Introduction
Increased migration and mobility of labor mean that there are increased numbers of bilingual and multilingual individuals. Increasingly then, aphasia, which is one of the most common outcomes following stroke, will be found in bilingual and multilingual individuals. Such a development poses a challenge for support services and a challenge to researchers to improve theoretical understanding of the various patterns of language recovery to develop effective treatments.
A number of different patterns of recovery have been described since the early work of the French neurologist Pitres (1895/Reference Pitres and Paradis1983). The recovery of both languages in line with their premorbid proficiency may well be the most common (i.e., parallel recovery; see Paradis, Reference Paradis, Stemmer and Whitaker1998, Reference Paradis2004), but published case reports indicate other patterns (Fabbro, Reference Fabbro1999; Paradis, Reference Paradis2004). These include the differential recovery of the two languages with selective recovery of just one language as the extreme. Ribot (Reference Ribot1882) supposed that the first acquired language would be least affected, whereas Pitres proposed that the language most used at the time of aphasia onset would be best recovered. But such conjectures cannot explain the full set of reported patterns. Recovery may also reveal a temporal pattern, namely, successive recovery (i.e., after the recovery of one language, the other language recovers), alternating recovery (i.e., the language that was first recovered is lost again due to the recovery of the language that was not first), and alternating antagonistic recovery (i.e., on one day the patient is able to speak in one language, while the next day only in the other). Finally, the aphasic individual may show pathological mixing of two languages (i.e., the elements of the two languages are involuntarily mixed during language production).
Patterns of recovery are likely to reflect many factors, including those relating to the languages in question, their acquisition and usage prior to the onset of aphasia, and the nature of the damage to the neural networks underpinning their representation and control. In lieu of deep theory to account for the different patterns by, for example, parameterizing the relevant variables, researchers have adopted different approaches to understanding single cases.
A traditional “localizationist” view argued that the loss of one language occurs because the bilingual's languages are represented in different brain areas or even in different hemispheres. A focal brain lesion within a language-specific area could then affect one language only, leaving the other language intact (Albert & Obler, Reference Albert and Obler1978). The claim that different brain areas mediate syntax in L1 and in L2 lies within this tradition (e.g., Ullman, Reference Ullman2001, Reference Ullman and Sanz2005). An alternative, dynamic view attributes the different patterns of language recovery to alterations in the system of language control (e.g., Abutalebi & Green, Reference Abutalebi and Green2007; Green, Reference Green1986; Green & Price, Reference Green and Price2001; Paradis, Reference Paradis, Stemmer and Whitaker1998, Reference Paradis2004).Footnote 1 In fact, Pitres himself (1895/Reference Pitres and Paradis1983) proposed a “dynamic” explanation of language recovery in bilingual aphasia: Language recovery could occur only if the lesion had not entirely destroyed language areas, but only temporarily inhibited them through a sort of pathological inertia.
I frame this review in terms of that approach. I emphasize the need to distinguish, when discussing the bilingual brain, between devices responsible for language representations (i.e., the language network) and devices involved in controlling these representations (i.e., language control).
An Adaptive Network View of the Neural Representation of Languages and Their Control
From a neurocomputational point of view, it makes sense that the procedures designed to process specific aspects of a linguistic message (e.g., word order, morphology and syntax, prosody) are utilized to process such signals in a second language (Green, Reference Green, van Hout, Hulk, Kuiken and Towell2003, Reference Green, Kroll and de Groot2005). In consequence, it is reasonable to infer that acquisition of a second language will lead to its representation in a common network of regions. This view has implications for the nature of the network. First, the neural representation of each language must be distinctive in some fashion. Second, a shared network must adapt to the demands of each language (e.g., De Bot, Verspoor, & Lowie, Reference De Bot, Verspoor and Lowie2007; Green, Reference Green, Kroll and de Groot2005; Hernandez, Li, & MacWhinney, Reference Hernandez, Li and MacWhinney2005). A bilingual, in Grosjean's (Reference Grosjean1998) felicitous phrase, is not “two monolinguals in one body.”
Normal proficient bilingual speakers are able to use each of their languages appropriately and switch, and perhaps translate, between them. It follows that within shared processing areas there may be distinct but interleaved microanatomical circuits for each language (see Paradis, Reference Paradis2004, p. 127) that can be selectively activated or neurons whose functional role (e.g., activating an L1 word form or an L2 word form) changes with different input patterns (i.e., with what else is co-active). The demonstration of such circuits, or such neural dependency, remains a task for the future, but current neuroimaging research is consistent with the broad position that languages are represented in shared processing regions (e.g., Abutalebi, Cappa, & Perani, Reference Abutalebi, Cappa and Perani2001). For instance, the meanings of words in different languages (English and Japanese) have a common substrate (e.g., Crinion, Turner, et al., Reference Crinion, Turner, Grogan, Hanakawa, Noppeney and Devlin2006). There is evidence too of common regions mediating syntax (e.g., Golestani et al., Reference Golestani, Alario, Meriaux, Bihan, Dehaene and Pallier2006). In addition, there are circuits involved in language control outside the classical language areas (Abutalebi & Green, Reference Abutalebi and Green2007; Crinion, Turner, et al., Reference Crinion, Turner, Grogan, Hanakawa, Noppeney and Devlin2006).
Neuropsychologically, the contrast between the language network and circuits involved in its control would be supported by cases in which there is normal comprehension and production in L1 and L2 but a difficulty in controlling the use of one language over another. S.J. is one clear example (Fabbro, Skrap, & Aglioti, Reference Fabbro, Skrap and Aglioti2000). S.J.'s lesion included the left prefrontal cortex, part of anterior cingulate and the left striatum. S.J. showed normal comprehension in both Italian and Friulian and intact clausal processing in both languages. However, S.J. was unable to avoid switching into Friulian (his L1) even when addressing an Italian speaker who spoke no Friulian. Likewise, when required to speak Friulian only, S.J. would switch into Italian (his L2) (see Riccardi, Fabbro, & Obler, Reference Riccardi, Fabbro and Obler2004 for an alternate case of a patient with Wernicke's aphasia who nonetheless switched appropriately).
Current neuroimaging data on normal bilinguals cannot be interpreted too far. Neuroimaging data mostly cover group results where individual variability is treated as noise. But such variability is really critical to our understanding of language in the brain because it points to the different ways in which a given language task can be performed (see Green, Crinion, & Price, Reference Green, Crinion and Price2006). The failure to address this adequately may be one reason why some researchers are unimpressed by the achievements of neuroimaging research (see Paradis, Reference Paradis2004, chap, 6, for a critique of the enterprise, and De Bot, Reference De Bot2008, for sustained doubts). Individual variability will not be unconstrained, and by examining it, we will be in a position to identify subgroups of normal bilingual speakers.
When we think of the consequences of stroke for a bilingual, individual variability is clinically crucial. In a recent meta-analysis of neuroimaging studies on bilinguals, Indefrey (Reference Indefrey2006) noted that although there may be no regions that are exclusively necessary for word production in one language across individuals, there do appear to be sites that are only necessary for word production in one language. Relevant data come from studies using electrical stimulation to map eloquent language areas prior to surgery (e.g., Lucas, McKhann, & Ojemann, Reference Lucas, McKhann and Ojemann2004). Interference of object naming at a given site is interpreted as indicating that the site is necessary for production in that language. A stimulation-sensitive site in naming in one language does not imply that one language is represented in a distinct region from another. Stimulation can disrupt the circuit for producing that word.
One region that is reliably activated more strongly for L2 than for L1 naming is the left-inferior frontal cortex, which, in the data of Lucas et al. (Reference Lucas, McKhann and Ojemann2004), only contains L1-specific sites. Indefrey (Reference Indefrey2006) offered a plausible interpretation of such data: Individuals seek to adapt existing representations or processes to produce words in L2. The network adapts to the demands of L2.
The extent to which proficient bilinguals converge in the way in which they process language with the processing profile of monolingual speakers of that language remains contentious, but aspects of nativelike performance can be achieved early in the process of acquisition. Understanding the nature of such adaptation is important for developing a causal account of recovery patterns—an account that currently eludes us. We turn now to consider the impact of the mode of acquisition of a language on its neural representation or, in our terms, on the adaptive response of the network.
Effects of Different Acquisition Trajectories
A standard assumption is that a second language acquired early may be represented rather differently from one acquired late. For example, a language acquired in the home may be acquired implicitly, whereas one acquired at school may be learned with explicit reference to identified regularities in the language and by means of explicit vocabulary learning. The manner or age of acquisition may modulate the way in which L2 is represented neurally compared to L1. The effects of stroke may then have differential effects for L1 compared to L2.
Both the wide-ranging neurolinguistic theory of bilingualism proposed by Paradis (Reference Paradis2004) and the account of L2 representation proposed by Ullman (Reference Ullman and Sanz2005) rely on the notion of two distinct memory systems. Both proposals agree that words are represented in one system, and implicit grammatical competence in another, and that there are differences in the representation of syntax in L2 compared to L1.
What grounds are there for believing that there are distinct memory systems in the brain that may mediate different aspects of a person's knowledge of language? Traditionally, the key data in favor of distinct memory systems derived from neuropsychological dissociations. The classic dissociation evidence that is most pertinent here is the finding of impressive skill-learning in patients despite profound amnesia, that is, despite their marked inability to recall words, sentences, faces, or scenes.
Evidence of dissociation does not require that we postulate distinct systems. Dissociations may arise because of different processing demands on a unitary system. So, for instance, differences between skill learning and recognition memory may reflect differences in the demands of the tasks at study and test (Roediger, Reference Roediger and Tulving2000). Undeniably, different memory tasks require different processes, but the crucial issue is whether there are different forms of memory that are mediated by different systems. Current evidence suggests that there are.
Explicit Versus Implicit Memory
One construal of the dissociation evidence holds that patients with amnesia are selectively impaired in their ability to consciously recollect prior study episodes. They have a deficit in explicit memory, and the hippocampal system, damaged in amnesic patients, is selectively involved in such memory (Graf & Schacter, Reference Graf and Schacter1985). In contrast, skill learning does not involve explicit memory but invokes implicit memory.
Declarative and Procedural Memory
Another prominent account of the classical dissociation evidence, and one we focus on, distinguishes between declarative and procedural memory (e.g., Cohen & Squire, Reference Cohen and Squire1980). Declarative memory supports the memory for words, facts, and events including the memory for arbitrary relations (e.g., between individuals and their phone numbers). Procedural memory, by contrast, supports the acquisition and expression of a skill. Amnesia on this account is a selective deficit of declarative memory occasioned by damage to a hippocampal-based memory system.
These two construals are frequently confounded because any explicit memory test involves a declarative memory for relations. However, Ryan, Cohen, and colleagues (see Ryan & Cohen, Reference Ryan and Cohen2003 for a review) used an eye movement procedure (i.e., an implicit task) and showed that patients with amnesia were selectively impaired in their memory for relations but only when there was a delay between viewing a scene and subsequently viewing an altered version of it.
Declarative and Procedural Memory: Implications for L2 Processing
Both Paradis (Reference Paradis2004) and Ullman (Reference Ullman and Sanz2005; see also an earlier article, Ullman, Reference Ullman2001) proposed that the vocabulary of a language, whether that of L1 or of L2, is represented in a declarative memory system along with other facts of the world. In contrast, grammatical and morphological processes, at least for L1, are mediated by the procedural system that handles other cognitive and motor skills. For L2, the declarative system plays a key role in the representation of grammar. Paradis (Reference Paradis2004) emphasized the role of conscious, metalinguistic knowledge of L2 in guiding the acquisition, through practice, of implicit grammatical competence (p. 50) and in the role of such metalinguistic knowledge in supporting recovery of L2 (p. 87).
Conceptually, as Paradis (Reference Paradis2004) argued convincingly, the procedural and implicit aspects of language use are fundamentally distinct from any metalinguistic knowledge a person has of the language. Individuals may possess metalinguistic knowledge of L2 that they do not possess of their L1. In contrast, their knowledge of L1 may be entirely implicit and embodied in computational procedures that allow them to speak it and of which they are unaware. It is plausible to infer that such knowledge is represented in distinct systems. Specifically, metalinguistic knowledge (“knowing that”) is subserved by a declarative memory system to which one has conscious access and that is supported by a medial-temporal (hippocampal) system. This aspect of the proposal relates to the claims of Graf and Schacter (Reference Graf and Schacter1985). In contrast, implicit linguistic competence (“knowing how”) refers to a set of computational procedures. It is mediated by a procedural memory system as proposed by Cohen and Squire (Reference Cohen and Squire1980) that encompasses the striatum, cerebellum, and frontal structures.
Selective damage to such systems will have consequences for the recovery of languages poststroke. To the extent a speaker has metalinguistic knowledge of L2 (i.e., explicit and conscious declarative knowledge of it), but not of L1, then such knowledge may be available to compensate for loss in the case of L2, but not in the case of L1. Consider the impact of damage to frontal, basal ganglia, and cerebellar circuits following stroke. On the assumption that individuals will have more metalinguistic (i.e., explicit and conscious declarative knowledge) of a later learned language, Paradis (Reference Paradis2004, p. 87) predicted that a later learned language is more likely to be affected in an older patient. Individuals with memory deficits will cease to be able to recruit their metalinguistic knowledge to compensate for any loss.
Ullman (Reference Ullman2001) developed his view of the implications of the procedural/declarative distinction for L2 processing from research on the processing of regular and irregular English verbs. The supposition was that the production of the past tense of a regular verb (e.g., “walk”) is computed by the procedural system whose properties can be captured by a rule such as add “-ed” to the stem (“walk”), whereas the past tense of an irregular verb (e.g., teach → taught) must be retrieved from the declarative memory system that will also store other idiosyncratic combinations (such as idioms).
On the assumption that a frontostriatal network (i.e., one that includes the basal ganglia) mediates the use of grammatical rules (i.e., computes regular forms for production and decomposes such forms for comprehension), and a temporoparietal memory system subserves the storage of irregular verbs, patients with striatal dysfunction should show abnormal performance on regular but not on irregular verbs.
However, Longworth, Keenan, Barker, Marslen-Wilson, and Tyler (Reference Longworth, Keenan, Barker, Marslen-Wilson and Tyler2005) found no association between striatal dysfunction and selective impairment in the ability to form regular past tense in patients with Parkinson's disease and genetically proven Huntington's disease. Such data undermine the attempt to link the processing of regular forms to a specific component of the procedural system. Further, prior studies have assumed that regular and irregular verbs differ solely in the way in which their past tenses are computed, but recent data indicate that they differ also in their semantic properties, and that when they are matched on these properties, there is no evidence of different processing mechanisms in their production (see Basnight-Brown, Chen, Hun, Kostić, Feldman, Reference Basnight-Brown, Chen, Hun, Kostić and Feldman2007). An alternative view of the role of the basal ganglia in language processing is that they have an inhibitory function (see, e.g., Abutalebi & Green, Reference Abutalebi and Green2007; Longworth et al., Reference Longworth, Keenan, Barker, Marslen-Wilson and Tyler2005) and so modulate both syntactic and lexico-semantic processing (see, e.g., Kotz, Frisch, von Cramon, & Friederici, Reference Kotz, Frisch, von Cramon and Friederici2003).
As applied to bilingual speakers, a key implication of the declarative/procedural distinction is that when an L2 is acquired relatively late, the declarative system will play an important role in the production of syntactic forms that rely on the procedural system in L1. Such a view predicts differential effects of brain damage on L1 and L2. Damage to the procedural system will impair L1 more than L2. In Ullman (Reference Ullman and Sanz2005; see also Ullman, Reference Ullman2001) an increase in L2 proficiency can be associated with increased use of the procedural system. But this means that for high proficiency bilinguals, there is no motivation to predict that damage to the procedural system will impair performance more for L1 than for L2. There is also a computational question. How does grammatical processing increase its dependence on the procedural system? The suggestion proposed by Paradis (Reference Paradis2004) provides an answer: Conscious metalinguistic knowledge can guide the practice of L2 syntactic forms, and so, in principle, give rise to implicit L2 syntactic competence over time.
The single adaptive network view makes a slightly different claim: Individuals use, right from the start of L2 acquisition, existing systems mediating the combinatory process of syntax or morphology. Changes in proficiency are associated with shifts from more controlled to more automatic processing (Abutalebi & Green, Reference Abutalebi and Green2007). In fact, current neuroimaging data and event-related potential (ERP) data do indicate rapid convergence with the neural patterns shown by native speakers (see Golestani et al., Reference Golestani, Alario, Meriaux, Bihan, Dehaene and Pallier2006, for relevant functional magnetic resonance imaging [fMRI] data and Osterhout, McLaughlin, Pitkanen, Frenck-Mestre, & Molinaro, Reference Osterhout, McLaughlin, Pitkanen, Frenck-Mestre and Molinaro2006, for relevant ERP data), though the extent and rapidity of such convergence may depend on the languages involved (see Chen, Shu, Liu, Zhao, & Li, Reference Chen, Shu, Liu, Zhao and Li2007). In any case, the observation of different patterns of activation in performing a language task needs to be understood in the context of likely variations within normal monolingual speakers.
This adaptive network view does allow for one critical difference in the representation of L1 and L2. In line with the proposals of Paradis (Reference Paradis2004), individuals may have conscious, metalinguistic knowledge for L2 that they do not possess for L1. If such metalinguistic knowledge is indeed the source of advantage for the differential or selective recovery of L2 over L1, then this view clearly predicts that the processing of L2 will differ from that of normal proficient bilinguals who are using implicit procedures of the procedural system. Testing such a prediction requires a neuroimaging study using a speech production task in a patient with differential recovery of L2 over L1. Unfortunately, no such study exists, to my knowledge.
Neuropsychological Studies of Bilingual Patients with Aphasia
The notion of a single adapted language network that mediates the representation of the meanings of words and grammar in each language presumes that common underlying principles of organization apply for each language. Where both languages can be produced, a common pattern of effects should be found subject to any structural differences between the languages, even if one language is recovered better than another. A problem in retrieving the names of objects or actions (anomia) is a common feature of aphasia, and in this review we focus on such studies.
Distinct Representations for L1 and L2?
Kiran & Tuctenhagen (Reference Kiran and Tuctenhagen2005) tested a 53-year old, Spanish-English bilingual female patient who was born in Mexico, moved to the United States at age 21, and then married a monolingual English-speaking individual. She suffered a left hemisphere stroke (middle cerebral artery) 9 months prior to testing. Premorbidly proficient in both languages, testing (first in one language and then the other in the same session) revealed a significant advantage for naming words to definition if they referred to concrete entities, for example, “sword” (53%–63% accuracy) as opposed to abstract entities, for example, “doctrine” (0–3% accuracy). They proposed that the retrieval of concrete word forms was spared because of conceptual specificity of the mental representation of such words. In contrast, the content of the definitions provided insufficient contextual information for the retrieval of abstract words (see Van Hell & De Groot, Reference Van Hell and De Groot1998). Consistent with the single adapted network, there was no effect of language. Nor in this study was there any effect of cognate status.
De Diego Balaguer, Costa, Sebastián-Gallés, Juncadella, and Caramazza (Reference De Diego Balaguer, Costa, Sebastián-Gallés, Juncadella and Caramazza2005) analyzed the responses of two bilingual patients with aphasia on a morphological transformation task. The contrast between regular and irregular verb forms exists in both Spanish and Catalan, and so it should be the case according to the declarative/procedural proposal that Spanish and Catalan agrammatic patients with anterior lesions should be poorer at producing the forms of regular verbs rather than those of irregular verbs. Premorbidly, J.M. was a fluent Spanish-Catalan bilingual male with Spanish as L1, whereas M.P. was a fluent Catalan-Spanish bilingual female with Catalan as L1 and educated in Spanish as L2. In the transformation task, the patients completed an auditorily presented sentence frame with a suitable spoken verb form. For instance, in response to a statement, glossed in English, as “Today I eat, yesterday I. . .”, they would complete with the verb form “ate.” Both in Spanish and Catalan, the patients were better at producing regular compared to irregular forms. Accordingly, De Diego Balaguer et al. argued that these two bilingual patients with agrammatism had a deficit at the level of morphosyntactic processing independent of the regularity or irregularity of the verb. As in the classical view, frontal structures are involved in morphosyntactic processing and not just rule-based processing. Regular verbs may be easier to produce for a number of reasons. One possibility is that regular transformations are more frequent than irregular transformations. However, the interpretation of these data is complicated by the fact that in Catalan and Spanish, both the irregular and regular stem must be inflected according to person and number (Paradis, Reference Paradis, Stemmer and Whitaker2008). It is therefore more grammatically complex than the regular form. As Paradis (Reference Paradis, Stemmer and Whitaker2008) noted, predictions derived from the procedural/declarative distinction must reflect the structural properties of the language in question. As indicated earlier, matching the semantic properties of regular and irregular verbs will also be important in future studies (see Basnight-Brown et al., Reference Basnight-Brown, Chen, Hun, Kostić and Feldman2007).
The results of De Diego Balaguer et al. (Reference De Diego Balaguer, Costa, Sebastián-Gallés, Juncadella and Caramazza2005) are relevant to the single network view: Common tissue is recruited to perform comparable functions in the two languages, although, as De Diego Balaguer et al. (p. 221) noted, it remains to be seen whether selective deficits would be observed in patients speaking two languages that differ markedly in their morphosytnax (as in the case of English and Spanish).
Other research has examined evidence for the representation and processing of verbs and nouns. Verbs are arguably more representationally complex than nouns and so potentially more vulnerable to damage (see Black & Chiat, Reference Black and Chiat2003 for an analysis of the multiple factors contributing to noun–verb dissociations). Poncelet, Majerus, Raman, Warinbaire, and Weekes (Reference Poncelet, Majerus, Raman, Warinbaire and Weekes2007) asked three patients to name actions and objects (from the Action and Object naming battery of Druks & Masterson, Reference Druks and Masterson2000) first in L1 and then a week later in L2. Overall, naming in L1 was better than in L2, and objects were named better than actions. Regression analyses established that imageability was a more potent predictor for object naming than for action naming (cf. Kiran & Tuctenhagen, Reference Kiran and Tuctenhagen2005).
Although our focus is on studies of aphasic patients, it is noteworthy that the study of a bilingual patient with Alzheimer's disease, L.P.M., also revealed comparable deficits in her two languages, but she showed the reverse dissociation to the three patients reported by Poncelet et al. (Reference Poncelet, Majerus, Raman, Warinbaire and Weekes2007). L.P.M had greater difficulty in naming objects compared to actions (Hernández, Costa, Sebastián-Gallés, Juncadella, & Reñé, Reference Hernández, Costa, Sebastián-Gallés, Juncadella and Reñé2007) L.P.M. was a 74-year-old, right-handed, Catalan-Spanish female speaker, highly proficient in both languages premorbidly, who had acquired both languages early in her life. In Catalan, L.P.M. showed severely impaired performance in using nouns compared to verbs. So, for example, she was better at naming the action associated with a depicted object (e.g., “brushing”) compared to naming the object of the action (e.g., “broom”). A similar result was obtained in Spanish: L.P.M. found it more difficult to retrieve nouns compared to verbs.
Taken together, these findings of grammatical-category-specific deficits in both languages suggest that common principles underlie the representation of words in the two languages and that a common neural tissue underlies both. Hernández et al. (Reference Hernández, Costa, Sebastián-Gallés, Juncadella and Reñé2007) noted two caveats. First, Spanish and Catalan, along with other Romance languages, have similar grammatical and morphological properties, and so it remains to be determined whether or not selective deficits exist when bilinguals have learned languages that differ more markedly in their grammatical and morphological properties. Second, L.P.M. acquired both languages early, and so it is possible that the common principles of organization are restricted to circumstances in which this is the case. We do not consider this likely, given our theoretical position and neuroimaging data that indicate convergence with native speakers even for late learners of that language. A further relevant factor affecting the retrieval of a verb in aphasia is likely to be its precise nature.
Kambanaros and van Steenbrugge (Reference Kambanaros and van Steenbrugge2006) asked 12 late Greek-English bilinguals (aged between 60 and 84 years) with anomic aphasia to complete a verb and noun picture-matching task and a verb and noun picture-naming task. Comprehension relative to a matched Greek-English control group was spared in both languages, but verb and noun production in picture naming was worse in both languages, with greater impairment in the production of verbs in Greek (L1) and in English (L2). Production of instrumental verbs such as “to cut” that reference an instrumental noun (e.g., knife) was better than that for noninstrumental verbs such as “to read” that do not. Such an effect may reflect the coactivation of the verb lemma via the instrumental noun leading to easier access to the phonological representation of the verb. On the assumption that the lexeme or word form of the instrumental noun is also activated, the name or phonological relation between the noun and verb should also affect the ease of production. Kambanaros & van Steenbrugge (Reference Kambanaros and van Steenbrugge2006) also contrasted instrumental verbs that shared a phonological relation with the instrumental noun (e.g., to hammer vs. hammer) with those that did not (to sweep vs. broom). The former were significantly more difficult to produce in L2, the weaker language. Conceivably, such an effect reflects the need to select the correct form and reflects the problem in handling competitive effects in the weaker language (see Kambanaros & van Steenbrugge, Reference Kambanaros and van Steenbrugge2006, p. 175).
If common organizational principles mediate the representation of words in each language, as we have argued, then individuals should show comparable patterns of effects modulated by the precise properties of the language. The results of a single case study reported by Alexiadou and Stavrakaki (Reference Alexiadou and Stavrakaki2006) are consistent with this idea. K.S., a 32-year-old, lately bilingual patient with Broca's aphasia, premorbidly fluent in Greek (her L1) and English (her L2), completed a sentence anagram task and a grammaticality judgment task in English and in Greek. Alexiadou and Stavrakaki observed comparable effects in the two languages and attributed differences to the inflectional properties of the two languages.
A similar conclusion can be drawn from a study of compound nouns. Jarema, Perlak, & Semenza (Reference Jarema, Perlak and Semenza2007) tested two right-handed chronic aphasic bilingual patients, who were premorbidly proficient in English and French before a left cerebrovascular accident (CVA) 5 years previously. They repeated, read, and translated compound nouns (e.g., avion-suicide/suicide plane) or adjective-noun compounds (e.g., feuille morte/dead leaf). In French, such compounds are right-headed (i.e., the adjective follows the noun), whereas in English they are left-headed (i.e., the adjective precedes the noun). Analysis of the data for both patients showed fewer errors (e.g., phonemic paraphasias, semantic paraphasias, and deletions of the first or second constituent) for the first constituent (the head) in French compared to the second constituent. There was no difference for English. The authors concluded that bilinguals with aphasia are sensitive to the internal structure of compounds in a language-appropriate manner. They interpreted the null effect of constituent position in English as reflecting a combination of a first-position effect and an effect of the head's position. In these data there was no significant effect of cognateness.
Anomia in bilingual speakers provides an opportunity and a threat to communicative success. It provides an opportunity in the sense that the person can switch into their other language in the event of a failure of lexical access in the current language. It provides a threat in that switching into the other language to circumvent a problem in word retrieval can only be successful if the addressee speaks the other language. However, a speaker may also use such retrieval deliberately to cue the retrieval of the correct word. When speakers are aware of such a possibility and uses it, then they can filter out the nontarget words and so avoid inappropriate intrusions. This was true for the patient in the next study.
Goral, Levy, Obler, and Cohen (Reference Goral, Levy, Obler and Cohen2006) examined the factors affecting word retrieval in a patient with trilingual aphasia. E.C. was a 46 year-old, right-handed male who suffered a left frontotemporoparietal infarct 4 years prior to their study. Premorbidly fluent in Hebrew (his mother tongue), English (L2), and French (L3), and with a functional knowledge of Spanish, Italian, and German, E.C. continued to use French with his immediate family in Geneva, and English and Hebrew with friends. On test (using relevant portions of the Bilingual Aphasia Test, Paradis & Libben, Reference Paradis and Libben1993), E.C. revealed mild impairments in Hebrew and English and mild to moderate impairment in French. In normal conversation when he was unable to immediately retrieve a word, E.C. would retrieve the word in another language with the expectation that it might cue the correct retrieval of the target word. He voluntarily avoided actually producing the nontarget word. For purposes of their study, Goral et al. (Reference Goral, Levy, Obler and Cohen2006) asked E.C. to produce such “intrusions” as part of prompted naturalistic conversation. Overall incidence was highest for French, then English, and lowest for Hebrew. Further, English was the primary source of interference in French and in Hebrew. French was the primary source of interference in English. In short, the least well-recovered language (French, L3) suffered the most intrusions from the second acquired language (English). Such a pattern suggests that involuntary intrusions may reflect the acquisition history with later learned languages linked more closely together (cf. Kroll & Stewart, Reference Kroll and Stewart1994). Difficulty in securing lexical access and in controlling an intended language is the most likely cause of the recovery pattern observed in the final case I report in this section.
García-Caballero et al. (Reference García-Caballero, Garcia-Lado, González-Hermida, Area, Recimil and Rabadán2007) reported the case of a 91-year-old, right-handed, Galician-Spanish bilingual female who stopped speaking her L1 (Galician) and started speaking Spanish (L2) following damage to the structures of the basal ganglia. (Although the head of caudate was apparently spared, there was right capsuloputaminal atrophy and small lesions of subcortical white matter.) Her everyday behavior indicated mild disinhibition such as asking her guests for coffee rather than offering it to them. Language testing showed spared comprehension in both L1 and L2. She was fluent in L2 but not in L1 and showed impaired repetition in L1, sometimes translating into L2. When spoken to in L1, she replied in L2; although in context, such behavior is not entirely inappropriate, as both languages are spoken. Verbal fluency was impaired as reflected in a reduced ability to name instances of a given semantic category. One year after initial testing, her spontaneous speech in L1 remained minimal, and she showed a small disadvantage in auditory comprehension in L1, but she was able to name objects in her L1.
García-Caballero et al. (Reference García-Caballero, Garcia-Lado, González-Hermida, Area, Recimil and Rabadán2007) proposed that the data from the study of this patient were consistent with the differential representation of L2 (citing Ullman, Reference Ullman2001). In line with Ullman's account, they proposed that L1 may be represented more implicitly with the structures of the basal ganglia playing a key role. Certainly, formal education in her L1, Galician, was prohibited under the Franco regime in Spain, and so the patient never learned to read or write in her L1 and was not exposed to Spanish (her L2) until 8 years of age. However, she was relatively proficient in her L2—it was the language she used for reading and watching television, and so, according to Ullman (Reference Ullman2001, Reference Ullman and Sanz2005), there should be increased reliance on the procedural system. If so, L2 should be impaired along with L1. There are reasons to doubt then that the selective recovery of L2 in this patient reflects the declarative representation of L2 grammar, especially as the neuroimaging data cited earlier indicate use of a common substrate in the acquisition of L2 grammar early in the process of acquisition.
It is conceivable that her production in L2 benefitted from metalinguistic knowledge of L2. However, García-Caballero et al. (Reference García-Caballero, Garcia-Lado, González-Hermida, Area, Recimil and Rabadán2007) noted that her mild disinhibition might reflect disruption of frontosubcortical circuits, and indeed, in our view this case is better interpreted as indicating a loss of control of L1 (i.e., an inability to select it) that may have been occasioned by a temporary problem in accessing words in L1 (i.e., a failure to inhibit L2 competitors) and a decision to talk only in L2 that led to improved restitution of function in that language.
Imaging Studies of Recovery
The single case studies discussed in the preceding section are consistent with the neuroimaging data on normal subjects suggesting a single adapted language network controlled by frontal–subcortical circuits. However, it is clearly desirable to determine if recovery is, or is not, supported by premorbid neural regions involved in language processing. Studies on monolingual aphasic patients continue to examine the role of the right hemisphere in language recovery. Functional imaging studies in normal individuals confirm activation in right hemisphere homologues of Broca's or Wernicke's area in naming and listening tasks. However, it has proved difficult to establish whether in aphasic patients such activation is really supporting recovery or hindering it. Saur et al. (Reference Saur, Lange, Baumgaertner, Schraknepper, Willmes and Rijntjes2006) studied a group of monolingual patients at three time points poststroke (after 2, 12, and 320 days), using a paradigm designed to activate classical language areas. At 12 days poststroke, patients activated right inferior frontal gyrus (the homologue of Broca's area) more than matched normal controls, and activation in this region correlated with a compound aphasia score. Patients continued to improve by the final time point, although activation in this region had declined. Interpretation is problematic (see Hillis, Reference Hillis2006), as changes in the compound aphasia score could reflect recovery of different language functions or of language control.
Evidence that right inferior frontal gyrus (IFG) does not take over function derives from research using repetitive transcranial magnetic stimulation (rTMS). rTMS involves focusing magnetic pulses on the scalp and can temporarily disrupt the function of the underlying neural tissue. Winhuisen et al. (Reference Winhuisen, Thiel, Schumacher, Kessler, Rudolf and Haupt2007) found that rTMS disrupted verb generation performance when applied over left IFG both at 10 days and 8 weeks poststroke, but did not disrupt it when applied over right IFG at either time point. Such a result suggests that right IFG does not take over function but may support it. Comparable studies have yet to be carried out with bilingual aphasic patients.
It is difficult to study recovery of function using fMRI or positron emission tomography (PET) in the first week poststroke as local vasculature dilates maximally and so alters normal neurovascular coupling. One approach that circumvents this difficulty correlates changes in blood flow to an area with recovery of function. Hillis et al. (Reference Hillis, Kleinman, Newhart, Heidler-Gary, Gottesman and Barker2006) showed that recovery of naming was dependent on the restoration of blood flow to three main areas of the cortex that included Broca's area (Brodmann area 44/45), Wernicke's area (Brodmann area 22), and one further area, a posterior inferior part of the temporal lobe (Brodmann area 37). Of the three areas, Brodmann area 37 (the largest) is probably the most important as 50–60 patients who failed to restore blood flow to this area showed no improvement in naming. In contrast, single-word auditory comprehension improvements correlated with restoration of blood flow to left Wernicke's area only. However, other areas are likely to be critical to auditory comprehension because damage in Wernicke's area may disrupt functioning in other, more anterior areas usually spared by stroke (see Crinion, Warburton, Lambon-Ralph, Howard, & Wise, Reference Crinion, Warburton, Lambon-Ralph, Howard and Wise2006).
There is still a dearth of studies of bilingual patients with aphasia that chart the course of recovery using neuroimaging or perfusional changes. One ground-breaking study is by Mariën, Abutalebi, Engelborghs, and De Deyn (Reference Mariën, Abutalebi, Engelborghs and De Deyn2005). They reported that improved functioning of the frontosubcortical circuit coincided with relief of symptoms. E.M. was a 10-year-old, right-handed boy with English (L1) and with Dutch as L2. Oral proficiency was good in both languages premorbidly. Following a left thalamic lesion, L1 recovered, but fluent aphasia persisted in L2 despite speech therapy in L2 and greater exposure to L2. Five months poststroke, recurrent bleeding led to fluent aphasia in both L1 and in L2. Further, when speaking to monolingual speakers of either language, E.M. showed pathological language switching and mixing. L2 intrusions were common in a task requiring the use of L1 only. Perfusion imaging showed a decrease in blood flow in the left frontoparietal and temporal regions and a region of the left basal ganglia (the left caudate nucleus). Remission of the symptoms of language mixing and switching was associated with increased blood flow to left frontal lobe and left caudate nucleus. In this later phase, where switching occurred, it was under conscious control as a means to overcome word-finding difficulties (see patient E.C., in the study by Goral et al., Reference Goral, Levy, Obler and Cohen2006). However, perfusion abnormalities remained in the left temporoparietal areas and left thalamus. Indeed, E.M. continued to display fluent aphasia in L1 and in L2 and could not perform the Stroop task. He also showed impaired translation at the single-word and sentence level in both directions. Such data support the view of the necessity of frontal–basal ganglia circuits in the control of language. Further work is required to understand the precise nature of translation problems. For instance, it would be theoretically significant if translation was impaired even when the patient was able to name a given object in each language, because that would support the view that translation involves distinct circuits of control.
Treatment Studies
Studies to date with monolinguals and bilinguals suggest that recovery and learning is often specific to the items trained. On the positive side, studies indicate that behavioral intervention can be effective in the chronic stage (i.e., months or years past the onset of aphasia), suggesting continuing plasticity.
In the treatment of bilingual aphasic patients, one of the central issues is the extent to which training in one language generalizes beyond that treatment set and to the translation equivalents in the other language and, in particular, to cognates. In fact, the outcome remains unclear and may reflect the timing of an intervention, the extensiveness of training, and the circumstances of language use. I consider studies that have looked at behavioral changes following an intervention, then consider work that relates behavioral change to neural substrate.
Hinckley (Reference Hinckley2003) did show generalization to novel items. Semantic and phonemic cues were used to aid picture naming in L1 and L2 in a patient who had suffered a left hemisphere stroke 4 months previously, which had induced transcortical motor aphasia (deficits in naming and in expressive language but relatively preserved repetition in both languages). Treatment effects in Spanish and English were established using untrained items from standard tests, the psycholinguistic assessment of language processing in aphasia (PALPA) (Kay, Coltheart, & Lesser, Reference Kay, Coltheart and Lesser1992) and the Bilingual Aphasia Test (BAT) (Paradis & Libben, Reference Paradis and Libben1993). There was improvement in both languages, and the greatest effects were for L1 (Spanish) compared to English (learned late in adolescence).
Kiran and Edmonds (Reference Kiran and Edmonds2004) also found a degree of generalization to novel items. They examined two premorbidly right-handed English-Spanish bilinguals with aphasia following onset at least 8 months previously. Unfortunately, no lesion site information is supplied but premorbidly, patient 1 was more proficient in English, whereas patient 2 was balanced. They used a semantic-based naming treatment that drew attention to the attributes of objects. Treatment effects varied with premorbid levels of proficiency. For the premorbidly balanced bilingual, training generalized to translation equivalents in the other language and to semantically related items. Performance on unrelated control items did not vary. For the English-dominant patient, untrained semantically related items benefitted, but there was no generalization to Spanish either for the translation equivalents or for semantically related items. The study suggests that treatment effects are rather restricted, and the generalization of any benefit is modulated by prior levels of proficiency. However, for purposes of interpretation, it is also important to know about current language usage.
Kohnert (Reference Kohnert2004) showed generalization for cognates (words that look and sound similar across languages) but not for noncognates. Kohnert treated D.J., a 62 year-old, Spanish-English bilingual male with severe nonfluent aphasia, following a single left embolic CVA a year previously. D.J. lived with his bilingual Spanish-English wife and son (primarily English speaking). Conversational utterances were infrequent and restricted to 1–3 words, mainly in Spanish. In contrast to premorbid use, D.J. mixed utterances with unintended intrusions of words or phrases from the other language. On test, D.J. showed marked deficits in naming and in expressive language but relatively preserved repetition in both languages (the classic symptoms of transcortical motor aphasia). Cognitively, and arguably consistent with unintended intrusions, he showed poor performance on the Wisconsin Card Sorting task (0/7). Kohnert devised two successive treatments: a cognitive treatment that trained D.J. in nonverbal skills (e.g., card-sorting tasks, and single-digit addition and subtraction) and a cognate treatment, which targeted 20 Spanish-English translations. Half of these were cognates (rosa/rose) and half were noncognates (escoba/broom). The intervention first in Spanish, then in English, involved tasks such as identifying pictures of the training items from distractors and completing a cloze test (“She sweeps the floor with a ____”). In addition to the training set, D.J. was tested on confrontational naming with 20 Spanish-English word pairs on four occasions; that is, before and after treatment in each language.
The cognitive training improved test performance speed and accuracy. D.J. also showed modest gains, relative to baseline, in confrontation naming in both Spanish and English tested using items that were not part of the treatment schedule. Enhanced information processing or general stimulation may have contributed to this.
The cognate training improved performance on untrained items in both Spanish and English. Naming cognates posttreatment in Spanish carried over to naming the same pictures in English a week later. There was no advantage, however, having previously named the noncognate items. Kohnert argued that exploiting cross-language lexico-semantic links might be a way to maximize treatment gains especially in severe cases of aphasia (Paradis, Reference Paradis and Paradis1993).
As discussed earlier, bilinguals can use their other language to aid communication in the event of a failure of lexical access. But this possibility is not always advantageous if the addressee does not speak that language. Ansaldo, Marcotte, Scherer, and Raboyeau (Reference Ansaldo, Marcotte, Scherer and Raboyeau2008) report on the case of E.L., a 56-year-old, Spanish-English bilingual male with anomia and pathological language mixing and switching between languages following a left subcortical lesion. Ansaldo et al. used the best preserved ability (translation from English into Spanish) to overcome inappropriate language switching (i.e., switching into English when addressing a monolingual Spanish speaker) to solve the lexical access problems in L1 (Spanish). The method, called “switch-back through translation,” worked by the speech and language therapist prompting E.L. immediately after an inappropriate switch from Spanish into English with the words “que quiere decir ..?” (in English, “which means..?) and so prompting E.L. to return to Spanish—the language of the conversation. Fading out by the therapist and self-cuing by E.L. resulted in significant improvements.
Treatment can be effective even with chronic patients, but it is likely to be constrained to items in the trained set or to their semantic associates or cognates. In the next section I consider what is known about the neural basis of treatment effects.
Examining the Effects of an Intervention Using fMRI
Recent work examining treatment effects on speech output in monolingual patients has reported change in neural activity in tissue surrounding the lesion in the left hemisphere, in right-hemisphere homologues and bilaterally (see Crinion & Leff, Reference Crinion and Leff2007, for a review), and in non-language-related areas (see Fridriksson et al., Reference Fridriksson, Moser, Bonilha, Morrow-Odoma, Shaw and Fridriksson2007). However, as Crinion and Leff noted, larger numbers of patients are needed to establish the generality of any findings linking lesion site and behavior and to rule out the possibility of differences in premorbid lateralization patterns.
An important line of work in bilinguals is to examine neural correlates of an intervention by using fMRI. Unfortunately, work of this type is still very rare with bilingual aphasia patients. In a pioneering example, Meinzer, Obleser, Flaisch, Eulitz, and Rockstroh (Reference Meinzer, Obleser, Flaisch, Eulitz and Rockstroh2007) examined treatment effects in C.Q., a highly educated, right-handed, 35-year-old man, premorbidly highly proficient in both German and French that he had acquired early. C.Q. also spoke English and Italian. C.Q. suffered a left-hemisphere stroke that damaged large parts of the frontotemporal and parietal lobes and the posterior part of the insula, and also included the putamen. Poststroke, C.Q. was rehabilitated in German (3 hours per week), but 32 months poststroke, C.Q. showed syntactic impairments and word-retrieval problems in German. Meinzer et al. tested C.Q. in an overt fMRI picture-naming task in both German and French prior to, and after, short-term intensive language training in German (constraint-induced aphasia therapy [CIAT]; see Pulvermueller et al., Reference Pulvermueller, Neininger, Elbert, Mohr, Rockstroh and Koebbel2001).
Improved performance was restricted to pictures named in the training set. Out of a set of 80 pictures, C.Q. named 34 pictures correctly prior to training compared to 54 pictures posttraining. Semantic paraphasias decreased from 22 to 10. In contrast, there was no performance improvement on the untrained French items pre/post: 4/6. The results revealed a bilateral (frontotemporal) increase in neural activity associated with improved naming ability following short-term intensive training only in the trained (German) language. No differences were found for picture naming in French. It is worth noting that 26/80 pictures were cognates, but there was no transfer here (Meinzer et al., Reference Meinzer, Obleser, Flaisch, Eulitz and Rockstroh2007, p. 1254) contrary to Kohnert (Reference Kohnert2004). A further important point, and one deserving of further study, is that during this intensive training, my understanding is that the patient continued to use German exclusively outside the clinic. Conceivably then French remained inhibited. Certainly French testing always followed German testing, and so the French names would be in competition with the German names. I emphasize the need to take account of the precise nature of language use because this helps determine the way in which language control is exercised.
Future Directions
Research to date on bilingual speakers with aphasia has yet to get the heart of an important aspect of normal language use: our construal of events. We use language to refer to states of affairs and to event sequences in the world, but such events can be construed in different ways, and such construals are constrained by language. English, for instance, emphasizes the motion aspects of an event, whereas Spanish emphasizes the resultant state. A difficulty in retrieving verbs may reflect a difficulty in retrieving the word form. Alternatively, it may reflect a deficit at the level of construal, that is, at the level of conceptual preparation (Levelt, Roelofs, & Meyer, Reference Levelt, Roelofs and Meyer1999) or ‘thinking for speaking” (Slobin, Reference Slobin, Gumperz and Levinson1996). Deficits at the level of “thinking for speaking” may be especially problematic for bilingual aphasic patients where the languages they speak differ in the construal of events that they demand. Further, problems in word retrieval may compound the difficulty of construing events in a language-relevant manner (see Dean & Black, Reference Dean and Black2005, also for novel techniques for exploring “thinking for speaking”).
Much more work is needed to understand which regions and networks mediate recovery. Understanding the relationship between changes in regional blood flow poststroke and the recovery of function is a vital step. The work of Hillis et al. (Reference Hillis, Kleinman, Newhart, Heidler-Gary, Gottesman and Barker2006) in a large sample of monolingual patients and the single case study of a bilingual child by Mariën et al. (Reference Mariën, Abutalebi, Engelborghs and De Deyn2005) are exemplary. Functional imaging studies are also important because they can help tackle theoretical predictions that are difficult to address in another way. For instance, if bilingual aphasics with better recovery of L2 compared to L1 make use of L2 metalinguistic knowledge in producing sentences in L2, then the neural pattern should differ from that evoked when speaking in L1. On the other hand, a proper interpretation of any such finding depends on establishing the variety of patterns in normal bilingual speakers (see Green et al., Reference Green, Crinion and Price2006). Only such information allows us to rule out the possibility that unusual patterns poststroke reflect reorganization of function rather than an example of an unusual, but normal, premorbid pattern.
I focused in this review on neuropsychological and neuroimaging studies of single cases. But an important goal is to establish the nature of the correlations between lesion site and behavioral deficit. Lesions can now be described at the voxel level (just as in functional imaging research) and voxel-based data can be related to continuous behavioral data (see Bates et al., Reference Bates, Wilson, Saygin, Dick, Sereno and Knight2003, for a pioneering example with monolingual aphasic patients). Voxel-based morphometry (VBM) identifies differences in gray or white matter based on structural MRI images. It has been used to examine the neuroanatomical bases of the different pattern of deficit in semantic dementia and herpes simplex encephalitis (e.g., Noppeney et al., Reference Noppeney, Patterson, Tyler, Moss, Stamatakis and Bright2007), and it would be useful to extend it to examine the effect of gray-matter changes on task performance in bilinguals. To the extent a set of regions is necessary for performance of a task, such as generating instances of a semantic category or generating words beginning with a particular letter or phoneme, gray-matter density should be predictive of relative differences in task performance in both languages of a bilingual aphasic patient. Current data are in line with this expectation (e.g., Grogan, Crinion, Ali, Green, & Price, Reference Grogan, Crinion, Ali, Green and Price2007).
One value of this larger-sample approach is that the relationship between gray-matter density and performance on a task can be used to estimate recovery. It could then provide an indicator of the extent of long-term support that might be needed and also offer a way to assess the effectiveness of any intervention. Success would be relative to the predicted recovery without the intervention.
As in the case of monolingual patients with aphasia, small amounts of therapy are unlikely to be effective. On the basis of a meta-analysis, Bhogal, Teasell, Foley, and Speechley (Reference Bhogal, Teasell, Foley and Speechley2003) concluded that trials involving less than 52 hours of treatment never result in improvements, whereas those with 60 hours or more always showed evidence of improvement.
Which aspects of treatment are efficacious is an important theoretical and practical matter. It would seem necessary to explore alternative ways to deliver treatment (other than therapist delivered) if large-scale trials are to be conducted. Pharmacological adjuncts to therapy may also improve the effectiveness of treatment (e.g., Berthier et al., Reference Berthier, Green, Higueras, Fernandez, Hinojosa and Martin2006) and offer a further way to explore the dynamics of recovery when their use is combined with an examination of perfusional changes and fMRI studies to examine the nature of the plastic changes during the performance of language tasks.
Conclusions
A detailed neurocomputational understanding of the causal basis of different patterns of recovery remains to be developed. However, current data are consistent with the view that the acquisition of another language involves adapting an existing network. Different languages are represented in shared neural areas with common organizing principles. It is also important to consider how individuals control the use of their languages. Certain patterns of deficit reflect problems of control rather than of deficits of representation. There is still a dearth of studies examining neural changes over the course of recovery, and treatment studies also remain sparse. But there is every prospect that progress can be made by combining neuropsychological and neuroimaging techniques and using these to examine the adaptive response of the brain to recovery and treatment.
ACKNOWLEDGMENTS
I thank the Wellcome Trust for support and Cathy J. Price, Jenny Crinion, and Alice Grogan for useful comments. Special thanks to Michel Paradis for his insightful comments on a previous draft.
