LEXICAL DEVELOPMENT

Much of the early research taking a psycholinguistic approach to adult L2 learning asked how words were acquired and attached to their respective concepts. The history of that work has been reviewed in previous literature (e.g., Kroll & Tokowicz, Reference Kroll, Tokowicz, Kroll and De Groot2005). The focal questions in the initial work on the development of the lexicon concerned the relative independence of the two languages, the degree to which words in each language accessed the same or different meanings, and the process by which learners acquire skills and automaticity in recognizing and producing words in a newly acquired language. As we noted at the outset and as we will review in the next sections, a major discovery about lexical access in bilinguals and multilinguals is that it is language nonselective, with alternatives in each language activated when only one language is recognized or produced. That discovery required a revision to models that assumed a clear separation between the two or more languages. It also cast new light on the idea that the native or dominant language functions only as a temporary learning strategy on which words in the new language can lean until they are able to function autonomously.

An issue that framed early accounts of late L2 lexical learning was whether adult learners are able to access the meaning of L2 words independently of the first language (L1) translation. Some studies suggested that direct access to meaning was possible for even learners at relatively early stages of acquisition (e.g., Duyck & Brysbaert, Reference Duyck and Brysbaert2004; Potter, So, Von Eckardt, & Feldman, Reference Potter, So, Von Eckardt and Feldman1984). Other studies claimed that early in adult L2 learning there was reliance on the L1 translation that gradually gave way to independence in the L2 once learners became sufficiently skilled (e.g., Kroll & Stewart, Reference Kroll and Stewart1994; Sunderman & Kroll, Reference Kroll, Bobb and Wodniecka2006). On this view, the L1 translation equivalent was seen as a mediator that enabled indirect access from the L2 word to the concept. Although there is still debate about the precise role that the L1 translation equivalent may play during early stages of late L2 learning, what is now clear is that the activation of the L1 translation equivalent remains even once individuals become skilled speakers of the L2. What changes is its function. In the section below on neuroscience evidence on the bilingual and multilingual lexicon, we review the logic of the new studies that use ERPs to reveal the earliest time course of processing to examine this issue at a finer grain than possible in behavioral studies alone (e.g., Guo, Misra, Tam, & Kroll, Reference Kroll, Gollan, Ferreira, Goldrick and Miozzo2012; Thierry & Wu, Reference Thierry and Wu2007). As we will see, the new evidence suggests that skilled bilingual readers are indeed able to access the meaning directly of words in the L2, but that the translation equivalent in the L1 becomes available only once they understand the L2 word.Footnote ¹

We return to discuss the nature of the neuroscience evidence in the final section of this chapter. Behaviorally, there is also new evidence to suggest that even proficient bilinguals activate the L1 translation of the L2 word and even under conditions in which the L1 itself is not present. Morford, Wilkinson, Villwock, Piñar, and Kroll (Reference Morford, Wilkinson, Villwock, Piñar and Kroll2011) examined the performance of deaf bilinguals who use American Sign Language (ASL) as their primary language, or L1, and who read written English as their L2. These deaf bimodal bilinguals were highly proficient in written English as the L2. They performed a semantic judgment task in which they had to decide whether two English words were related in meaning. Notably, no ASL was present in the task. Unbeknownst to the participants, some of the English word pairs had ASL translations that were related in form. Morford et al. found that the time to make a semantic relatedness decision for the English words was affected by the form similarity or difference in the ASL translation, suggesting that there was implicit access to the translation even when it was logically not required to perform the task in English alone.

Linck, Kroll, and Sunderman (Reference Linck, Kroll and Sunderman2009) investigated the role of the translation equivalent for learners who were immersed in the L2 environment during a study abroad experience. Native English speakers who were intermediate learners of Spanish were tested while they were living in Spain for a semester. Their performance was compared to a matched group of classroom learners of Spanish who had no immersion experience. Linck et al. used a translation recognition task in which a word was presented in Spanish, the L2, followed by a word in English, the L1. The task was to simply decide whether the L1 word was the correct translation of the L2 word. Like earlier studies (e.g., Talamas, Kroll, & Dufour, Reference Talamas, Kroll and Dufour1999; Sunderman & Kroll, Reference Kroll, Bobb and Wodniecka2006), the critical conditions were those in which the two words were not the correct translations but in which the L1 word was either similar in form to the translation or semantically related to the translation. They found that learners with classroom experience only were affected by both types of lures. They were slower to reject words similar to the translation and also words semantically related to the translation compared to unrelated controls. In contrast, the immersed learners were unaffected by words similar to the translation but showed a large semantic interference effect. Together with other results of that study, Linck et al. concluded that under conditions of language immersion, learners appear to inhibit the L1. The behavioral methods used in this study do not allow a precise conclusion with respect to the locus of inhibition, but they suggest that under conditions of language immersion there may be enhanced control over the activation of the language not in use.

The recent evidence for the activation of the translation for even highly proficient bilinguals has two important theoretical implications. First, it suggests that neither side of the traditional debate about the role of the translation equivalent was correct. One side of that debate argued that the translation was functionally irrelevant to L2 performance because even learners at early points in acquiring L2 skill could understand the meanings of words in the L2 directly without mediation via the L1 (e.g., Brysbaert & Duyck, Reference Bogulski and Kroll2010; Potter et al., Reference Potter, So, Von Eckardt and Feldman1984). The other side of the debate argued that the translation equivalent was crucial only as a mediator to enable learners to acquire meaning for L2 words until a point at which they were skilled and could process the L2 word directly (e.g., Kroll & Stewart, Reference Kroll and Stewart1994). The new data show that the translation continues to play a role long after learners become highly proficient in the L2, but that its function may change with increasing L2 skill. The activation of the translation equivalent in proficient bilinguals, however briefly and however task dependent (an issue that has not yet been examined in detail), has another implication that is perhaps even more important than the first. If bilinguals and multilinguals routinely activate the L1 translation when processing the L2, even if only implicitly, then the nature of language processing becomes fundamentally changed for bilinguals and multilinguals relative to monolinguals.Footnote ² Not only does the processing of L2 words become more automatized in ways that have been characterized in the past literature (e.g., Segalowitz & Hulstijn, Reference Segalowitz, Hulstijn, Kroll and De Groot2005), but the processing of L1 words changes as well. In the next section of the chapter we discuss in some detail how the open nature of the lexical architecture reflects the resulting cross-language interactions. But first we consider the implications of these changes for learning new words from another language.

One approach to examine new learning experimentally is to introduce a restricted vocabulary within the context of a training study. Many past studies have taken this approach in an attempt to simulate the initial stages of vocabulary acquisition (e.g., see the volume by Healy & Bourne, Reference Healy and Bourne1998). These studies contrast different methods of teaching new words and the effects of different word types on immediate and later measures of retention (e.g., De Groot & Keijzer, Reference De Groot and Keijzer2000). Critically, these conditions have been examined for monolingual or native speakers who are learning words in an unknown language. More recently, this approach has been extended to compare the performance of monolingual learners of words in a new language with bilingual learners for whom the new words represent a third language (L3). A few early studies demonstrated that bilinguals were at an advantage in word learning (Papagno & Vallar, Reference Papagno and Vallar1995; Van Hell & Mahn, Reference Van Hell and Mahn1997). More recently, the bilingual advantage in word learning has been replicated and generalized to other training paradigms and to early as well as late bilinguals (e.g., Kaushanskaya & Marian, Reference Kaushanskaya and Marian2009a, Reference Kaushanskaya and Marian2009b). Bogulski and Kroll (under review) replicated the bilingual advantage but demonstrated that the advantage was restricted to only those bilinguals who learned the novel L3 vocabulary via their L1 translations. When the L3 was trained via the L2, there was no difference between bilinguals and monolinguals, suggesting that there is a special role of the native language for initial stages of new lexical acquisition. They speculated that bilinguals learn to inhibit their native language to speak the L2, a topic we will discuss below, and that experience then generalizes to learning new words. When bilinguals have to learn new words via the L2, their experience with the L1 is then hypothesized to be less helpful because it is the dominant L1 that is actively inhibited, not the L2. It will remain to be seen whether future studies provide support for this hypothesis. But what seems apparent is that the way in which bilinguals negotiate the potential competition across their two languages may give rise to a set of phenomena that affect not only language processing itself but also learning.

CROSS-LANGUAGE ACTIVATION DURING WORD RECOGNITION

Bilinguals activate lexical alternatives across all of the languages they know when they read words in one language alone. Parallel activation is particularly evident when the words overlap across languages, such as cognate words, which share orthography, phonology, and meaning (e.g., piano in English and Spanish), or homographs, which share only orthography and phonology while differing in their meaning (e.g., pan, which is a container for cooking in English but a baked food in Spanish). These language-ambiguous words show differential processing patterns compared to words that have similar lexical properties but little cross-language overlap (e.g., mesa, which means table in Spanish). If a bilingual or multilingual could access words selectively in one language alone, then the presence of overlap should have no consequence for processing. Indeed, monolingual speakers show no evidence for differential processing of language-ambiguous words such as cognates and homographs, suggesting that it is the knowledge of two languages that changes the way that these words are processed, not some set of correlated lexical attributes. Likewise, a budding literature suggests that polyglots activate representations of multiple languages in parallel. A major question in multilingual word recognition is whether a single language alternative can be selected from the myriad of activated words, and if so, what cues may aid in enabling a language selection. The recent research shows that language-selective access is difficult to achieve.

The primary evidence for parallel activation of languages during visual word recognition comes from the processing of language-ambiguous words (e.g., Dijkstra, Reference Dijkstra, Kroll and De Groot2005). Interlingual homographs tend to produce interference, being processed more slowly than unambiguous controls, whereas cognates tend to produce facilitation, being processed more rapidly than controls. In each case, the general assumption has been that the activation of word forms in both languages creates competition when there is conflict, as in the case of homographs where each language maps onto different semantics, and convergence when the cross-language activation can be resolved to the same meaning, as in the case of cognates. Cognate facilitation is often more robust than homograph interference (e.g., Dijkstra, Van Jaarsveld, & Ten Brinke, Reference Dijkstra, Van Jaarsveld and Ten Brinke1998).

Parallel activation has been observed at many different levels of linguistic representation, including the orthography, phonology, and semantics of a word. The recent studies show that these effects are not categorical but continuous, such that the more two words overlap in their written form, the greater the co-activation (e.g., Duyck, Van Assche, Drieghe, & Hartsuiker, Reference Duyck, Van Assche, Drieghe and Hartsuiker2007; Schwartz, Kroll, & Diaz, Reference Schwartz, Kroll and Diaz2007; Van Assche, Drieghe, Duyck, Welvaert, & Hartsuiker, Reference Van Assche, Drieghe, Duyck, Welvaert and Hartsuiker2010; Van Assche, Duyck, Hartsuiker, & Diependaele, Reference Van Assche, Duyck, Hartsuiker and Diependaele2009). Cross-language activation can also be seen for bilinguals whose languages do not share the same writing system, indicating that the phonology is responsible for the observed interactions. Different-script bilinguals (e.g., Chinese and English) and different-modality bilinguals (e.g., one signed language and the other written), show sensitivity to the language not in use when words in the two languages overlap in their phonological properties (e.g., Morford et al., Reference Morford, Wilkinson, Villwock, Piñar and Kroll2011; Thierry & Wu, Reference Thierry and Wu2007). Morford et al. showed that when deaf readers make judgments about English words in the absence of overt sign, there is apparent activation of the sign translation and momentary conflict when the sign translations of English word resemble one another in form but the meaning of the English words is distinct. Bilinguals likewise show sensitivity to the degree of phonological overlap during a visual recognition task; processing speed is slowed if there are differences in phonology present (e.g., base in English and Spanish, which differ in stress; Schwartz et al., Reference Schwartz, Kroll and Diaz2007; Van Assche et al., Reference Van Assche, Drieghe, Duyck, Welvaert and Hartsuiker2010), suggesting that the phonology of the unintended language is activated during reading. As mentioned earlier, parallel activation of the semantics is evident in homograph recognition; the conflict in meaning across the two readings of an interlingual homograph produces a cost to processing.

Nonselectivity is a pervasive phenomenon; it is observable whether the native language or the L2 is in use, across a variety of language pairs, and in many different tasks. Language co-activation is not a marker of lack of fluency. Influence of the native language on the L2 has been reported for even highly proficient L2 speakers (Dijkstra et al., Reference Dijkstra, Van Jaarsveld and Ten Brinke1998; Dijkstra, Grainger, & Van Heuven, Reference Dijkstra, Grainger and Van Heuven1999; Duyck et al., Reference Duyck, Van Assche, Drieghe and Hartsuiker2007; Libben & Titone, Reference Libben and Titone2009; Schwartz & Kroll, Reference Kroll, Bobb and Wodniecka2006; Schwartz et al., Reference Schwartz, Kroll and Diaz2007; Van Assche et al., Reference Van Assche, Drieghe, Duyck, Welvaert and Hartsuiker2010). Typically, the native language is more dominant than the L2, allowing it to exert its influence quite easily. Yet, co-activation is not restricted to influence of the L1 on the L2. Nonnative languages can come to influence the way the native language is processed given sufficient proficiency in the L2. For example, Van Assche et al. (Reference Van Assche, Duyck, Hartsuiker and Diependaele2009) found cognate effects for Dutch-English bilinguals who were reading in Dutch. Likewise, Schwartz et al. showed that English-Spanish bilinguals were sensitive the degree of phonological overlap of words in Spanish while they were reading words in English, their L1 (see also, e.g., Schwartz & Kroll, Reference Kroll, Bobb and Wodniecka2006; Van Assche et al., Reference Van Assche, Duyck, Hartsuiker and Diependaele2009; Van Hell & Dijkstra, Reference Van Hell and Dijkstra2002). Regardless of whether processing is in the L1 or L2, lexical alternatives become activated in parallel while a single language is being read. Evidence is much the same for speakers of an L3. Van Hell and Dijkstra found that native speakers of Dutch who knew both English and French showed cognate facilitation to both Dutch-English cognates as well as Dutch-French cognates, indicating that even a relatively more distant language (the L3) becomes activated during L1 processing. Nonselective word recognition has been demonstrated with a variety of language pairs including Dutch and English (Duyck et al., Reference Duyck, Van Assche, Drieghe and Hartsuiker2007; Van Assche et al., Reference Van Assche, Duyck, Hartsuiker and Diependaele2009; Van Assche et al., Reference Van Assche, Drieghe, Duyck, Welvaert and Hartsuiker2010; Van Hell & De Groot, Reference Van Hell and De Groot2008; Van Hell & Dijkstra, Reference Van Hell and Dijkstra2002), Dutch and German (Lemhöfer, Dijkstra, & Michel, Reference Lemhöfer, Dijkstra and Michel2004), Spanish and English (Schwartz & Kroll, Reference Kroll, Bobb and Wodniecka2006; Schwartz et al., Reference Schwartz, Kroll and Diaz2007), Polish and English, Polish and German (Szubko-Sitarek, Reference Szubko-Sitarek2011), French and English (Libben & Titone, Reference Libben and Titone2009; Titone, Libben, Mercier, Whitford, & Pivneva, Reference Titone, Libben, Mercier, Whitford and Pivneva2011), and Chinese and English (Thierry & Wu, Reference Thierry and Wu2007). Parallel activation of languages that occurs independent of language proficiency, age of acquisition, or the language pairs under investigation speaks to the interactivity of the bilingual lexicon, and by extension the interactivity in language processing in general.

Many different methods have been used to study language co-activation. The main conclusion is that nonselectivity is relatively task-independent, though different tasks can provide information about the time course of processing. Cognate facilitation and homograph inhibition are observed in a variety of measures. Initial research was conducted with behavioral measures such as lexical decision (Dijkstra et al., Reference Dijkstra, Van Jaarsveld and Ten Brinke1998; Van Hell & De Groot, Reference Van Hell and De Groot2008; Van Hell & Dijkstra, Reference Van Hell and Dijkstra2002), translation (Sanchez-Casas, Davis, & Garcia-Albea, Reference Sanchez-Casas, Davis and Garcia-Albea1992; Van Hell & De Groot, Reference Van Hell and De Groot2008), word association (Van Hell & Dijkstra, Reference Van Hell and Dijkstra2002), and word naming (e.g., Schwartz & Kroll, Reference Kroll, Bobb and Wodniecka2006; Schwartz et al., Reference Schwartz, Kroll and Diaz2007). More recent studies find language co-activation in measures such as eye tracking (Duyck et al., Reference Duyck, Van Assche, Drieghe and Hartsuiker2007; Libben & Titone, Reference Libben and Titone2009; Titone et al., Reference Titone, Libben, Mercier, Whitford and Pivneva2011; Van Assche et al., Reference Van Assche, Drieghe, Duyck, Welvaert and Hartsuiker2010; Van Assche et al., Reference Van Assche, Duyck, Hartsuiker and Diependaele2009) and ERPs (Midgley, Holcomb, & Grainger, Reference Midgley, Holcomb and Grainger2011). Behavioral studies tend to measure the aggregate result of processing. Hence a cognate effect in a lexical decision task provides no information about the point at which both languages started to become activated. It could theoretically be the case that parallel activation occurs late in the process of word recognition, almost as if bilinguals translated words between their two languages, or both languages might become activated initially at the point at which orthography begins to be decoded. More time-sensitive methods, such as ERPs and eye tracking, indicate that parallel activation is not solely a late process. Both language alternatives become activated early in processing. For example, in eye tracking, cognate effects are observable in first fixation duration and gaze duration measures, assumed to reflect initial lexical access. They also show that both languages remain activated throughout the time course of processing (e.g., Duyck et al., Reference Duyck, Van Assche, Drieghe and Hartsuiker2007; Libben & Titone, Reference Libben and Titone2009; Van Assche et al., Reference Van Assche, Duyck, Hartsuiker and Diependaele2009; Van Assche et al., Reference Van Assche, Drieghe, Duyck, Welvaert and Hartsuiker2010), suggesting that the intended language may never actually be selected categorically or that the selection is not observable without an even more sensitive measure.

The bilingual interactive activation plus (BIA+) model was proposed by Dijkstra and Van Heuven (Reference Dijkstra and Van Heuven2002) to account for language nonselectivity. The model is based on the interactive activation model of McClelland and Rumelhart (Reference McClelland and Rumelhart1981). BIA+ is split into two subsystems: the word recognition system and the task schema. The word recognition system is responsible for determining which lexical candidates are activated during word recognition and in what language those candidates belong. It contains a series of nodes representing orthography, phonology, semantics, and language nodes. Orthographic input activates nodes representing orthographic representations, and this activation spreads to higher levels of representation. The task schema is responsible for the maintenance and execution of task goals (e.g., make a lexical decision). The model achieves nonselectivity by making the following two assumptions. First, within the word recognition system, the words of each language are assumed to be stored in an integrated lexicon. This shared storage allows activation to spread between words in both languages. Second, it assumes a degree of modularity between the subsystem responsible for word recognition and the subsystem responsible for task goals and decisions (the task schema). The word recognition system provides feed-forward activation to a task schema responsible for making decisions about the task at hand (e.g., a lexical decision). However, the task schema is prevented from providing any kind of input into the word recognition system. In this manner, the task goals (e.g., read only in Spanish) cannot influence lexical activation patterns. Any task effects observed in experiments (e.g., elimination of the homograph effect in contexts where the semantics are unimportant) are assumed to come into play following word recognition. The BIA+ model assumes that word recognition is nonselective from the earliest stages of lexical access, consistent with the eye tracking studies reviewed above.

Much of the work on the interaction between languages during word recognition has been conducted with bilinguals as participants.Footnote ³ Substantially less work has focused on multilingual speakers, likely because of the increase in complexity as additional languages are added into experiment. Lexical variables must be stringently controlled between additional languages, and for proficient speakers of multiple languages, it is also more difficult to assess the relative proficiency and dominance of each language. The extant research on multilingual word recognition has converged on much the same interpretation as the bilingual research. Cognate facilitation has been observed among speakers of three languages. (e.g., Lemhöfer et al., Reference Lemhöfer, Dijkstra and Michel2004; Szubko-Sitarek, Reference Szubko-Sitarek2011; Van Hell & Dijkstra, Reference Van Hell and Dijkstra2002). Both the L1 and L2 become activated during L3 processing (Lemhöfer et al., Reference Lemhöfer, Dijkstra and Michel2004; Szubko-Sitarek, Reference Szubko-Sitarek2011). Van Hell and Dijkstra (Reference Van Hell and Dijkstra2002) found facilitation in the L1 for cognates with the L2 but also for cognate words with the L3, if the L3 was sufficiently proficient. These results indicate that all of the known languages in which there is a high level of proficiency become active when bilinguals or trilinguals are recognizing words in one language alone. A relatively open question in the literature is the extent to which multiple languages are genuinely activated in parallel. If this is the case, then words that share characteristics between three languages would be predicted to be processed differentially relative to words that overlap between two languages only. The available research suggests that this may be the case (Lemhöfer et al., Reference Lemhöfer, Dijkstra and Michel2004; Szubko-Sitarek, Reference Szubko-Sitarek2011), attesting to the interactivity of the multilingual lexicon. More research is needed to elucidate the differences between bilingual and multilingual word recognition.

The existence of language nonselectivity and information about which lexical codes are activated in parallel are well documented. Given the pervasiveness of the phenomenon, recent research has considered how lexical selection occurs. Without a mechanism for selection, bilinguals might be expected to demonstrate confusion between their two languages. Yet bilinguals are not typically confused; they only rarely make slips of the tongue (Poulisse, Reference Poulisse1999). To investigate language selection, research has focused on the question of whether there are factors that can function as a language cue that might allow for language-selective access.

One potential source for a language cue may be present in the cross-linguistic differences between two language pairs; languages often differ on many facets. Cognates, for example, often exhibit slight differences in their realization between two languages, despite their overall shared form. Cognates often have distinct phonology in each language (e.g., the cognates base and base in English and Spanish) and can also lack perfect orthographic overlap (e.g., the cognates ship and schip in English and Dutch). Likewise, cognate translations related by phonology may be written in entirely different scripts, as is the case between Hebrew and English or Chinese and English. In the most extreme case, one of the two languages in a pair may lack a system of writing entirely, such as is the case for ASL-English bilinguals. Counterintuitively, these structural differences do appear not to function as a language cue. As mentioned earlier, Morford et al. (Reference Morford, Wilkinson, Villwock, Piñar and Kroll2011) found that ASL-English bilinguals were facilitated in judging that two English words were semantically related if the two words also shared similar hand-shape (i.e., phonological) forms in ASL, suggesting that ASL became activated during the processing of English words, a context where ASL was not perceptually relevant. While the form of the bilingual's two languages may influence processing, it does not eliminate nonselectivity entirely. More research is necessary before we can conclude that bilinguals do not exploit language-specific features to allow language-specific lexical access.

A second potential cue may be present in aspects of the language context and in higher order linguistic representations such as the syntax or semantics in which words are typically embedded. Despite the presence of rich context in naturalistic language use, the early experimental evidence for nonselectivity came almost entirely from tasks in which words were presented in isolation (e.g., Dijkstra et al., Reference Dijkstra, Van Jaarsveld and Ten Brinke1998; Dijkstra et al., Reference Dijkstra, Grainger and Van Heuven1999; Schwartz et al., Reference Schwartz, Kroll and Diaz2007; Van Hell & Dijkstra, Reference Van Hell and Dijkstra2002). An obvious question was whether sentence context itself would override the nonselectivity observed in isolated word recognition. Quite counterintuitively, recent research suggests that the mere presence of a sentence context alone seems to be ineffective in allowing a bilingual to select one language during comprehension. When bilinguals process language-ambiguous words within a coherent sentence context, the effects of the language not in use remains, as if the words had been presented out of context. For example, Van Assche et al. (Reference Van Assche, Duyck, Hartsuiker and Diependaele2009) reported cognate effects while Dutch-English bilinguals read sentences in their native language, Dutch. Although the sentences appeared in only one language and that language was their native and more dominant language, there was a persistent effect of English, the L2, on the processing of Dutch, the L1. Although effects tend to be more robust in the L2, which tends to be more vulnerable to the influence of the L1, the overall pattern of these findings have been replicated in a number of studies with different language pairs.

One aspect of sentence context that does appear to function to accomplish language selection is highly predictable semantic constraint. When a sentence is highly predictable in its interpretation, cross-language effects are diminished to the point where they are no longer observable. For example, Schwartz and Kroll (Reference Schwartz and Kroll2006) asked Spanish-English bilinguals to read sentences in which cognates and noncognate controls were embedded. One set of sentences was low constraint, in that the critical target word, a cognate or control, was not predictable on the basis of the initial context. Another set of sentences was highly predicable. To illustrate, in a sentence like “When we entered the hall we saw a piano in the corner of the room,” the cognate word “piano” is not predictable given the surrounding context, hence it has a low semantic constraint. When the same cognate is placed in the sentence “Before playing, the composer wiped the keys of the piano at the beginning of the concert,” it becomes highly predictable given the preceding context. Schwartz and Kroll found that in low-constraint sentences, there was cognate facilitation similar to out of context presentation. In contrast, following high semantic constraint, cognate facilitation was eliminated, suggesting that word recognition became language-selective. The results were virtually identical in both English and Spanish. Schwartz (Reference Schwartz2003) further found that high-constraint sentences functioned to eliminate cross-language phonological modulation that had been observed in isolated word recognition (e.g., Schwartz et al., Reference Schwartz, Kroll and Diaz2007). This type of interaction between semantic constraint and language co-activation has been documented in a handful of other studies (e.g., Chambers & Cooke, Reference Chambers and Cooke2009; Libben & Titone, Reference Libben and Titone2009; Titone et al., Reference Titone, Libben, Mercier, Whitford and Pivneva2011; Van Hell & De Groot, Reference Van Hell and De Groot2008), although there is remaining debate about the presence and locus of the semantic constraint effects (e.g., see Van Assche et al., Reference Van Assche, Drieghe, Duyck, Welvaert and Hartsuiker2010).

Grammatical information also seems to function as a potential language cue. Sunderman and Kroll (Reference Sunderman and Kroll2006) found that lexical form interference could be eliminated in a translation recognition task (i.e., decide whether two words are translations of one another) when the two words differed in their grammatical class. Likewise, Baten, Hofman, and Loeys (Reference Baten, Hofman and Loeys2010) demonstrated that word class interacted with the degree of language co-activation for words embedded in a sentence context. A facilitatory homograph effect was present when participants were required to make a lexical decision to target words, but only when the meaning of the homograph shared grammatical class with its translation. For example, when the Dutch-English homograph brief was used as an adjective in an English sentence (brief is a noun meaning letter in Dutch), no homograph interference was observed, suggesting that higher order grammatical properties such as word class can provide information that can aid bilinguals in selecting the target language. In each of these examples, it is not clear whether the locus of selection occurs early or late in processing. Early selection would occur if information about the language to be used was able to guide lexical access to only the language in use. Late selection would involve activation of both language alternatives regardless of the presence of language cues pointing to one language alone and only later in processing would the language-appropriate alternative be chosen. Although more evidence overall suggests a late point of selection, consistent with the predictions of the BIA+ model, identifying the precise locus at which selection will require that studies use methods such as eye tracking and ERPs that permit a sensitive analysis of the early time course of processing.

In sum, it is apparent that bilinguals and multilinguals activate the representations of all the languages they know. It is unclear whether a lexical alternative in a single language is eventually selected, whether this is necessary in comprehension, and what factors may function to allow for selection. In some respects, it is not that surprising that word recognition is language nonselective since the information presented to readers (and to listeners, although we have focused here on visual word recognition) is largely not under the control of the individual. We turn now to lexical production, a process that necessarily is under the speaker's control.

CROSS-LANGUAGE ACTIVATION DURING WORD PRODUCTION

The act of planning to speak a single word engages a set of component processes that include the intention to describe a thought, name an object, or translate a word; the encoding of the concept to be described; access to the possible words that correspond to that concept; specification of the phonology associated with the target words; and execution of an articulatory plan. How a single word is eventually selected is a problem that has been studied intensively in the last 25 years (e.g., Levelt, Reference Levelt1989). If bilinguals were able to specify in advance the language that they intend to speak, then the characterization of the speech-planning process proposed for monolingual speakers would also apply. However, the recent evidence on speech planning in bilinguals, like the evidence on word recognition, shows that information about both languages is active, at least momentarily, when bilinguals plan to speak even a single word in one language alone (e.g., Costa, Reference Costa, Kroll and de Groot2005; Hanulovà, Davidson, & Indefrey, Reference Hanulovà, Davidson and Indefrey2011; Kroll et al., Reference Kroll, Bobb and Wodniecka2006). The presence of cross-language activation in production is counterintuitive because speech planning should be under the control of the speaker, and the selection of the language to speak should, in theory, be able to be specified in advance. The focus of the recent research has been to identify the locus of cross-language interactions and the mechanisms that permit bilinguals to select the language to be spoken.

Two classes of models have been considered to account for the way in which bilinguals select a single word to speak. One alternative proposes that selection is language specific (e.g., Costa, Miozzo, & Caramazza, Reference Costa, Miozzo and Caramazza1999; Finkbeiner, Almeida, Janssen, & Caramazza, Reference Finkbeiner, Gollan and Caramazza2006; Finkbeiner, Gollan, & Caramazza, Reference Finkbeiner, Almeida, Janssen and Caramazza2006). On this view, there may be activation of words within the language not in use, but the activation of those words does not automatically make them candidates for selection. In effect, the activation of lexical candidates in the language not to be spoken does not create competition for selection. This solution to the selection problem draws heavily on mechanisms that enable the bilingual to attend only to candidates within the target language. We have argued elsewhere (e.g., Kroll, Guo, & Misra, Reference Misra, Guo, Bobb and Kroll2012) that this so-called mental firewall across the two languages would logically require that bilinguals are able to exploit cues to language status that enable them identify the language in use. Those cues may be related to linguistic features of the two languages but also to features of the context in which the two languages were acquired and are used. For example, bilinguals whose two languages are structurally distinct might be expected to have an easier time identifying the target and nontarget languages. Yet the evidence suggests that even very obvious features that might distinguish the two languages are not exploited in this way.

Studies of language production in different-script bilinguals (e.g., Hoshino & Kroll, Reference Kroll, Bobb, Misra and Guo2008) or different-modality bilinguals (e.g., Emmorey, Borinstein, Thompson, & Gollan, Reference Gollan, Montoya, Cera and Sandoval2008) show that despite apparent differences in the bilingual's two languages, there is persistent activity of the language not in use. To illustrate, Hoshino and Kroll (Reference Hoshino and Kroll2008) asked Japanese-English and Spanish-English bilinguals to name pictures whose names were either cognates or not across the two languages. For Japanese-English bilinguals, there can be no orthographic overlap in the names; only the phonology can be shared. For Spanish-English bilinguals there can be overlap at both the orthographic and phonological levels. They found identical cognate effects for both groups, and no cognate effects for monolingual English speakers naming the same pictures. The results suggest that the presence of any cross-language features that are shared, regardless of the presence of those that are not, is sufficient to create the observed interactions. The phonology of the language not in use seems to be activated regardless of the intention to speak the target language and regardless of whether the two languages are easily separable.

If information in both languages becomes activated when bilinguals intend to speak one language only, then a later mechanism must be engaged to eventually select words in the language to be produced and to control production to avoid inadvertent errors of language. Observation and recent empirical evidence suggest that bilinguals rarely make the error of speaking the language not in use. Although older bilinguals are more likely to make language errors than younger bilinguals, these errors are much lower than what might be expected (e.g., Gollan, Sandoval, & Salmon, Reference Gollan, Sandoval and Salmon2011).

The other class of models assumes that there is competition for selection, with candidates in both languages available, but subsequent inhibition of the language not in use and particularly of the stronger or more dominant language (e.g., Green, Reference Green1998; Guo, Liu, Misra, & Kroll, Reference Guo, Liu, Misra and Kroll2011; Misra, Guo, Bobb, & Kroll, Reference Kroll, Gollan, Ferreira, Goldrick and Miozzo2012; Kroll, Bobb, Misra, & Guo, Reference Kroll, Bobb, Misra and Guo2008; Levy, McVeigh, Marful, & Anderson, Reference Levy, McVeigh, Marful and Anderson2007; Linck et al., Reference Linck, Kroll and Sunderman2009; see Philipp, Gade, & Koch, Reference Levy, McVeigh, Marful and Anderson2007, for evidence on trilinguals). The evidence for inhibition initially came from the demonstration of processing costs in language switching. Meuter and Allport (Reference Meuter and Allport1999) showed that bilinguals suffer larger switch costs when switching languages from the L2 into the L1 than the reverse. The result might be seen as counterintuitive because the L1 is normally more available than the L2. However, the asymmetry in switch costs can be understood if we assume that in order to name the L2, the activated L1 has to be suppressed. The larger switch costs on the subsequent naming trial then reflects a spillover effect, with L1 requiring more resources following this momentary suppression. There has been extensive debate about this interpretation of the asymmetric switch costs that questions whether an inhibitory mechanism is required and, if so, whether it characterizes the performance of all L2 speakers (e.g., Costa & Santesteban, Reference Costa and Santesteban2004; Finkbeiner, et al., Reference Finkbeiner, Gollan and Caramazza2006; Gollan & Ferreira, Reference Gollan and Ferreira2009; Kroll & Gollan, in press). What is clear in the empirical evidence is that the L1 is differentially slowed relative to the L2. Similar effects are seen when the language of naming is mixed, with uncertainty about which of the two languages will be required (e.g., Christoffels, Firk, & Schiller, Reference Christoffels, Firk and Schiller2007; Guo et al., Reference Guo, Liu, Misra and Kroll2011; Kroll, Dijkstra, Janssen, & Schriefers, Reference Kroll, Dijkstra, Janssen and Schriefers2000). What is unclear is whether slowing of the L1 reflects an inhibitory process that then imposes demands on general mechanisms of cognitive control or accommodation to the presence and activation of the two languages more generally (e.g., see Gollan, Montoya, Fennema-Notestine, & Morris, Reference Gollan, Montoya, Fennema-Notestine and Morris2005, for evidence on slower L1 naming in bilinguals relative to monolinguals).

One claim in the recent literature is that the demands on speech planning for bilinguals create a deficit in lexical access that is particularly apparent in spoken production. Gollan and colleagues (e.g., Gollan, Montoya, Cera, & Sandoval, Reference Gollan, Montoya, Cera and Sandoval2008; Gollan et al., Reference Gollan, Montoya, Fennema-Notestine and Morris2005; Gollan et al., Reference Gollan, Sandoval and Salmon2011) have argued that the observed costs to the dominant language can be understood with respect to frequency. Bilinguals divide their time over the two languages so that, all other things being equal, they necessarily speak each of their languages less frequently than monolinguals. This so-called weaker links hypothesis or frequency lag hypothesis gains support from studies that compare the performance of bilinguals and monolinguals. Bilinguals have more tip-of-the-tongue states than monolinguals, they are slower to name pictures in their native language, and they perform more poorly than monolinguals on measures of verbal fluency in which they are asked to generate as many exemplars of a category as they can name in a fixed time period (see Kroll & Gollan, in press, for a recent review). The question that this alternative raises is whether the observed costs to bilingual lexical access in production are due to reduced functional frequency for both languages, including the dominant L1, or whether they can be explained as a consequence of cross-language competition and subsequent selection. Many of these phenomena can be explained equally well by either alternative. It is quite possible that both alternatives are correct in characterizing different aspects of the speech-planning process. The frequency lag alternative is a relatively passive process that reflects the representational availability of words in each language. The competition for selection alternative is an active process that requires that cognitive control mechanisms be recruited for even the simplest task of producing the name of a single object. As we will see in the next section on neuroscience evidence on bilingual lexical access, recent electrophysiological and neuroimaging studies reveal differences between bilingual and monolingual both structurally and functionally in addition to differences in the brain activity that is engaged during the planning of speech in the L1 versus L2.

THE NEURAL BASIS OF BILINGUAL AND MULTILINGUAL LEXICAL PROCESSING

A theme in the earlier sections of this chapter is that the languages known by a bilingual or multilingual speaker are activated in parallel when any one language is understood or spoken. As noted in the introduction, there has been an upsurge of recent research using neuroscience methods to investigate the time course and localization of language processing in the brain. The new evidence from these neuroscience studies, using ERPs and fMRI, has illuminated the presence of cross-language activation of lexical information and also its consequences. Critically, the use of electrophysiological measures has provided a highly time-course-sensitive source of information. With ERPs specifically, the very earliest processes that are active in the brain when a speaker comprehends a word or sentence or plans to speak in one language can be identified prior to the speaker having any conscious experience of that process. In contrast to the high temporal resolution permitted by ERPs, fMRI methods allow spatial resolution that permits investigations of the cortical representation of language.

The results of neuroscience studies sometimes converge with the behavioral evidence we have reviewed in the previous sections, but they also provide unique information that is potentially unavailable in the behavioral record alone. In the brief review that follows, we illustrate these new contributions by focusing on results that provide unique information about bilingual and multilingual lexical processing. The selective nature of our review should not mislead the reader. There is, in fact, a great deal of evidence that supports the major findings and conclusions of the behavioral studies. We reference other recent reviews that can be consulted for more detail within each of the topics we discuss.

Late L2 learning is typically considered difficult, especially at initial stages of learning and especially for the grammar and phonology (e.g., Johnson & Newport, Reference Johnson and Newport1989; Piske, MacKay, & Flege, Reference Piske, MacKay and Flege2001). The assumption in the past literature has been that lexical knowledge is relative easy to acquire and unaffected by age of acquisition. However, recent studies suggest that there are effects of age of acquisition of the L2 that influence not only the grammar and phonology, but also the lexicon (e.g., Isel, Baumgaertner, Thrän, Meisel, & Büchel, Reference Isel, Baumgaertner, Thrän, Meisel and Büchel2010; Izura & Ellis, Reference Izura and Ellis2004).

One approach is to examine the neural basis of lexical acquisition by exploiting techniques that reveal brain activity in real time to identify changes that occur during the very earliest phases of L2 exposure. In a seminal study using ERPs, McLaughlin, Osterhout, and Kim (Reference McLaughlin, Osterhout and Kim2004) provided evidence to show that the brain starts processing and learning at very early stages of language acquisition, critically before learning is detectable in behavior. In this study, the focal ERP component was the N400, a negative wave that finds its maximum at around 400 ms after stimulus presentation. In previous studies, the N400 has been shown to reflect lexical processing and word meaning (e.g., see Kutas & Federmeier, Reference Kutas and Federmeier2011, for a review of the past 30 years of research on the N400). McLaughlin et al. compared native English speakers enrolled in an introductory French course with a control group of nonlearners. ERPs were longitudinally recorded at three points in time (following 14, 53, and 138 mean hours of instruction) while participants performed a lexical decision task (i.e., deciding whether a target word is a real word or not in French) on a prime-target pair of either semantically related French words (chien-chat, dog-cat), semantically unrelated French real words (chien-table, dog-table), or a word/nonword pair (chien-nasier, dog-nasier). Results showed that nonlearners did not show any differences behaviorally, nor did they show any modulation of the N400 component in the ERP record. Critically, for the French learners, despite the failure to observe behavioral improvement during this early stage of instruction, a significant modulation of the N400 for nonwords was seen almost immediately, after only 14 hours of instruction. The result is a dramatic demonstration that the brain may precede behavior in revealing the consequences of initial lexical learning.

Similar patterns of dissociations between ERP and behavioral data have been reported in other studies of early L2 lexical learning (in comprehension) using semantic categorization tasks (e.g., Alvarez, Holcomb, & Grainger, Reference Alvarez, Holcomb and Grainger2003). Modulations in the ERP signal in the absence of a change in behavioral performance have also been observed in early stages of grammatical processing (Tokowicz & MacWhinney, Reference Tokowicz and MacWhinney2005).

Raboyeau et al. (Reference Raboyeau, Marie, Balduyck, Gros, Demonet and Cardebat2004) investigated L2 lexical production in native French speakers who were trained to learn novel words in English. They used positron emission tomography (PET),Footnote ⁴ testing participants at two time points (T1 and T2). Results showed differential areas of activation for French (L1) and English (L2). French naming (in L1) was correlated primarily with activation in the left frontal temporal network, while English naming (in L2) activated a network of areas comprising the left frontal cortex, that is, BAs 4/6 (areas related to speech output), the right cerebellum and the left insular cortex (reported to be activated during speech gesture learning), and the right medial temporal regions that have been hypothesized to reflect the involvement of episodic memory during verbal learning. Interestingly, the anterior cingulate cortex (ACC), an area crucially involved in attentional and cognitive control, was also found to be active, suggesting that naming in the L2 requires attentional control to possibly inhibit the relative L1 lexical competitors.

In a subsequent study using fMRI, Raboyeau, Marcotte, Adrover-Roig, and Ansaldo (Reference Raboyeau, Marcotte, Adrover-Roig and Ansaldo2010) investigated the role of different learning phases and word type during an L2 longitudinal lexical learning task in French native speakers exposed to novel Spanish vocabulary. To assess the effect of the learning phase, participants underwent two fMRI sessions, the first one five days after the learning section (early learning phase) and a second one approximately 14 days after the first learning phase (consolidation phase). To examine the role of word type, words included French-Spanish cognates and noncognates. Participants were asked to perform a picture naming task (overt speech) while in the scanner, both in French (L1) and in Spanish (L2). Behaviorally, the results showed that during the early learning phase, participants were able to correctly name a subset of the new items, with cognates being faster and more accurate than noncognates. During the consolidation phase, performance was at ceiling for all word types. Like the previously reported PET data, the fMRI data showed that areas of the brain involved in attentional control were activated (e.g., the ACC). The result again suggests that to access and speak the L2, speakers may need to recruit additional cognitive resources that allow them to negotiate potential competition from the more dominant language. Activation in Broca's area was also observed for the successful retrieval of cognates during the early learning phase, suggesting that cognate retrieval may engage phonological processing, whereas activation in areas related to semantic processing was observed for the retrieval of noncognates. Taken together, these results converge with the conclusions reached by the studies utilizing ERPs in showing that, at very early stages of L2 lexical learning, it is possible to observe neural and electrophysiological changes that may not observable in behavior. See Osterhout, Mclaughlin, Pitkänen, Frenck-Mestre, and Molinaro (Reference Osterhout, Mclaughlin, Pitkänen, Frenck-Mestre and Molinaro2006) for a review of neuroscience methods that reveal the changes in the brain for the earliest stages of lexical and morpho-syntactic learning in late L2 learners.

The use of neuroimaging techniques (ERPs and fMRI) has also been informative with respect to testing the prediction of models of the bilingual lexicon that were initially based on evidence from behavioral studies. In the earlier sections of this chapter, we discussed in some depth the predictions of the BIA+ model (Dijkstra & Van Heuven, Reference Dijkstra and Van Heuven2002) regarding language nonselectivity. Van Heuven and Dijkstra (Reference Van Heuven and Dijkstra2010) have recently reviewed the neuroimaging studies that provide further support for the model. Here we focus on the debate surrounding the revised hierarchical model, or RHM (Kroll & Stewart, Reference Kroll and Stewart1994). As noted earlier, there has been disagreement in the past behavioral literature as to whether L2 learners rely on the L1 translation equivalent to mediate access to the meaning of L2 words. The RHM proposed that this lexically mediated route to meaning was critical early in learning because it provided a means to exploit the existing lexical-semantic networks in place for the L1. Critics of the model (e.g., Brysbaert & Duyck, 2010) have argued that all L2 speakers, regardless of their proficiency, are able to process L2 words without L1 mediation. The behavioral studies provide mixed evidence, with some studies suggesting that the L1 translation is active, and others suggesting not. Critically, a recent set of studies using ERPs and behavioral methods (e.g., Morford et al., Reference Morford, Wilkinson, Villwock, Piñar and Kroll2011; Thierry & Wu, Reference Thierry and Wu2007) have challenged the claims of the RHM and also of its critics. These new studies show that even highly proficient bilinguals activate the L1 translation of L2 words when they are reading in L2 context without the overt presence of the L1. The result is a problem for the RHM because the model initially claimed that the L1 translation was only used as a mediator during early stages of learning, when the connections between form and meaning for the L2 were not already established. Proficient bilinguals should have no need for this sort of mediation, and indeed, they would not be able to speak fluently if required to access the translation equivalent as they read or listen to speech in the L2. But the result is also a problem for critics of the RHM who claim that the L1 translation is not required at all.

Thierry and Wu (Reference Thierry and Wu2007) tested relatively proficient Chinese-English bilinguals immersed in the L2 in the United Kingdom. Participants were required to decide whether a pair of L2 words in English were semantically related or not. Unbeknownst to the participants, some of the related and some of the unrelated word pairs, if translated into Chinese, contained Chinese characters that were shared. If these relatively proficient bilinguals were able to read in L2 without accessing the L1 translation, then no effect of the hidden Chinese characters was expected. However, if they did access the L1 translation, then a conflict should have resulted when the characters matched in Chinese but the English words were semantically unrelated or when the characters in Chinese mismatched but the English words were semantically related. Thierry and Wu's results showed a reduction in the N400 component for judgments in which the Chinese translations were shared. The results were interpreted as showing a semantic priming effect from the L1 to the L2. Critically, they suggest that the L1 translation continues to be activated in even highly proficient bilinguals. The study by Morford et al. (Reference Morford, Wilkinson, Villwock, Piñar and Kroll2011) reviewed earlier as evidence for language nonselectivity reported a similar result in a behavior version of this task. The Morford et al. study also asked whether the L1 translation is activated when bilinguals read in the L2. But perhaps even more surprising than finding these effects in Chinese-English bilinguals, Morford et al. found a similar result for deaf readers of English as the L2, where the L1 translation in sign language takes a completely different form. Together, these recent studies point to an important role of the translation equivalent for late bilinguals regardless of the similarity of the two languages.

As we noted above, these recent findings demonstrating activation of the L1 translation for highly proficient bilinguals are problematic for all accounts of lexical development in the L2. Guo et al. (Reference Guo, Misra, Tam and Kroll2012) performed an ERP study to determine whether highly proficient bilinguals are indeed using the translation equivalent for the same purpose that learners are hypothesized to use it, for example, to access the meaning of L2 words. The participants in the Guo et al. study were Chinese-English bilinguals immersed in English in the United States. They were very similar to the bilinguals tested in the Thierry and Wu (Reference Thierry and Wu2007) study. The study used a translation recognition task, in which participants had to decide whether a Chinese word was the correct translation of an English word. The critical conditions in the experiment occurred when the two words were not translation equivalents but included a critical distractor. The distractor word could be related in form to the correct translation, related in meaning to the correct translation, or completely unrelated to the correct translation. Based on past behavioral results, Guo et al. predicted that if bilinguals access the L1 translation equivalent to understand the meaning of the L2 word, then distractors similar to the correct translation should produce interference in both the behavioral task (i.e., longer Reaction times (RTs) than the unrelated distractors) and in the ERP record. Like the earlier ERP studies we have reviewed, the results of this experiment showed a dissociation between the behavioral and ERP data. When the interval between the two words was relatively long, 750 ms, in the behavioral data there were robust and similar effects observed for distractors related in lexical form and meaning to the translation. In the ERP data, a different pattern of results was obtained. Interference for the distractors related to the form of the translation was indexed by an early P200 component, whereas semantic interference was indexed by a later N400 component for the distractors related in meaning to the translation. In a later time window, both types of distractors produced a late positive component (LPC) observed between 500 and 700 ms. When the interval between the two words was shortened to 300 ms, the behavioral pattern remained the same, and the effects in the ERP data for the semantic distractors remained, but the early component for the translation form distractors disappeared. The results suggest that relatively proficient bilinguals do indeed access the L1 translation when reading words in L2 but that access to the translation follows rather than precedes access to the meaning of the L2 word. In this case, the aggregate nature of the response time measures indexing behavior was not sufficiently sensitive to reflect the differences apparent in the ERP data.

Our brief review of the neuroscience of lexical processes has focused on ERP measures because they provide information about the earliest time course in processing for what is typically a process that unfolds rapidly. It is beyond the scope of the present chapter to review in detail all of the ERP studies on bilingual word recognition and production (for recent reviews, see Kroll et al., Reference Kroll, Gollan, Ferreira, Goldrick and Miozzo2012; Van Hell & Kroll, Reference Van Hell, Kroll, Altarriba and Isurin2013). A number of fMRI studies have also documented the consequences of cross-language competition for language processing (e.g., see Van Heuven & Dijkstra, Reference Van Heuven and Dijkstra2010, for a review of the evidence related to the BIA/BIA+ models).

In our review of bilingual word production, we focused on nature of the selection process that enables a bilingual to select a given word to be planned for speech in the intended language. This issue has also been the focus of recent fMRI studies that attempt to characterize the brain networks that enable control of bilingual speech (e.g., Abutalebi & Green, Reference Abutalebi and Green2007). The competition that is hypothesized to result during the selection of lexical candidates for production is thought to be regulated in a network of brain structures associated with executive function and inhibitory control (e.g., the left prefrontal cortex, LPFC, and the anterior cingulate cortex, ACC). For example, Abutalebi et al. (Reference Abutalebi, Annoni, Zimine, Pegna, Seghier, Lee-Jahnke and Khateb2008) reported that this network was activated differentially when bilinguals choose between words to speak in one of their two languages compared to choosing a word from among others within one language alone. Other studies have demonstrated differential involvement of areas associated with cognitive control when bilinguals switch from one language to the other (e.g., Guo et al., Reference Guo, Liu, Misra and Kroll2011; Wang, Xue, Chen, Xue, & Dong, Reference Wang, Xue, Chen, Xue and Dong2007). The claim is that the parallel activity that we have documented throughout this chapter produces changes in bilingual brains that affect the domain general functions that underlie cognitive control. To illustrate, in a very recent paper, Abutalebi et al. (Reference Abutalebi, Della Rosa, Green, Hernandez, Scifo, Keim and Costa2012) showed that the ACC appears to be changed by bilingual experience. When placed in a context in which they are required to monitor conflict, not only do bilinguals outperform monolinguals, but they also appear to require fewer neural resources than monolinguals to resolve the same level of conflict.

In studies of cognitive control and its neural basis, bilinguals have been shown to outperform their monolingual counterparts on tasks that engage executive function, such as task switching, ignoring irrelevant information, and resolving conflict (e.g., Bialystok, Craik, Green, & Gollan, Reference Bialystok, Craik, Green and Gollan2009; Garbin et al., Reference Garbin, Sanjuan, Forn, Bustamante, Rodriguez-Pujadas and Belloch2010; Luk, Anderson, Craik, Grady, & Bialystok, Reference Luk, Anderson, Craik, Grady and Bialystok2010; Rodriguez-Fornells, Rotte, Heinze, Noesselt, & Muente, Reference Rodriguez-Fornells, Rotte, Heinze, Noesselt and Muente2002). These components of executive function have been hypothesized to enable bilinguals to control the selection of the intended language in the appropriate context. But the control that is exercised during bilingual language processing is thought to then spill over into cognitive functioning more generally. To what extent are these effects of bilingualism on executive function a consequence of lexical experience? Some studies have identified significant correlations between aspects of lexical performance, such as language switching, and aspects of cognitive performance, such as task switching (e.g., Prior & Gollan, Reference Gollan, Sandoval and Salmon2011). These correlations are consistent with the interpretation that language processing and cognitive processing draw on the same networks. But correlational analyses do not provide a causal account of how different aspects of language processing map onto their respective cognitive consequences. Other lexical research demonstrates a role for inhibitory processes, with active inhibition of the more dominant language when words are spoken in the less dominant language (e.g., Linck et al., Reference Linck, Kroll and Sunderman2009; Meuter & Allport, Reference Meuter and Allport1999; Misra et al., Reference Misra, Guo, Bobb and Kroll2012). A clear issue for ongoing research is to determine the scope and time course of these processes and to identify their consequences for behavior and for the neural processes that support it.