Second language acquisition (SLA) is the development of an additional language through experience, instruction, interaction, and use. It involves changes in how people perceive speech, recognize words, build grammatical representations, connect expressions with meanings, comprehend messages, and produce language under real-time constraints. Learning a second language therefore means reorganizing several interacting cognitive systems, from attention and memory to prediction, motor planning, and social communication. The language may be a person's second, third, or later language; the term L2 is a research convention rather than a claim that all multilingual lives follow one sequence. Ellis (2005); Hulstijn (2005).
The central psychological question is how exposure becomes usable linguistic knowledge. Hearing a word once can support recognition without allowing reliable production; correctly completing a grammar exercise can coexist with difficulty understanding the same structure in fast conversation. A learner may speak fluidly about familiar topics while continuing to struggle with unfamiliar accents or written registers. These patterns arise because language proficiency comprises partially separable processes, knowledge systems, and performance conditions. Long-term acquisition is demonstrated by retained learning, transfer to new examples, and increasingly flexible performance, rather than by success on the exercise used during training. Ellis (2005); Hu (2026).
What Second Language Acquisition Means in Psychology
Acquisition, learning, proficiency, and performance
Acquisition refers to relatively durable changes in a learner's language representations and processing capacities. Learning is the broader process by which experience produces those changes, whether a person intentionally studies a rule or incidentally extracts a pattern while listening. Proficiency is a description of current competence across specific tasks and domains. Performance is what a learner actually does in a particular situation, under the influence of time pressure, fatigue, familiarity, anxiety, and conversational context. These distinctions matter because an immediate test can reward memorization or deliberate rule application without showing that knowledge can be accessed automatically in ordinary communication. Ellis (2005).
A second language is often distinguished from a foreign language by whether the surrounding community uses it, but psychologically the more informative variables are opportunities for meaningful exposure, interaction, feedback, and practice. Someone learning English primarily online may encounter extensive authentic discourse; someone living in an English-speaking city may have limited opportunities to use English. Heritage-language acquisition, simultaneous bilingual development, and second-language learning overlap in mechanisms while differing in histories of exposure and social identity. Research designs should describe those histories instead of treating labels as interchangeable.
The components of the developing system
Pronunciation, phonological categories, vocabulary, morphology, syntax, semantics, pragmatics, discourse, literacy, and interactional timing develop along partly distinct trajectories. Receptive knowledge usually differs from productive control; grammatical accuracy differs from fluency; comprehensibility differs from the presence or absence of a foreign accent. A comprehensive assessment therefore samples different modalities and communicative conditions. Growth in one component may facilitate another: improved word recognition frees resources for sentence interpretation, and greater familiarity with grammatical constructions makes prediction and speaking more efficient. In'nami, Hijikata, and Koizumi (2022).
The Cognitive Architecture of Learning an Additional Language
Attention and selection from continuous input
Everyday language contains more information than a novice can process. Learners must identify task-relevant contrasts, segment speech, recognize recurring words, connect expressions to situations, and filter competing interpretations. Attention influences which aspects of an utterance become available for encoding; existing knowledge influences what attracts attention in the first place. The same English sentence may be heard as an undifferentiated stream by a beginner and as a set of familiar constructions by an advanced learner. This difference is a consequence of prior learning, not merely of effort.
A learner can understand the overall meaning of a sentence while overlooking a low-salience grammatical ending. Instruction may redirect attention to a form–meaning relationship, but attention alone does not guarantee acquisition. The representation must be encountered, used, and retrieved under sufficiently varied conditions. The extent to which conscious noticing is necessary for all forms of language learning remains debated. Hulstijn (2005); N. C. Ellis (2005).
Statistical learning, frequency, and distribution
Language environments provide patterned information. Sounds co-occur in predictable sequences; particular verbs favor certain constructions; words appear in recurring semantic and social contexts. Through repeated experience, learners become sensitive to such distributions. Frequency matters because repeated encounters create opportunities to strengthen mappings, while diversity matters because meeting a form in different contexts can help identify what remains stable across instances. Neither raw frequency nor a single exposure statistic explains every learning outcome: salience, meaning, similarity to existing knowledge, spacing, and task demands affect how patterns are encoded.
This is a foundation for usage-based and distributional explanations. The learner gradually extracts generalizations from examples, and those generalizations can become productive beyond the instances previously encountered. Researchers disagree about the degree to which abstract rules, construction-specific representations, and predictive computations best describe the resulting knowledge. These theories can make overlapping predictions, so studies must vary input and test transfer to unfamiliar sentences. N. C. Ellis (2005); Oikawa and Uchihara (2026).
Working memory, long-term memory, and retrieval
Working memory temporarily maintains and manipulates language information during comprehension and production. Learning a new sentence can require holding an unfamiliar word form while interpreting its context, selecting a meaning, and planning a response. As familiar elements become easier to recognize, the learner can devote more processing capacity to the rest of the message. Working-memory measures relate to some language outcomes, but those correlations do not establish a fixed maximum proficiency. A meta-analysis of L2 reading found an association around r = .30, with measurement choices moderating the relationship. In'nami, Hijikata, and Koizumi (2022).
Long-term learning entails changes in the accessibility and organization of sound patterns, word forms, meanings, constructions, and procedures. Retrieval strengthens access to information under conditions that resemble future use; spaced encounters counter the rapid decline that often follows massed study. A meta-analysis of 48 experiments involving 3,411 participants found advantages for spaced practice, particularly for delayed retention when spacing was longer. The optimal interval depends on the desired retention horizon and material, and equal and expanding schedules performed comparably in that synthesis. Kim and Webb (2022).
Automaticity, procedural control, and cognitive load
Novice performance is often deliberate: a speaker selects a word, checks a rule, then constructs an utterance. Practice can make common linguistic operations faster and more coordinated, reducing the need to consciously supervise each step. This helps explain why someone may know a rule during an untimed exercise yet fail to implement it while speaking. Automaticity is specific to practiced representations and contexts; rapid production of rehearsed phrases does not automatically confer equally flexible competence with novel structures.
Skill-acquisition perspectives emphasize repeated, goal-directed performance and feedback. Usage-based perspectives emphasize the accumulation and abstraction of experience, while interactionist perspectives emphasize opportunities to interpret and negotiate meaning. Each explains different aspects of the developing system. A productive research question asks which mechanism predicts a particular learning outcome under specified conditions, rather than declaring a single account sufficient for all components. Norris and Ortega (2000); Mackey (2020).
How Input Becomes Linguistic Knowledge
Comprehensible input and accessible form–meaning connections
Meaningful input supplies the evidence from which linguistic generalizations can be learned. A message needs to be sufficiently accessible that a learner can map some of its expressions to events, referents, or intentions. Visual context, familiarity with the topic, slower presentation, and knowledge of related words may all support interpretation. Understanding a message, however, does not imply that every sound, morpheme, or syntactic pattern in it has been encoded. Input-based explanations are strongest when they specify the information available to the learner and how exposure changes later processing.
For example, the learner might hear a new expression while a speaker points to a familiar object. This provides a plausible referent, but durable knowledge requires distinguishing that expression from similar alternatives and using it beyond the original scene. Multiple encounters with varied speakers and contexts can improve generalization. A 2026 study of incidental vocabulary learning found that contextual diversity benefited delayed meaning recognition, while the pattern also depended on learners' prior vocabulary knowledge; more variation is therefore not uniformly helpful at every proficiency level. Oikawa and Uchihara (2026).
Interaction as contingent evidence
Conversation permits learners to request clarification, check a hypothesis, receive a reformulation, or revise an utterance. Its cognitive value is partly its contingency: the response depends on what the learner has just understood or attempted to say. Meaning negotiation can direct attention to a missing word or grammatical distinction while preserving a communicative goal. Interaction also supplies temporal and social information about how language works in actual exchanges. Researchers investigate these effects with task-based comparisons, recorded interaction, and delayed posttests. Mackey (2020).
Interaction quality cannot be reduced to the number of conversational turns. A learner might exchange many short, predictable messages without processing new forms. Another might participate in a demanding task that creates repeated opportunities to understand a contrast and reformulate language. Proficiency, interlocutor support, topic knowledge, and power relations all shape the experience. Studies of interactive learning therefore need to separate amount of exposure from the features of the interaction that may have caused the observed gain.
Production, hypothesis testing, and noticing gaps
Speaking and writing require the learner to select and sequence linguistic forms rather than simply recognize them. Trying to express a meaning can expose an incomplete representation: a person knows what they intend to say but discovers that they cannot retrieve the required tense, article, pronunciation, or connective. Searching for the form, asking for help, and reformulating the utterance create occasions for learning. Production is also a way of testing whether a hypothesized construction communicates the intended meaning.
The strongest test of a production activity is whether it improves subsequent comprehension or flexible production beyond the practiced examples. Repeating a correct sentence immediately after a model shows uptake, but retained learning requires delayed measurement. Output practice should therefore be analyzed in relation to input, feedback, and transfer rather than treated as an isolated ingredient. Mackey (2020); Lyster and Saito (2010).
Implicit Learning and Explicit Knowledge
Two distinctions that researchers must keep separate
Implicit learning describes change that may occur without deliberate rule discovery; explicit learning involves conscious attention, hypothesis formation, or instruction. Implicit knowledge refers to linguistic competence that can be deployed with limited introspective access; explicit knowledge refers to information a learner can consciously report or analyze. The process used to learn and the kind of knowledge measured afterward are related but separate research questions. A lesson designed explicitly can contribute to later fluent processing, and incidental experience can coexist with conscious awareness of some regularities. Hulstijn (2005); N. C. Ellis (2005).
A timed grammaticality judgment is often used as a proxy for relatively automatic knowledge, whereas an untimed judgment or an explanation task is often used to probe explicit knowledge. These measures remain imperfect. Timing can alter strategy, grammaticality judgments may be affected by familiarity, and even speech produced under time pressure can recruit explicit knowledge. Rod Ellis developed a test battery that made these operational questions central. A 2026 critical examination likewise underscores the continuing interpretive difficulty of mapping tasks to a clean implicit–explicit division. R. Ellis (2005); Tamura and Fukuta (2026).
Instructional consequences of the distinction
Explicit explanation can help a learner notice otherwise overlooked contrasts and form a useful initial hypothesis. Meaningful exposure and practice can then support more rapid access to the relevant patterns. The effectiveness of explicit instruction is well documented for specified learning targets, but evidence is less decisive about universal long-term superiority across all language abilities and testing formats. The classic Norris and Ortega meta-analysis found substantial benefits of focused instruction and a relative advantage for explicit treatments in its sampled literature, while warning that design and measurement heterogeneity constrain generalization. Norris and Ortega (2000).
There is no single recipe that translates metalinguistic explanation directly into automatic competence. Learners may need many varied opportunities to recognize and use a form in meaning-bearing contexts. The relationship between explicit and implicit systems is an active theoretical debate: some proposals permit transfer or interaction between them, while others assign a more constrained role to conscious knowledge. Current evidence supports studying that interaction at the level of particular tasks, processing conditions, and language structures. N. C. Ellis (2005); Tamura and Fukuta (2026).
Vocabulary Acquisition and Lexical Organization
From an encounter to a usable word
Knowing a word is a multilayered achievement. It includes recognizing its sound or written form, connecting it with meanings, identifying its grammatical behavior, understanding common collocations and register, and retrieving it efficiently during production. Initial recognition can be fragile: the learner may identify a new word on a multiple-choice test yet be unable to use it in a conversation. Vocabulary develops through repeated form–meaning encounters, attention to contextual constraints, retrieval, and differentiation from similar words.
Incidental vocabulary learning is especially important because reading, listening, and interaction expose learners to words that are not the explicit targets of a lesson. Context can support a plausible inference but can also encourage guessing that is insufficiently precise. Multiple contextual examples help establish which meaning features belong to the word itself and which belong to a particular sentence. The 2026 contextual-diversity study illustrates why learner proficiency must be considered when varying examples. Oikawa and Uchihara (2026).
Cross-language connections and lexical competition
A learner's existing language knowledge affects the acquisition of additional word forms. Cognates may be easier to learn because related words share sound, spelling, or meaning; false cognates may invite a misleading association. Cross-language similarity is consequently a resource and a source of interference, depending on the actual structure and task. An experimental 2026 reading study found immediate learning for cognates, noncognates, and false cognates, with the strongest learning for cognates and the weakest for false cognates; contextual information helped, while an additional awareness intervention did not produce the expected enhancement. Silva and colleagues (2026).
As lexical knowledge expands, learners build associations among words in the new language and across languages. Recognition becomes sensitive to frequency, semantic neighborhoods, and competing candidates. The resulting lexicon is dynamic: using an L2 word can activate related concepts and sometimes similar forms in another known language. Such cross-language activation is an important explanation for processing differences, while the broader question of bilingual language control belongs to its own specialized research intent.
Speech Perception, Pronunciation, and Phonological Learning
Learning to hear distinctions that matter
An L2 learner brings perceptual expectations shaped by prior language experience. Acoustic differences that are informative in one language may not map neatly onto the phonological categories of another. Learning to perceive unfamiliar contrasts involves weighting acoustic cues, recognizing the variability introduced by speakers and contexts, and linking sound categories to lexical meaning. A useful test asks whether the learner can generalize beyond the voices, words, and recordings used in training.
Pronunciation is also a sensorimotor learning problem. A speaker coordinates articulatory plans, auditory feedback, rhythm, and stress while trying to communicate. Improvements in segmental accuracy can develop differently from improvements in comprehensibility or fluency. A persistent accent alone is not evidence of inadequate linguistic knowledge. Instruction that offers perceptually varied examples and feedback may improve particular targets, but the strength and transfer of these effects depend on what was practiced and measured. Saito (2021).
Orthography can shape auditory learning
Reading and listening do not remain independent when learners acquire a language through print. Written forms can help identify lexical segments and stabilize unfamiliar words, yet the spelling system can also encourage inaccurate expectations about sound. The consequences depend on the relation between orthography and phonology in the languages involved, when written forms are introduced, and the learner's prior literacy experience. A study published in the 2026 volume of Studies in Second Language Acquisition examined how written input influences the encoding of L2 phonological contrasts among Arabic-speaking learners of English, illustrating an active empirical question rather than a single universal benefit of text. Studies in Second Language Acquisition, 2026, volume 48, issue 2.
Grammar, Constructions, and Real-Time Comprehension
From stored expressions to productive patterns
Grammar learning requires more than memorizing labels for word classes or reciting rules. Learners must map forms to functions, build representations of word order and morphology, track agreement and reference, and interpret unfamiliar combinations. Some constructions are acquired as frequent multiword sequences before their more abstract properties become fully productive. Competing theories differ in how they explain the relation between stored examples and abstract grammatical representations, but all must account for generalization and characteristic learner errors.
The familiar distinction between comprehension and production matters here. A learner can use context to interpret an unfamiliar tense while still being unable to produce it fluently. Conversely, a memorized pattern may be produced correctly without flexible understanding of its constraints. Researchers therefore include comprehension tasks, elicited production, novel-item transfer, and delayed retention measures rather than equating any one score with grammatical competence. R. Ellis (2005).
Parsing under time pressure
Sentence comprehension depends on incremental decisions. The learner must identify boundaries, maintain earlier constituents, integrate newly encountered words, and revise interpretations when subsequent input makes an initial analysis implausible. Slower lexical access can increase processing demands even when the grammatical rule is known. A person may consequently succeed on an untimed written test but misinterpret an ordinary sentence delivered at conversational speed. Working-memory demands and lexical familiarity are important candidate explanations, although task demands must be assessed independently. In'nami, Hijikata, and Koizumi (2022).
Different theories assign different roles to expectations about upcoming language. Surprisal-based approaches describe costs associated with the probability of linguistic input, while memory-based approaches emphasize maintaining and retrieving dependencies. These models are not interchangeable, and their comparative explanatory power depends on experimental predictions. SLA research can test whether growing proficiency changes reliance on semantic cues, syntactic information, and prior expectations without assuming that adult L2 comprehension simply reproduces first-language patterns.
Feedback and the Correction of Errors
Recasts, prompts, and explicit information
Corrective feedback provides information after a learner produces a form that departs from the target pattern. A recast offers a reformulated version; a prompt invites the learner to supply the alternative; metalinguistic feedback directs attention to a rule or contrast. Each may affect the learner's next utterance, but immediate repair and sustained acquisition are different outcomes. Feedback can direct attention to low-salience features, encourage retrieval, and help distinguish competing hypotheses. Lyster and Saito (2010); Brown (2021).
A classroom meta-analysis by Lyster and Saito found beneficial and enduring average effects of oral corrective feedback, with larger effects for prompts than recasts in the studied conditions. That result concerns particular classroom studies and tasks; it does not establish that one type of feedback is preferable for every language feature, age group, or communicative goal. A later synthesis highlighted the potential moderation by grammatical complexity. Comparisons must also separate the frequency of feedback from how explicitly the learner is told what was wrong. Lyster and Saito (2010); Granena and Yilmaz (2021).
Timing, durability, and transfer
The best time to correct a learner depends on the relevant task and on what the outcome measure captures. In one experimental study, immediate feedback supported greater development of an English past-tense target than delayed feedback among seventh-grade learners under the specific intervention tested. Other designs yield different patterns because feedback timing interacts with attention, task continuity, and learner knowledge. The scientific inference is that timing is a manipulable condition with measurable effects, not that all errors should be interrupted at once. Fu and Li (2022); R. Ellis (2021).
Useful evaluation distinguishes recognition of the corrected form, spontaneous self-repair, delayed accuracy, and application to new structures. Some errors disappear during a carefully rehearsed task and return when attention must be divided. For these reasons, feedback studies should report the linguistic targets, baseline proficiency, tasks, comparison groups, follow-up interval, and missing data, rather than treating a short-term increase in accuracy as comprehensive acquisition.
Individual Differences and Learning Trajectories
Language aptitude and its components
Language aptitude is a family of capacities associated with success on particular language-learning tasks, including sensitivity to sound patterns, analytic learning, and aspects of memory. It is not a single intelligence score or an immutable designation of who can learn languages. A meta-analysis reported a moderate association between measured aptitude and L2 grammar learning, approximately r = .31 across its combined evidence, with variation across age and instructional conditions. Another synthesis found that aptitude measures overlap with some cognitive abilities but are distinguishable from motivation and general intelligence. S. Li (2015); S. Li (2016).
Measurement is crucial because conventional tests often emphasize explicit or analytic learning. A task that rewards noticing a rule need not measure the capacity to acquire patterns incidentally under continuous exposure. Relationships may also change as proficiency develops, because novices and experienced users face different bottlenecks. A 2026 study on aptitude and proficiency directly investigates this moderation and cautions against assuming a fixed predictor effect throughout learning. Pavlekovic and Roehr-Brackin (2026).
Motivation, emotion, and opportunities to learn
Motivation can influence the activities people choose, the persistence of their engagement, and the amount and quality of language input they encounter. Enjoyment and anxiety may shape participation, especially in speaking tasks that expose learners to social evaluation. Their measured associations with achievement do not, by themselves, establish a causal pathway from a questionnaire score to linguistic competence: proficiency can change confidence, and classroom conditions can influence both affect and learning. Research is strongest when it examines these relationships longitudinally and compares competing explanations.
A longitudinal study of Japanese learners of English followed speaking development alongside motivation, enjoyment, and anxiety across approximately eighteen months, offering evidence that affective and linguistic trajectories need not move together in simple ways. The practical implication is to support sustained, meaningful participation and measure the resulting language development, rather than interpreting momentary confidence as a direct proxy for knowledge. Saito and colleagues (2025).
Exposure histories, inequality, and social context
Learners differ in schooling, literacy opportunities, language distance, time available, interlocutor networks, economic resources, identity, and access to feedback. These experiences can influence observed proficiency independently of cognitive aptitude. A person using the target language daily for work and relationships faces a different distribution of linguistic demands from a student who encounters it mostly in examinations. Research that treats years of study as interchangeable with hours or quality of experience risks attributing contextual differences to stable individual characteristics.
Social acceptance also shapes whether learners attempt unfamiliar forms, seek clarification, and continue participating after misunderstandings. These circumstances are legitimate parts of language acquisition even when a laboratory mechanism is being studied. A robust model describes both the cognitive change within an individual and the distribution of experiences that make that change possible.
Age, Sensitive Periods, and Development
Starting age is related to eventual attainment in many naturalistic samples, particularly after prolonged exposure. The causal interpretation remains contested because age of onset covaries with schooling, input quantity, immigration history, identity, continued language use, and development of cognitive strategies. Large-scale analyses can estimate patterns of attainment but cannot by themselves randomize age or isolate a unique neurobiological transition. Hartshorne, Tenenbaum, and Pinker analyzed responses from hundreds of thousands of English speakers and modeled an age-related change in grammar-learning ability; the interpretation of such population-level patterns remains an active theoretical dispute. Hartshorne, Tenenbaum, and Pinker (2018).
Age influences several domains differently. Younger learners may ultimately develop particular phonological or grammatical patterns through extended exposure, while older learners can show advantages in the initial stages of deliberate vocabulary and rule learning. The proposition that all aspects of second-language learning close at one exact birthday is too strong for the available evidence. Furthermore, evidence from immigrant immersion populations cannot be transferred automatically to school programs with much less exposure. The critical-period hypothesis requires its own separate analysis of measures, sampling, and alternatives; here the relevant conclusion is that age interacts with the learning system rather than replacing it as an explanation. Hartshorne, Tenenbaum, and Pinker (2018); Vanhove (2013).
Learning Conditions: Classroom, Immersion, and Digital Practice
Structured instruction and naturally occurring exposure
Classrooms can make rare forms frequent, highlight subtle contrasts, and provide systematic feedback. Naturalistic environments can make communication consequential and expose learners to varied speakers, registers, and pragmatic demands. Their relative usefulness depends on what must be learned and how the environment is structured. An hour of conversation dominated by predictable routines is different from an hour of challenging exchange with opportunities for clarification; an explanation of a grammar rule is different from varied practice using that rule for a communicative purpose.
Focused instruction has demonstrable benefits for selected linguistic targets, but studies differ in treatment duration, instructional intensity, comparison groups, and assessment measures. A credible account avoids describing classroom learning as a wholly separate cognitive system. Both classroom and out-of-class experience supply input and activity that can change attention, memory, representations, and procedural control. Norris and Ortega (2000); Brown (2021).
Retrieval, spacing, and varied practice
A practical learning cycle alternates meaningful comprehension with opportunities to retrieve and use recently encountered forms. Spacing practice across days allows repeated access after some forgetting and often improves delayed retention relative to concentrating equivalent exposure in a single session. The interval should match the learning goal: a next-day classroom quiz and a requirement to remember vocabulary months later represent different targets. Research does not support one universal spacing algorithm for all languages or contents. Kim and Webb (2022).
Variation can also support transfer when learners are ready to identify the invariant pattern beneath surface changes. Changing speakers, genres, tasks, and contexts challenges reliance on one memorized example. Beginners may need enough stable support to establish initial form–meaning correspondences before exposure becomes highly variable. The delayed vocabulary-recognition findings of Oikawa and Uchihara illustrate how variation interacts with prior knowledge. The most informative performance check uses new items and a delayed assessment, not merely repeated success with familiar prompts. Oikawa and Uchihara (2026).
Sleep and consolidation
Learning continues to involve memory processes after a practice session. A systematic review and meta-analysis of novel-word learning studies in adults found an overall advantage of sleep relative to wakefulness across recall, recognition, and lexical-integration outcomes. This literature includes experimental word-learning paradigms; its results are relevant to consolidation but do not establish that sleep alone produces full second-language competence. The evidence supports treating rest and the timing of practice as conditions that can influence memory, alongside exposure, retrieval, and use. Schimke and colleagues (2021).
Artificial Intelligence Tools and Second Language Learning
The learning mechanism is in the learner
AI-supported tools may provide opportunities for explanation, practice, feedback, and interaction outside formal lessons. A generative chatbot can produce example sentences, simulate a conversation, explain vocabulary in different contexts, and offer immediate prompts for revision. These affordances can change the quantity and structure of practice. A learner's gains depend on how those opportunities are used and measured, including whether the system supplies accurate information and whether the tasks require independent retrieval.
A 2025 meta-analysis of 41 experimental or quasi-experimental studies reported a positive average association between generative-chatbot interventions and L2 outcomes; heterogeneity in comparison conditions, intervention lengths, targeted skills, and educational contexts constrains broad causal generalization. Another systematic review of 30 empirical studies identified opportunities in speaking and writing alongside limitations in nuanced interaction. Research on artificial-language systems should separate educational efficacy from claims about what those systems internally understand. Producing convincing language provides no direct test of human-type comprehension or subjective experience. Li, Wang, and Yang (2025); Wiboolyasarin and colleagues (2025).
Designing informative assessments of AI-assisted study
The main causal question is whether access to a tool produces durable learning relative to a credible alternative. A study should distinguish receiving a fluent answer from independently producing or understanding the target language later. Appropriate comparisons might hold practice time constant, assess whether feedback is accurate, and use delayed tests completed without the assistant. Automated ratings require validation against the linguistic construct being measured. Interactions with a system also raise practical concerns about data privacy and unreliable corrections, especially when highly fluent output masks errors.
The promise of adaptation is a hypothesis that requires testing. Personalization based on a learner's performance may help direct attention to useful targets, yet an algorithm can also overfit to repeated tasks or reward superficial fluency. Long-term outcomes across different populations and languages are more informative than short interventions evaluated with the same item formats used during training. Li, Wang, and Yang (2025); Bao and colleagues (2025).
How Scientists Study Second Language Acquisition
Experimental, longitudinal, and observational evidence
Experimental designs manipulate training or exposure conditions while measuring outcomes under controlled tasks. Random assignment strengthens causal interpretation when implementation and follow-up are adequate. Longitudinal studies trace individual change and can reveal nonlinear trajectories, but observed correlations between motivation, age, exposure, and achievement require careful control of alternative explanations. Naturalistic corpora offer ecological coverage of actual language use, while laboratory tasks help isolate candidate mechanisms. Each approach is informative for different questions.
Converging measures are especially valuable. Self-paced reading can reveal timing differences; eye tracking can identify attention and anticipation during processing; lexical decisions and priming tasks probe access; elicited imitation and production tasks sample linguistic representations; comprehension tests assess communicative interpretation. None is a transparent window onto a single mental mechanism. Researchers must evaluate task validity, reliability, baseline proficiency, and practice effects before interpreting group differences. R. Ellis (2005); In'nami, Hijikata, and Koizumi (2022).
Transfer, delayed outcomes, and methodological quality
A short-term intervention can improve performance through familiarity with the test format. Stronger studies measure retention and transfer to different speakers, topics, or grammatical constructions. An experimental effect should be interpreted relative to the comparison condition: outperforming a no-practice group differs from outperforming an equally intensive alternative. Sampling, missing observations, researcher degrees of freedom, and selective publication further affect the credibility of results.
Meta-analysis can synthesize evidence but does not remove the limitations of included studies. Hu's 2026 second-order review examined the reporting and methodological practices of 224 L2 meta-analyses and identified areas requiring improvement. This is a reminder to assess how studies were selected, whether effect sizes were dependent, how moderators were specified, and how publication bias was handled before treating an average effect as a universal law. Hu (2026).
Distinguishing evidence levels
Established: additional-language learning changes cognitive representations through experience; exposure and practice matter; focused instruction can improve targeted outcomes; spaced practice is useful for delayed retention in the conditions studied. Supported but context-sensitive: corrective feedback, input variation, aptitude, and working memory show relationships with outcomes that depend on tasks and participants. Preliminary: the relative advantages of particular generative-AI implementations across long-term skills and learner populations. Contested: a unique implicit–explicit interface theory, a single age boundary for all language domains, and the reduction of all acquisition to one computational mechanism. Philosophical questions about artificial understanding must be evaluated on their own grounds rather than inferred from chatbot performance. Norris and Ortega (2000); Kim and Webb (2022); Hu (2026); Tamura and Fukuta (2026).
Practical Implications for Learning and Teaching
A scientifically grounded learning program establishes an understandable communicative goal, supplies repeated access to meaningful examples, draws attention to difficult distinctions, creates opportunities to retrieve and produce language, and tests retention outside the original lesson. It observes progress across comprehension and production instead of assuming that an isolated vocabulary or grammar score represents general proficiency. For novice learners, accessible input and structured support reduce avoidable processing demands. As knowledge grows, more varied materials and less predictable tasks support flexible use.
Teachers can prioritize feedback according to its relevance to the task and the learner's present knowledge. Learners can use retrieval and spaced review while continuing to listen, read, speak, and write in meaningful contexts. A productive study plan tracks skills separately: understanding speech from new speakers, recalling words without cues, producing grammatical forms in conversation, following an extended explanation, and making oneself understood. The value of any technique is established by improvement on the target ability after a meaningful delay, preferably with unfamiliar items. Kim and Webb (2022); Lyster and Saito (2010).
Progress is also shaped by the continuity of language use. Periods of low access or changed communicative demands can alter how readily particular forms are retrieved. Fluctuation on a difficult day does not necessarily indicate that the underlying representation has vanished. Long-term plans therefore distinguish maintaining access to established knowledge from introducing entirely new content. When a program claims that a method is uniquely efficient, the relevant questions are what was learned, compared with which alternative, on which delayed assessment, and by whom.
Clinical and Developmental Boundaries
Differences in accent, grammatical accuracy, fluency, or time needed to learn an additional language occur across typical learners. They are not, by themselves, diagnostic indicators. A developmental language disorder, a reading disorder, hearing impairment, or an acquired language impairment requires its own developmental history and appropriate professional assessment. Tests administered in an unfamiliar language can underestimate an individual's abilities; interpretation must consider language exposure, educational opportunity, and proficiency in each relevant language.
SLA research can inform how language experience affects performance, but it does not replace clinical differential diagnosis or prescribe treatment for dyslexia, aphasia, stuttering, or other conditions. Research on bilingual identification, language disorders, and specialist interventions belongs to appropriately qualified clinical and educational domains. The scientific task of this article is to explain acquisition mechanisms and the conditions under which additional-language knowledge develops.
Frequently Asked Questions
Can adults acquire a second language successfully?
Yes. Adults can acquire vocabulary, grammar, comprehension, and communicative skills through sustained experience and instruction. Age is related to some patterns of eventual attainment, especially in naturalistic populations, while starting age, learning intensity, language use, and linguistic domain interact. A statistical age trend does not determine any individual's ceiling. Hartshorne, Tenenbaum, and Pinker (2018).
Does immersion automatically make someone fluent?
Immersion changes access to input and opportunities for use; its benefits depend on whether interactions are comprehensible, varied, and demanding enough to produce learning. Repeated contact with the same narrow set of routines can support those routines without producing balanced proficiency across domains. Structured practice and feedback may complement immersion.
Is grammar study necessary, or can exposure do everything?
Meaningful exposure is central, and focused instruction can improve targeted learning. Explicit explanations are useful for some forms and learners, while fluent access requires substantial experience with language in context. The relative contributions are best assessed for a specific skill and learning environment rather than presented as a universal either-or choice. Norris and Ortega (2000); Tamura and Fukuta (2026).
Why can I read a language better than I can speak it?
Reading and speaking place different demands on recognition, lexical retrieval, planning, and timing. During reading, information can often be revisited; spontaneous speech requires constructing and articulating a message as it unfolds. Progress in one modality may precede progress in another, particularly when practice opportunities differ.
What makes vocabulary last in memory?
Repeated meaningful encounters, retrieval, and spacing support durable access. Research on spaced practice favors distributing learning across time, with interval choices informed by the desired retention period. Connecting a word to varied contexts can improve generalization, though the amount of variation should match prior knowledge. Kim and Webb (2022); Oikawa and Uchihara (2026).
Can AI conversation partners replace human interaction?
AI systems can provide additional practice and some forms of feedback. Current outcome studies support potential benefits under particular intervention conditions, while evidence about long-term transfer, nuanced social interaction, and comparative effectiveness remains developing. Learners should evaluate independent performance, the accuracy of corrections, and opportunities for authentic communication. Li, Wang, and Yang (2025).
Conclusion
Second language acquisition is a cumulative transformation of how a person perceives, represents, retrieves, interprets, and uses language. Exposure provides structured evidence; attention selects what can be encoded; statistical learning and experience strengthen patterns; working and long-term memory support access; communication and feedback reveal gaps; and practice builds more flexible performance. The relative contribution of each mechanism varies with the language feature, task, learner, and social environment. The scientific standard for a successful explanation is a precise account of what changes, under which conditions, and how that change survives time and generalizes to new use.
Related Articles
Cluster Language, Thought and Psycholinguistics — The scientific parent chapter situates acquisition within the complete architecture of human language processing.
Word Recognition: How the Mind Identifies Spoken and Written Words — How learned sound and spelling patterns become accessible lexical candidates.
Sentence Processing: How We Understand Grammar in Real Time — How linguistic representations support incremental comprehension under processing demands.
References
Bao, W., Wang, T., Zhang, L., Yusop, F. D., and Ruan, X. (2025). A systematic review of AI in second language acquisition using the expanded SAMR model (2015–2024). Discover Computing. https://doi.org/10.1007/s10791-025-09833-6.
Brown, D. (2021). Meta-analysis and research synthesis. In The Cambridge Handbook of Corrective Feedback in Second Language Learning and Teaching. Cambridge University Press. https://doi.org/10.1017/9781108589789.009.
Ellis, N. C. (2005). At the interface: Dynamic interactions of explicit and implicit language knowledge. Studies in Second Language Acquisition, 27(2), 305–352. https://doi.org/10.1017/S027226310505014X.
Ellis, R. (2005). Measuring implicit and explicit knowledge of a second language: A psychometric study. Studies in Second Language Acquisition, 27(2), 141–172. https://doi.org/10.1017/S0272263105050096.
Ellis, R. (2021). Explicit and implicit oral corrective feedback. In The Cambridge Handbook of Corrective Feedback in Second Language Learning and Teaching. Cambridge University Press. https://doi.org/10.1017/9781108589789.017.
Fu, M., and Li, S. (2022). The effects of immediate and delayed corrective feedback on L2 development. Studies in Second Language Acquisition, 44(1), 2–34. https://doi.org/10.1017/S0272263120000388.
Granena, G., and Yilmaz, Y. (2021). Corrective feedback and grammatical complexity: A research synthesis. In The Cambridge Handbook of Corrective Feedback in Second Language Learning and Teaching. Cambridge University Press. https://doi.org/10.1017/9781108589789.036.
Hartshorne, J. K., Tenenbaum, J. B., and Pinker, S. (2018). A critical period for second language acquisition: Evidence from 2/3 million English speakers. Cognition, 177, 263–277. https://doi.org/10.1016/j.cognition.2018.04.007.
Hu, Y. (2026). Evaluating methodological and reporting practices of L2 meta-analyses: A second-order review. Studies in Second Language Acquisition, 48(3), 599–628. https://doi.org/10.1017/S0272263126101764.
Hulstijn, J. H. (2005). Theoretical and empirical issues in the study of implicit and explicit second-language learning: Introduction. Studies in Second Language Acquisition, 27(2), 129–140. https://www.cambridge.org/core/journals/studies-in-second-language-acquisition/article/theoretical-and-empirical-issues-in-the-study-of-implicit-and-explicit-secondlanguage-learning-introduction/4F499EC1AE0886A01AB226B47BDA2F85.
In'nami, Y., Hijikata, Y., and Koizumi, R. (2022). Working memory capacity and L2 reading: A meta-analysis. Studies in Second Language Acquisition, 44(2), 381–406. https://doi.org/10.1017/S0272263121000267.
Kim, S. K., and Webb, S. (2022). The effects of spaced practice on second language learning: A meta-analysis. Language Learning, 72(1), 269–319. https://doi.org/10.1111/lang.12479.
Li, M., Wang, Y., and Yang, X. (2025). Can generative AI chatbots promote second language acquisition? A meta-analysis. Journal of Computer Assisted Learning, 41(4), e70060. https://doi.org/10.1111/jcal.70060.
Li, S. (2015). The associations between language aptitude and second language grammar acquisition: A meta-analytic review of five decades of research. Applied Linguistics, 36(3), 385–408. https://doi.org/10.1093/applin/amu054.
Li, S. (2016). The construct validity of language aptitude: A meta-analysis. Studies in Second Language Acquisition, 38(4), 801–842. https://doi.org/10.1017/S027226311500042X.
Lyster, R., and Saito, K. (2010). Oral feedback in classroom SLA: A meta-analysis. Studies in Second Language Acquisition, 32(2), 265–302. https://www.cambridge.org/core/journals/studies-in-second-language-acquisition/article/abs/oral-feedback-in-classroom-sla/4999EE1C8379B2BF026B148EAF373CA1.
Mackey, A. (2020). Doing meta-analytic and synthetic research on interaction, feedback, tasks, and L2 learning. In Interaction, Feedback and Task Research in Second Language Learning. Cambridge University Press. https://doi.org/10.1017/9781108589284.007.
Norris, J. M., and Ortega, L. (2000). Effectiveness of L2 instruction: A research synthesis and quantitative meta-analysis. Language Learning, 50(3), 417–528. https://doi.org/10.1111/0023-8333.00136.
Oikawa, G., and Uchihara, T. (2026). The effects of contextual diversity on incidental vocabulary learning. Studies in Second Language Acquisition, 48(3), 757–783. https://doi.org/10.1017/S0272263126101818.
Pavlekovic, R., and Roehr-Brackin, K. (2026). Finding the tipping point: Proficiency as a moderator of aptitude in L2 learning. Studies in Second Language Acquisition, First View. https://www.cambridge.org/core/journals/studies-in-second-language-acquisition/article/finding-the-tipping-point/633E6ABE9A50FBF3789500ABFACB64E9.
Saito, K. (2021). Effects of corrective feedback on second language pronunciation development. In The Cambridge Handbook of Corrective Feedback in Second Language Learning and Teaching. Cambridge University Press. https://www.cambridge.org/core/books/abs/cambridge-handbook-of-corrective-feedback-in-second-language-learning-and-teaching/effects-of-corrective-feedback-on-second-language-pronunciation-development/ADBECD931EB45C0A2B57B3A6A53F0E78.
Saito, K., Dewaele, J.-M., In'nami, Y., and Abe, M. (2025). Disentangling the causal role of motivation, enjoyment, and anxiety in second language speech learning: A final report. Studies in Second Language Acquisition, 47(2). https://www.cambridge.org/core/journals/studies-in-second-language-acquisition/issue/1EB9BDFE27D402F8FAA290B7D67B831C.
Schimke, E. A. E., Angwin, A. J., Cheng, B. B. Y., and Copland, D. A. (2021). The effect of sleep on novel word learning in healthy adults: A systematic review and meta-analysis. Psychonomic Bulletin and Review, 28(6), 1811–1838. https://doi.org/10.3758/s13423-021-01980-3.
Silva, B., Foryś-Nogala, M., Kutyłowska, K., Szewczyk, J., and Otwinowska, A. (2026). Learning cognates, false cognates, and noncognates contextually through reading: Do contextual support and cross-linguistic awareness matter? Studies in Second Language Acquisition, 48(3), 808–832. https://www.cambridge.org/core/journals/studies-in-second-language-acquisition/article/learning-cognates-false-cognates-and-noncognates-contextually-through-reading-do-contextual-support-and-crosslinguistic-awareness-matter/B5305384F8C944F38A1B132134AA176E.
Tamura, Y., and Fukuta, J. (2026). A critical examination of the relationship between explicit and implicit knowledge and language teaching. Studies in Second Language Acquisition, First View. https://doi.org/10.1017/S0272263126101946.
Vanhove, J. (2013). The critical period hypothesis in second language acquisition: A statistical critique and a reanalysis. PLOS ONE. https://pmc.ncbi.nlm.nih.gov/articles/PMC3723803/.
Wiboolyasarin, W., Wiboolyasarin, K., Tiranant, P., Jinowat, N., and Boonyakitanont, P. (2025). AI-driven chatbots in second language education: A systematic review of their efficacy and pedagogical implications. Ampersand, 14, 100224. https://doi.org/10.1016/j.amper.2025.100224.