task the difference between because “the control processes required to per- the Simon effects obtained for monolinguals and form the Simon task are involved in all trials, not bilinguals should become smaller. just those explicitly containing conflict” (Martin- Rhee & Bialystok, 2008, p. 82). The reason is The combined studies provided clear and that the task incessantly requires a high level of consistent answers to the questions posed: First, vigilance on the part of the participants because excepting the population of young adults, they never know in advance what type of trial, specifically university undergraduates, among all congruent or incongruent, will be presented next. participant populations the bilinguals generally Therefore, in addition to a smaller Simon effect in showed reliably smaller Simon effects and/or per- bilinguals, bilinguals were expected to generally formed better overall on the Simon task than respond more rapidly than monolingual controls, monolinguals, indicating less disruption from on both congruent and incongruent trials. misleading information, and thus superior inhibi- tory control. The authors regarded the absence of To address these questions, across a number of an effect of bilingualism in the university under- studies Bialystok and her colleagues had mono- graduates as a ceiling effect: For these partici- lingual children and young-adult, middle-aged, pants inhibitory control is at its peak so that and older-adult monolinguals and matched bilingualism cannot boost it any further. Second, groups of balanced bilinguals all perform the bilinguals were hampered less than monolinguals Simon task as described above. To be able to by an increase in working memory load, suggest- gather fine-grained knowledge on the relation ing that the beneficial effect of bilingualism between bilingualism and cognitive control, in extends beyond improved inhibitory control to addition to the congruent–incongruent mani- the improvement of other aspects of cognitive pulation the researchers included three further functioning. Third, as predicted, aging was manipulations. First, they looked at the effect of

396 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS associated with less-efficient executive control, as control earlier and maintain their ability to evidenced both by relatively large detrimental control those functions longer than mono- effects of increases in memory load and by linguals. Given the fundamental centrality increases in the Simon effect in the older-adult of these executive processes to our everyday groups. Fourth, both these negative effects of cognitive life, this is an altogether promising aging were considerably smaller for bilinguals outcome for bilinguals. than for monolinguals, suggesting that bilingual- ism indeed attenuates the adverse effects of aging If these salutary effects of bilingualism are indeed on executive control (and see Bialystok, Craik, & caused by the incessant use of more general, non- Freedman, 2007, for evidence that bilingualism bilingualism-specific, control processes, expertise delays the onset of symptoms of dementia). in other domains that exploit these same control Fifth, in the majority of cases the possibility was processes should similarly affect performance on excluded that the beneficial effects of bilingualism the Simon task. A final result of the above studies simply concerned a speed advantage by showing confirms this hypothesis: Bialystok et al. (2005b) that the magnitude of the Simon effect (the pri- split up a participant group of young adults, mary signature of inhibitory control) always dif- college undergraduates, into a subgroup of fered between monolinguals and monolinguals, youngsters who spent much time playing video except after extended practice. Sixth, practice games on the computer and a subgroup who reduced, and ultimately annihilated, the dif- played computer games infrequently. Recall that ferences between bilinguals and monolinguals: the group of young adults was the only group After extended practice the adverse effect of that, overall, did not show an effect of bilingual- increased load was equally large in both groups ism on performance in the Simon task. Again a and the Simon effect was reduced to zero for both clear-cut result emerged, and one that will be groups. reassuring for all those parents who have worried about the possibly harmful effects of excessive All in all, these findings constitute compelling computer game play on their (especially male) evidence that the same executive processes are offspring: Performance on the Simon task was involved in managing the control over two lan- considerably better in the group who played guages as in performing the Simon task and, video games frequently, suggesting that gaming plausibly, other conflict tasks that require task- improves executive control (see also Bialystok, irrelevant information to be inhibited or ignored. 2006). Similary, Green and Bavelier (2003) have Furthermore, the memory load data suggest shown that playing action video games modifies that bilingualism also boosts memory capacity. visual selective attention. Whatever mental havoc (Bialystok, 2008, qualifies the latter conclusion by excessive gaming may cause, apparently there are suggesting that this holds specifically for memory beneficial effects on cognition as well. What this tasks that are primarily based on executive con- result also demonstrates is that there are other trol. Memory tasks that are based primarily on ways to become an expert in inhibitory control verbal recall are performed more poorly by than having the good fortune to grow up in a bilinguals than monolinguals, presumably as a bilingual home. consequence of bilinguals’ lesser prior experience with, and therefore lesser vocabulary skills in, the Evidence from other tasks language in which the recall test is administered; see p. 389.) Apparently, bilingualism yields bene- Child studies. Although the conclusion seems fits across a larger set of skills that all appeal to warranted that bilingualism boosts cognitive con- general executive processes. As summarized and trol, the exact nature of the benefit and the concluded by Bialystok (2007, pp. 220–221): processes and mechanisms that underlie this advantage are not yet fully understood. To aug- The research [. . .] reveals compelling evi- ment and refine our knowledge about these dence that bilinguals develop executive

7. COGNITIVE CONSEQUENCES OF BILINGUALISM 397 processes and mechanisms, researchers have inhibiting attention to the distracting cue, but not begun to compare the performance of bilinguals on tasks that require the inhibition of a habitual and monolinguals on other tasks that require response. In other words, inhibitory processes some form of cognitive control different from are not generally enhanced in bilingual children. the requirements posed by the Simon task. This The authors (Martin Rhee & Bialystok, 2008, way the limits of the bilingual advantage can p. 85) attributed this selective advantage to the be pinpointed and the exact sources of the nature of the specific bilingual experience, which advantage can be charted step by step. To this resembles the demands of bivalent tasks: end Martin-Rhee and Bialystok (2008) not only administered the Simon task but also a For bilinguals, their two linguistic systems Stroop-like task, the “day–night” task, in which function as bivalent representations, offering their participants, monolingual and bilingual different, potentially competing response children of about 4.5 years old, were instructed to options to the same intention or goal. To say “night” to a picture of a bright sun and “day” manage this conflict, bilinguals must attend to a picture of a dark moonlit sky. to the relevant language system and ignore the unwanted system to assure fluency in The choice of the day–night task was based on speech production. a categorization of inhibitory control tasks into two types, each associated with a specific type of The data suggest that practicing this form of stimulus display: tasks that require “interference attention control through bilingualism then suppression” and those that require “response transfers to increased levels of attention control inhibition”. The former type of tasks are associ- in similar non-language tasks. ated with so-called “bivalent displays”, the latter with “univalent” displays. In bivalent displays This analysis receives support from a study there are two perceptual features that may that assessed the generality of the bilingual converge on one and the same response (on advantage by administering a much larger battery congruent trials) or that each summons a dif- of tasks, nine in all, that were all hypothesized ferent response tendency (on incongruent trials). to index some aspect of executive functioning and In univalent displays there are two response that all involve some form of inhibition (Carlson options to one and the same stimulus feature, & Meltzoff, 2008). The participants were three a habitual one and an arbitrary one. The Simon groups of kindergarten children: (1) a group of task is an instance of an interference-suppression Spanish–English bilingual children who had task because the color and the position of the been exposed to both languages from birth. (2) stimulus concern two potentially competing fea- A group of L1 English children who attended tures and the display is thus bivalent. Correct task immersion elementary schools where they were performance requires that the task-irrelevant fea- instructed in English for half the day and in either ture (position) is suppressed. The “day–night” Spanish or Japanese for the rest of the day. At the task (as the classical Stroop task) is an instance of time of testing this immersion experience had a response-inhibition task: The habitual response lasted about 6 months. (3) A control group of L1 to provide the common name of the picture must English children attending traditional English be inhibited in order to come up with the schools. Despite the fact that the children in the requested response, the one based on an arbitrary bilingual group had relatively low scores on a test agreement. of verbal ability and had parents with relatively low education levels, their scores on all tests The bilingual advantage turned out to be equaled those of the other groups. But when limited to the Simon task and not to extend to verbal ability and parental education level were the day–night task. These data thus suggest that statistically controlled for, the composite bilingual children outperform their monolingual executive-functioning score based on all nine peers on tasks that require attention to be selec- subtests together showed superior performance tively directed to specific cues in conflict situations,

398 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS for the bilingual group as compared to the other only does this finding show that advantageous groups. The performance of the immersion and effects of bilingualism on cognitive control may control groups did not differ from one another. take a relatively short period of bilingual experience to become manifest, but it also shows Each of the individual tests also suggested an that for a bilingual advantage (at least the one advantage for the bilingual group, but the effect tapped in this task) to develop no experience in was only statistically significant on a subset of bilingual output is required. After all, infants this them. It appeared that the shared feature of the age do not produce speech yet (except in the form tests in this subset was that they all required of babbling; see Table 2.1, p. 21). It thus seems inhibition of attention to misleading contextual that a relatively short period of just perceiving information. This hypothesis was confirmed and processing speech input in two languages by the results of a factor analysis on the data already has a positive effect on cognitive control. of all nine tests. This analysis revealed two dis- tinct factors that the researchers interpreted as Adult studies. As mentioned earlier, the com- measuring the children’s ability to manage con- mon bilingual advantage in the Simon task had flicting attentional demands and the ability to not turned up in a comparison of monolingual control impulses, respectively. The bilingual and bilingual young adult college undergraduates. advantage exclusively occurred on the tasks The reason may be that young adults in general that loaded on the first of these two factors. In are at a peak of their executive control abilities conclusion then, it appears that the bilingual and that the null effect of bilingualism reflects advantage does not apply to all circumstances asymptotic performance in both groups. That that require inhibitory control but to circum- bilingualism can nevertheless boost executive stances in which distracting information must control even in this age group was recently be inhibited or ignored. A further conclusion to demonstrated by Costa, Hernández, and draw from Carlson and Meltzoff’s (2008) study is Sebastián-Gallés (2008). The participants in this that the bilingual advantage, not surprisingly, study were all young adults and consisted of a depends on the degree of bilingualism. This group of early highly proficient Catalan–Spanish follows from the fact that the immersion group, bilinguals and a control group of Spanish mono- after 6 months of L2 experience, did not yet linguals matched with the bilinguals on age and show any advantage in executive functioning as educational level. The researchers had all of them compared with the control group. perform the “attentional network task” (ANT) developed by Fan, McCandliss, Sommer, Raz, Still, a recent study in which infants were and Posner (2002). This task has been designed tested suggested that for other aspects of to reflect three types of attentional processes executive function a bilingual advantage may that are assumed to each be taken care of by a not take more than about half a year of bilingual different set of brain structures: executive control experience to occur (Kovács & Mehler, 2009). In (monitoring performance and resolving conflict), this study a group of 7-month-old bilingual alerting (to become and stay alert), and orienting infants raised with two languages from birth and (selecting a subset of information from sensory a group of matched monolingual 7-month-olds input). In the ANT task participants must indi- were trained to anticipate a reward stimulus (a cate whether an arrow (→ or ←) that is embedded puppet) on one side of a screen. This training between four flanker arrows, two on each side, phase was followed by a number of trials in which points towards the right or the left. In a congru- the reward stimulus appeared on the other side of ent condition the flanker arrows point in the same the screen. The bilingual infants, but not their direction as the target arrow (→→→→→ or monolingual age-mates, managed to redirect their ←←←←←); in an incongruent condition the anticipatory looks to this other position on the pointing direction differs between the target and screen after the puppet’s position switch, a find- the flankers (→→←→→; or ←←→←←). Just as ing the authors attributed to superior cognitive control in the bilingual infants. Importantly, not

7. COGNITIVE CONSEQUENCES OF BILINGUALISM 399 the incongruent condition in the Simon task, the into the nature of the bilingual advantage by incongruent condition usually produces slower contrasting two ways in which the human control responses than the congruent condition. This system may safeguard itself against outputting congruency effect is attributed to the fact that a unintended responses: through “active” or conflict must be resolved in the incongruent con- “reactive” inhibition. In the authors’ termin- dition; namely, between the opposite pointing ology, active inhibition is executed if some central directions of target and flankers. The attention inhibitory system directly inhibits (deactivates) process assumed to be involved in resolving the memory nodes associated with non-target conflict is executive control. The involvement of responses (e.g., the node for Spanish mesa, the second process, alerting, is manipulated by “table”, when a Spanish–English bilingual presenting or not presenting an alerting cue (an intends to speak English, not Spanish), thus mak- asterisk) prior to target presentation. The third ing these responses less available. Conversely, process, orienting, is manipulated by having reactive inhibition is operative if the current or not having the cue indicate the position where goal, for instance, the intention to speak English, the target subsequently appears. Costa and his boosts the activation in the memory nodes repre- colleagues wondered which one(s) of these senting the targeted responses (here, the nodes manipulations (congruency, the presence or representing English words). Through lateral absence of a cue, cue position) affects the per- inhibition along inhibitory connections between formance of bilinguals and monolinguals dif- targeted and non-targeted memory nodes (e.g., ferently, thus suggesting the differential efficiency between the nodes for table and mesa) the of one or more of the three different attention activation level in the non-target nodes is then mechanisms involved. lowered, making the associated responses less available. According to this view, inhibitory In agreement with the analogous findings of effects do not result from actual inhibition Bialystok and her collaborators with the Simon exerted by an inhibitory control system but from task, the bilinguals suffered less interference from a control mechanism that works through facilita- incongruent flankers than the monolinguals did, tion. The process of lateral inhibition between again suggesting that they are better at resolving word nodes representing translation equivalents conflict in a situation where there is misleading presupposed by this view is more often assumed information. It is particularly noteworthy that in models of bilingual word processing (see this finding materialized even though all par- p. 177). These two views on bilingual language ticipants were young adults, and thus at the peak control are illustrated in Figure 7.11. (Notice that of a human’s executive skills. Also, as in the the present authors’ definition of active and Simon task studies, the bilinguals were faster reactive inhibition deviates from our use of the overall than the monolinguals. A further finding terms “proactive” and “reactive” control in of interest was that bilinguals benefited more Chapter 6. The term “proactive” control was used from the presence of an alerting cue than mono- there to refer to a process whereby the intention linguals, indicating that the alerting mechanism to speak one language and not the other boosts works more efficiently in bilinguals. The third the activation of the memory nodes in the tar- manipulation (cue position) did not differentiate geted language, thereby rendering them more between bilinguals and monolinguals, suggesting available. In other words, the process involved is that the third attentional process, orienting, does the first step in what the present authors call not benefit from bilingual experience. To summar- “reactive inhibition”. In contrast, “reactive con- ize, it appears from this study that the executive trol” as we used the term concerned a process control and alerting components of the brain’s operating on the imminent output of the lan- attentional network work more efficiently in guage system, preventing elements of the non- bilinguals than in monolinguals. selected language from emerging in speech.) A second recent study, by Colzato, Bajo, Van Colzato and collaborators examined the den Wildenberg, and Paolieri (2008), probed

400 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS Two possible ways in which the human control system may bilingual advantage. The bilinguals in question safeguard itself against outputting unintended responses: did not master two spoken languages, as those (1) The current goal boosts the activation of the targeted in the studies discussed above, but one spoken memory nodes (a). The activated targeted nodes then language and one sign language. Like “unimodal” inhibit non-targeted nodes through reactive inhibitory bilinguals (who master two spoken languages), connections (b). (2) A central inhibitory system directly such “bimodal” bilinguals have acquired two inhibits non-targeted memory nodes (c). From Colzato et al. full linguistic systems (see pp. 53–54 for a sub- (2008). Copyright © 2008 American Psychological Society. stantiation of the claim that sign languages are fully fledged languages). But unlike unimodal nature of the bilingual advantage by comparing bilinguals, bimodal bilinguals can produce words the performance of adult balanced Dutch– from both languages simultaneously, one through English bilinguals and Spanish monolinguals on vocal speech, the second through signing. In three non-verbal cognitive control tasks that, other words, the physical restriction that words according to the authors, differ from one another can only be uttered vocally one at the time forces in the type of inhibition they appeal to, active or lexical selection in unimodal bilinguals, whereas reactive. I will refrain from explaining these tasks the physical possibility of speaking and signing at and their rationale because doing so would take the same time does not do so. An investigation by an inordinate amount of space. Instead I will Emmorey, Borinstein, Thompson, and Gollan confine myself to mentioning the authors’ con- (2008a) showed that the simultaneous production clusion that the pattern of results was consistent of words across two modalities in bimodal with a reactive-inhibition account of bilingual bilinguals is not only a theoretical possibility but control. More precisely, the authors concluded in fact happens in actual practice. that bilinguals are better than monolinguals in maintaining action goals, which boost activation Emmorey, Luk, Pyers, and Bialystok (2008b) in the goal-relevant memory representations. wondered whether the bilingual advantage in cognitive control might result from the fact that To conclude this review of studies that unilingual bilinguals are always forced to select attempted to delineate the set of circumstances one language in speech and that knowing two under which bilinguals outperform monolinguals languages per se does not contribute to the on non-verbal control tasks, one such study must advantage. If so, unlike unimodal bilinguals, not be overlooked. It included bilinguals of a bimodal bilinguals should not exhibit better type ignored so far, and provided important performance than monolinguals on non-verbal new information on the limits and source of the control tasks. The researchers examined this hypothesis by comparing the performance of monolinguals, unimodal bilinguals, and bimodal bilinguals on two “flanker tasks” and one control task. One of the flanker tasks was similar to the one included in Costa et al.’s study (2008; see above), except that the stimuli were built not from arrows but from chevron heads: The stimuli in the congruent condition consisted of a series of chevrons all pointing in the same direction, one of them, printed in red, being the target, the other the flankers (e.g., <<<<<). In the incongruent condition the flankers pointed in the opposite direction from the target (<< ><<). On each trial the participants had to indicate, by pressing one of two buttons, whether the target chevron pointed towards the right or the left. In the

7. COGNITIVE CONSEQUENCES OF BILINGUALISM 401 second flanker task the target, again a red advantage in cognitive control results from the chevron, was flanked on either side by two red unimodal bilingual’s experience in controlling flankers. In a “go” condition the flankers were two languages in the same modality. As men- diamonds (◊◊>◊◊). In a “no-go” condition they tioned above, the authors attributed the effect were red Xs (XX>XX). In the “go” condition the specifically to the requirement of output participants had to indicate in what direction control in unimodal bilinguals. Notice, however, the chevron pointed by pressing the appropriate that the results of the study performed by button. In the “no-go” condition they had to Kovács and Mehler (2009; see the above section withhold their response. Finally, on each trial in on child studies) with preverbal infants as par- the control task a single chevron was presented ticipants suggest that input control might and the participants had to indicate whether it contribute to the unimodal bilinguals’ advantage pointed to the right or the left. as well. The results provided clear support for the All in all, it can be concluded that the com- hypothesis that the superior cognitive control in bined child and adult studies discussed above unimodal bilinguals does not result from their clearly suggest that (unimodal) bilingualism knowing two languages but from the requirement brings superior cognitive control that is already to unrelentingly and inevitably select one lan- manifest in very early childhood and lasts in guage: Unimodal but not bimodal bilinguals adulthood. These studies have also begun to responded faster than monolinguals on both reveal which subcomponents of the multifaceted congruent and incongruent trials in the first skill of cognitive control are the ones to benefit of the above flanker tasks and on the “go” trials from the bilingual experience. While this evidence in the second. On the control task, in which was being gathered, other researchers started to there was no conflict to resolve or response to map the brain areas involved in this cognitive withhold, all three groups responded equally advantage. The results from the pertinent studies fast. The authors concluded that the bilingual will be presented in Chapter 8. SUMMARY • The multiple languages of a bilingual or multilingual all interact with one another both during acquisition and use. The inevitable consequence of this fact is that the linguistic utterances of bilinguals and multilinguals differ from those of monolingual speakers of the languages involved. In other words, multilingualism does not equal multiple monolingualism. • The cross-linguistic influence of a multilingual speaker’s currently non-selected languages on the currently selected one is determined by a number of factors: the typological distance between the non-selected languages and the selected language; the foreign language status, strength, and proportion of recent use of the non-selected languages; the order of acquisition of the selected and non-selected languages; the multilingual’s current language mode; the speaker’s proficiency in the selected language. • Degree of loss of L2 vocabulary acquired at some point in the past but not used recently depends on the level of L2 proficiency originally attained, the duration of elapsed time since acquisition, the spacing of the training sessions during acquisition, and specific features of the learning materials. Part of the vocabulary originally acquired beyond some minimal threshold level seems to be immune to loss and can still be retrieved from memory after decades of non-use. • The savings paradigm is a highly sensitive technique to detect residues of seemingly lost knowledge acquired in the remote past. It involves the brief relearning of materials (e.g., a set of L2 words) likely to be known in the past but seemingly lost and the (equally brief) learning of

402 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS matched materials unlikely to be known before (e.g., another set of L2 words). The test scores on a subsequent cued recall test for the former materials are typically substantially higher than those for the latter materials even when the test is administered after decades of disuse of the former materials. This finding suggests that prior knowledge is not lost but unavailable, and that it takes little to reactivate it. • The L1 can be completely lost and replaced by an L2 when at a young age the L1 learner is abruptly cut off from further L1 input, as sometimes happens in adoption. Such “catastrophic forgetting”, presumably caused by a process of retroactive interference from the L2, has successfully been imitated in computational modeling. • A comparison of studies on L1 and L2 loss suggests that an unused L1 is more susceptible to forgetting than an unused L2. However, to legitimately draw this conclusion studies need to be designed that rule out an explanation of the differential loss of L1 and L2 in terms of a number of potentially confounding variables: the amount of continued exposure to the corroded language in between initial learning and later testing; the types of linguistic knowledge that are examined during testing; the originally acquired level of fluency in the corroded language; the test methods used to assess the degree of loss. Only if differential loss is observed for L1 and L2 if these variables have been controlled for can it be concluded that L1 is more easily lost than L2. • As compared with the speech of monolingual language users, bilingual speech in both L1 and L2 is characterized by an accent in all domains of language: phonology, grammar, and semantics. There are two possible sources of these accents: memory structures that differ between monolinguals and bilinguals or response competition caused by activated structures in the non-response language in bilinguals. • Languages differ from one another on many structural dimensions including whether and how they mark grammatical gender, number, and tense. Cross-linguistic studies have shown that the specific linguistic structures encountered by native speakers of a language affect their thought processes and the content of their conceptual structures. Bilingual studies have shown that a switch of language may change bilinguals’ processes of thought, and that the content of thought in at least some conceptual domains may differ between bilinguals and monolinguals. • A form of bilingualism called “additive” is advantageous for cognitive functioning. This form of bilingualism emerges when an L2 is added onto the native language instead of gradually replacing it. A form of bilingualism called “subtractive” is detrimental for cognitive functioning. In this form of bilingualism, due to social pressure and educational policy the use of the native language is discouraged and, consequently, it is gradually replaced by L2. • So-called “metalinguistic” tasks appeal to the language user’s ability to step back from the meaning-conveying function of language and to reflect on the structural aspects of linguistic expressions. Metalinguistic ability has been analyzed in two component abilities: analysis of representational structures and control of selective attention. It is hypothesized that bilinguals outperform monolinguals on metalinguistic tasks that appeal to the second of these abilities. • The Simon task is a non-verbal perceptual-motor task that assesses the ability to inhibit or ignore irrelevant spatial information, a skill that requires cognitive control. Bilinguals of all age groups, excepting young adults, outperform matched monolinguals on this task, suggesting superior cognitive control in bilinguals. Furthermore, the adverse effects of aging on cognitive control are smaller in bilinguals than in monolinguals, suggesting that bilingualism attenuates the detrimental effect of aging on cognitive control. These results suggest that bilingualism boosts cognitive control. The absence of a bilingual advantage in young adults is probably due to the fact that at this age cognitive control is at its peak and cannot be boosted further by bilingualism.

7. COGNITIVE CONSEQUENCES OF BILINGUALISM 403 • The finding that bilingualism brings along superior performance on non-verbal tasks requiring cognitive control suggests that language control in bilinguals exploits more general processes and mechanisms of cognitive control. • Two forms of cognitive control can be distinguished: (1) The ability to direct attention selectively to specific information in conflict situations while inhibiting or ignoring misleading contextual information. (2) The ability to suppress habitual responses, in other words, to control impulses. Studies that compared monolingual and bilingual children and adults on tests tapping the first of these two types of ability suggest that it is boosted by bilingualism. Studies that compared monolingual and bilingual children on tasks tapping the second ability suggest that it is not affected by bilingualism. In addition, it has been shown that bilingual adults react to alerting signals in the environment more efficiently than monolingual adults. • There are two ways in which the human control system may safeguard itself against outputting unintended responses: (1) It may directly deactivate memory nodes associated with non-target responses, thus making these responses less available. (2) It may boost the activation in the memory nodes that mediate the targeted responses. Through lateral inhibition along inhibitory connections between targeted and non-targeted memory nodes the activation level in the non- target nodes is then lowered, making the associated responses less available. • Unlike bilinguals mastering two spoken languages (“unimodal” bilinguals), bilinguals who master one spoken and one sign language (“bimodal” bilinguals) are not better than monolinguals at non-verbal tasks requiring cognitive control. This suggests that the bilingual advantage manifested by unimodal bilinguals does not result from knowing two languages per se, but from the fact that these bilinguals are forced to always select one language in speech. This control requirement is less stringent in bimodal bilingualism because the distinct motor systems involved in signing and speaking allow simultaneous production in both modalities.



8 Bilingualism and the Brain INTRODUCTION AND PREVIEW examined language processing in monolingual language users. (More precisely, these studies In the previous chapters, the way bilinguals and investigated native language processing in lan- multilinguals process language has been dealt guage users who were considered monolingual, with in detail and much has also been said about while it was usually ignored that the language the type of knowledge structures that underlie users in question might also have mastered one or bilingual language processing. But one important more other languages to at least some degree.) and topical research area has clearly been Monolingual studies have also revealed the underexposed so far—the study of the brain specific areas in the left hemisphere that play a structures that subserve bilingual and multi- role in language processing. In addition, they lingual functioning. This neglect is remedied in have shown that certain types of language pro- this final chapter. The emergence of functional cessing—among others, the processing of indirect neuroimaging techniques in the last decade of forms of language such as metaphors or irony— the twentieth century boosted research in this recruit areas in the right hemisphere. field and the results of the pertinent studies have refined and extended the conclusions drawn from Two questions have dominated—and continue earlier neuropsychological and behavioral studies to dominate—the study of the bilingual brain. probing into the brain basis of bilingualism. The first is what parts of the brain are recruited when bilinguals process language and, specific- It is an established fact that the human brain’s ally, whether these might differ from the regions two hemispheres are not equally involved in involved in monolingual language processing. A language processing, but that in the vast majority substantial portion of the older studies address- of the human population the left cerebral hemi- ing this question have focused on language later- sphere predominantly subserves linguistic alization in bilinguals, trying to find out whether behavior; in other words, language is “lateralized” language is left-lateralized to the same degree as it in the left hemisphere in most people. This con- is in monolinguals or whether perhaps language is clusion has primarily been based on studies that processed more bilaterally in bilinguals. The sec- ond main question posed in the study of the 405

406 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS bilingual brain is what brain regions are involved lines of study. The first concerns clinical neuro- in bilingual language control; that is, the bilin- psychological studies on patients with language gual’s ability to maintain the use of one of his or disorders resulting from brain damage caused by, her languages in circumstances where unilingual for instance, a stroke, a brain tumor, trauma, or output is required and to switch between the lan- some degenerative brain disease. Deficits in guages if such is demanded by the circumstances. language comprehension and/or production due to neurological damage are collectively known as In two separate sections, this chapter provides aphasia. Since the pioneering work of Paul Broca the results of studies that have addressed these (1824–1880) and Carl Wernicke (1848–1905) it is questions: what are the language areas in the known that lesions in specific parts of the brain bilingual brain and what brain areas are specific- lead to specific language disorders, suggesting ally involved in bilingual control? To set the stage that specific parts of the brain subserve different for presenting the bilingual work, the so-called aspects of linguistic functioning. Although one “classical language areas” will first be introduced: and a half centuries of aphasia research still has the brain regions in the left hemisphere that about not settled the debate on the exact function of 150 years ago were first found to be involved in each of the various brain structures involved in language processing. In this same section the language representation and processing, the brain various views on the functions these areas have in areas identified by Broca and Wernicke as com- language processing will be presented and some ponents of the language system feature in all con- attention given to the contribution of the right temporary views on the brain’s language network hemisphere to language processing. (see further on). As mentioned above, functional neuroimaging In the era of Broca and Wernicke, the dam- studies have contributed importantly to our cur- aged brain area causing the language disorder rent knowledge about the neural substrate of could only be identified post mortem, during bilingualism. I will present the essence of the autopsy. Nowadays the afflicted area can be techniques used in these studies (and of a number detected in vivo by means of a number of non- of structural imaging techniques) in the methods invasive brain-scanning techniques, collectively part of the Methods and Nomenclature section. known as “structural imaging” techniques, with The common behavioral methods to examine lan- which the brain’s anatomical structures can be guage lateralization will also be presented in this revealed, intact and damaged. (“Non-invasive” part. The second (nomenclature) part of this sec- means that the technique does not require that tion describes the conventions used in labeling some substance, for instance some fluid or brain areas. Without some basic knowledge of measuring device, is inserted in the body.) If a these conventions, grasping the core message in damaged structure is detected, the specific charac- neuroimaging publications can be extremely chal- teristics of the functional disorder then point to lenging for the reader. In neuroimaging studies, in the functional role of this structure. One of these principle, the same tasks are used as all those techniques is computed tomography (CT), an employed in the behavioral studies on bilingual- advanced application of the X-ray technique ism that were reviewed in the preceding chapters, which reconstructs three-dimensional images of which renders a separate task section superfluous. brain structures from the two-dimensional images produced by the conventional X-ray technique. METHODS AND NOMENCLATURE CT exploits the fact that the amount of radiation absorbed by neural tissue varies with tissue Methods density, with high-density tissue absorbing more radiation than low-density tissue. A second Knowledge about the way the brain represents technique to reveal brain structures is magnetic and processes language is gathered in two main resonance imaging (MRI), which exploits the magnetic properties of the elements in organic

8. BILINGUALISM AND THE BRAIN 407 tissue by temporarily changing the orientation area located in the inferior part of the left of certain atoms (e.g., hydrogen) by means of a frontal lobe (see p. 420). Damage to this area strong magnetic field and then detecting the mag- typically results in telegraphic speech: simple netic field that is generated when the atoms return short sentences that lack inflections and words to their original position. The specific charac- with syntactic functions such as articles, pre- teristics of these magnetic fields vary with the positions, pronouns, and auxiliaries. It is there- density of the hydrogen atoms in the various fore tempting to conclude that Broca’s area is the brain tissues, thus enabling the detection of brain site of grammar. However, alternative accounts structures. The spatial resolution of MRI scans is are possible, for instance that these symptoms far better than that of CT scans: MRI scans are a manifestation of a working memory deficit provide images of brain structures that are less while grammatical knowledge is intact (see than 1 mm apart, whereas brain structures must pp. 421–422). To discover the functionality of the be about 5 mm apart for them to be discriminated various parts of the brain involved in language on CT scans. MRI scans reveal the distinction representation and processing, studying the between gray matter (regions of the nervous effects of brain lesions on language use must system that primarily contain the cell bodies) and therefore be complemented with examinations of white matter (areas of the nervous system that how the healthy brain processes language. This primarily contain axons; these axons are coated concerns the second main line of study referred to with myelin, a whitish substance). A third above: The measurement of brain activity while structural imaging technique, diffusion tensor neurologically intact people perform language imaging, uses an MRI scanner to provide tasks. The measurement techniques in question, images of white matter pathways in the brain. It collectively known as “functional imaging” measures the density and motion of water in the techniques, either detect the brain’s electrical axons. (For more details about these techniques activity (or correlates thereof) or its metabolic see Gazzaniga, Ivry, & Mangun, 2009, and activity while participants perform the task at Kolb & Whishaw, 2001, which provided the basis hand (a language task or a task that taps into for the above summary of these neuroimaging another cognitive domain). In addition to being techniques.) used with neurologically intact people, these techniques can also be exploited in clinical popu- Conclusions on the role of particular brain lations to see how electrical or metabolic activity structures in linguistic functioning that are deviates from the patterns observed in healthy based on brain damage must be met with caution. individuals. The reason is that the disorder caused by the damaged structure does not unambiguously Electrical and magnetic signals point at the structure’s normal function. In terms of an analogy used by Gazzaniga et al. (2009, One of the techniques that exploit the brain’s p. 148): electrical activity while it is busy performing some task concerns the registration of “event-related [. . .] allowing the spark plugs to decay or potentials” (ERPs). It makes use of the fact cutting the line distributing the gas to the that the electrical activity of the brain is modu- pistons will cause an automobile to stop lated by sensory, motor, and cognitive events. running, but this does not mean that spark In the participants of these studies, an electro- plugs and distributors do the same thing; encephalogram (EEG) is registered that records rather, their removal has similar functional the electrical activity of thousands or millions consequences. of neurons in the cortex by means of a number of recording electrodes placed over the scalp. To illustrate the ambiguity with a specific lan- A reference electrode is placed at some dis- guage disorder, consider one of the characteristic tance from these electrodes and the potential symptoms of damage to Broca’s area, a brain

408 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS Electrode positions and names according to the 10-20 left hemisphere, have odd numbers; those placed electrode system. F = frontal; T = temporal; P = parietal; O = on the right part of the scalp have even numbers. occipital; C = central; FP = frontal pole; Z = 0/zero. Electrodes placed on the midline of the head are labeled with a Z (for 0/zero, to avoid confu- differences between the recording electrodes and sion with the letter O). The letters F, P, T, the reference electrode are recorded. While the and O refer to the four main parts of the cerebral EEG is registered, once in a while a stimulus of a hemispheres, called lobes, above which the specific type is presented (the so-called “event”). electrodes are positioned: the frontal, parietal, This produces a small voltage change in the EEG temporal, and occipital lobe. The upper part signal, the ERP. The voltage change caused by a of Figure 8.2a provides a “lateral” view of single trial is too small to be detected, but when a the four lobes of the left hemisphere, looking number of stimuli of the same type are presented, at them from the side of the brain, as illustrated and the voltage changes caused by all of them in Figure 8.2b. The lower part of Figure 8.2a are averaged, the voltage changes caused by the provides a “medial” view of the four lobes of critical stimuli can be separated from the back- the right hemisphere, looking at them from the ground noise and can be related to a specific middle of the brain, as illustrated in Figure 8.2b. cognitive process (such as syntactic or semantic The electrodes with the letters Fp are located analysis). above the “frontal pole”, the most anterior area of the frontal lobe. The electrodes labeled with The electrode positions and names used in a C lie above the “central sulcus” or “central EEG/ERP registration are standardized. fissure”, the cleft that divides the frontal and Figure 8.1 illustrates the original 10-20 inter- parietal lobes. (In addition to lateral and medial national system (Jasper, 1958), which includes 19 views of the brain, journal articles often present electrodes. illustrations of the brain looking at it from above or below; these views are called “dorsal” and All electrode names consist of one or more “ventral”, respectively.) The “lateral sulcus”, the letters in combination with a number. Electrodes cleft that divides the frontal and temporal lobes, placed on the left part of the scalp, above the is also called the “Sylvian fissure”. As we shall see, areas on both sides of this sulcus (so called “perisylvian areas”) are usually recruited during language processing. The 19 electrodes in the original 10-20 standard system suffice to detect variations in the EEG pattern that are associated with dif- ferent states of consciousness (such as an alert state, drowsiness, and deep sleep) and to detect deviations from these various normal states caused by congenital abnormalities or afflictions such as epilepsy. However, ERP research often requires a higher density of electrodes. Exten- sions of the 19 electrodes system, up to 256 electrodes, have therefore been developed in which additional electrodes are placed in between and beyond those in the original system and the electrodes in the original system are accordingly renumbered. The ERP signal incited by a particular stimulus (the “event”) and extracted from the

8. BILINGUALISM AND THE BRAIN 409 (a) The frontal, parietal, occipital, and temporal lobes of the cerebral cortex as seen from a lateral view of the left hemisphere (top) and from a medial view of the right hemisphere (bottom). (b) The terms “lateral” and “medial” explained.

410 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS EEG consists of a wave pattern of positive and a negative polarity that typically peaks at 400 ms negative peaks called “components” (see pp. 43 after stimulus onset. It is evoked by every content and 249 for illustrations). Different components word in a sentence, is generally thought to index reflect different cognitive processes and each semantic processing, and is modulated by various component is characterized by a number of factors such as how well the word fits the context features. The first characteristic of a component is semantically and its frequency of occurrence its polarity; that is, whether it concerns a positive (Kutas & Hillyard, 1980). The latency of a or negative voltage change relative to the base- particular component can vary as a function of line. Positive and negative voltage changes are factors such as the age of the participants and generally (not always), and counterintuitively, their level of expertise in the domain tapped by plotted downward and upward, respectively. The the experimental task. In other words, the latency second of a component’s characteristics is its part of a particular component’s name (e.g., 400) latency, the time duration, expressed in milli- is not a perfectly reliable marker of the time it seconds, between stimulus presentation and the takes the cognitive process underlying this com- moment the component reaches its peak voltage ponent to develop. An alternative to naming level. The third characteristic is the component’s components after their polarity and latency is amplitude; that is, the size of the voltage change to use names that refer to the cognitive functions at its maximum. Finally, the topography of a the components are assumed to index. For component, often visualized in a so-called instance, the P600 is also called the syntactic topographical map, refers to where on the skull’s positive shift (SPS) because it is assumed to reflect surface (at what electrode positions) the associ- syntactic processing. A drawback of this naming ated electrical activity is detected. If two different convention is that the initial hypothesis regarding stimulus types (e.g., nouns versus function words the component’s function may eventually turn such as articles or prepositions) show different out to be incorrect, or that the component is not topographical maps, this suggests that the two only aroused by the cognitive operation suggested stimulus types are processed by different neural by its name but by other cognitive processes as systems. However, the exact structures involved well. are not revealed by the associated topographical maps. The reason is that electric signals are The specific combination of values on the distorted during the trajectory from their neural above-mentioned four features that typify indi- source to the scalp through the brain tissue in vidual ERP components provides information between the two. ERP components can be on the nature, strength, and development of the “exogenous” or “endogenous”. Exogenous com- cognitive operations that underlie the various ponents reflect physical features of the stimulus components. For instance, if two of an ERP such as its intensity, whereas endogenous signal’s components differ in polarity and components index the cognitive effects of the topography, this suggests that the two com- stimulus. ponents are generated by qualitatively different cognitive processes (e.g., syntactic vs. semantic ERP components are often named after their analysis) that are subserved by different neural polarity and latency so, for instance, the P600 is a networks. A component’s latency and amplitude component with a positive polarity that typically indicate how the underlying cognitive process peaks at 600 ms after the onset of the stimulus. develops over time and to what extent it is It is evoked by a number of syntactic anomalies implicated, respectively (Hagoort & Ramsey, such as a violation of grammatical number 2004; Hahne & Friederici, 2001). As compared agreement between a sentence’s subject and with behavioral measures (such as registering the verb (e.g., Hagoort, Brown, & Groothusen, 1993; time it takes participants to produce a particular Osterhout & Holcomb, 1992) and by syntactic vocal or manual response to a stimulus), an ambiguities (e.g., Kotz, Holcomb, & Osterhout, advantage of the ERP methodology is that 2008). In contrast, the N400 is a component with components show up in the signal without

8. BILINGUALISM AND THE BRAIN 411 participants having to respond overtly to the advantage is due to the fact that, unlike electric stimulus. The participants’ task may, for instance, signals, magnetic signals are not distorted on simply be to listen to sentences. Not requesting an their way from a deeper neural source through overt response guarantees that the ERP signal is more superficially located brain tissue to the not contaminated by any contribution of such a scalp (Gazzaniga et al., 2009). However, a dis- response to the signal and provides a relatively advantage is that MEG can only detect the neural pure marker of the underlying cognitive process activity flowing parallel to the skull’s surface of interest. A further advantage of the ERP whereas EEG can also detect neural activity methodology is its high temporal resolution, flowing at a right-angle to the skull (e.g., Hagoort which means that the cognitive process of interest & Ramsey, 2004). Because in the sulci of the leaves its mark on the signal within milliseconds brain in particular the activity flows parallel after its onset. In this respect the ERP technique to the surface of the skull, it is especially the compares favorably with techniques that measure activation flow in the sulci that is detected by the brain’s metabolic activity during cognitive means of MEG. processing (see below). A specific disadvantage of the ERP technique has already been mentioned, Hemodynamic signals namely that the signal that it produces has a low spatial resolution. This means that the electrical Whereas EEG and MEG measure neural activity activity detected at the scalp does not provide directly, the remaining functional imaging tech- reliable information about the exact location of niques to introduce here, positron emission tom- the signal’s neural source, the brain structure ography (PET) and functional magnetic resonance that causes it. As we will see below, in this respect imaging (fMRI), do so indirectly by inferring the common techniques to detect the brain’s neural activity in a specific brain region from the metabolic activity are superior to ERP measure- increased blood supply to that region when it is ment, but also magneto-encephalography (MEG) involved in some cognitive task. The increased fares better than ERP measurement in source blood supply provides active neurons with the localization. additional energy, in the form of oxygen and glucose, that they need to do their specialized MEG is related to EEG/ERP but it measures work. The medical term for the dynamic regula- the magnetic correlate of the electrical activity in tion of the blood flow in the brain is “hemo- the neurons upon the presentation of a critical dynamics” and the change in blood flow related to stimulus (again called the “event”). It was neural activity is called the “hemodynamic mentioned above that in ERP research a voltage response”; hence the name of the signals pro- change caused by a single trial is too small to duced by these techniques: “hemodynamic be detected but that averaging over a number of signals”. trials of the same type reveals a voltage change that can be related to a cognitive process. To produce a PET image of the parts of the Similarly, in MEG studies the magnetic field brain that are active while a specific task is being aroused by an individual stimulus is too small performed, a small amount of water labeled with to be observed, but averaging over a series of a radioactive element (an “isotope”) is injected similar trials produces a measurable event-related into the bloodstream while the participant is magnetic field. The magnetic activity is registered engaged in the task. In cognitive activation by means of tens to hundreds of sensors in a studies most often a radioactive form of oxygen, helmet-like MEG scanner that surrounds the oxygen-15, is used, which subsequently accumu- participant’s head. The temporal resolution of lates in different brain areas in direct proportion the MEG signal is as good as that of the ERP to the amount of blood supply to those areas. The signal and, in comparison to ERP measurement, nuclei of oxygen-15 isotopes are unstable and MEG has the advantage that it is relatively easy emit a positron from their nucleus during decay. to localize the signal’s neural source. This This positron collides with an electron in the

412 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS blood. During this collision the mass of positron from the blood supply to the same region in the and electron is converted into two photons that experimental condition. A statistically significant move in opposite directions and are simul- difference in blood supply to one or more regions taneously detected by two of the photon in the two conditions indicates that these brain detectors that encircle the participant’s head and regions are recruited by the cognitive component that together constitute the PET camera. The under study. simultaneous detections are counted and the total count during 1 minute following the injec- Just as PET, fMRI makes use of the fact that tion is converted into an image of the blood active parts of the brain are supplied with more flow in the brain: The larger the simultaneous blood than inactive parts, but whereas PET photons count, the more collisions between requires the injection of radioactive elements into positron and electron in the blood must have the bloodstream, fMRI is non-invasive. It exploits taken place and, hence, the more blood there the physical properties of elements that occur must have been in the region where the collisions naturally in blood, specifically the magnetic took place. This, in turn, provides an indirect properties of iron-holding hemoglobin in the measure of neural activity in that region. The blood cells, and makes use of the MRI scanners measure in which the results of PET studies are that are also used for structural MRI (see above). usually expressed is the so-called “regional cere- Hemoglobin transports oxygen through the bral blood flow” (rCBF); that is, the distribution blood. The oxygen is absorbed by active brain of the blood supply in the brain. Because it areas and as a consequence the hemoglobin takes relatively long—around 60 seconds—to becomes deoxygenated. Deoxygenated hemo- get a clear image of the increase in rCBF as a globin has so-called “paramagnetic” properties consequence of increased neural activity, as and can be detected by the fMRI equipment. compared with the ERP method described More precisely, the equipment measures the above, the temporal resolution of PET is rather ratio between oxygenated and deoxygenated poor. Its spatial resolution is relatively good how- hemoglobin. This ratio is called the BOLD ever, revealing active brain regions of a volume of signal (for “blood oxygenation level dependent”). 5–10 mm3. Even though active areas consume extra oxygen, this ratio is higher, not lower, in brain areas PET studies typically compare changes in with increased neuronal activity. In other metabolic activity in specific brain areas in at least words, the proportion of oxygenated hemo- two conditions: an experimental condition and globin is relatively large in active brain tissue some control or baseline condition. The control even though active tissue absorbs a relatively condition can be a resting state in which the par- large amount of oxygen. This phenomenon— ticipants do not have to perform any task, or it paradoxical at first sight—is explained by assum- may involve a task that differs from the experi- ing that the absorption of oxygen by active mental task in one or more critical respects. For cells results in an immediate increase of blood example, in a bilingual study the comparison flow to that part of the brain, so much so that not might be between picture naming in L1 (the all of the additionally supplied oxygen can be control condition) and L2 (the experimental con- absorbed, and a local surplus of oxygen is cre- dition) or between word naming in L1 (control) ated. Like PET studies, fMRI studies typically and translating L1 words into L2 (experimental). compare brain activity in an experimental condi- Ideally, the control and experimental conditions tion and some well-chosen control condition only differ from one another in the specific that is assumed to involve the same cognitive pro- cognitive component that is being examined. The cessing components as the experimental task blood supply to a specific region measured in the except for the one critical cognitive component control condition—as inferred from the photon being examined. The subtraction procedure sub- count based on positron–electron collisions sequently reveals the brain area(s) recruited by taking place in that region—is then subtracted this component.

8. BILINGUALISM AND THE BRAIN 413 A clear advantage of fMRI as compared with language lateralization in monolinguals and PET is that the fMRI method’s non-invasive bilinguals has also been investigated, and con- character permits more trials per participant than tinues to be, by means of older methods. The PET, where the fact that radioactive material has oldest one is the post mortem study of the to be injected causes a constraint on the number site of brain lesions in bilingual aphasic patients. of data points that can be collected from an indi- If language lateralization differs between mono- vidual participant. This allows for more advanced linguals and bilinguals, the lesion site should experimental designs with fMRI (see e.g., differ between monolingual and bilingual aphasic Gazzaniga et al., 2009, and Hagoort & Ramsey, patients. For instance, if the specific form the 2004, for details). Furthermore, the spatial monolingual–bilingual lateralization difference resolution of fMRI scans is better than that of takes is that both languages of a bilingual are PET scans, revealing structures of 3 mm3 in more often subserved by the right hemisphere volume. In comparison to EEG/ERP and MEG, in bilinguals than in monolinguals (see the temporal resolution of fMRI is poor, but pp. 429–432 for a number of different views on better than of PET, taking a couple of seconds. how language lateralization might differ Finally, for an image to be created by means of between monolinguals and bilinguals), the inci- PET, a participant must perform a particular task dence of so-called “crossed aphasia”—aphasia for at least 40 seconds, and most fMRI studies following a lesion in the right instead of the left employ a design in which the task is performed hemisphere—should be larger in bilinguals than during 20 to 30 seconds and the fMRI signal is in monolinguals. summed over this time duration. However, with fMRI it is also possible to measure the BOLD In addition to the lesion paradigm and the signal time-locked to a single stimulus, a method recent neuroimaging techniques, three non- that is known as “event-related fMRI”. Just as in invasive behavioral techniques to examine ERP studies, averaging over a number of repeti- cerebral lateralization of language (and other tions of the stimulus produces a measurable cognitive functions) exist that emerged around signal. 1970: the “dichotic listening paradigm”, the “tachistoscopic viewing paradigm”, and the Further methods “verbal–manual interference paradigm”. In the dichotic listening paradigm two different In the introduction to this chapter I mentioned auditory inputs, each for instance consisting of a that, in the vast majority of monolingual lan- series of words, are presented simultaneously, guage users, the left hemisphere plays a pre- one to the left ear, the other to the right ear. The dominant role in language processing; that is, stimuli presented to the right ear are predomin- language is lateralized to the left hemisphere. In antly processed by the left hemisphere and those addition I introduced the idea that hemispheric presented to the left ear are predominantly pro- lateralization in bilinguals, for one or both cessed by the right hemisphere. Immediately after languages, is less pronounced than in mono- stimulus presentation the participants have to linguals—in other words, that language is more recall as many as possible of the items presented bilaterally localized in bilinguals. Since its to both ears. Left hemisphere specialization or emergence around 1990, functional neuroimaging dominance for language processing would show has become an increasingly popular method to from better recall performance for the stimuli pre- test this idea. Different language lateralization in sented to the right ear than for those presented to monolinguals and bilinguals would show from the left ear. different distributions of neural activity in the two hemispheres when participants from these In the tachistoscopic viewing paradigm visual two populations perform one and the same stimuli such as written words are very briefly and language task. The possibility of differential randomly presented in the left or right visual field while participants fixate their eyes on a fix- ation point in the center of the field. This manner

414 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS of stimulus presentation can nowadays be pro- left-index-finger tapping conditions thus provides grammed on a computer, but because these were a measure of the degree of cerebral asymmetry of not widely available when this paradigm was language. first developed a piece of equipment called a “tachistoscope” was used for this purpose, hence All three paradigms can be used to test the the paradigm’s name (the word tachistoscope hypothesis that language lateralization differs derives from the Greek words for rapid and between monolinguals and bilinguals by simply vision). The visual information presented in the comparing the performance of monolinguals and right visual field is sent directly to the left bilinguals in an experiment that implements hemisphere, whereas the information presented one of these paradigms. In addition, all three in the left visual field is sent to the right hemi- paradigms provide information on the precise sphere. The degree of involvement of each hemi- nature of the lateralization difference. For sphere in language processing is inferred from instance, if language is less lateralized in the recognition score (and recognition speed) bilinguals than in monolinguals (that is, the for the words presented to the one or the other two hemispheres are more evenly involved in lan- visual half-field. Left hemisphere dominance guage processing in bilinguals), the size difference for language would show from higher recognition between the interference effects in the right and scores for items presented in the right visual left tapping conditions should be smaller in field than for items presented in the left visual bilinguals than in monolinguals. In contrast, if field. the cerebral lateralization of language is as asymmetric in bilinguals as in monolinguals but it Finally, the verbal–manual interference is the right rather than the left hemisphere that paradigm exploits the fact that in most people dominates language processing in bilinguals, the the left hemisphere not only dominates language interference caused by a concurrent verbal task use but also controls the dominant right hand. on the rate of finger tapping should be largest In experiments that implement this paradigm, when the tapping must be done with the left participants are first asked to tap a key as rapidly index finger. Yet further possibilities (e.g., left as possible with one of their index fingers for lateralization for the L1 and right lateralization some length of time and then to do this with the for the L2) can also be inferred from the pattern other index finger. The tapping rate for each index of interference effects to emerge. A limitation finger is measured. Following this control condi- of all three techniques is that, unlike the neuro- tion the participants perform the finger-tapping imaging techniques, they do not provide any task once again but this time they concurrently information on the exact locus of processing have to perform some verbal task. If language within a hemisphere. primarily resides in the left hemisphere, finger tapping with the right index finger, controlled by Just as the various functional neuroimaging the left hemisphere, should exhibit strong inter- techniques, the above three behavioral paradigms ference from the concurrent verbal task and the to localize language (and other cognitive func- tapping rate should be lower than in the control tions) in the brain can be used, and in fact are condition. Although interference from the con- predominantly used, with healthy participants. A current verbal task will also emerge when tapping couple of further techniques are invasive and are is done with the left index finger (performance sometimes applied to neurologically damaged under dual-task conditions is generally poorer patients prior to neurosurgery. One of them is than when the participants can concentrate on a direct electrical stimulation of the brain. It was single task), the interference effect (the difference developed in the 1940s by the Canadian neuro- between the tapping rates in the control and surgeon Wilder Penfield and has been used since experimental conditions) should be largest in during neurosurgical operations to precisely the right-finger tapping condition. The size of the demarcate the cortical areas that subserve specific interference effect in the right-index-finger- and cognitive functions. This way decisions can be made concerning the best way to perform the

8. BILINGUALISM AND THE BRAIN 415 surgery, thus reducing the risk of negative side disconnecting the two hemispheres in this way, effects. The technique involves the application the communication between the two hemispheres of a weak electrical current to the brain of the is impaired. Differential involvement of the two conscious patient by means of a bipolar electrode. hemispheres in language processing can then be This causes a transient inhibition of the function examined by presenting language materials, for subserved by the stimulated brain region. While instance words or sentences, in either the left the current is applied, the patient must perform visual field or the right visual field and asking some cognitive task, for instance provide the the patient to perform certain language tasks names of depicted objects. If task performance (e.g., provide the meanings of the words). The fails during electrical stimulation but is technique can obviously be applied with both unhampered otherwise it can be concluded that monolinguals and bilinguals who have undergone the stimulated brain area is normally involved in this type of surgery, but given the low incidence the cognitive function tapped by the task. By of split-brain patients the chances are small applying currents systematically to neighboring that sufficiently large numbers of matched mono- areas, the borders of the area subserving the func- lingual and bilingual patients will be found to tion can be charted. Ojemann and his colleagues inform the differential lateralization debate. (e.g., Calvin & Ojemann, 1994; Ojemann & Whitaker, 1978) have employed electrical stimula- Finally, transcranial magnetic stimulation tion to locate the brain tissue involved in language (TMS) is a non-invasive technique that creates processing in both monolinguals and bilinguals transient virtual brain lesions by briefly disrupt- (see Fabbro, 1999, for a more detailed description ing neural processing. It involves the generation of the technique). of a strong electrical current in a device held by the researcher against the participant’s head. A second technique used with patients is the This current generates a magnetic field, which Wada test, named after Juhn Wada, who first passes through the skin and skull and causes the used it around 1960 to determine which hemi- neurons in the underlying brain region to fire. By sphere was dominant for language in patients stimulating the neurons in this way, their normal who were about to undergo neurosurgery. The activity is disrupted briefly, as indicated from barbiturate amobarbital is injected to one performance failures on tasks that tap into the hemisphere via an artery (the carotid artery) cognitive function subserved by the stimulated that leads into this hemisphere. This causes a region. The method was originally developed for temporary anesthesia of this hemisphere which clinical purposes but can be used with neuro- lasts several minutes, and a language test is logically intact people as well. A recent report on administered within this period to determine the two bilinguals who received a TMS treatment for role of this hemisphere in language processing. major depression (Holzheimer, Fawaz, Wilson, & Some time later the procedure is repeated, now Avery, 2005) provided a first indication that the injecting the artery that leads into the other technique might be useful for examining language hemisphere. Comparing the performance of the processing in bilinguals. patient on the language test in the two conditions reveals which of the two hemispheres is dominant Nomenclature for language. In publications on cognitive neuroscience the Functional specialization of the two hemi- various regions of the cerebral cortex are labeled spheres can also be examined on “split-brain” according to three conventions. The first concerns patients. These patients have undergone surgery a labeling of the main divisions of the cerebral in which a large part of the corpus callosum, a hemispheres formed by the brain’s main gyri white matter tract that connects the two hemi- and sulci. Figure 8.2a above illustrated this con- spheres with one another and that consists of vention, except that the complete nomenclature hundreds of millions of axons, is cut through to makes more fine-grained divisions than in terms relieve the symptoms of epilepsy. As a result of

416 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS The main gyri and sulci of the cerebral cortex as seen from a lateral view of the left hemisphere. of just the four main lobes and the main sulci “left” and “right” refer to the left and right hemi- shown there. These subdivisions are shown in spheres, not to the left or right part of a specific Figure 8.3, which provides a lateral view of the region in either the left or right hemisphere. left hemisphere. The second way of labeling specific brain areas To extend the degree of precision in indicating is according to the so-called “cytoarchitectonics” specific brain areas, conventional names are used of the brain; that is, in terms of the micro- for the top, bottom, front, and back of the brain anatomy of the different cell types and their as a whole and of its anatomical subregions. The organization and distribution in the brain. The top part is referred to as either “superior” or person best known for this approach is Korbinian “dorsal”; the bottom part is called “inferior” or Brodmann, who about a century ago (in 1909) “ventral”; the front is labeled “anterior” or “ros- subdivided the cerebral cortex into 52 areas tral”; the back is named “posterior” or “caudal”. based on minute microscopic examination of the This convention is illustrated in Figure 8.4a for cell structures in the cortex. Up to this day the cerebral cortex as a whole and in Figure 8.4b the numbering system introduced by Brodmann is for the left temporal lobe. Notice that, as a con- used widely in neuroscience and the pertinent sequence of this convention, the orientation of publications refer to the areas that he dis- different subregions indicated by the same name, tinguished by means of numbered “Brodmann say, superior, may differ (cf. the orientation of the areas” (BAs). Figure 8.5a shows Brodmann area called superior in Figures 8.4a and 8.4b). areas on a lateral view of the left hemisphere. According to this naming convention, a specific Other Brodmann areas are located inside the brain area might thus be referred to as the brain; these are shown in Figure 8.5b on a medial “left posterior (or “caudal”) middle frontal view of the right hemisphere. According to this gyrus” and another might be called the “anterior naming convention, Broca’s area (in the left (or “rostral”) left middle temporal gyrus”. inferior frontal gyrus) would be indicated by Alternatively, acronyms of the full names are BAs 44 and 45, and a second area well known used, such as LIFG for “left inferior frontal for its involvement in language processing, gyrus” or STG for “superior temporal gyrus”. It Wernicke’s area (in the left superior temporal is important to realize that in these designations gyrus) would be referred to as BA 22. Most

8. BILINGUALISM AND THE BRAIN 417 Brodmann areas occur symmetrically in both Conventional names for the hemispheres. top, bottom, front, and back of the cerebral cortex as a Finally, the third convention is to refer to brain whole and of its subregions, areas according to their function. For instance, as illustrated on a lateral the “motor cortex” (BA 4, on the anterior side view of (a) the left of the central sulcus) is recruited during the hemisphere and (b) the left execution of movements, the “somatosensory temporal lobe. cortex” (BAs 1, 2, and 3, posterior to the central sulcus) represents sensory information such as lated with speech devoid of syntax, one might smells and touch, and the “primary visual cortex” consider calling it the “syntax cortex”, but as (BA 17, in the medial posterior part of the already alluded to above, the syntactic problems occipital lobe) processes visual information. exhibited by most people with Broca’s aphasia When the convention was described above to may have a different cause than corrupted name ERP components after their function it was syntactic knowledge, such as a shortage of com- mentioned that it involves the risk that the initial putational resources (e.g., Kolk & Heeschen, hypothesis regarding the component’s function 1990) or a general problem of unifying smaller may later turn out to be wrong. Labeling brain linguistic entities into larger wholes (Hagoort, areas according to the function they are thought 2005). A further comment on labeling brain areas to serve involves the same risk. For instance, by their function is that many brain functions because damage to Broca’s area is often corre- are subserved by networks consisting of different non-neighboring brain regions. The consequence would be that the chosen name should be assigned to each of these regions. A couple of further technical terms help to grasp the essence of neuroimaging publications. These publications often illustrate where neural activity is located on brain slices that cut up the brain according to three different orientations:

418 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS Brodmann areas as seen from (a) a lateral view of the left hemisphere and (b) a medial view of the right hemisphere. “coronal”, “sagittal”, and “axial”. These are illus- LANGUAGE AREAS IN THE BRAIN AND trated in Figure 8.6. Axial orientations are also THEIR FUNCTION known as “transversal” or “horizontal”. Two fur- ther terms that one encounters in these publica- The classical language areas: tions are “voxels” and “regions of interest” Broca’s area, Wernicke’s area, and (ROIs). Voxels are cube-shaped elements in which their connection brain signals are recorded. ROIs are the brain regions the researcher focuses on in the data As already mentioned, from post mortem investi- analyses, and usually cover many voxels. They are gations of the brains of aphasic patients it has chosen on the basis of established knowledge and long been known that there are at least two areas informed hypotheses about the brain regions that in the left hemisphere that are recruited during might be involved when the participants perform language use. Broca (1861) described a patient the specific experimental task being used. For who was completely unable to speak while his instance, in studies on language processing the comprehension was relatively good. Post mortem selected ROIs will typically include Broca’s area examination of this patient’s brain showed that and Wernicke’s area.

8. BILINGUALISM AND THE BRAIN 419 Conventional names for the three types of cut-through sections of the brain used in neuroimaging studies. he had a lesion in the middle posterior part of connected directly as well as indirectly via the the inferior frontal gyrus in the left hemisphere concept center. (BAs 44 and 45). About a decade later Wernicke (1874) reported on a patient who exhibited the Lichtheim’s model has been extremely opposite behavioral pattern, demonstrating poor influential in aphasia research for over a century, understanding of spoken language while at the serving as a productive framework for advancing same time producing fluent but unintelligible hypotheses on various forms of aphasia. Damage speech. An autopsy on this patient’s brain showed to Broca’s area was thought to lead to a pattern a lesion in the left superior temporal gyrus (BA of symptoms that are now collectively known as 22). The brain’s language areas identified in “Broca’s aphasia”. Consistent with Lichtheim’s these pioneering studies are now called Broca’s hypothesis that Broca’s area stores the motor area and Wernicke’s area and, together with the representations of words, the speech of patients “arcuate fasciculus” (see below), they are often with lesions in this area is very effortful, non- called the classical language areas. About another fluent, and poorly articulated. A further charac- decade later, Lichtheim (1885) developed a func- teristic of the speech of these patients is that tional model of language that encompassed three the sentences they utter are telegraphic and language centers in the brain, two of them being “agrammatic”; that is, they are simple and short Broca’s area and Wernicke’s area. According and devoid of grammatical markers and function to this model, Broca’s area stores the motor words. Because Lichtheim’s model deals with representations of words and Wernicke’s area the storage of various types of word knowledge, represents the auditory forms of words (that is, not syntax, it cannot readily account for this their phonological representations). The third particular symptom of Broca’s aphasia. A final language center, the “concept center”, was salient characteristic of the linguistic behavior of thought to store conceptual representations. In Broca’s aphasic patients is that their comprehen- the model the auditory and motor areas were sion is relatively good, just as it was in Broca’s original patient, although they typically run into

420 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS comprehension problems when the sentences to concept center and Broca’s area; the latter to process are grammatically complex. The general damage to brain structures between the concept hypothesis that emerged from this behavioral center and Wernicke’s area (see Fabbro, 1999, for pattern was that Broca’s area supports speech a description of the associated functional dis- production. In contrast, “Wernicke’s aphasia”, orders). Figure 8.7 shows the brain locations caused by damage to Wernicke’s area, is charac- of the classical language areas: Broca’s area, terized by the opposite behavioral pattern: As Wernicke’s area and the arcuate fasciculus that already noticed by Wernicke, damage to this area connects them. results in fluent but nonsensical speech accom- panied by severe comprehension problems. This In between Broca’s and Wernicke’s seminal led to the hypothesis that Wernicke’s area is the publications, the idea that a bilingual’s or multi- site of speech comprehension. lingual’s multiple languages might be represented in different brain areas was suggested for the first Lichtheim’s model and more recent revisions time (Scoresby-Jackson, 1867), was rejected of it (Geschwind, 1970) have served as a frame- about three decades later by Pitres (1895), and re- work to account for other forms of aphasia as emerged again in the 1970s (see Paradis, 1995, for well. For instance, “conduction aphasia”, charac- a brief history). It has stayed with us ever since terized by relatively good auditory compre- and the debate on language representation in hension and fluent speech but the inability to bilinguals and multilinguals is currently gathering repeat speech, has been attributed to damage momentum now that non-invasive functional to the arcuate fasciculus, a bundle of fibers that neuroimaging techniques have become available connects the language areas in the temporal that can spot brain activity while living persons (Wernicke) and frontal (Broca) lobes. If all of perform language tasks. these areas are damaged, “global aphasia” emerges, characterized by overall poor linguistic Functional equivocality of the classical performance. Other forms of aphasia that have language areas been explained in terms of the model are “transcortical motor aphasia” and “transcortical The view that Broca’s and Wernicke’s areas sensory aphasia”. The former has been attributed subserve speech production and comprehension, to damage to the brain structures between the respectively, is no longer widely accepted because The classical language areas in the brain: Broca’s area (BAs 44 and 45), Wernicke’s (BA 22), and the arcuate fasciculus, the connection between Broca’s and Wernicke’s areas.

8. BILINGUALISM AND THE BRAIN 421 a substantial amount of evidence against this the validity of this hypothesis. On the one hand, division of labor between these two brain areas studies manipulating the semantics of the pre- has accumulated during the past decades. This sented stimulus materials have shown that these counterevidence comes from both lesion studies manipulations not only influence the activation in and, since around 1990, increasingly from neuro- Wernicke’s area but also in Broca’s area and, on imaging studies in which language processing in the other hand, stimulus materials that obviously neurologically healthy people is examined. In require syntactic analysis do not always activate a review of the latter type of studies, both PET Broca’s area. A further review of neuroimaging and fMRI, Stowe, Haverkort, and Zwarts (2005) studies that searched for the neural substrate of concluded that it is beyond any doubt that the syntax in the normal population (Kaan & Swaab, left inferior frontal gyrus (encompassing Broca’s 2002) similarly led to the conclusion that Broca’s area) and the superior temporal gyrus (encom- area is not always involved in syntactic processing passing Wernicke’s area) are indeed essential and, conversely, that many parts of the temporal for language use but that the hypothesis that the lobe, including Wernicke’s area, also support former subserves language production and the syntactic processing. latter language comprehension is not supported by the evidence. The production/comprehension A number of lesion studies also provide model predicts that tasks that only involve evidence against the idea that Broca’s area is comprehension and tasks that only involve the site of syntax. It has for instance been shown production selectively activate Wernicke’s area that patients performing poorly on tests of and Broca’s area, respectively. Yet many of the mopho-syntax shared damage to Wernicke’s area studies that employed either comprehension or but not to Broca’s area, while some patients production tasks showed activation in both with a lesion in Broca’s area did not show severe areas. This result emerged both from studies problems with morpho-syntax (Dronkers, that examined the processing of single words and Wilkins, Van Valin, Redfern, & Jaeger, 1994). from those investigating sentence processing. Furthermore, patients with lesions in Broca’s The authors therefore concluded that Broca’s area who exhibit agrammatic symptoms do not and Wernicke’s areas contribute in some way consistently do so under all circumstances but to both comprehension and production (Stowe especially under computationally taxing circum- et al., 2005). stances (Haarman & Kolk, 1994). The depend- ency of agrammatic symptoms on computational A second view on the different functional load suggests that Broca’s area is not the roles of Broca’s area and Wernicke’s area, which site where grammatical knowledge is stored, became popular around 1970, is that the former because if it were, the linguistic expressions of is the site of syntactic knowledge, supporting the afflicted persons should invariably be agram- syntactic processing in both production and matic, irrespective of the current computational comprehension, whereas the latter stores word demands. All in all it appears that the view meanings and subserves lexical semantic process- that Broca’s and Wernicke’s areas subserve ing. Consistent with this view is the finding that syntactic and semantic processing, respectively, the speech of people with Broca’s aphasia is does not fare any better in accounting for the characterized by a lack of grammar. It also agrees evidence than the view that they support speech with the observation that many of these patients production and comprehension, respectively. cannot fully grasp the meanings of sentences if full understanding cannot be solely based on the Stowe et al. (2005) discussed a third view on meanings of the constituent words combined the function of Broca’s area, one that is consistent with general world knowledge but requires with the variability of agrammatical performance grammatical analysis as well (e.g., Caramazza & in patients with lesions in this area. It holds that Zurif, 1976). However, the neuroimaging data Broca’s area supports some type of computation reviewed by Stowe et al. (2005) also cast doubt on on knowledge units retrieved from knowledge stores elsewhere in the brain and that it maintains

422 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS these knowledge units for the duration of the uses the term “unification” to refer to the process computation. In other words, according to this of integrating the information retrieved from the view, Broca’s area is a working memory system lexicon, be it syntactic, semantic, or phonological that temporarily stores and processes informa- information, into a representation of multiword tion, be it syntactic or of any other type, linguistic expressions. Finally, the control component sees or non-linguistic. A specific version of this view is to it that the language system operates in agree- advanced by Hagoort (2005), who approached ment with the current communicative circum- the question of which brain areas are committed stances and the goals of the interlocutors. For to language from a different perspective than instance, it enables a bilingual (or bi-dialectal) is mostly adopted. He started out by noting speaker in unilingual settings to produce output that researchers generally address this question in the selected language (or dialect) or to switch from an experimental-task perspective, trying languages in translation settings and a mono- to find out what brain areas are involved during lingual speaker to select the contextually the execution of specific language tasks (e.g., verb appropriate register (e.g., formal or casual; child generation, picture naming, word reading). To directed speech, “motherese”, or adult directed be able to make better sense of the steadily speech). In short, according to Hagoort’s design increasing body of data gathered in perspective, linguistic behavior involves a specific neuroimaging studies, he then proposed a “design type of computation on specific types of lin- perspective” instead; that is, an approach that guistic knowledge structures with a specific goal starts out by charting the various cognitive in mind. components that enable language use, both com- prehension and production. After these com- Assuming that the language system indeed ponents have been identified, hypotheses are put consists of these three functional components, forward about the brain regions that subserve what are the brain areas that subserve them? them. On the basis of a review of the neuroimaging literature, Hagoort (2005) pointed to the left Adopting this approach, Hagoort (2005) inferior frontal gyrus (including Broca’s area, hypothesized the existence of three such com- BAs 44 and 45, but also covering BA 47) as the ponents, which he called “memory”, “unifica- unification site and localized the memory tion”, and “control”, and that together he function in a large region in the left temporal lobe referred to as the MUC framework. In his ter- (including Wernicke’s area, BA 22, in the superior minology, “memory” refers to long-term memory temporal gyrus but also covering more anterior information, specifically information in the and inferior regions: BA 38, also called the mental lexicon. Importantly, Hagoort adheres to “temporal pole”, and BA 21 in the middle tem- the view that the mental lexicon not only contains poral gyrus). Interestingly, different subregions information on the phonological, phonetic, within the left inferior frontal gyrus appear to syntactic, and conceptual properties of words, be specialized for unification of different types but also so-called “structural frames” that enable of information: phonological, syntactic, and grammatical analysis of sentences. This is in semantic (see e.g., Stowe et al., 2005, and Paulesu line with a class of linguistic theories, called “lexi- et al., 1997). Finally, Hagoort localized the con- calist”, that no longer assume a distinction trol function in a network in the frontal lobe between lexical items on the one hand and syn- containing the anterior cingulate cortex (ACC; tactic rules on the other hand (e.g., Jackendoff, BA 24 and BA 32) and the dorsolateral prefrontal 2002; Vosse & Kempen, 2000). Instead, all gram- cortex (DLPFC, BA 46 and BA 9). Later on matical information, including the knowledge (pp. 435–437). I will provide more details on that enables syntactic parsing, is assumed to be these two regions, which are generally assumed stored in the mental lexicon. In addition to to be involved in executive control and to be storing information, the memory component is domain independent (that is, not specific for thought to be involved in retrieving it. Hagoort language). They also appear to play a crucial role

8. BILINGUALISM AND THE BRAIN 423 in bilingual language control; that is, the ability The brain regions assumed to of a bilingual to produce unilingual output subserve the memory (M), whenever such is required, mixed language output unification (U), and control (C) under circumstances where language mixing is components in Hagoort’s common and natural and does not hamper com- (2005) MUC model of munication, and to translate between the two lan- language use. guages whenever the bilingual speaker wishes to do so. visual cortex (BA 17 in the posterior part of the occipital lobe) will be involved when visual lan- With the exception of the ACC, the brain’s guage material is input; the motor cortex (BA 4) language-processing regions that are assumed in will be activated in all tasks that require speech the MUC model are indicated in Figure 8.8, output; and regions in the right hemisphere which shows a lateral view of the left hemisphere will be more heavily involved while reading (the ACC lies within the brain and can be seen in logographic print than while reading alphabetic Figure 8.9, p. 437). These regions coincide with print. language regions revealed in three further recent reviews of the neuroimaging literature on lan- So far I have presented three views on the func- guage processing (Démonet, Thierry, & Cardebat, tion of Broca’s area (it is committed to speech 2005; Indefrey & Levelt, 2004, which specifically production; it is the site of syntactic knowledge; it focuses on the neural substrate of speech produc- is some sort of working memory system) and two tion; Stowe et al., 2005), so there can be no doubt views on the role of Wernicke’s area (it subserves that they are all crucial for language processing. comprehension or lexical-semantic processing). Depending on the exact characteristics of the These views by no means exhaust the set of task and stimulus materials used in a specific hypotheses about the function of these lan- study (e.g., auditory or visual, recognition/ guage areas that have been proposed. A further comprehension or production; spoken or written hypothesis concerning the role of a large region in output; alphabetic or logographic reading the left temporal lobe, roughly covering the brain materials) additional areas will be activated. For area that Hagoort (2005) assigned the memory instance, the primary auditory cortex (BA 41 in function and including Wernicke’s area, was the posterior part of the superior temporal gyrus) advanced by Damasio, Grabowski, Tranel, will be activated with auditory input; the primary Hichwa, and Damasio (1996). These researchers hypothesized that the left temporal lobe is com- mitted to word retrieval and, furthermore, that different parts of it are specialized for retrieving different types of words. To test these ideas they presented a group of patients with lesions throughout the brain with a set of photographs showing faces of well-known people, animals,

424 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS or tools, and asked them to name the entities hypothesized that this area stores and retrieves on the photographs. The performance of the various aspects of lexical information, including patients was compared with that of a control the phonological forms (names) and syntactic group of neurologically healthy people and the and conceptual specifications of all the words patients showing abnormal naming performance known by the language user. In contrast, relative to this control group (30 out of 127) were Damasio and her collaborators assumed that singled out for further examination. It was obvi- the three subregions they delineated are what ous that the naming deficit was not caused by they called “intermediary” regions that do not damaged conceptual knowledge because the themselves contain the names of persons, ani- patients’ responses clearly showed that they knew mals, and tools, but knowledge on how to the depicted entities perfectly well. For instance, reconstruct these names, and that the regions in presented with the picture of a skunk a patient question mediate between activated concepts might say: “Oh, that animal makes a terrible smell elsewhere in the brain and the representation of if you get too close to it; it is black and white, and the corresponding names, again elsewhere in the gets squashed on the road by cars sometimes.” brain. Brain scans were subsequently made to locate the damaged brain areas. Of the 30 selected To provide yet a further example of the multi- patients (those who had shown abnormal naming plicity of ideas about the function of Broca’s and behavior), 29 had damage to the left temporal Wernicke’s areas and the surrounding brain lobe, thus indicating that word retrieval is indeed tissue, I will call to mind a three-step model of subserved by this brain area. But more spectacu- sentence comprehension presented elsewhere in larly, and confirming the second hypothesis, the this book (p. 215): early local, automatic, syntactic scans revealed that the exact site of the lesion was structure building, followed by lexical-semantic correlated with a specific word retrieval deficit: integration, followed by syntactic revision and Patients with a lesion in the most anterior part integration (Hahne & Friederici, 1999). These of the left temporal lobe, the temporal pole, three processing steps are reflected in three dif- exhibited impaired naming of persons; those with ferent ERP components: the ELAN (for early a lesion in the middle of the left inferior temporal left anterior negativity), N400, and P600, which gyrus showed impaired naming of animals; show up in the ERP signal about 150–200 ms, finally, those with a lesion in the posterior part of 300–500 ms, and around 600 ms, respectively, the left inferior, middle, and superior temporal after the onset of the critical stimulus and are gyri exhibited impaired naming of tools. A sub- characterized by a different topography. ERP sequent PET study with brain-intact participants research on aphasic patients has shown that these corroborated the results of the lesion study: components are modulated in aphasic patients The brain region activated while the participants (e.g., Friederici, von Cramon, & Kotz, 1999; see named persons was the one that was damaged in Kotz & Friederici, 2003, for a review). Together the patients who exhibited abnormal performance with fMRI studies of patients with lesions in in naming persons, and the same correspondence frontal and anterior temporal regions in the between the patient data and those of the brain- left hemisphere and in the basal ganglia (e.g., intact participants held for naming animals and Friederici, Hahne, & von Cramon, 1998), these tools. results have led to the suggestion that early syntactic structure building is subserved by a part As mentioned, the three left temporal lobe of Broca’s area in the left inferior frontal gyrus areas that Damasio et al. (1996) showed to be (BA 44) as well as by the anterior part of the left involved in word retrieval together largely superior temporal gyrus (Friederici & Kotz, coincide with the area that subserves the memory 2003), and that later syntactic integration and component in Hagoort’s (2005) MUC model. revision is supported by the basal ganglia and a Yet it is obvious that in the MUC model it more posterior part of the left superior temporal serves a different function. Recall that Hagoort gyrus (Wernicke). (The basal ganglia are a

8. BILINGUALISM AND THE BRAIN 425 collection of subcortical brain structures that contained by declarative memory. Ullman posits play an important role in the control of move- that in languages that are learned later in life the ment and that also appear to be involved in lan- two memory systems play a different role than in guage control in bilinguals; see further on.) In L1 because “later exposure to language may other words, this model holds that different types impair the ability of the procedural memory of syntactic processing are supported by different system to learn or compute aspects of grammar. parts of the left temporal lobe and that a part of Instead, linguistic forms that are computed Broca’s area is involved in one of these types of grammatically in procedural memory in L1 may syntactic processing. As such, it differs from all of depend largely on declarative/lexical memory in the views on the function of the inferior frontal L2” (Ullman, 2001, p. 117). The later the age of and superior temporal areas discussed above. first exposure to the L2, the larger the role of declarative memory in grammatical processing A final illustration of the great diversity of is assumed to be. views on the function of the classical language areas in the left hemisphere concerns the When earlier on in this chapter I introduced declarative/procedural memory model developed the various methods that are used to identify by Ullman and his colleagues (Ullman, 2001; the language areas in the brain, I mentioned the Ullman et al., 1997). It builds on a common dis- need for complementing studies of damaged tinction in theories on memory; namely, between brains with those examining the brains of neuro- a system that is involved in the learning, repre- logically intact people because of the equivocality sentation, and use of facts and events on the of the clinical evidence. The above discussion, one hand (“declarative” memory) and a system based to a large extent on neuroimaging studies involved in the learning and control of motor that investigated language processing in healthy and cognitive skills and habits on the other hand people, demonstrates that this source of evidence (“procedural” memory). The learning and suffers from the same indeterminacy. Both lines recall of information in declarative memory is of study clearly designate the same brain regions accessible to consciousness, whereas information as those that are unmistakably involved in lan- in procedural memory is non-conscious. For guage processing. However, neither one discloses this reason, declarative and procedural memory the exact role of each of them and, interestingly, are also referred to as “explicit” and “implicit” some of the proposed accounts (Hagoort, 2005; memory, respectively. According to Ullman, Ullman, 2001) assume a type of operation for a lexical knowledge—that is, knowledge of the specific area which suggests that it is committed sounds and meanings of words—is part of to other, non-linguistic, domains of cognition as explicit declarative memory and primarily rooted well rather than being specialized for language in temporal lobe structures, including Wernicke’s processing. For instance, the left inferior frontal area. In contrast, grammar, a rule-based aspect gyrus was considered to be a domain-general of language, is thought to be a component of working memory system or a procedural memory procedural memory and to be subserved by left system, and it is well known that the control frontal structures and the basal ganglia, including system assumed in Hagoort’s MUC model Broca’s area. Particularly relevant in view of our is involved in executive control in general interest in bilingualism is Ullman’s suggestion— (pp. 435–437). Further on we will see that it also which he underpins with lesion data—that a plays a major role in language control in dysfunction of procedural memory can be com- bilinguals. pensated by an increased dependence on declara- tive memory in the sense that linguistic structures The involvement of the right that are normally computed by the procedural hemisphere in language processing system (e.g., regular past tenses of verbs) are memorized by declarative memory. Alternatively, So far the focus has been on brain areas in the these structures are constructed by explicit rules left hemisphere, in agreement with the general

426 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS view that of the two hemispheres the left one is lobe, both left and right, subserves vision. This specialized for language processing. This view, account is supported by the fact that the known as language lateralization, has its origins activation in the occipital lobe disappears when in Broca’s finding (1861) that a circumscribed the sentence-reading condition is compared with lesion in the left frontal cortex, with the rest of a visual control condition (and the occipital lobe the cerebral cortex undamaged, resulted in a activation is thus removed in the subtraction severe speech production disorder. After verifying procedure). that the observed functional disorder did not concern an isolated idiosyncratic case, Broca con- Providing a further example, Tan et al. (2001) cluded that the language function is subserved have shown that several areas in the right exclusively by the left hemisphere. As related by hemisphere are strongly activated during the pro- Fabbro (1999), Broca’s finding and its theoretical cessing of logographic Chinese characters. The implications caused quite some excitement at the authors ascribed this effect to the fact that the time, because since Aristotle all classic theories right hemisphere is specialized in processing had assumed hemispheric symmetry of brain visual-spatial information and it is very detailed function. It follows from Broca’s finding that this visual-spatial processing that is required to be view of cerebral symmetry cannot be correct able to tell Chinese logographic characters apart. because a circumscribed lesion in one hemisphere Importantly, the Chinese characters also acti- could then never result in the complete loss of a vated the common language areas in the left specific function. After all, the homologous hemisphere, again pointing at the specialist lan- area in the intact hemisphere should continue to guage function of these areas. Similarly, support this function. Buchweitz, Mason, Hasegawa, and Just (2009) compared the brain activation of Japanese L1 Since those early days, the idea that the left speakers reading Japanese syllabic hiragana vs. hemisphere is specialized in language processing Japanese logographic kanji, and showed that in the majority of people (about 90–95% of right- reading kanji was associated with a relatively handed people and about 70% of left-handed high level of activation in occipito-temporal areas people) has received support from a wealth of of the right hemisphere. (Conversely, reading studies, including neuroimaging studies and hiragana showed more activation in areas in both studies that employed the various behavioral the left and right hemisphere that are associated paradigms designed to study cerebral asymmetry with phonological processing.) (the dichotic listening, tachistoscopic viewing, and verbal–manual interference paradigms pre- But less obvious demonstrations of right sented earlier). Nevertheless, there is plenty of hemisphere activation during language processing evidence to suggest that the right hemisphere is have also been shown, in both the temporal lobe involved in language processing as well. Some of and the frontal lobe. In reviewing the pertinent these sources of evidence can easily be reconciled studies, Stowe and her colleagues (2005) con- with the notion of left hemisphere specialization cluded that the right hemisphere is involved in for language. For instance, as compared with a the processing of lexically ambiguous words and control condition in which participants do not indirect forms of language use such as metaphors, have to perform any task, during the reading suggesting that the right hemisphere provides an of sentences a large area of the occipital lobe of alternative interpretation when the initially con- both the left and the right hemisphere is activated structed meaning turns out to be incompatible (e.g., Stowe et al., 2005). Rather than concluding with contextual information. Furthermore, it that this finding shows the right occipital lobe is appears that neural tissue in the right hemisphere involved in language processing, the more obvious is recruited when for some reason or other the conclusion is that this activation reflects the processing demands increase; for instance, task’s visual component: Reading involves vision because the linguistic material to process is com- and it is a well-established fact that the occipital plex or the input speech is presented in a com- pressed fashion and therefore hard to decipher. In

8. BILINGUALISM AND THE BRAIN 427 fact, right hemisphere activation during the pro- LANGUAGE AREAS IN THE BILINGUAL BRAIN cessing of lexical ambiguities or indirect language just mentioned may also reflect computational Introduction load. Further support for a computational load account of right hemisphere involvement, also Soon after Broca’s seminal discoveries the mentioned by these authors, concerns the fact question was raised of whether a bilingual’s two that a number of special populations (elderly languages (and a multilingual’s multiple lan- people, those who stutter, people with autism or guages) are subserved by the same brain areas as schizophrenia, and—especially important in view those committed to language in monolingual of our interest in bilingualism—second language speakers. If all of a person’s languages recruit learners), often show more right hemisphere neurons in a particular brain region, damage to activation than the standard population. These this region should have an equal impact on all of special groups of people can all be argued to have them. If, instead, the different languages are sub- fewer cognitive resources to spare than required served by different brain areas, a lesion should for the language task at hand and additional right provoke a disorder of the language that is hemisphere processing may compensate for this normally subserved by the afflicted area while insufficiency. the other language(s) should remain unaffected. As it happened, soon after Broca’s influential This view on right hemisphere involvement publication, Scoresby-Jackson (1867) reported during language processing strongly resembles the case of a bilingual patient who exhibited the one that Michel Paradis (1997, 2004) has selective loss of one language following brain fervently advocated to account for evidence of injury. This led him to speculate that the two differential right hemisphere involvement in languages of a bilingual are stored in different bilinguals as compared with monolinguals (see cortical areas. Specifically he speculated that a also, e.g., Fabbro, 1999, 2001). He, however, bilingual’s native language is stored in Broca’s added the relevant new perspective that right area whereas later languages are stored in a brain hemisphere involvement under specific circum- area anterior to Broca’s area (Fabbro, 2001). A stances in no way jeopardizes the idea that couple of decades later Pitres (1895) rejected this language is processed exclusively by the left account on logical grounds. In his reasoning he hemisphere because, according to him, the com- departed from the assumption that there are four pensatory processes involved are entirely non- brain areas that subserve language, namely two linguistic. In his words: “Increased right sensory centres, for auditory and visual images, hemisphere involvement [. . .] does not reflect and two motor centers, for graphic and phonetic the representation or processing of the language motor images (Pitres, 1895, as cited by Ijalba, system [. . .], but, on the contrary, whatever Obler, & Chengappa, 2004; see also Paradis, 1995, nonlinguistic competence is substituted for it” 1997). He then argued that selective loss of one (Paradis, 1997, p. 338). The non-linguistic com- language would imply that the aphasia-causing petence he is referring to is the ability to exploit event had accidentally damaged all four of these pragmatic and metalinguistic knowledge in regions while at the same time leaving the corre- communication. In other words, whether or not sponding areas supporting the other language one regards right hemisphere activation during intact. That brain trauma could ever result in language processing as evidence against the view such a specific distribution of lesions across the that language is exclusively subserved by the left brain’s language areas is highly unlikely. So what hemisphere depends on one’s definition of lan- else might have caused the selective loss of one guage. If pragmatic and metalinguistic knowledge language as observed by Scoresby-Jackson, and is excluded from the definition, right hemisphere since then by many others? activation during language use is fully consistent with the view that only the left hemisphere is involved in language processing.

428 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS As an alternative to an explanation in terms of After Pitres (1895), the view that all of a the selective destruction of the neural substrate multilingual’s languages are subserved by the of the apparently lost language, Pitres (1895) same brain regions dominated the field for many suggested that this language was in fact not lost decades (see Paradis, 1995, 1997, for reviews) and but unavailable due to a functional impairment. many researchers shifted their interest towards Specifically, he hypothesized that it was inhibited the question of what factors might determine the because of what he called “inertia” of the lan- exact recovery pattern exhibited in bilingual guage centers. This hypothesis provides a natural aphasia: the order in which, or the age at which, explanation of the fact that many aphasic patients the languages were acquired; the frequency of who exhibit selective loss immediately after use of each of the languages prior to the insult; the insult subsequently show a recovery of the the level of proficiency originally gained in each affected language within a time span too short language (see Ijalba et al., 2004, for a discussion). for it to have been relearned within the interval Around 1970, however, after the introduction of a between its apparent loss and return (this set of non-invasive techniques to study cerebral recovery pattern in bilingual aphasia is known as lateralization of cognitive functions in both “sequential recovery” or “successive recovery”). neurologically intact individuals and people with This fast recovery clearly suggests that the lan- brain damage (the dichotic listening, tachisto- guage was not lost but only temporarily unavail- scopic viewing, and verbal–manual interference able. Similarly, selective permanent loss (known paradigms introduced before; pp. 413–414), an as “selective recovery”) can be explained in terms upsurge of studies on language localization in of permanent inhibition of one of the languages the bilingual brain can be witnessed. This time the rather than in terms of its selective loss. This specific question being posed was whether lan- explanation of recovery patterns in terms of a guage might be lateralized differently in bilinguals functional deficit is common in contemporary and monolinguals (instead of recruiting different views on bilingual aphasia (e.g., Green, 1986; neural tissue within the same, left, hemisphere) Paradis, 2001; see also p. 295). The relevant point and, if so, whether this holds for both of a to stress here is that the occurrence of selective bilingual’s languages or for just one of them. The loss of one language does not enforce the conclu- suggestion that there might be a higher incidence sion—drawn by Scoresby-Jackson (1867)—that of crossed aphasia (aphasia caused by a lesion the multiple languages of bilinguals and multi- in the right hemisphere) in bilinguals than in linguals are subserved by neural tissue in spatially monolinguals (Gloning & Gloning, 1965; but see different brain regions. Instead, it is compatible Karanth & Rangamani, 1988, for counterevi- with the view that a bilingual’s two languages dence) presumably contributed to this renewed recruit different circuits of neurons within the interest in language localization in bilinguals. same brain regions. In the words of another After all, cases of crossed aphasia indicate that scholar from the olden times: “[. . .] within the language is subserved by neural structures in the same area, the same elements are active, though in right hemisphere of the afflicted patients, and a different combinations and interacting with a dif- higher incidence of the phenomenon in bilinguals ferential linguistic constellation” (Minkowski, would thus suggest that the right hemisphere 1927, p. 229, as cited by Fabbro, 2001, p. 212). On more often subserves language in bilinguals than this account, the (temporary) loss of one lan- in monolinguals (see Solin, 1989, for a review). In guage may reflect the (temporary) inhibition of the next section I will discuss the results of studies the neural circuit that subserves this language that examined language lateralization in within (one of) the common language area(s) in bilinguals by using the common behavioral later- the left hemisphere and suggests that the brain alization paradigms. Functional neuroimaging mechanism that regulates the activation and studies that looked at language localization in inhibition of this neural circuit is damaged (see bilinguals will be discussed afterwards. pp. 437–440 for details).

8. BILINGUALISM AND THE BRAIN 429 Language lateralization in bilinguals: for the second language (5), as well as similar lan- Evidence from behavioral lateralization guage localization in bilinguals and monolinguals paradigms (6). To reconcile a specific pattern of results with the (most popular) hypothesis that the right In two commentaries on a large number of hemisphere is more involved in language pro- studies in which language lateralization in bi- cessing in bilinguals than in monolinguals, the linguals was examined by means of the dichotic hypothesis has been narrowed down ad absurdum listening, tachistoscopic viewing, and verbal– to extremely restricted subgroups of people “such manual interference paradigms, Paradis (1990, as proficient female late acquirers in informal set- 2003) concluded that this body of studies has not tings, provided they keep their eyes closed [. . .] or advanced our knowledge of language lateraliza- block one nostril” (Paradis, 1990, p. 578). If this tion in bilinguals “one bit” but just produced “a is indeed the state of affairs in bilingual language clutter of inherently uninterpretable contra- lateralization research, it is easy to see why dictory results” (Paradis, 2003, p. 441) and that Paradis has likened the whole enterprise of trying successive studies aimed at clarification of the to identify language lateralization differences issue only created more confusion. He based between monolinguals and bilinguals to a futile this harsh judgment on his observation that the search for the mythical Loch Ness Monster experimental methods used in these studies lack (Paradis, 1992, 2003) and urged the bilingual validity. To be valid, he argued, four conditions research community “to move on to more pro- must be fulfilled: (1) It must be demonstrated that ductive research” (Paradis, 1990, p. 576). the instrument (e.g., the dichotic listening pro- cedure) does indeed measure degree of cerebral In an attempt to create order in this apparently laterality (e.g., that in monolinguals language is chaotic body of experimental data, Vaid and her more lateralized than in bilinguals). (2) It must be colleagues conducted meta-analyses on selected clearly defined what aspect of language is alleged bilingual lateralization studies (Hull & Vaid, to be differently lateralized in bilinguals and 2006; Vaid & Hall, 1991). They noted that the monolinguals, for instance, grammar or vocabu- considerable variability of the experimental lary. (3) It must be demonstrated that the stimuli results is not all that surprising given the large that are used do indeed tap into that particular variety in participant characteristics (e.g., in L2 aspect of language. (4) It must be explicitly onset age and fluency level) and in paradigm and argued why it is that for this specific aspect of task characteristics across studies. Meta-analyses language lateralization should differ between provide a way to detect “underlying patterns monolinguals and bilinguals (Paradis, 2003). across large quantities of disparate data points by standardizing statistical outcomes and According to Paradis, the prototypical study minimizing the influences of researcher bias, on bilingual language lateralization does not meet paradigm bias, procedural bias, and reliance on these requirements. As a consequence, a gamut particular methodologies” (Hull & Vaid, 2006, of divergent data patterns emerged from the p. 439). In addition, they provide a means to individual studies that have led to equally many evaluate the magnitude of the effects of different different, and often contradictory, claims about variables and to compare these effects. language lateralization in bilinguals as compared with monolinguals. To mention just a few of the As theoretical starting point, Vaid and her 14 listed by Paradis (2003): Greater right hemi- colleagues reviewed the literature and identified sphere involvement is assumed (1) for both lan- the five most common hypotheses on brain guages of (all) bilinguals; (2) only for the second lateralization in bilinguals (Hull & Vaid, 2005; language of bilinguals; (3) only in late bilinguals; Vaid & Hall, 1991). Their meta-analyses should (4) only in early bilinguals; (5) only in second subsequently reveal which one of these hypoth- languages acquired informally. But also greater eses received support from the data. According to left hemisphere involvement has been claimed the L2 hypothesis, the right hemisphere is more involved when bilinguals process their L2 than

430 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS when they process their L1. When they process the studies that had included a monolingual their L1 the left hemisphere is involved to an comparison group, the language of this control equal extent as in language processing by mono- group was the bilingual participants’ L1 in some linguals. The balanced bilingual hypothesis states cases but their L2 in other cases. In still other that during language processing proficient cases the data of the bilinguals’ two languages bilinguals exploit their right hemisphere more were collapsed and then compared with those of than monolinguals and that this holds for both the monolingual control group. languages. The stage of L2 acquisition hypothesis (Obler, 1981) posits that during the initial stages In the 2006 study these problems were of L2 acquisition the right hemisphere is rela- addressed by exclusively including studies that tively much involved in processing this language encompassed a bilingual–monolingual com- and that left hemisphere involvement gradually parison and in which the language of test taking increases with increasing L2 fluency. The manner was always the monolinguals’ one and only of L2 acquisition hypothesis posits that if the L2 language and the bilinguals’ L1. Furthermore, was acquired primarily in an informal manner— as in the 1991 study, only studies that tested that is, in naturalistic communicative contexts brain-intact participants were included. A total involving interaction with other interlocutors— of 23 studies met all the selection criteria and there is more right hemisphere involvement than were included in the meta-analysis. In view of when it was acquired in formal settings. Finally, the different hypotheses on bilingual language the age of L2 acquisition hypothesis (Vaid & lateralization presented above, one consequence Genesee, 1980) states that the closer in time the of the selection procedure should be emphasized acquisition of an L2 is to L1 acquisition, the at this point: Even though the majority of the more similar the lateralization pattern for hypotheses focus on the locus of bilinguals’ L2, the two languages should be. In other words, early the design of the study can only reveal lateraliza- bilinguals should show a similar lateralization tion patterns for their L1. The variables examined pattern for their two languages, whereas late in the meta-analysis were language experience bilinguals should show a disparate pattern for (monolingual vs. bilingual), experimental para- their two languages. digm, and, among the bilinguals, L2 proficiency (proficient vs. non-proficient) and age of L2 The results of the earliest meta-analysis acquisition (early, before age 6, vs. late, after age (Vaid & Hall, 1991), which included a total of 6). The experimental paradigms that were com- 59 studies testing brain-intact participants, sug- pared concerned the three common lateralization gested that, overall, language lateralization is paradigms introduced before: dichotic listening, similar in bilinguals and monolinguals and that tachistoscopic viewing (called the “visual hemi- the left hemisphere dominates language process- field” paradigm in this specific study), and verbal– ing in both populations. However, a comparison manual interference (called the “dual-task” para- within bilinguals that took age of L2 acquisition digm here). Due to the relatively small number and level of L2 proficiency into account showed of selected studies, no potentially important that level of L2 proficiency did not modulate the finer-grained distinctions between the studies pattern of lateralization, but that age of L2 were examined, such as the type of stimuli (e.g., acquisition did, the early bilinguals exhibiting less words or sentences) and the exact task used left hemisphere lateralization (that is, more (e.g., word recall or word identification). bilateral hemispheric involvement) than the late bilinguals. But, as noted by Hull and Vaid (2006), A first noteworthy finding emerging from the this early meta-analytic study suffered from a 2006 meta-analysis was that the three lateraliza- couple of shortcomings: Three-quarters of the tion paradigms produced different sets of results, included studies had not directly compared lan- thus indicating that Paradis’ (1990, 2003) concern guage laterality in bilinguals and monolinguals about the validity of the paradigms used to but had tested bilinguals only. Furthermore, of examine lateralization of language was justified. Specifically, the dichotic listening paradigm

8. BILINGUALISM AND THE BRAIN 431 showed left hemisphere dominance in both could not be tested. One of them is the idea monolinguals and bilinguals, whereas the visual that the right hemisphere is more involved in hemifield and dual-task paradigms resulted in L2 processing than in L1 processing (the “L2 data patterns suggesting bilateral involvement hypothesis”). A second is that this holds espe- and less left hemisphere lateralization in both cially for the initial stages of L2 acquisition (the groups. A further finding was that, in agreement “stage of L2 acquisition hypothesis”). A third is with the common view on language lateralization, that the two languages of early bilinguals show monolinguals as a group (and collapsed across the same lateralization pattern while those of late paradigms) were moderately left hemisphere bilinguals show a more disparate lateralization dominant for language. Additional subgroup pattern (the “age of L2 acquisition hypothesis”). analyses among the bilinguals revealed that early Further meta-analyses that examine the laterali- bilinguals, who were all fluent bilinguals, showed zation patterns in both L1 and L2 of various bilateral hemispheric involvement (in their L1; bilingual populations (early and late learners; see above), whereas late bilinguals were left fluent and non-fluent bilinguals) will have to be hemisphere dominant (in their L1) and did not performed to complement the above findings. differ from monolinguals. Both of these findings converged with the results obtained in the earlier To conclude this discussion of the behavioral meta-analysis (Vaid & Hall, 1991). Finally, when bilingual language lateralization studies, three age of L2 acquisition was controlled, no effect of comments are in order. The first two echo Para- L2 proficiency (on L1 laterality) materialized. dis’s skepticism with regard to the three main In other words, L2 acquisition age on its own paradigms used in this field and the way they sufficed to explain lateralization differences are implemented in individual studies: The fact among the bilinguals’ L1. that the three paradigms produce different data patterns indicates that they do not all measure The most interesting of these results is the the same thing. Therefore, to be able to make any finding that the early bilinguals’ L1 seems to be legitimate claims about language lateralization on more bilaterally localized than the one and only the basis of data gathered by means of these language of monolinguals. As noted by the paradigms we would first need to know which one authors, this finding “presents a compelling of them measures what it purports to measure. argument that there is something special about The second comment concerns the nature of the early exposure to multiple languages that affects stimulus materials used in these studies. Most neurofunctional organisation” (Hull & Vaid, of them have used word stimuli and, indeed, this 2006, p. 459). That the L1 of late bilinguals is also held for the majority of studies included in localized the same way as the language of the above meta-analyses. As argued by Paradis monolinguals—namely, predominantly in the left (2003), of the various components of the lan- hemisphere—is less of a surprise, because at guage system, phonology, morphology, grammar, least during the first 6 years of life the language and vocabulary, the latter qualifies least as a acquisition history of the late bilinguals in the component specific to (human) language. He sub- above study was identical to that of mono- stantiated this claim with a number of arguments; linguals. The moment the late bilinguals were first for instance, that vocabulary is subserved by a exposed to dual-language input, the development memory system different from the one that sub- of the L1 and the brain were both well on the way. serves the other language components (namely by declarative and procedural memory, respectively; As mentioned, the above study was designed to see e.g., Ullman, 2001, discussed above). Further- reveal the lateralization pattern for the bilingual more, anthropoids such as gorillas and chimpan- participants’ L1, not their L2. As a corollary, zees are quite successful at learning vocabulary somewhat ironically in view of the authors’ goals but fail to learn any grammar. It has been claimed to contrast the above hypotheses about language that the same holds for children growing up lateralization in bilinguals, the majority of the deprived of linguistic input (pp. 48–53). For these predictions derived from these hypotheses in fact

432 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS and other reasons Paradis concluded that by pre- and L2; see p. 427). In contrast, the two languages senting isolated words as stimuli, these studies of both early and late bilinguals activated the have used the least suitable type of stimulus to same region within Wernicke’s area. The spatial examine the location of language in the brain. separation between the two languages in Broca’s A final comment to make here is that, even if area in late but not early bilinguals led the the above three lateralization paradigms are at authors to conclude that age of language acquisi- all suitable to reveal differential hemispheric tion affects the functional organization of lan- involvement in language processing in mono- guage in the brain. A similar conclusion was linguals and bilinguals, they are surely unsuitable drawn in a recent trilingual study in which the to disclose within-hemispheric differences between analyses were also concentrated on Broca’s and these two populations. In this respect, functional Wernicke’s area (Bloch et al., 2009). Interestingly, neuroimaging techniques are more promising. It in this study it was also found that in early is to these that I will now turn. bilinguals a third language, acquired late, activated exactly the same brain regions as the Localizing language in the bilingual two early languages did. brain: Functional neuroimaging studies But soon after Kim et al.’s (1997) seminal In line with the conclusions drawn from the above study, other publications reported the conclusion meta-analyses of studies using the behavioral that both languages recruit the same brain lateralization paradigms (Hull & Vaid, 2005; regions, not only in early bilinguals but also in Vaid & Hall, 1991), from the first moment neuro- late ones. For instance, Chee, Tan, and Thiel imaging studies on the neural substrate of bi- (1999) examined brain activation in early and late lingualism started to appear in the literature the adult Mandarin–English bilinguals while they possibility was pondered that the age at which a performed various silent word generation tasks bilingual speaker is first exposed to an L2 might in response to visual cues (e.g., stimulus cue: cou; affect bilingual cerebral language organization. response: couple). Whereas Kim and colleagues For this reason, Kim et al. (1997) tested both a had only chosen Broca’s and Wernicke’s areas as group of adult early bilinguals, exposed to two regions of interest, Chee and collaborators languages during infancy, and a group of adult focused their analyses on other brain areas as late bilinguals, first exposed to an L2 in early well, in both the left and right hemisphere, that adulthood. They used fMRI as the neuroimaging previous studies had shown to be involved in lin- technique and focused their data analyses on guistic behavior. They found that word generation Broca’s area in the left inferior frontal gyrus and in Mandarin and English, in early and late Wernicke’s area in the left superior temporal bilinguals alike, aroused exactly the same pattern gyrus (see Figure 8.7). The task performed by of brain activation, involving dorsolateral pre- the participants while being scanned was silent frontal areas (that are typically associated with sentence generation; specifically, they were asked bilingual language control and executive control to describe events that occurred the previous in general; see pp. 435–446), inferior frontal areas day. A rather spectacular interaction between age (including Broca’s area), the supplementary of acquisition on the one hand and pattern of motor area (an area anterior to the central sulcus brain activation on the other occurred: In early that is involved in motor actions and thus also bilinguals the two languages gave rise to acti- in speech), and occipital and parietal regions in vation in exactly the same subregion within both hemispheres. This common activation Broca’s area, whereas in late bilinguals L1 and L2 pattern for the two languages is particularly activated neighboring locations within Broca’s noteworthy in view of the fact that Chinese and area (in fact supporting, in a slightly different English use different writing systems and other form, Scoresby-Jackson’s, 1867, speculations studies have shown that the processing of Chinese regarding differential cortical localization for L1 characters recruits different cortical regions than the processing of alphabetic symbols (e.g., Tan

8. BILINGUALISM AND THE BRAIN 433 et al., 2001, 2003; see also Buchweitz et al., 2009 manipulated the age of acquisition variable while for a similar result in a comparison of Japanese keeping the L2 proficiency level constant, at a syllabic hiragana and Japanese logographic high level, across the two age of acquisition con- kanji). ditions: before age 4 (early) and after the age of 10 (late). The task to be performed by the partici- In agreement with the latter results, Illes et al. pants was listening to stories, the same task as (1999) found similar patterns of activation, in left used in the earlier study. This time, not only the and right frontal regions, when adult fluent late early bilinguals but also the late ones showed English–Spanish bilinguals performed a semantic similar activation patterns for the two languages categorization task to series of visually presented and the authors concluded that “attained pro- words (does the word refer to a concrete or an ficiency is more important than age of acquisition abstract entity?). For both English and Spanish as a determinant of the cortical representation of words, the activation in these areas was larger L2” (Perani et al., 1998, p. 1841). than the activation observed when, in a separate control condition, non-semantic decisions had to In subsequent reviews of studies that be made to these same words (is the word printed employed both comprehension and production in upper case or lower case?). A pattern of con- tasks, at both the word and sentence level, and verging activation in largely the same brain probing different linguistic domains (semantics regions for different languages has also been and grammar), it was concluded that, generally, demonstrated for multilingualism, in a study attained L2 proficiency is a stronger determinant wherein Dutch–English–French trilinguals, all of the cerebral organization of language than age relatively fluent late learners of English and of acquisition (Abutalebi, 2008; Abutalebi, French, performed a word fluency task, a picture- Cappa, & Perani, 2001, 2005). The authors naming task, and a reading comprehension task furthermore concluded that grammatical process- (Vingerhoets et al., 2003) ing may be excepted from this general conclusion, because in at least one study, by Wartenburger et What may have caused the different pattern of al. (2003), a high level of L2 proficiency did not results between these studies and, specifically, why preclude the occurrence of an age of acquisition did age of acquisition not have the same effect in effect while participants performed a grammatical all studies? One cause of the deviant results might task. In this study three groups of Italian– be the use of different tasks. Plausibly, different German bilinguals were presented with sentences activation patterns between a bilingual’s two that were grammatically and semantically correct languages only hold for specific language tasks, or contained a grammatical or semantic violation tapping specific aspects of language. But the and were asked to perform correctness judgments above studies may also accidentally have differed while fMRI scans were made. Three groups with respect to one further variable, one that has participated: a group of simultaneous early repeatedly been shown to determine the pattern bilinguals with a high level of proficiency in both of brain activation when bilinguals process their languages; a group of late bilinguals who were native or their second language: the level of all equally proficient in L2 (and L1) as the early attained L2 proficiency in late bilinguals. Perani et bilinguals; a group of late bilinguals with a lesser al. (1998) hypothesized that the differential acti- level of L2 proficiency. Whereas the pattern of vation of a number of brain regions during L1 brain activation for the semantic judgments and L2 processing they had observed in an earlier largely depended on the participants’ L2 pro- study might have been due to the fact the par- ficiency level, age of acquisition mainly affected ticipants in that study had not only been late the activation pattern observed during the gram- bilinguals, but were at the same time not pro- matical judgment task: The activation pattern ficient in their L2. To find out whether perhaps observed for the late but proficient bilinguals per- the lesser proficiency in L2 than in L1 had been forming the semantic task closely resembled that the cause of the different activation patterns of the early bilinguals, in both languages, whereas for the two languages, the researchers now

434 LANGUAGE AND COGNITION IN BILINGUALS AND MULTILINGUALS the late but less-proficient bilinguals showed a data. For this reason, Indefrey (2006) conducted a pattern of activation over a more extended set of meta-analysis of 30 hemodynamic experiments brain areas, in both hemispheres, when they pro- (PET and fMRI) that had all compared acti- cessed their second language. In contrast, whereas vation patterns for L1 and L2 while making use the grammatical task engaged exactly the same of one of three types of tasks: tasks that require pattern of brain activation for the two languages the production of words, semantic decisions to in the early bilinguals, in both groups of late written words, or the grammatical and semantic bilinguals more extended brain areas in, espe- analysis of written or spoken input sentences. cially, the left hemisphere were activated during In addition to this variation in tasks, three par- grammatical processing in L2 as compared with ticipant variables were taken into account in the L1, and this pattern was largely the same for these meta-analysis: L2 age of acquisition, L2 pro- two participant groups. ficiency, and amount of L2 exposure. Indefrey’s main goal was to identify reliable neural dif- So far it seems that age of acquisition selec- ferences between the activation patterns during tively influences bilingual cerebral organization of L1 and L2 processing. The differences that grammar, whereas attained level of L2 proficiency emerged were coded according to a reference selectively influences bilingual cerebral organiza- system the author had used earlier in a study that tion of semantics. Further studies, however, attempted to determine the brain regions involved suggest that this conclusion does not do justice in speech production (Indefrey & Levelt, 2004) to all of the experimental evidence. To mention and that contains 114 brain regions together just one of the studies that have shown results covering the whole brain. inconsistent with this conclusion, Xue, Dong, Jin, Zhang, and Wang (2004) compared the brain Despite the large number of brain regions activity for L1 and L2 processing in late second focused on, across all experiments only 15 dif- language learners performing a semantic ferences between activated regions for L1 and L2 judgment task. The participants were 10–12 years emerged from the analysis and these all concerned old, native speakers of Chinese who began differences in the strength of activation, not the learning L2 English when they were 8 years old location of activation. Of these differences, and, importantly, had all only reached a low level 13 concerned reliably stronger activation in a of proficiency in English. Despite the fact that a particular area during L2 processing, but this dif- semantic task was used (which the above results ferential activation only occurred for specific have shown to be sensitive to L2 proficiency), tasks and specific bilingual subgroups. (In the the same pattern of activation occurred for remaining two cases stronger activation in L1 strong L1 and weak L2, suggesting that “there are was observed.) All three participant variables shared neural substrates for semantic processing appeared to play a role in the activation differ- of L1 and L2 even when one is at a very low L2 ences between L1 and L2 observed in the word proficiency level” (Xue et al., 2004, p. 791). production tasks. In contrast, differential acti- Apparently, in addition to age of acquisition and vation in L1 and L2 appeared only to be related to attained level of proficiency, there are other L2 proficiency in the tasks requiring semantic factors that affect the pattern of brain activation; decisions to words and only to L2 age of acquisi- for instance, the amount of exposure to the tion in the tasks that tapped syntactic processing L2 (see e.g., Perani & Abutalebi, 2005, and in sentence comprehension. The latter two Vingerhoets et al., 2003). findings agree with the task-specific effects of L2 age of acquisition and L2 proficiency in As we have seen above (p. 429), if a specific Wartenburger et al.’s (2003) study discussed scientific debate cannot be settled on the basis of above. But perhaps the most noteworthy finding individual studies, meta-analyses can provide a was that no L2-specific regions of activation way out of the indeterminacy by creating order in were discovered, but only strength of activation the variability of potentially important factors differences within regions activated for both L1 and taking these into account in analyzing the

8. BILINGUALISM AND THE BRAIN 435 and L2. A further finding of interest was that formance but might show a negative corre- these strength of activation differences between lation after years of practice. The positive L1 and L2 primarily showed up in the posterior correlation might simply reflect [. . .] the part of the left inferior frontal gyrus (IFG; degree of effort put into the task. By Broca’s area). Temporal lobe areas (including contrast, a negative correlation might be Wernicke’s area) were on the whole not more explained by assuming that the neural activated in L2 than L1. The functional roles structures support the linguistic process that Indefrey assigns to the left posterior IFG and involved in the task more effectively [. . .]. to temporal lobe regions are non-lexical com- positional and lexical processes, respectively In other words, if only little activation in a (cf. the “unification” and “memory” functions specific brain area is observed during task per- attributed to these areas in Hagoort’s, 2005, formance this does not necessarily indicate the MUC model; see p. 422). If this functional inter- area is not central for task performance. To the pretation of these areas is correct, the activation contrary, it may be that the area is extremely strength differences between L1 and L2 pro- central for the task at hand but has been involved cessing in the left posterior IFG suggest more often enough to do its job extremely efficiently. A effortful, less efficient, compositional processes similar relation between amount of practice on (e.g., grammatical analysis) in L2 than in L1. a task and activation in a specific brain area has been assumed for the extent (rather than the If the functional neuroimaging studies dis- strength) of activation: “[. . .] it is generally cussed above have made one thing clear it is accepted that the cortical area that subserves a that, to be able to make sense of the data, many skill gets less extensive as the skill gets more variables need to be taken into account because automatized” (Paradis, 1995, p. 212). they all might influence the (strength and/or locus of the) pattern of differential brain activation for Two final remarks are in order here. One is L1 and L2: the specific task the participants have that in neuroimaging studies that search for to perform while being scanned (comprehension differences in the locus and strength of brain acti- or production; auditory or visual; at the word vation during L1 and L2 processing, surprisingly level or sentence level; one that taps semantic few words are spent on discussing the implication processing, grammatical processing, or both), and of the results for the differential lateralization participant characteristics such as L2 acquisition debate. Whatever the reason for this textual age, L2 proficiency, and amount of L2 exposure. lacuna, generally the data seem to provide little Other factors that presumably play a role in dif- support for the view that a bilingual’s two ferential activation for L1 and L2 is whether or languages are differentially lateralized. Instead, not they employ the same writing system (e.g., the same areas in the left hemisphere seem to Buchweitz et al., 2009) and, correlated with subserve both. The second remark is a related this, whether a bilingual’s two languages are one, namely that in these neuroimaging studies typologically close or distant. monolingual control groups are generally miss- ing, so that it is impossible to determine whether One specific aspect of the neuroimaging data language is differently lateralized in monolinguals that needs to be better understood to be able to and bilinguals. make sense of them is what it is exactly that causes a region to be activated at different levels LANGUAGE CONTROL IN BILINGUALS under different circumstances. That this question is not trivial shows from the following quote from Chapter 6 dealt with the question of how Indefrey (2006, p. 297): bilinguals manage to either produce unilingual speech or to switch between their two languages Once an area becomes involved in a lin- depending on the prevailing circumstances. guistic task, its activation level might be initially positively correlated with per-