RESEARCH DOSSIER
Bilingual Language Processing and Code-Switching
Abstract. Bilingual language processing is the coordinated perception, comprehension, retrieval, and production of linguistic information across multiple learned languages. Code-switching adds the structural and interactional organization of language alternation to the scientific study of lexical accessibility and response control. This fundamental research monograph synthesizes foundational sociolinguistic and psycholinguistic research, experimental naming and translation paradigms, eye-movement studies, event-related potentials, neural accounts of language control, spontaneous conversational corpora, developmental evidence, and critical reviews of purported executive-function advantages. The analysis distinguishes parallel lexical activation from language selection, externally cued switching costs from voluntary conversational alternation, and language-specific adaptation from hypothetical transfer to general cognition. Recent 2025–2026 research extends the account to sentence predictability, syntax-driven intrusion errors, language-switching effects on attention and memory, and working-memory modulation during reading. The synthesis evaluates theoretical alternatives, sampling and measurement biases, ecological validity, replication, and causal limits. It identifies a research program integrating explicit processing models with the linguistic, cultural, and social conditions of bilingual communication. Human–Artificial language mediation is considered as a prospective interdisciplinary research setting with separately testable hypotheses.
Keywords: bilingual language processing; code-switching; bilingualism; psycholinguistics; language control; parallel lexical activation; lexical access; language-switching costs; cued language switching; voluntary switching; BIA+; adaptive control hypothesis; language dominance; eye tracking; neural language control; cognates; conversational code-switching; bilingual executive function; cross-linguistic activation
Bilingual Language Processing as a Scientific Object
Bilingual language processing concerns the perception, recognition, comprehension, selection, planning, and production of linguistic forms by people who have experience with two or more languages. Its central scientific question is how multiple linguistic repertoires become available during a particular communicative act, including situations in which only one language is overtly used. Bilingualism describes a multidimensional experience rather than a uniform cognitive category. Age of acquisition, habitual use, literacy, domain of proficiency, language similarity, interactional setting, and the status of each language in a community jointly shape the observations. The cross-language eye-movement evidence of Spivey & Marian (1999) makes this question experimentally tractable: an unspoken language can influence looking behavior even when a single language supplies the spoken instruction.
Code-switching is the situated alternation or combination of linguistic resources associated with different languages within or across utterances and conversational turns. Its organization can be studied at several levels: where a switch occurs in a clause, which lexical and grammatical material is selected, how interlocutors interpret the alternation, and how the speaker controls the production sequence. Poplack (1980) established the value of analyzing attested bilingual speech with attention to grammatical patterning, while Gumperz (1982) analyzed switching as an interactional contextualization resource. Neither conversational alternation nor a slowed reaction in a switching experiment has a fixed psychological meaning independent of task, setting, and communicative purpose.
The present research dossier integrates psycholinguistic processing with conversational organization. It examines how non-target words become activated, how intended-language responses are selected, how language context changes control demands, and why switching patterns differ across listening, reading, spontaneous dialogue, and laboratory picture naming. A processing account requires evidence about time and mechanism; a conversational account requires evidence about turns, identities, and recipient design. The review therefore treats naming latency, eye movements, event-related potentials, speech errors, and natural conversation as distinct observations with complementary inferential roles. The broad theoretical objective is an empirically constrained explanation of flexible language use rather than a single scalar measure of bilingual competence.
Defining Language Dominance, Proficiency, and Experience
Language dominance indicates a relative advantage in a specified domain and measurement context. A speaker may name pictures faster in one language, read technical material more fluently in another, and prefer a different language for intimate conversation. First language, strongest language, and most frequently used language therefore need separate operational definitions. Researchers who use dominance as an explanatory variable should report its basis: standardized naming performance, vocabulary assessment, self-reported skill, language-history interview, or an experimentally derived measure. In Gollan et al. (2005), picture naming and picture classification distinguished lexical retrieval efficiency from broader semantic categorization; a bilingual difference in naming did not license an inference of generalized weakness in conceptual knowledge.
Exposure history is a trajectory involving household practices, school instruction, migration, professional use, social networks, and changing opportunities to practice. Two individuals classified as bilingual can differ dramatically in the distribution and frequency of their language use. Balanced bilingualism is likewise task-relative; equivalent ratings for speaking and reading do not guarantee equivalent lexical accessibility under speeded conditions. Byers-Heinlein & Lew-Williams (2013) discusses early bilingual development as a normal family of experiences while stressing that separate language exposure and knowledge profiles matter. Studies that collapse distinct populations into one binary bilingual category lose precisely the experience variables that language-control models should explain.
A scientifically interpretable comparison specifies languages and varieties, proficiency methods, contexts of use, script systems, community norms, and the linguistic demands of the outcome task. Linguistic similarity can ease some cognate decisions while amplifying competition in others. Dominance may reverse across the lifespan when daily interaction changes, and participants may experience patterned language choices rather than random alternation. A useful research design measures bilingual history continuously and tests interactions between experience and task conditions. Group labels can support description, but mechanistic claims require evidence connecting measured experience to particular activation, control, or conversational outcomes.
A Historical Genealogy of Code-Switching Research
The study of bilingual alternation developed across traditions that initially emphasized different scientific problems. Sociolinguistic and anthropological research documented code-switching in actual communities, where choice of linguistic variety could index participation, stance, solidarity, or shifts in the activity being conducted. Gumperz (1982) examined conversational switching through discourse strategies and situated interpretation rather than isolated word lists. Poplack (1980) used recorded Spanish-English speech to investigate structural distributions and the relationship between bilingual competence and grammatical form. These research programs matter because the phenomenon first became visible as a richly organized social practice before cognitive experimentation began isolating switching decisions at millisecond timescales.
The grammatical tradition asked what structural constraints permit elements of different languages to coexist in one clause. Myers-Scotton (1993) proposed a matrix-language account specifying how an organizing language supplies morphosyntactic frames and how embedded material enters them. Auer (1998) gathered interaction-centered investigations emphasizing sequential context, meaning, and identity. The traditions need to be reconstructed in their historical context: a distributional regularity from a specific speech community and a universal grammatical proposal are different kinds of claim. Corpus annotation decisions, definitions of borrowing, and the languages sampled affect whether a proposed constraint survives comparison.
Psycholinguistic theories subsequently connected bilingual representations with the mechanisms of lexical choice. Kroll & Stewart (1994) used category interference in translation and picture naming to investigate asymmetric mappings among lexical and conceptual representations. Green (1998) offered an inhibitory-control account of selecting an intended language, and Meuter & Allport (1999) tested switching costs in cued numeral naming. Later eye-tracking and neural studies showed that comprehension can involve covert activation of a language never spoken in the trial. The research history thus connects corpus-based observation, formal linguistic theory, cognitive experimentation, and neuroscience, each revealing a different level of organization.
Conversational Functions and the Grammar of Switching
In natural conversation, speakers can alternate languages when introducing reported speech, addressing particular participants, signaling a change in frame, elaborating a concept, or responding to a previously used code. The function of a switch must be established from its sequential position, surrounding turns, participants' interpretations, and local linguistic practices. Gumperz (1982) provided influential analyses of contextual cues, while work brought together in Auer (1998) examined the interactional accomplishments of language alternation across settings. An analyst who assigns a universal social function to every switch risks importing an interpretation that the speakers themselves did not display.
The structure of an utterance constrains and enables switching through lexical category, morphosyntactic dependencies, word order, agreement, and the phonological properties of the languages involved. Poplack (1980) examined patterns of grammatical alternation, and Myers-Scotton (1993) articulated explicit predictions for within-sentence combinations. The evidence is best treated as a field of testable structural generalizations, not as proof that one constraint applies to all language pairs. Comparing patterns across typologically diverse languages requires aligned definitions of morpheme, insertion, alternation, constituent, borrowing, and conventionalized mixed usage.
Conversational code-switching also exposes limits of laboratory generalization. A speaker responding to a colored screen cue may switch without communicative motivation; a speaker telling a story may select a bilingual phrase because it advances a socially shared activity. Fricke & Kootstra (2016) documented priming of code-switching within spontaneous bilingual dialogue, giving a bridge between structural repetition and context-sensitive choice. The productive research strategy is to compare matched linguistic materials across conversational and experimental settings while preserving the fact that natural interaction supplies goals, expectations, and partners absent from many highly controlled experiments.
Operationalization and the Logic of Bilingual Evidence
A research program becomes cumulative when it states precisely what its dependent variables represent. Switch cost is conventionally the difference in response time or error rate between switch and repetition trials under a particular cue structure. A cognate effect is a contrast between related and control lexical items after matching confounds such as frequency and orthography. A code-switching frequency score counts occurrences under an annotation scheme, which is not a direct measure of cognitive effort. Eye fixations index time-sensitive visual attention and lexical competition only when a task connects looks to a specific stimulus set. Meuter & Allport (1999) and Spivey & Marian (1999) exemplify why these distinct outcomes answer different mechanistic questions.
Convergent methods are especially important because the same behavioral slowing can have several causes. Switching might invoke response-set reconfiguration, lingering suppression, expectancy violation, low-frequency word retrieval, or an interaction between proficiency and discourse context. Neural signals identify correlated processing stages and networks but do not by themselves establish an inhibitory mechanism. Thierry & Wu (2007) demonstrated covert first-language effects in event-related potentials under second-language-only stimulation; the inference of parallel activation was supported by an explicit hidden-translation manipulation. Causal conclusions about suppression demand experiments that dissociate inhibition from expectancy, task difficulty, and lexical accessibility.
The present synthesis is a critical narrative research monograph based on independently identifiable original studies, major theoretical accounts, and evidence reviews. It is neither a registered systematic review nor a new meta-analysis. Search coverage emphasizes experimental foundations, conversational research, and newer work through October 2026; accessible abstracts support only claims they actually report. Each study is evaluated by its materials, participants, contrast conditions, outcome measures, ecological context, and alternative explanations. The guiding principle is to preserve the link between an observed result and the precise explanatory proposition it can support.
Parallel Activation in Spoken Language Comprehension
Spoken comprehension begins with a continuously unfolding acoustic signal whose initial segments can support several candidate words. For bilingual listeners, the candidate set may include forms from both languages even when only one is relevant to the instruction. Spivey & Marian (1999) measured gaze to objects whose names competed across languages and found looks consistent with partial activation of an irrelevant-language lexical candidate. The observation challenges a strict input-switch model in which the unused lexicon is completely inaccessible at the earliest recognition stage. It does not imply that every word in both languages is activated equally; activation reflects acoustic overlap, learned mappings, language experience, and unfolding context.
Eye tracking in the visual world paradigm links time-locked spoken instructions to attention among pictured referents. Its explanatory force depends on the available distractors, timing of the critical sound, baseline item preferences, and whether participants treat the display as a constrained search environment. Marian et al. (2003) combined eye-tracking and brain-imaging evidence to address shared and distinct aspects of bilingual processing. Such converging results support partially overlapping representations and language-specific neural organization without requiring a simplistic picture of two fully separate lexicons. Spoken recognition should be analyzed as a competition-sensitive process that draws on both input-driven evidence and contextual expectations.
Not every cross-language fixation is a direct measurement of inhibitory failure. A listener can resolve the correct referent efficiently while briefly considering another candidate, and an increase in competitor looks may reflect differences in word frequency or phonotactics. Proper evaluation compares within-language and cross-language competitors, controls gaze latency, and distinguishes early activation from later decision processes. When a discourse partner consistently speaks one language, listeners can adjust prior expectations without erasing accumulated knowledge of the other. The resulting picture is one of graded access under contextual regulation, with eye movements serving as a window into the time course rather than a complete mechanistic explanation.
Bilingual Visual Word Recognition and the BIA+ Architecture
Written word recognition introduces a different configuration of evidence: letters, orthographic neighborhoods, phonological codes, and semantic representations can converge on multiple lexical entries. The BIA+ framework of Dijkstra & van Heuven (2002) represents bilingual word recognition through interacting orthographic, phonological, and semantic information, together with a distinction between a word-identification system and task or decision processes. This architecture explains why language membership cannot always be settled before candidate words become active. It also encourages a careful distinction between effects of immediate linguistic context and the instructions that determine what response a participant makes after recognizing the item.
Cross-language cognates offer a powerful manipulation because related forms may benefit from overlapping lexical and semantic structure. In sentence reading, Van Assche et al. (2009) recorded eye movements and found cognate facilitation even when bilingual participants read in their native language. The result supports influence of another language during an ordinarily automatized task. Interpretation requires matching orthographic similarity, frequency, predictability, and morphological structure. A cognate advantage should not be generalized to all translation pairs or all sentence environments. It reflects the organization of shared information under a particular language-pair and task configuration.
The BIA+ distinction between access and decisions is a theoretical proposal that can be challenged by evidence about when task cues influence early stages. Schwartz et al. (2025) reported eye-movement results from Spanish-English sentence reading suggesting that a language color cue can reduce switching costs, including evidence relevant to early lexical activation. This creates a discriminating test of models that sharply isolate access from nonlinguistic control. The productive interpretation is model comparison: estimate which aspects of a visual signal are modifiable by context, at what temporal resolution, and whether observed effects survive matched predictability and proficiency. Evidence for context-sensitive access refines the architecture rather than eliminating the necessity to distinguish recognition from response selection.
Lexical Similarity, Cognates, and Language-Specific Constraints
Language pairs differ in their mapping between sound, script, word form, and meaning. Spanish-English bilinguals encounter numerous partially similar forms; bilinguals using nonoverlapping writing systems encounter different orthographic cues, while phonological or semantic relations can still cross languages. Consequently, the magnitude and even the direction of cross-language interference depend on the items sampled. Van Assche et al. (2009) provides a particularly informative case because cognate facilitation appeared during naturalistic sentence reading, revealing consequences of additional-language knowledge without explicit language switching. A universal theory of bilingual lexical processing must therefore parameterize linguistic similarity rather than treating all bilingual participants as sharing the same representational geometry.
Some lexical items invite distinct sources of competition. Homographs can be identical in spelling but different in meaning across languages; cognates share meaning and form to varying degrees; translation equivalents can share a concept while differing entirely in surface structure. These relations should be annotated continuously, including pronunciation and distributional frequency, rather than simply assigned to binary categories. In Dijkstra & van Heuven (2002), orthographic, phonological, and semantic activation have different roles in the model; that separation allows experiments to target which information drives facilitation or conflict. The independent manipulation of script, meaning, and word form helps constrain how language nodes and contextual cues contribute.
Language-specific constraints also matter at a longer scale. A bilingual reader interpreting a sentence uses syntax, discourse expectations, and register knowledge to predict what forms are plausible next. The same unexpected foreign-language word may be processed differently in a bilingual conversation, a classroom reading task, or a passage with repeated switches. Studies must report whether sentence materials make switches predictable and whether participants have practiced the mixed register. A failure to match frequency or expectation could be mislabeled a language-control cost. The most informative experimental comparisons cross language membership with contextual probability, thereby separating the presence of a non-target language from the surprisal of a specific item.
Translation, Lexical Retrieval, and Conceptual Mediation
Translation tasks expose the architecture connecting lexical forms and meaning. Kroll & Stewart (1994) varied semantic category blocking during picture naming and translation and found direction-dependent interference, consistent with differences in the strength of interlanguage links and conceptual mediation. Such findings helped establish the importance of translation direction, proficiency, and representational access. When a participant translates from a strong language to a weaker one, retrieval may depend on different links than translation in the opposite direction. Yet translation is itself a task: the researcher requests an explicit form in a specified language, so response preparation and monitoring contribute alongside lexical representation.
A common conceptual representation does not require identical translation equivalents or a one-to-one mapping between words. Languages differ in lexicalization, grammatical categories, pragmatic connotations, and usage frequencies; bilingual concepts may be tuned through experiences available in particular linguistic communities. Picture-naming experiments can minimize some translation-specific demands, and comparisons with nonverbal categorization can probe the separability of concept retrieval from word production. Gollan et al. (2005) found that bilinguals' naming differences were not paralleled by equivalent impairment in picture classification. This dissociation helps isolate a lexical-access component while reminding investigators that classification tasks are themselves limited proxies for the range of semantic knowledge.
Direction and task context influence interpretation of any lexical disadvantage. Slower dominant-language naming in bilingual participants could reflect reduced use of the language, increased competition, stimulus familiarity, or an experimentally induced monitoring strategy. A study must distinguish these candidates through matched frequency distributions, language-specific practice, mixed versus unmixed blocks, and multiple nonverbal baselines. Gollan & Ferreira (2009) showed that voluntary use of either language can sometimes facilitate naming relative to constrained conditions, demonstrating that imposing a single response language changes the performance environment. Lexical representations cannot be fully explained without specifying the goals under which they are retrieved.
Language Selection in Speech Production
Speech production requires the retrieval of meanings, lexical candidates, grammatical dependencies, phonological forms, and articulatory plans under a particular communicative goal. For a bilingual, the intended response language is one dimension of this planning process. Selection can be modeled through the relative activation of alternatives, the adjustment of response thresholds, inhibitory operations, or combinations of those mechanisms. Green (1998) formalized an influential inhibitory-control model in which task schemas participate in controlling competing language representations. Its importance lies in explicit predictions about how a response language might be prepared and how residual control can influence later trials, rather than in any claim that every bilingual utterance requires equal suppression.
Production research commonly uses picture naming, number naming, translation, or reading aloud. These tasks differ in how much semantic retrieval they demand and in how directly the target word is available. Kroll & Stewart (1994) showed that the semantic organization of stimuli can alter translation and naming, emphasizing the role of conceptual and lexical mappings. In ordinary discourse, the speaker has a communicative plan and may choose the most appropriate linguistic resource before lexical retrieval reaches a final form. Experiments that constrain language selection artificially isolate some mechanisms while suppressing this freedom. A comparison across tasks is therefore essential for interpreting which form of control is operative.
Language selection also interacts with speech monitoring. A speaker can prepare a response in a preferred language, detect an unintended intrusion, or repair a word after articulation begins. These processes have distinct time courses and should be measured separately. Production errors offer evidence about control only when researchers can distinguish accidental intrusions from socially appropriate switches. The inferential challenge is to explain when the planned language is maintained through inhibition, when a lexical form wins through accessibility, and when an alternation is chosen because it supports the interaction. Modern experiments combine eye movements, acoustic onset timing, and error classification to separate those possibilities.
Cued Switching, Reaction-Time Costs, and Asymmetry
In a standard cued language-switching experiment, participants alternate naming responses according to arbitrary visual symbols or other instructional cues. The switch cost compares trials that require changing the response language with trials that repeat the preceding language. Meuter & Allport (1999) demonstrated asymmetrical switching costs in numeral naming, with a larger cost when returning to the dominant language than when moving toward the weaker language. This pattern was consistent with residual suppression of the dominant language during weaker-language production. The historical result remains a landmark because it transformed control theories into trial-level predictions that can be experimentally tested.
Asymmetric costs are not unique evidence for one mechanism. Preparation time, cue interpretation, language dominance, task-set inertia, stimulus repetition, and response thresholds can affect the contrast. Bobb & Wodniecka (2013) critically examined what asymmetric switching costs do and do not show about inhibition. A rigorous analysis considers the relationship between switch direction and dominance, whether cue identity changes across repeats, and whether the same items occur at comparable frequencies. The possibility of residual inhibition becomes stronger when alternative decision-level accounts are tested rather than merely assumed away. Symmetric costs in another participant group do not automatically disprove inhibition, because the model may predict different patterns when proficiency is more balanced.
The switch-cost measure is conditional on an experimental sequence. Its value may change after a long run in one language, with advance cueing, and when participants can predict the next switch. Researchers should analyze error rates together with latencies to rule out speed-accuracy tradeoffs. They should also distinguish local switch cost from global mixing cost, which contrasts mixed-language blocks with single-language blocks and may reflect sustained monitoring. An adequate theory predicts both types of cost under specified conditions. Treating any single latency difference as the amount of inhibition in the bilingual mind would exceed the measurement's explanatory scope.
Voluntary Switching and Context-Dependent Effort
Natural conversation frequently permits bilingual speakers to choose the language that makes a response more accessible, socially appropriate, or efficient. Gollan & Ferreira (2009) investigated voluntary picture naming in young and older Spanish-English bilinguals and found that switching patterns differed from arbitrary cued tasks. Participants often mixed languages even when switching had a measurable cost, and allowing a preferred language sometimes facilitated responding. The design demonstrates a crucial property of cognitive control: the cost of an externally imposed switch is not the same quantity as the cost of a self-selected, goal-sensitive alternation.
Blanco-Elorrieta & Pylkkänen (2017) directly contrasted laboratory switching with more naturalistic language use through neural and temporal measures. Along with the review by Blanco-Elorrieta & Pylkkänen (2018), this evidence challenges the inference that switching is inherently effortful in all contexts. The nature of a cue, the predictability of language demands, and whether a bilingual partner shares a mixed repertoire can alter recruitment of control mechanisms. Voluntary switching can also occur preferentially on accessible words, yielding a selection effect in which observed switches look easier because the speaker chooses them at favorable moments. This requires joint modeling of the switch decision and subsequent performance.
Comparisons between voluntary and cued conditions must consider that self-selection changes the distribution of words, language ratios, and sequence predictability. A difference in observed cost cannot be interpreted as pure strategy unless these design features are measured or controlled. Conversely, experiments that eliminate choice may overstate the relevance of response-set reconfiguration to everyday discourse. The strongest account predicts when externally cued control becomes necessary, when opportunistic lexical selection is beneficial, and when conversational conventions generate predictable alternation. The phenomenon is flexible precisely because multiple interacting control policies can support successful language use.
The Adaptive Control Hypothesis and Interactional Context
The adaptive control hypothesis of Green & Abutalebi (2013) proposes that bilingual language control is shaped by recurring interactional contexts. Single-language contexts, dual-language contexts in which different speakers or settings call for different languages, and dense code-switching contexts are predicted to place different demands on goal maintenance, interference suppression, response selection, monitoring, and switching. The proposal connects laboratory mechanisms with real-world practices by treating language context as part of the causal environment. It generates a research agenda: measure the actual ecology of language use and test whether specific control profiles are associated with that experience.
Adaptive control should be distinguished from the strong claim that all bilinguals develop globally superior executive abilities. The first hypothesis concerns adaptation of language-related processing to different demands; the second requires consistent transfer to nonlinguistic tasks. Blanco-Elorrieta & Pylkkänen (2018) argued that ecological mismatch can explain apparent disagreements between experiments and lived switching behavior. Experience measures must be collected prospectively or with carefully validated retrospective instruments, not inferred from a participant's citizenship or language labels. Dense mixing might reduce the need for prolonged language-wide suppression while increasing demands for local integration, but such predictions require direct testing in matched populations.
The model also raises important causal questions. People whose language control differs at baseline may select social settings with more or less mixing; training environments may affect language behavior without changing general executive control; and socioeconomic, educational, and community variables may confound associations. A convincing test would track switching contexts over time, independently characterize linguistic demands, and compare predictions from adaptive control with learning and selection accounts. Green & Abutalebi (2013) provides an explanatory framework, while observed correlational differences between speaker groups remain conditional evidence until alternative developmental pathways are addressed.
Neural Organization of Bilingual Language Control
Neuroimaging studies implicate distributed frontal, cingulate, subcortical, and language-related systems during language selection. Abutalebi & Green (2016) reviews neural adaptation and reserve models, linking selection demands to the structure and activity of control networks. Blanco-Elorrieta & Pylkkänen (2017) showed that the temporal and neural profile of switching depends on whether the task resembles controlled laboratory alternation or naturalistic use. Such findings place selection in an interacting network rather than assigning a dedicated language-switch center to a single anatomical region. Networks support multiple cognitive operations, and activation by a switching task may reflect error monitoring, conflict, preparation, or general task difficulty.
A mechanistic interpretation requires contrast conditions carefully matched in linguistic and motor demands. A speech-switch trial can produce more activity than a repetition trial because it requires a different lexeme, a changed phonological plan, or increased uncertainty. The neural response is therefore informative when paired with behavioral measures, explicit temporal modeling, and comparisons that isolate candidate control processes. Abutalebi & Green (2016) discusses adaptation in bilingual control systems, but cross-sectional imaging cannot alone determine whether language experience caused an anatomical difference or whether other correlated experiences contributed. Distinguishing plasticity from selection requires longitudinal, intervention, or naturally occurring change designs.
The integration of neural and behavioral evidence should remain sensitive to individual differences. Language acquisition age, daily practice, dominance, and the social meaning of each language may shift the balance of automatic retrieval and active control. Imaging studies with small or unusually homogeneous samples require special caution when drawing population-level conclusions. The research object is the organization of distributed selection processes under changing task demands. A neural network finding becomes scientifically explanatory when it yields testable predictions for who switches, when a switch is easy or difficult, and how representations are used during communication.
Sentence Context, Predictability, and Comprehension Switch Costs
Reading a sentence that contains an unexpected change of language differs from choosing a new naming language for the next trial. The reader integrates lexical form, grammar, discourse coherence, and predictions about what information should appear. A delayed fixation on a switched word can therefore reflect surprise at its form, the need to update a language expectation, or additional semantic integration. Chen & Christianson (2026) investigated Chinese-English bilingual reading with eye tracking and reported results that assigned substantial importance to word-form surprisal and narrative context in observed costs. The study provides a recent test of the assumption that encountering a different language automatically requires a discrete language-wide switching operation.
Context must be represented at more than one temporal scale. The preceding word can supply local syntactic constraints, while a paragraph establishes its discourse topic and language conventions. Repeated switches can create a new expectation for mixed-language material, and a speaker's identity may provide information about probable choices. Schwartz et al. (2025) found that language color cues affected eye-movement measures of comprehension in mixed Spanish-English sentences, suggesting that advance context may penetrate processes conventionally described as automatic lexical access. The results become especially informative when early eye-movement outcomes are compared with later regressions and when lexical overlap is independently manipulated.
A strong experimental design separates the language of a word from its probability in the given material. For example, switch and no-switch items can be matched on predictability by constructing appropriate discourse and exposure conditions; frequency, length, and cognate status can be included as independent factors. Researchers should also report whether switch costs are measured on the critical word or spill over into following words. Wang et al. (2026) found that working-memory load changed the profile of Chinese-English sentence-level switch effects. Such findings favor an integrated account in which language membership interacts with memory demands, linguistic expectations, and the temporal position of the measurement.
Grammatical Planning and Language Intrusion Errors
A language intrusion occurs when a bilingual unexpectedly produces a word in a language different from the one required by the task. Reading aloud of mixed-language passages makes such errors observable because the printed target specifies the intended form. Intrusions occur disproportionately on some grammatical categories and under particular dominance conditions, raising questions about how lexical retrieval is coordinated with syntax. Quinn et al. (2026) examined eye movements during successful and unsuccessful switches in Spanish-English paragraph reading. The study separated content words from function words and compared regressions, gaze behavior, and production outcomes to locate potential monitoring operations.
The experiment reported that dominant-language function words were especially vulnerable to intrusions even when readers showed greater monitoring of those targets. This matters for theories that assume a single selection mechanism applies uniformly to all lexical classes. Function words are integrated closely with sentence structure and are often processed rapidly, while content words can be semantically chosen earlier in planning. Quinn et al. (2026) therefore supports a distinction among syntactic planning, lexical activation, and monitoring, with evidence that attention can reveal an impending conflict without always preventing a production error. The result depends on the read-aloud procedure and cannot automatically be equated with naturally occurring code-switching.
The temporal order of preparation, eye movement, and articulation remains central. A regression to an earlier word could reflect advance monitoring, reconsideration after speech onset, or checking following an error. Synchronizing ocular and vocal measurements provides a way to distinguish these functions. Experiments can further alter predictability and grammatical structure to test whether syntactic organization changes the likelihood of intrusion independently of word frequency or language dominance. This approach links a classic production-control question to the structure of connected speech and gives more specific predictions than a general assertion that bilingual speakers suppress an unwanted language.
Structural Priming in Spontaneous Bilingual Dialogue
Structural priming is a tendency to reuse previously encountered linguistic patterns. In bilingual conversation, this can include the likelihood of alternating languages and the form of a mixed utterance. Fricke & Kootstra (2016) examined priming in spontaneous bilingual dialogue and reported that both switching behavior and the structure of code-switched utterances were susceptible to preceding linguistic experience. The result joins a cognitive account of recent structural activation with an interactional account of conversational alignment. It suggests that a language switch should sometimes be explained by accessibility inherited from the immediately preceding exchange rather than by a deliberate, isolated decision.
The distinction between within-speaker adaptation and partner-driven coordination is essential. A bilingual may repeat a structure because the representation has been recently activated, because a partner's phrase makes the pattern socially appropriate, or because the interaction moves into a familiar topic. Auer (1998) foregrounds the sequential and social conditions of switching, while Fricke & Kootstra (2016) gives a controlled bridge into experimentally testable repetition effects. To infer a structural mechanism, studies need comparably coded preceding and target utterances, controls for lexical repetition, and attention to who produced each prime. Real conversation offers ecological richness but also creates selection and dependency among observations.
A general model of bilingual production should account for both stable language distributions and short-term change. The speaker's long-term repertoire determines what is possible; current partners and topic make some choices probable; recently heard structures alter local availability. The result is a dynamic system in which switching is often predictable once these interacting timescales are represented. A future research program could use densely sampled dialogue, speaker-level random effects, explicit structural codes, and out-of-sample prediction. The core empirical aim is to identify which contextual variables change the probability of switching rather than inferring a general cognitive trait from switch counts alone.
Code-Switching, Attention, and Memory for Discourse
Code-switching can affect the listener's allocation of attention and the subsequent availability of information. Salig et al. (2025) investigated the effect of hearing switches embedded in stories and reported better attention to and memory for material near switches among bilingual listeners. The finding reframes a potential comprehension interruption as a possible orienting event under appropriate conditions. It does not imply that code-switching universally improves learning or that every bilingual listener benefits: the relevant contrasts involve particular story materials, language histories, and the timing of recalled information. The mechanism might involve distinctiveness, expectancy, contextual meaning, or interaction among these factors.
Attention and memory must be distinguished in the evidence chain. Self-reported concentration after a switch is not identical to an independently measured change in encoding. Later recall may depend on the salience of the switched item, its semantic relation to the narrative, and whether the switch cues an important boundary. Salig et al. (2025) reports both attentional and retention outcomes, while the broader memory literature shows why a retrieval advantage can emerge through several encoding and contextual pathways. A decisive design would hold narrative importance, prosody, lexical novelty, and overall predictability constant while varying language alternation and participant experience.
The implications extend to teaching, interpretation, media, and natural conversation, where a switch may signal a change in relevance. Yet applied recommendations require more than one experimental setting. What helps a bilingual listener attend to a story may disrupt someone who lacks the relevant language knowledge, and an attention effect over a few moments need not persist as a long-term learning benefit. The research question concerns how an alternation interacts with an established repertoire and task goal to shape selective processing. A mechanistic theory should predict both facilitated memory and interference under independently characterized circumstances.
Discourse Corpora, Transcription, and Interactional Evidence
Natural bilingual corpora provide evidence unavailable in isolated response-time tasks: turn taking, repairs, prosody, interlocutor relationships, audience design, code alternation across clauses, and conventional community norms. Poplack (1980) and the studies collected by Auer (1998) illustrate why recorded interaction remains indispensable for defining the phenomenon. Corpus research begins with transparent sampling of speakers, genres, settings, and recording conditions. Without these, an apparent linguistic regularity can reflect one community's speech practice or the researcher's sampling method rather than a general principle of bilingual cognition. The social meaning of language choice requires attention to actual participant behavior.
Transcription is an analytic operation. Decisions about where one language ends and another begins become difficult with shared vocabulary, borrowed forms, similar varieties, or integrated loanwords. Myers-Scotton (1993) addresses the relationship of structural constraints and borrowing, showing why categories should be stated before counting tokens. Reliable analysis requires explicit coding schemes, trained annotators, disagreement resolution, and documentation of ambiguous cases. Automatic language identification can assist large-corpus work but should be validated on mixed utterances, not only monolingual benchmark text. A classification system that ignores grammatical context may systematically mislabel the very phenomenon it is intended to explain.
Conversational analyses and quantitative models answer complementary questions. Sequential analysis examines what a particular switch accomplishes at a moment in interaction; a mixed-effects corpus model estimates how properties of language, speaker, and preceding discourse predict switching across many events. Combining them can show why a frequent pattern exists and how participants use it. Gumperz (1982) supplies an interpretive tradition, while Fricke & Kootstra (2016) shows that interactional switching can exhibit experimentally recognizable priming. The methodological aim is a dialogue between fine-grained social meaning and reproducible formal measurement.
Bilingual Development and the Emergence of Language Choice
Children acquiring two languages develop linguistic knowledge through experience with caregivers, peers, schools, and communities. Their patterns of choice and alternation depend on which languages are available to which partners and on the vocabulary and grammatical structures they have learned. Byers-Heinlein & Lew-Williams (2013) reviews evidence on early bilingual development and common claims about supposed developmental deficits or universal advantages. Research on code-switching in children must distinguish a deliberate communicative adaptation from an incomplete lexical entry, a recently learned pattern, or a conventional family practice. Age of exposure alone is insufficient to determine the causes of an observed switch.
Developmental measurement is challenging because performance can vary strongly across language and domain. A child may know a concept but retrieve the available name in one language more readily, and vocabulary inventories restricted to one language undercount the full repertoire. Longitudinal observations can measure how language choice changes with new educational and social environments. Kroll & Stewart (1994) supplies a model of asymmetric mappings among forms and concepts that can guide specific developmental hypotheses, although the original adult translation experiments are not direct evidence about every child's learning trajectory. Separate lexical growth, pragmatic appropriateness, and switching skill rather than assuming they all move together.
Developmental conclusions also depend on population sampling. Families differ in the status of their languages, opportunities for literacy, and exposure to institutional expectations favoring a particular code. Multilingual education research should therefore record not just test scores but frequency and type of input, interlocutor repertoire, and task language. Byers-Heinlein & Lew-Williams (2013) emphasizes that children can build two linguistic systems through ordinary experience. The most useful questions concern conditions supporting vocabulary growth, metalinguistic flexibility, and interactional competence, while treating language-specific variability as a meaningful research outcome rather than a deviation from one prescribed bilingual profile.
Individual Differences and Dynamic Language Experience
A bilingual speaker's processing profile varies across tasks, settings, and periods of life. Language dominance can change after relocation, professional specialization, or a shift in daily conversation; frequency of language mixing can change when partners or social opportunities change. Gollan & Ferreira (2009) showed that voluntary switching behavior is sensitive to dominance and age in the tested Spanish-English sample. A repeated-measures design is particularly valuable because it separates stable between-person differences from within-person adjustments to context. The scientific object becomes a dynamic distribution of response policies rather than a permanent placement on a bilingual-versus-monolingual scale.
Task-specific measures help reveal which component of experience matters. Picture naming captures lexical retrieval under visual prompting; a self-report questionnaire measures perceived use or competence; conversation recordings show actual alternation; eye tracking measures momentary competition during comprehension. These outcomes should be modeled together only after their measurement properties have been established. Gollan et al. (2005) illustrates why slower naming does not imply poorer conceptual classification. Similarly, many daily switches do not by themselves establish stronger or weaker domain-general inhibitory control. An individual profile must preserve the differences among representation, retrieval access, selection, and social choice.
Researchers should also separate experience from social labeling. Heritage speakers, late learners, immigrant multilinguals, classroom learners, and speakers using regional varieties can have overlapping skills but different opportunities for code-switching. The causal pathways include practice, partner expectations, register specialization, and linguistic similarity. A longitudinal framework would measure these features repeatedly and test whether changes in actual language use predict subsequent shifts in lexical or switching behavior after controlling for proficiency and environment. Such a framework is compatible with the adaptation proposal of Green & Abutalebi (2013) but requires independent evidence to adjudicate the specific mechanisms.
The Bilingual Executive-Function Advantage Controversy
A prominent research claim holds that continual management of multiple languages enhances executive functions outside the language domain. Bialystok et al. (2004) reported smaller Simon-task conflict effects in tested bilingual groups, including middle-aged and older participants, and stimulated a substantial research program. Such work offers a plausible transfer hypothesis: recurrent language selection could train monitoring, inhibition, or shifting processes used elsewhere. Yet the hypothesis concerns a general consequence of bilingual experience, whereas the existence of cross-language activation and context-dependent language control can be established within language tasks without that transfer. These propositions must be evaluated independently.
Evidence for a broad advantage became contested as independent studies tested additional executive tasks and populations. Paap & Greenberg (2013) reported no coherent bilingual advantage across multiple executive indicators in its comparisons. de Bruin et al. (2015) examined publication patterns and found that studies supporting an advantage were more likely to appear in journals than studies with results challenging it. Such asymmetries matter because the visible literature can overrepresent positive findings even when individual experiments are competently conducted. Group differences also depend on socioeconomic background, education, migration history, task reliability, and how bilingualism is defined. A single statistically significant interaction cannot establish a domain-general cognitive enhancement.
The large meta-analytic review by Lehtonen et al. (2018) synthesized 891 effect sizes from 152 adult studies and reported that small initial advantages in some executive domains were no longer supported after publication-bias correction. This conclusion is a finding about the surveyed adult literature and analytical choices, not a declaration that bilingual experience never affects any cognitive outcome. The scientific task is to replace a categorical advantage narrative with identifiable mechanisms, high-quality measurement, preregistered contrasts, and sufficiently powered designs. Language-specific adaptive control remains a productive hypothesis without being converted into an unsupported universal executive-function benefit.
Measurement Reliability, Replication, and Causal Inference
Bilingual processing experiments can be compelling yet sensitive to design details. A trial-level switching effect may depend on cue frequency, previous language, item repetition, cognate composition, response modality, and how error trials are removed. Eye-tracking inference depends on region definitions, fixation thresholds, baseline reading fluency, and treatment of skipped words. Bobb & Wodniecka (2013) illustrates why one asymmetry pattern has multiple possible explanations, while Schwartz et al. (2025) tests how a contextual cue can affect comprehension-level results. Strong studies specify competing hypotheses before examining the data and present models capable of distinguishing them.
The statistical unit requires attention. Trials are nested in speakers and crossed with words, sentences, and sometimes conversational partners. A model ignoring item variability can exaggerate the stability of a result across a population of lexical materials. Analyses should report participant recruitment, language-history instruments, exclusions, counterbalancing, missing observations, effect sizes, uncertainty intervals, and assumptions underlying mixed-effects models. Comparisons of bilingual and monolingual groups need prior consideration of sampling and equivalence on relevant nonlanguage variables. Lehtonen et al. (2018) demonstrates the importance of aggregate evidence and publication-bias assessment when interpreting an extensive but heterogeneous literature.
Causal inference requires manipulation or a defensible longitudinal strategy. Because bilingualism is an accumulated experience, random assignment to a lifetime linguistic trajectory is impossible; short training experiments may probe specific control mechanisms but do not reproduce all social and developmental pathways. Preregistration, shared materials, multi-lab replication, and transparent analytic code can make conclusions more robust. Researchers should distinguish direct mechanistic evidence within language tasks from correlations between language history and general cognition. Where only an article's abstract is accessible, the monograph attributes its reported design and result without inventing unreported randomization details, sample sizes, or statistics.
Cultural Context, Language Ideology, and Research Generalizability
Code-switching is interpreted through local conventions. Speakers may regard alternation as ordinary, prestigious, intimate, marked, or inappropriate depending on context and institutional power. Gumperz (1982) shows why conversational interpretation relies on shared contextual knowledge; Auer (1998) brings research across diverse interactional environments. A laboratory task cannot remove these social histories merely by labeling stimuli as language A and language B. Participants may perceive a mixed utterance as authentic dialogue, a school exercise, or an anomalous instruction, and the perceived genre can affect expectations as well as performance.
Cross-community comparisons should document social setting and language variety alongside psycholinguistic characteristics. A bilingual language pair with strong structural similarity may produce different types of lexical competition from a pair with different scripts, while a community in which switching is highly conventionalized supplies different pragmatic expectations from one that discourages mixing. Poplack (1980) provided community-grounded evidence of patterned alternation rather than a universal benchmark of bilingual behavior. The most defensible generalizations specify which features travel across communities and which effects are conditional on history, language, and activity.
Cultural and ethical adequacy also affects measurement. Standardized tests may privilege one dialect, one literacy history, or forms associated with high-status institutions; assigning a single proficiency score without those qualifications can misrepresent the individual. Researchers should obtain consent appropriate for naturalistic speech collection, protect potentially identifying recordings, and include multilingual experts in transcription decisions. Bilingual research has scientific value when its sampling includes varied linguistic ecologies and when the theory predicts the effects of those differences. This enlarges the validity of cognitive accounts while retaining a clear distinction between measured processing and social interpretations.
Recent Evidence and the Limits of Simple Switching Models
Research published in 2025 and 2026 increasingly targets the assumptions that made early switching models tractable. Schwartz et al. (2025) reported comprehension evidence relevant to whether global inhibition and general control influence lexical access when linguistic input includes switches. Salig et al. (2025) showed that hearing an alternation can affect attention to and memory for narrative content. Quinn et al. (2026) used synchronized reading and eye tracking to show partial monitoring even when bilingual readers produced unintended language intrusions. The value of these studies lies in differentiating processes that once appeared under a common label of switching cost: lexical access, prediction, syntactic planning, attention, and error monitoring.
Two 2026 sentence-comprehension studies further constrain interpretation. Wang et al. (2026) manipulated memory load and observed context-dependent switch outcomes in Chinese-English bilingual readers, while Chen & Christianson (2026) examined switch-related eye movements in relation to narrative consistency and unexpected word forms. Their findings should be evaluated as study-specific evidence because they used distinct stimuli, load manipulations, and outcome definitions. Taken together, they motivate experimental designs that distinguish an unexpected language membership from uncertainty about the next word and from general processing load. An observed switch cost becomes theoretically informative when its components can be separated rather than assigned to one global controller.
Contemporary work also emphasizes the difference between detection and successful correction. Quinn et al. (2026) reported that readers could direct additional visual monitoring toward error-prone switch targets without invariably preventing language intrusions. This is relevant to theories that equate noticing a conflict with exerting effective inhibition. Control may be staged across preparation, early lexical selection, speech planning, and later repair. Future replication should examine other language pairs, speech modalities, and spontaneous interactions, preferably with comparable measurement of target predictability. The underlying scientific advance is the movement from a single switch-cost outcome toward temporally resolved and context-dependent mechanistic hypotheses.
Computational Models and Multimethod Research Design
A computational theory of bilingual processing should map explicit representations, control policies, and contextual variables onto observable responses. The architecture in Dijkstra & van Heuven (2002) distinguishes lexical activation from task decisions, while Green (1998) proposes language-specific inhibitory operations and Green & Abutalebi (2013) predicts adaptation to recurring communicative settings. These are distinguishable model components rather than competing slogans. Formal implementations can estimate how lexical frequency, similarity, expected language, recent switches, and task instructions jointly affect the probability of a response and its latency. Fit to observed outcomes should be tested on held-out items, speakers, and contexts rather than on the data used to choose parameters.
Natural language interaction supplies a second layer of explanatory variables. Conversation corpora allow estimates of partner adaptation, topic transitions, and structural priming; psycholinguistic tasks provide tightly timed interventions; longitudinal sampling can assess changes after shifts in exposure. Fricke & Kootstra (2016) demonstrates a bridge between priming and conversational switching, and Blanco-Elorrieta & Pylkkänen (2017) motivates measuring the difference between laboratory and naturalistic control. A multilevel theory can connect these methods by predicting which observed conversational features alter the distribution of accessible lexical alternatives and therefore the frequency of selected switches. The causal commitments must remain explicit: fitting a dialogue model does not independently establish which neural process generated the pattern.
Measurement quality determines whether increasingly complex models add knowledge. A high-parameter architecture can fit reaction times without uniquely identifying inhibition, expectation, and response threshold. Researchers should use experimental contrasts designed to make these mechanisms predict different outcomes. Open corpus annotation, transparent handling of nested observations, standardized language histories, and replication with distinct linguistic communities increase the value of parameter comparisons. The methodological goal is an identifiable model that predicts effects across modalities while respecting genuine differences between reading, naming, conversation, and overt translation.
Human–Artificial Language Mediation and the Artificial Era
The Artificial Era introduces sustained interaction among multilingual human speakers and artificial language systems. Translation interfaces, speech recognition, multilingual assistants, and collaborative writing environments can change which languages people select, when they switch, and how their messages are interpreted. The established psycholinguistic literature supplies questions about expectations, lexical accessibility, language accommodation, and the consequences of mixed-language input. Gumperz (1982) and Auer (1998) remain relevant to human interactional meaning, while Dijkstra & van Heuven (2002) and Green & Abutalebi (2013) supply process-oriented hypotheses that can be applied cautiously to human users interacting through a computational mediator. The scientific target is measurable change in human language practices under a specified technological arrangement.
An artificial system that generates text in multiple languages does not thereby instantiate the experimentally demonstrated neural mechanisms of human bilingual control. Human switch costs, lexical competition, and language histories have definitions tied to tasks, acquisition, and observed behavior. Comparative research can create operational correspondences—language-choice consistency, response latency, interference from previous context, correction after intrusions—without assuming that identical external performance entails identical internal organization. A rigorous research design would specify the artificial architecture or model, record its prompt and interaction history, and compare matched tasks while preserving distinct explanatory vocabularies for human cognition and artificial processing.
Human–Artificial Relations adds further research questions: whether conversational partners adapt their switching patterns to a multilingual assistant, whether translations alter recall, and whether mixed-code output shapes trust or perceived competence. These are proposed empirical questions rather than established findings from the classical bilingualism literature. Longitudinal human-computer interaction studies, controlled output manipulations, and multilingual dialogue corpora can test them. The larger historical horizon places bilingual language processing within transformations of communication while maintaining the scientific identity of this dossier as a study of human bilingual comprehension, production, selection, and code-switching.
Critical Scientific Discussion and Open Problems
Several apparent contradictions in the literature dissolve once the dependent variables are separated. Non-target lexical activation is supported by comprehension research such as Spivey & Marian (1999) and Van Assche et al. (2009), but its existence does not dictate one unique mechanism of overt response selection. Large cued switch costs in Meuter & Allport (1999) coexist with lower or context-sensitive costs under voluntary choice in Gollan & Ferreira (2009). Grammatical and interactional organization documented by Poplack (1980) and Auer (1998) adds explanatory information unavailable in response-time sequences. A successful synthesis therefore explains how access, choice, monitoring, and social function interact instead of treating them as mutually exclusive descriptions.
The most important unresolved mechanistic question concerns when language-wide control influences early recognition rather than only later decisions. Dijkstra & van Heuven (2002) offers a particular separation of access and task response; Schwartz et al. (2025) reports evidence challenging a fully impermeable boundary. Related questions concern how syntactic planning shapes intrusion errors, whether switch costs follow from unexpected linguistic form, and how memory load changes cross-language processing. Quinn et al. (2026), Chen & Christianson (2026), and Wang et al. (2026) each provide specific constraints, yet their different populations and paradigms prevent simple pooling into a universal effect. Cross-method model comparison and replications are needed to identify which conclusions generalize.
The executive-function debate supplies an additional lesson in evidence evaluation. Bialystok et al. (2004) motivated a compelling transfer hypothesis, but Paap & Greenberg (2013), de Bruin et al. (2015), and Lehtonen et al. (2018) show why group differences, publication patterns, and aggregate estimates must be examined before making a broad causal claim. The field should prioritize preregistered, adequately powered tests with explicit experience measures, validated cognitive tasks, longitudinal designs, and multilingual population coverage. The objective is an explanatory science that describes what bilingual speakers actually do, distinguishes benefit from cost under specific conditions, and predicts when social context changes the apparent cognitive demand.
Conclusion
Bilingual language processing is the coordinated use of multiple learned linguistic repertoires during perception, comprehension, planning, and communication. Research supports the simultaneous availability of cross-language lexical information under many conditions, while also showing that the pathway from activation to overt language choice depends on the task, discourse, dominance, and social setting. Spivey & Marian (1999), Thierry & Wu (2007), and Van Assche et al. (2009) provide independent behavioral and neural evidence for effects of a language not overtly used. These findings establish parallel access as a central explanatory problem, although the strength and timing of that access vary with linguistic materials and experience.
Code-switching is both a structural property of multilingual discourse and a dynamically controlled response behavior. Historical corpus research, grammatical proposals, and conversational analysis show that switching can be patterned and meaningful; experiments demonstrate costs and adaptations when language choice is externally constrained. Poplack (1980), Meuter & Allport (1999), Gollan & Ferreira (2009), and Fricke & Kootstra (2016) capture complementary parts of this research object. The most informative contemporary studies probe the interface: how prediction, attention, grammatical planning, and monitoring affect a speaker's or listener's performance when languages alternate in connected speech.
The accumulated evidence favors a multidimensional account. Language experience shapes lexical representations and expectations, control processes regulate selection under particular demands, and interlocutors jointly organize the social conditions of alternation. Claims of generalized executive superiority require their own independent proof and cannot be inferred from successful multilingual communication. The continuing scientific program is to connect reliable primary observations, formal process models, ecologically representative dialogue, and longitudinal evidence in ways that yield discriminating predictions. Bilingual language processing and code-switching remain exemplary phenomena for studying how cognition operates through multiple linguistic systems in a socially structured world.
References
- Abutalebi, J., & Green, D. W. (2016). Neuroimaging of language control in bilinguals: Neural adaptation and reserve. Bilingualism: Language and Cognition, 19(4), 689–698. https://doi.org/10.1017/S1366728916000225
- Auer, P. (Ed.). (1998). Code-switching in conversation: Language, interaction and identity. Routledge. https://www.routledge.com/Code-switching-in-Conversation-LanguageInteraction-and-Identity/Auer/p/book/9780415158312
- Bialystok, E., Craik, F. I. M., Klein, R., & Viswanathan, M. (2004). Bilingualism, aging, and cognitive control: Evidence from the Simon task. Psychology and Aging, 19(2), 290–303. https://doi.org/10.1037/0882-7974.19.2.290
- Blanco-Elorrieta, E., & Pylkkänen, L. (2017). Bilingual language switching in the laboratory versus in the wild: The spatiotemporal dynamics of adaptive language control. Journal of Neuroscience, 37(37), 9022–9036. https://doi.org/10.1523/JNEUROSCI.0553-17.2017
- Blanco-Elorrieta, E., & Pylkkänen, L. (2018). Ecological validity in bilingualism research and the bilingual advantage. Trends in Cognitive Sciences, 22(12), 1117–1126. https://doi.org/10.1016/j.tics.2018.10.001
- Bobb, S. C., & Wodniecka, Z. (2013). Language switching in picture naming: What asymmetric switch costs (do not) tell us about inhibition in bilingual speech planning. Journal of Cognitive Psychology, 25(5), 568–585. https://doi.org/10.1080/20445911.2013.792822
- Byers-Heinlein, K., & Lew-Williams, C. (2013). Bilingualism in the early years: What the science says. LEARNing Landscapes, 7(1), 95–112. https://pmc.ncbi.nlm.nih.gov/articles/PMC6168212/
- Chen, D., & Christianson, K. (2026). Language-switching cost in Chinese-English bilingual reading comprehension: Evidence from eye-tracking. Journal of Psycholinguistic Research, 55, Article 132. https://doi.org/10.1007/s10936-026-10315-z
- de Bruin, A., Treccani, B., & Della Sala, S. (2015). Cognitive advantage in bilingualism: An example of publication bias? Psychological Science, 26(1), 99–107. https://doi.org/10.1177/0956797614557866
- Dijkstra, T., & van Heuven, W. J. B. (2002). The architecture of the bilingual word recognition system: From identification to decision. Bilingualism: Language and Cognition, 5(3), 175–197. https://doi.org/10.1017/S1366728902003012
- Fricke, M., & Kootstra, G. J. (2016). Primed codeswitching in spontaneous bilingual dialogue. Journal of Memory and Language, 91, 181–201. https://doi.org/10.1016/j.jml.2016.04.003
- Gollan, T. H., Montoya, R. I., Fennema-Notestine, C., & Morris, S. K. (2005). Bilingualism affects picture naming but not picture classification. Memory & Cognition, 33(7), 1220–1234. https://doi.org/10.3758/BF03193224
- Gollan, T. H., & Ferreira, V. S. (2009). Should I stay or should I switch? A cost-benefit analysis of voluntary language switching in young and aging bilinguals. Journal of Experimental Psychology: Learning, Memory, and Cognition, 35(3), 640–665. https://doi.org/10.1037/a0014981
- Green, D. W. (1998). Mental control of the bilingual lexico-semantic system. Bilingualism: Language and Cognition, 1(2), 67–81. https://doi.org/10.1017/S1366728998000133
- Green, D. W., & Abutalebi, J. (2013). Language control in bilinguals: The adaptive control hypothesis. Journal of Cognitive Psychology, 25(5), 515–530. https://doi.org/10.1080/20445911.2013.796377
- Gumperz, J. J. (1982). Conversational code switching. In Discourse strategies (pp. 59–99). Cambridge University Press. https://doi.org/10.1017/CBO9780511611834.006
- Kroll, J. F., & Stewart, E. (1994). Category interference in translation and picture naming: Evidence for asymmetric connections between bilingual memory representations. Journal of Memory and Language, 33(2), 149–174. https://doi.org/10.1006/jmla.1994.1008
- Lehtonen, M., Soveri, A., Laine, A., Järvenpää, J., de Bruin, A., & Antfolk, J. (2018). Is bilingualism associated with enhanced executive functioning in adults? A meta-analytic review. Psychological Bulletin, 144(4), 394–425. https://doi.org/10.1037/bul0000142
- Marian, V., Spivey, M., & Hirsch, J. (2003). Shared and separate systems in bilingual language processing: Converging evidence from eyetracking and brain imaging. Brain and Language, 86(1), 70–82. https://doi.org/10.1016/S0093-934X(02)00535-7
- Meuter, R. F. I., & Allport, A. (1999). Bilingual language switching in naming: Asymmetrical costs of language selection. Journal of Memory and Language, 40(1), 25–40. https://doi.org/10.1006/jmla.1998.2602
- Myers-Scotton, C. (1993). Duelling languages: Grammatical structure in codeswitching. Oxford University Press. https://doi.org/10.1093/oso/9780198240594.001.0001
- Paap, K. R., & Greenberg, Z. I. (2013). There is no coherent evidence for a bilingual advantage in executive processing. Cognitive Psychology, 66(2), 232–258. https://doi.org/10.1016/j.cogpsych.2012.12.002
- Poplack, S. (1980). Sometimes I'll start a sentence in Spanish y termino en español: Toward a typology of code-switching. Linguistics, 18(7–8), 581–618. https://doi.org/10.1515/ling.1980.18.7-8.581
- Quinn, J., Ferreira, V. S., & Gollan, T. H. (2026). Language switches can be monitored but not fully controlled: Eye-tracking evidence for syntax-driven language control. Psychonomic Bulletin & Review, 33, Article 184. https://doi.org/10.3758/s13423-026-02940-5
- Salig, L. K., Valdés Kroff, J. R., Slevc, L. R., & Novick, J. M. (2025). Hearing a code-switch increases bilinguals’ attention to and memory for information. Journal of Memory and Language, 143, Article 104647. https://doi.org/10.1016/j.jml.2025.104647
- Schwartz, A. I., Negron, J., & Scholl, C. (2025). The recruitment of global language inhibitory control and cognitive-general control mechanisms in comprehending language switches: Evidence from eye movements. Bilingualism: Language and Cognition, 28(2), 415–426. https://doi.org/10.1017/S1366728924000567
- Spivey, M. J., & Marian, V. (1999). Cross talk between native and second languages: Partial activation of an irrelevant lexicon. Psychological Science, 10(3), 281–284. https://doi.org/10.1111/1467-9280.00151
- Thierry, G., & Wu, Y. J. (2007). Brain potentials reveal unconscious translation during foreign-language comprehension. Proceedings of the National Academy of Sciences, 104(30), 12530–12535. https://doi.org/10.1073/pnas.0609927104
- Van Assche, E., Duyck, W., Hartsuiker, R. J., & Diependaele, K. (2009). Does bilingualism change native-language reading? Cognate effects in a sentence context. Psychological Science, 20(8), 923–927. https://doi.org/10.1111/j.1467-9280.2009.02389.x
- Wang, Y., Liu, X., & Sun, C. (2026). The modulating role of memory load on language switching in sentence comprehension: Evidence from eye movements. BMC Psychology, 14, Article 880. https://doi.org/10.1186/s40359-026-04871-1