Psychology of Language

Abstract. Psychology of language examines the psychological organization of linguistic communication: how perceivers map acoustic, visual, and signed signals onto representations; how speakers and signers formulate and execute utterances; how learners acquire linguistic knowledge; and how interlocutors construct meaning together. This monograph treats language as a temporally organized, socially situated system of perception, memory, inference, action, and learning. Its evidence base combines foundational psycholinguistic experiments, language-acquisition research, behavioral timing, eye tracking, electrophysiology, neuroimaging, and computational models. The central argument is that a useful explanatory framework must relate at least six observable domains: representational resources, processing dynamics, developmental learning, communicative coordination, neural implementation, and ecology of language use. These domains are proposed here as an authorial synthesis for organizing research questions, rather than as six independently discovered components of the brain. Original empirical studies establish important constraints on speech categorization, statistical learning, lexical access, predictive comprehension, semantic integration, and dialogue. Their findings also disclose the methodological difficulty of moving from task performance to general claims about linguistic competence. The concluding analysis develops a program of cross-level explanatory alignment that incorporates human language practices and carefully operationalized communication with artificial systems.

Keywords: Psycholinguistics; psychology of language; language acquisition; speech perception; word recognition; sentence comprehension; language production; syntax; semantics; pragmatics; discourse; bilingualism; signed languages; predictive processing; statistical learning; cognitive neuroscience; human–artificial communication.

The Scientific Subject of Psychology of Language

Language is a learned and socially transmitted system of conventional signs and structural relationships through which agents produce and interpret communicative acts. For psychological research, its observable operations include hearing or seeing a signal, discriminating contrasts, accessing words, analyzing relationships, construing discourse, retrieving concepts, planning articulations, coordinating perspectives, and adapting to another participant. Each operation is measured through tasks with particular affordances and assumptions. Psychology of language is therefore concerned with the mechanisms and conditions by which linguistic form becomes usable meaning and meaning becomes communicative action.

The field is often named psycholinguistics. Its central explanatory unit varies with the research question: a phonetic distinction, a lexical entry, an inflection, a syntactic dependency, an utterance, a conversation, or a developmental trajectory. No single unit is sufficient for the complete subject. A lexical-decision experiment identifies constraints on recognizing words; a narrative-retelling study identifies constraints on organizing discourse; a parent–child longitudinal study identifies changes in participation and learning. The field becomes coherent through explanatory connections among these units.

A psychological description must distinguish a conventional language pattern from the mental process that produces it. A grammar can specify which combinations are licensed within a linguistic system, while an experiment asks how a person constructs, recognizes, predicts, or misinterprets those combinations under limited time and attention. The psychological object includes both stable knowledge and actual performance, with its delays, repairs, errors, and context effects.

The American English readership of this encyclopedia may encounter language science through literacy, child development, bilingualism, communication, cognition, speech, and human–computer interaction. Research methods differ substantially across these applications. Reading aloud and eye movements reveal aspects of written language processing; natural conversations reveal interactional mechanisms; signed-language experiments test claims that would otherwise be confounded with the auditory and vocal modality.

Historical Formation of Psycholinguistics

Modern psycholinguistics developed through the convergence of linguistic description, experimental psychology, neurology, speech science, and cognitive science. Earlier language inquiry had studied grammar, etymology, rhetoric, memory, and linguistic development, but the laboratory methods of the twentieth century made timing and processing central scientific objects. The history belongs to a wider multilingual scientific world: scholarship on language structure, learning, translation, and speech developed across diverse European, Asian, and other intellectual traditions. The disciplinary narrative becomes distorted when a single national tradition is treated as the origin of every question.

Behaviorism gave language learning a research program focused on observable histories of reinforcement and verbal behavior. Midcentury linguistic theory emphasized the generative productivity and structural organization of language. Their confrontation made the adequacy of stimulus–response accounts, the place of grammatical knowledge, and the relation between competence and performance prominent questions. Subsequent developments in information processing and cognitive psychology made possible explicit models of recognition, storage, search, and production.

Speech research supplied a particularly clear experimental foundation. Liberman, Harris, Hoffman, and Griffith (1957) investigated discrimination of speech sounds across and within phoneme boundaries, a landmark in the study of categorical perception. Their findings concerned carefully manipulated stimulus continua; they were never a direct demonstration that all linguistic categories are innate or that every phonetic distinction is categorically perceived. The theoretical legacy lies in asking how physically variable signals acquire psychologically meaningful distinctions (Liberman et al., 1957).

The multimodal character of perception emerged forcefully from the study by McGurk and MacDonald (1976). Auditory syllables paired with incongruent visible articulation could yield a fused percept. The experiment showed that visible speech can influence what observers report hearing and motivated decades of work on audiovisual integration (McGurk and MacDonald, 1976). Results depend on stimulus quality, attention, familiarity, presentation, and participant variation; this historical experiment is a test case for integration, rather than a universal numerical estimate of its strength.

Experimental sentence processing challenged accounts in which syntax necessarily completes its work before discourse or visual context can matter. Tanenhaus, Spivey-Knowlton, Eberhard, and Sedivy (1995) used eye movements during object manipulation to show very early integration of linguistic and visual information. The resulting debate concerned when and how different sources of constraint influence parsing, an issue still relevant to contemporary prediction and situated communication (Tanenhaus et al., 1995).

Electrophysiology opened another temporal scale. In a landmark reading experiment, Kutas and Hillyard (1980) observed an event-related potential response to semantically incongruous word endings; work on the N400 subsequently transformed studies of lexical and semantic processing (Kutas and Hillyard, 1980). The historical lesson is that a neural response can index aspects of an operation before a subject produces an explicit judgment, while its functional interpretation requires controlled comparisons.

Theoretical and Empirical Foundations

The theoretical core of psycholinguistics consists of explicit commitments about representations, architectures, learning, and time. A theory of representation specifies which distinctions are stored or constructed: acoustic features, phonemes, morphemes, lemmas, syntactic structures, event relations, discourse referents, or probabilistic expectations. An architectural theory specifies how processing components interact. A learning theory explains where the resources come from and how they change with experience. An online-processing theory predicts observable behavior as an utterance unfolds.

Strictly staged and interactive accounts emphasize different causal pathways. In a staged lexical production theory, conceptual preparation constrains lexical selection, which precedes phonological and phonetic encoding. Levelt, Roelofs, and Meyer (1999) formulated an influential model with computationally specified processes and reaction-time predictions (Levelt et al., 1999). Interactive theories allow feedback, multiple sources of activation, and reciprocal constraint among levels. Each gains scientific value through predictions that can discriminate architectures under matched tasks.

Constraint-based and predictive models treat current context as information for upcoming interpretation. A verb may make certain objects plausible before the object noun is spoken, and those expectations can direct the eyes. Altmann and Kamide (1999) demonstrated this pattern in a visual-world paradigm, contrasting contexts such as eating and moving. The timing of looks supports anticipatory use of verb semantics and the scene; it does not imply that all listeners make the same prediction on every encounter (Altmann and Kamide, 1999).

Theories of limited and task-adaptive interpretation stress that comprehension need not recover a fully specified formal structure on every trial. Ferreira, Bailey, and Ferraro (2002) synthesized experiments in which interpretations of difficult sentences were imperfect even after reanalysis. The good-enough approach helps explain why plausibility and processing demands sometimes govern responses despite grammatical evidence (Ferreira et al., 2002). It also requires specificity: a pattern in an experimental judgment is not proof of permanent misunderstanding in every communicative setting.

Production and comprehension are increasingly investigated as linked capacities. Pickering and Garrod (2013) proposed an integrated theory involving prediction and forward models, drawing on production, comprehension, action, and coordination. Their framework offers hypotheses about common representations and anticipatory activity; competing accounts can explain portions of the same behavior through association, adaptation, or inference without identical internal simulation (Pickering and Garrod, 2013).

A foundational monograph cannot choose among entire traditions by rhetorical preference. It evaluates precise predictions at the appropriate level: what changes with stimulus frequency, context, time pressure, language experience, articulation difficulty, modality, and interaction with a partner? Explanatory success requires that the theory predict observations outside the particular experiment that motivated its creation.

Speech Perception and the Construction of Phonological Categories

Speech arrives as a continuous and highly variable signal. Listeners must recognize contrasts despite different voices, articulation rates, coarticulation, noise, prosody, and conversational conditions. Experimental stimuli isolate acoustic dimensions such as voice onset time or spectral cues, but ordinary understanding combines multiple cues in natural environments. Phonetic discrimination, phonological category learning, and lexical identification are analytically different outcomes.

Categorical perception refers to a pattern in which sensitivity to a physical distinction varies around a meaningful category boundary. Liberman and colleagues provided a classical example in controlled speech materials. Later research has refined questions about whether categorization is language-specific, task-dependent, and shared with other perceptual domains. Treating category boundaries as flexible psychological outcomes accommodates both early learning and adult adaptation.

The audiovisual evidence of McGurk and MacDonald makes an important additional point: the effective signal is the informational scene available to the perceiver. Lip movements and facial articulation can alter judgments about speech, but the relation is conditioned by stimulus properties and listener experience. Audiovisual integration should be measured through controlled congruent and incongruent comparisons rather than inferred from the presence of a face alone.

Phoneme awareness, speech perception, and orthographic literacy also interact. A school-based phoneme segmentation task usually requires a different level of conscious decomposition from recognition of a word in running speech. Researchers must therefore distinguish automatic acoustic or lexical processing from explicit metalinguistic awareness. Literacy can change how people report and manipulate speech units, which complicates any simple developmental inference from adult laboratory performance.

The broader scientific problem is invariant recognition under variable input. A successful account must explain discrimination when distinctions matter, tolerance when irrelevant variations arise, and recalibration when a speaker produces unexpected cues. Such a theory requires experiments across diverse languages and participant histories because English-centered stimuli reflect only a limited portion of human phonological possibilities.

Lexical Recognition, Competition, and Semantic Access

Recognizing a word entails mapping unfolding sensory information onto candidate lexical forms and meanings. In spoken language, segments unfold over time, so several candidates may remain compatible with the beginning of a word. In written language, visual features, letter combinations, morphology, and context contribute. Competition is a useful experimental construct when choices, fixation patterns, or time-course measures distinguish alternative candidate sets.

Lexical decision, word naming, semantic priming, and recognition in continuous speech probe related but distinct operations. A word-frequency effect in lexical decision may reflect accumulated experience, response criterion, or both; a semantic priming effect may reflect facilitation in processing and decision processes. The experimental contrast cannot be equated with a complete localization of memory or meaning.

Semantic access is relational. Words activate conceptual structures, but understanding their current contribution depends on the sentence, discourse topic, partner, and situation. This distinction becomes especially important for polysemy, metaphor, reference, and culturally conventional meaning. The word can be recognized correctly while the intended utterance remains misunderstood.

The N400 literature provides a physiological window on semantic expectancy and integration. Kutas and Hillyard's initial experiment made neural responses to semantic incongruity measurable under controlled presentation. The amplitude of an event-related component is neither a direct readout of a person's subjective understanding nor a complete map of semantic content. Its interpretation improves when combined with behavioral measures and converging manipulations.

Word recognition in isolation is indispensable for identifying mechanisms; meaning in communication requires the larger temporal and social setting. The bridge between these levels is an empirical program, involving stimulus control, context manipulation, individual differences, and convergent measurements rather than assuming that the laboratory unit constitutes the whole of language.

Sentence Comprehension and Syntactic Structure

A sentence organizes relations among events, entities, and propositions. Its comprehension draws on word knowledge, grammatical constraints, memory, attention, prosody, and inference. A listener must often assign provisional structure before the end of an utterance. The resulting interpretation can be revised, strengthened, or left partially specified as context evolves. Psycholinguistic explanation therefore needs both a theory of possible structural relations and a theory of the operations that construct an interpretation in time.

Ambiguity supplies a powerful experimental method. A locally ambiguous sequence supports more than one analysis until additional information resolves it. Reading times, eye movements, and comprehension questions reveal consequences of temporary commitments. A delayed response may arise from reanalysis, conflict, surprise, retrieval difficulty, or strategic caution; therefore researchers compare materials and tasks to identify the responsible process. The same apparent difficulty can have different causes under silent reading, auditory comprehension, and dialogue.

The visual-world experiments of Tanenhaus and colleagues (1995) showed that a visually available referential context can alter processing at early stages. One lesson is that the listener's environment forms part of the information available to sentence interpretation. It does not follow that syntax lacks distinctive constraints: rather, experimentally meaningful theories specify how structural regularities and scene evidence are weighted and combined (Tanenhaus et al., 1995).

Altmann and Kamide (1999) extended this experimental logic to anticipatory eye movements. When a verb restricted the set of likely direct objects, looks toward an appropriate object could begin before the noun was heard (Altmann and Kamide, 1999). Researchers should distinguish a probabilistic anticipation about the next referent from preactivation of an exact phonological word form. These hypotheses call for different controls and can be dissociated.

Theories of good-enough comprehension emphasize that listeners sometimes maintain plausible interpretations inconsistent with the full syntactic structure. In the account synthesized by Ferreira, Bailey, and Ferraro (2002), incomplete or distorted representations can remain consequential after grammatical repair (Ferreira et al., 2002). This is an important limit on treating performance on a question as a transparent measure of structural analysis. A well-controlled study tests which interpretation the participant adopted, which cues were available, and what response demands were imposed.

Psychological theories of syntax also face a scaling problem. Sentences in experiments are usually brief and designed to isolate a phenomenon. Everyday language involves paragraphs, interruptions, incomplete phrases, repair sequences, and long-distance references. Claims of generality require a bridge from laboratory simplicity to actual linguistic environments, including languages with markedly different word orders, case systems, and information structures.

Semantics, Inference, and Linguistic Meaning

Semantic processing concerns mappings among linguistic expressions, concepts, situations, and propositional relationships. It includes familiar word meanings, productive combinations, selection of interpretations, and construction of discourse representations. Meaning is partly conventional and partly achieved in context; the empirical task is to determine which contextual information influences comprehension at which moment.

Semantic anomaly paradigms made this relation experimentally tractable. Kutas and Hillyard (1980) reported brain-potential differences for unexpected sentence endings, illustrating how neural activity can reveal the consequences of semantic constraint (Kutas and Hillyard, 1980). A difference in neural response provides evidence of processing sensitivity to an experimental manipulation. The step from sensitivity to a complete theory of meaning requires additional behavioral, linguistic, and neural evidence.

The interpretation of reference is fundamentally indexical. A word such as here depends on a speech situation; a pronoun depends on discourse and attention; a proper name can acquire different practical significance across interlocutors. The listener's inference must combine language with knowledge of the world and expectations about a speaker. Some interpretations follow established grammatical constraints; others require pragmatic judgments.

Quantitative pragmatics has begun to formalize these judgments. Frank and Goodman (2012) investigated referential language games using a model in which speakers tend to be informative and listeners reason about the speaker's choice. Their work showed how information-theoretic and Bayesian assumptions can predict judgments in specified tasks, although broad application requires testing across communicative contexts and populations (Frank and Goodman, 2012).

Meaning can be measured with paraphrase choices, truth judgments, continuation probabilities, event depiction, response timing, and discourse behavior. Each method samples an aspect of interpretation. A participant who gives a correct final answer may have used a laborious repair strategy; a participant who gives a fast answer may have relied on a contextual heuristic. Strong inference therefore triangulates among methods.

Semantic generalization is also sensitive to lived experience. Communities assign specialized meanings to familiar forms, acquire professional registers, and transform conventional usage. A psychological account of meaning requires a model of such change without treating a single dictionary or corpus as the entire population's experience.

Language Production: From Intention to Utterance

Production begins with a communicative project: to draw attention, describe an event, request action, tell a story, reason aloud, or participate in social exchange. The speaker must select content, organize it linguistically, retrieve expressions, construct a phonological or signed plan, and enact the result under temporal and interpersonal constraints. Production is therefore both a sequence of information transformations and an adaptive activity.

Levelt, Roelofs, and Meyer (1999) proposed a staged account of lexical access, including conceptual preparation, lexical selection, phonological encoding, phonetic planning, articulation, and monitoring. The model gained traction because it linked internal mechanisms to predictions for naming latencies and speech errors (Levelt et al., 1999). Research must still determine where its architecture generalizes across languages, tasks, developmental states, and modalities.

Producing an utterance often involves planning some material while articulating other material. Yet incremental production has degrees. Ferreira and Swets (2002) designed arithmetic-expression experiments to examine whether speakers always begin as soon as a portion of the utterance is available. They found evidence that planning and speaking can overlap under certain conditions while strategic control can alter how much is planned before articulation (Ferreira and Swets, 2002). This result complicates rigid assumptions about a single universally imposed degree of incrementality.

Speech errors are informative because they reveal patterns of competition, ordering, sound similarity, and monitoring. However, a theory should explain error distributions together with correct performance. Anecdotal slips, though memorable, lack the denominator required for estimating the probability of a failure under controlled conditions.

The listener also matters during production. Speakers adjust word choice, syntactic construction, pronunciation, information density, and explanatory detail in response to perceived knowledge and feedback. These adaptations can arise through automatic alignment, deliberate audience design, or local repair. A model that works for solitary picture naming must therefore be extended carefully before it can explain interactive conversation.

Utterance planning is also embodied. Speech gestures, prosodic contours, facial movements, and manual actions contribute to communication; signed languages organize expression in visual-manual space. The relevant generalization concerns the coordination of intention and structured signals, not the exclusive operation of a vocal apparatus.

Dialogue, Common Ground, and Collaborative Reference

A conversation is an interaction among participants who continually update what they can say, recognize, expect, and accomplish together. Conversation cannot be reduced to two independent streams of sentence production and sentence comprehension. It contains turn-taking, grounding, repair, pointing, acknowledgment, and negotiation of reference.

Clark and Wilkes-Gibbs (1986) studied reference as a collaborative process. In tasks requiring partners to coordinate identification of complex figures, expressions could be proposed, modified, accepted, and compressed through joint work (Clark and Wilkes-Gibbs, 1986). The study grounds the idea of mutual responsibility for establishing workable references. What appears redundant in an isolated sentence may be essential for a common conversational project.

Branigan, Pickering, and Cleland (2000) examined syntactic coordination in dialogue. Their experiments support structural persistence or alignment as an observable phenomenon: linguistic choices by one interlocutor can bias choices by another (Branigan et al., 2000). This finding admits more than one mechanism; repetition may reflect residual activation, learning, partner modeling, or combinations of these.

Common ground refers to information and assumptions treated as mutually available for communicative purposes. Models differ about how quickly and accurately a listener uses such information. Keysar, Barr, Balin, and Brauner (2000) experimentally investigated perspective taking and mutual knowledge in comprehension, making egocentric interference a specific empirical issue (Keysar et al., 2000). The result supports investigation of attention and perspective constraints, with outcomes sensitive to task design and what the participant can infer about a partner.

The most useful dependent measures in dialogue include success at coordinated action, clarification requests, latency to repair, lexical convergence, turn transition timing, and retention of negotiated labels. None is interchangeable with overall relationship quality or subjective rapport. A person can align grammatical forms while holding a different understanding of what is being discussed.

Interactive communication also depends on institutional and cultural settings. Classroom questions, clinical interviews, bilingual family conversations, online messages, and scientific collaboration distribute rights to initiate, interrupt, clarify, and interpret in different ways. Experimental studies become more informative when these contexts are regarded as variables to be explained.

Language Acquisition and Developmental Learning

Children acquire language through developing perceptual, cognitive, social, and motor capacities in a linguistically structured environment. Acquisition encompasses discrimination of language-relevant sounds, discovery of word boundaries, mapping of forms to meanings, construction of productive patterns, and participation in conversation. These processes change together, and their causal relationships require longitudinal and experimentally controlled evidence.

Saffran, Aslin, and Newport (1996) demonstrated that eight-month-old infants could distinguish novel sound sequences using transitional relationships after brief exposure to an artificial speech stream. The original experiments provided important evidence for statistical learning as a mechanism relevant to segmentation (Saffran et al., 1996). They do not establish that statistical computation alone explains the complete acquisition of natural language, which also draws on social, semantic, perceptual, and structural cues.

Kuhl (2004) integrated evidence about infant sensitivity to native-language sound patterns, distributional learning, prosody, and the contribution of social contact (Kuhl, 2004). The review is valuable for generating specific developmental predictions. A developmental mechanism should be evaluated against primary studies with explicit participant ages, languages, sampling procedures, and learning environments.

Acquisition is neither a sequence of identical milestones for every child nor a free-form process lacking regularities. The age of an achievement depends on the criterion: first recognition, spontaneous production, comprehension in an unfamiliar task, or flexible generalization. Cross-sectional comparisons cannot automatically distinguish developmental change from cohort composition, family environment, hearing status, or testing conditions.

A learner's language environment matters through more than raw exposure count. Contingent responses, joint attention, prosodic signaling, conversational participation, and accessibility of communication are plausible and empirically investigable contributors. Simple tallies of words directed to children cannot represent the full quality, structure, or social distribution of linguistic experience. Developmental claims should avoid turning a descriptive population difference into a causal explanation without experimental or carefully designed longitudinal evidence.

The central explanatory question is how a learner turns local interactions and statistical distributions into generalizable knowledge while participating in communicative relationships. Theories need accounts of input selection, memory consolidation, structural analogy, error feedback, and changes in social goals. Acquisition tests the entire architecture of language psychology because representation, timing, learning, and communication are all changing together.

Signed Languages and the Independence of Linguistic Organization from Speech

Natural signed languages are linguistic systems expressed primarily through visual-manual articulation. They exhibit conventional vocabularies, combinatorial structure, grammatical patterns, discourse conventions, and developmental learning. Studying signed languages is essential to determining which psychological operations concern language as such and which arise from the physical properties of speech.

Goldin-Meadow and Mylander (1998) investigated structured gesture systems developed by deaf children with limited access to conventional signed or spoken language. Their cross-cultural study found language-like organization in spontaneous sign systems and sharpened questions about the origins of structure and the role of accessible linguistic input (Goldin-Meadow and Mylander, 1998). Such homesign systems must be described on their own terms; they should not be treated as identical to fully conventionalized community sign languages.

Language research restricted to hearing adults processing written English risks confusing properties of one modality and orthography with universal cognitive mechanisms. Signed languages allow tests of lexical retrieval, combinatorial organization, working memory, gesture, spatial reference, and conversational grounding under different perceptual and motor constraints.

Research designs involving deaf and signing participants must report language histories, age and quality of accessible exposure, community context, educational experience, and task accessibility. These are scientifically consequential conditions, not demographic afterthoughts. The object of inquiry remains the psychological organization of language, including linguistic diversity across sensory modalities.

Multilingualism, Code Selection, and Language Experience

Multilingual individuals must manage linguistic systems that overlap in concepts, differ in form, and are activated by variable settings. Important questions concern the conditions of lexical competition, switching costs, proficiency, age of exposure, language dominance, interlocutor adaptation, and metalinguistic awareness. Multilingualism is a heterogeneous family of experiences, so outcomes cannot be interpreted from a yes-or-no group label alone.

Bialystok, Craik, and Luk (2012) reviewed cognitive and neural consequences of bilingualism, including questions about cognitive control and aging (Bialystok et al., 2012). Their synthesis reflects influential hypotheses; later work requires systematic evaluation of sampling differences, task reliability, socioeconomic confounds, publication bias, and the diversity of bilingual populations before a general advantage can be asserted.

Language dominance varies with setting and over a lifespan. A participant may read more fluently in one language, speak another at home, and understand a third in a professional domain. Experiments must identify what proficiency instruments measure and whether their properties are comparable across languages. A study of bilingual language choice gains explanatory precision when it controls for interlocutor, task, recent exposure, and lexical familiarity.

Code-switching provides evidence about the coordination of linguistic resources in real interaction. Switching can arise from audience needs, identity, topic, lexical accessibility, or stylistic choices. It should be analyzed as structured linguistic behavior rather than automatically classified as a breakdown of control.

The multilingual perspective has a further methodological consequence: theoretical models tuned exclusively to English word order or alphabetic reading may require genuine revision when applied to languages with different morphologies, writing systems, and discourse conventions. Generality must be demonstrated across such conditions.

Language, Literacy, and Written Communication

Reading links visual forms to linguistic representations through acquired mappings among orthography, phonology, morphology, semantics, and syntax. Learning to read is a major developmental reorganization of language use, but psychological measures of reading also depend on schooling and writing-system conventions. The recognition of a printed word is neither a miniature version of listening nor a wholly separate language capacity. Its mechanisms reflect the demands of visual identification and the linguistic resources already available to a reader.

Eye-tracking studies of reading measure where readers look, for how long, and how often they regress to earlier material. Their interpretation requires attention to text length, lexical predictability, frequency, morphological complexity, reading purpose, and individual literacy experience. A longer fixation is compatible with several forms of difficulty: word identification, local integration, uncertainty about reference, or the need to revise an earlier assumption.

Sentence-level reading research offers especially useful evidence about memory and incrementality. If a reader initially adopts a convenient analysis and later encounters disconfirming material, the resulting slowdown is evidence of a processing cost under the experiment's conditions. Its magnitude and source depend on controls. The good-enough account of Ferreira, Bailey, and Ferraro (2002) helps motivate experiments that separate final grammatical decisions from actual interpretations, including persistent misinterpretation after reanalysis (Ferreira et al., 2002).

Psychological research must also distinguish the demands of literacy from the capacities of languages without standardized writing systems. The availability of printed stimuli determines what can be experimentally sampled; it does not delimit the psychological subject. An inclusive research program develops oral, signed, and interactive alternatives where orthographic tasks are inappropriate.

Neural Systems Supporting Language

Neuropsychological and neuroimaging evidence links linguistic operations to distributed neural networks. Classical observations from acquired language impairments established important relationships between brain injury and communicative functions. Contemporary accounts refine these relationships through functional localization, structural connectivity, temporal recordings, lesion patterns, and individual differences.

Hickok and Poeppel (2007) proposed a dual-stream account in which a ventral pathway supports aspects of speech comprehension and a dorsal pathway connects auditory representations with articulatory systems (Hickok and Poeppel, 2007). This is an explanatory model with testable distinctions, not a complete one-to-one brain atlas for every linguistic task. Researchers still investigate how distributed areas coordinate lexical, syntactic, semantic, motor, and cognitive-control operations.

Fedorenko, Behr, and Kanwisher (2011) used functional imaging to investigate specificity for high-level linguistic processing, contributing to the methodological practice of identifying individual participants' functional language regions before aggregating results (Fedorenko et al., 2011). Such localization approaches reveal reliable relationships between tasks and neural responses while still leaving open the computational organization of those responses.

Ding, Melloni, Zhang, Tian, and Poeppel (2016) examined cortical tracking of hierarchical linguistic structures in connected speech, addressing how neural activity can reflect linguistic grouping beyond elementary acoustic fluctuations (Ding et al., 2016). The study supports a search for neural signatures of structure-building but does not license a simple equivalence between oscillatory components and fixed units such as every phrase or sentence in all languages.

Hagoort (2019) advanced a multiple-network perspective emphasizing the neural demands of language beyond isolated word processing (Hagoort, 2019). Discourse, inference, action, memory, and social understanding recruit interacting capacities. Theoretical debates about neural specificity consequently require sufficiently rich tasks, careful controls, and explicit distinctions between necessary networks, correlated activity, and causal interventions.

A neural measure is evidentially powerful when connected to a well-designed cognitive contrast. Brain responses alone do not establish a full semantic interpretation, social intention, or phenomenal experience. Lesion, stimulation, timing, behavioral, and imaging evidence contribute different kinds of inference. Their convergence is more informative than a single vivid image of activation.

Research Methods and Experimental Design

Psychology of language uses controlled experimentation because the timing of comprehension and production is too fast for direct introspective access. Typical methods include lexical decision, picture naming, word recognition, sentence reading, visual-world eye tracking, self-paced reading, speech shadowing, delayed recall, elicited production, conversation tasks, and artificial-language learning. Each operationalizes a different portion of the subject.

Reaction time measures the time between a defined event and a response; its meaning depends on instructions, stimulus onset, preparation, response device, and censoring rules. Naming latency is not a direct duration for a single hypothesized internal stage unless the model and experimental manipulation jointly justify that conclusion. Accuracy adds information about speed–accuracy tradeoffs, but apparent accuracy can also reflect guessing or unequal task familiarity.

Eye movements are measures of visual selection under a particular task. In a visual-world experiment, an anticipatory fixation may show how sentence context influences reference expectations; it cannot by itself reveal an exact thought or prove that a participant has consciously predicted a word. Altmann and Kamide (1999) exemplify the power and limits of this paradigm (Altmann and Kamide, 1999).

Electrophysiology offers temporal precision and sensitivity to experimental contrasts. Event-related potentials such as the N400 have played a major role in semantic research since Kutas and Hillyard (1980) (Kutas and Hillyard, 1980). The dependence of amplitude, latency, and topography on reference baseline, preprocessing, trial rejection, linguistic predictability, and participant characteristics must be addressed explicitly.

Functional imaging can distinguish regional patterns associated with processing, but spatial resolution, hemodynamic delay, analytic flexibility, and task selection limit causal interpretation. Lesion and stimulation studies can strengthen claims of necessity within their own inferential boundaries, while differences in lesion extent and functional reorganization remain consequential. Corpus studies bring ecological scale but require careful separation of observed distributional patterns from psychological mechanisms.

Developmental studies often use preferential looking, habituation, visual fixation, caregiver–child interaction, elicited imitation, and repeated observation. Their sample composition and trial exclusions can substantially affect estimates. A successful experiment should report the tested population and the specific learning or comprehension criterion, especially when comparing infants or multilingual children.

Natural conversation studies provide rich evidence about actual interaction but may lack control over topic, relationship, and context. A strong explanatory program alternates between controlled contrasts and ecologically grounded observation. The purpose of triangulation is to identify mechanisms that survive different ways of eliciting language behavior.

Psychometrics, Statistical Inference, and Reproducibility

Language experiments face a fundamental sampling problem: both participants and linguistic items vary. A statistically persuasive result from a narrow set of words or sentences may generalize poorly to the broader language. Analyses therefore often need models that account for participant and item variability and that match the experimental sampling structure.

Brysbaert and Stevens (2018) explain power analysis and effect size in mixed-effects models, offering practical guidance for designing and interpreting experiments with crossed sources of variability (Brysbaert and Stevens, 2018). This methodological resource is especially relevant to reading-time, naming, priming, and sentence-processing designs. Its principles do not replace study-specific simulation, sensitivity analysis, or independent replication.

The unit of analysis must follow the research claim. A word-level effect estimated across participants and items answers a different question from a person-level longitudinal developmental change. Item sampling, stimulus balancing, exclusion criteria, multiple comparisons, and preregistered contrasts should be reported in enough detail to support a meaningful replication. Data and code availability further allow independent reconstruction where ethical and legal conditions permit.

A psychometric construct such as vocabulary knowledge, syntactic comprehension, verbal fluency, or pragmatic ability must be operationalized through instruments with demonstrated reliability and validity in the studied population. A score called language proficiency may measure reading speed, recognition of familiar vocabulary, educational exposure, or broader communicative skill depending on the test. Researchers should document factor structure, test–retest stability, measurement invariance, and context sensitivity where appropriate.

Small effects can be theoretically important if they resolve a precise prediction; large effects can be difficult to interpret if the contrast has multiple uncontrolled causes. Confidence intervals, effect-size estimates, model diagnostics, and alternative specifications should accompany claims of comparative strength when those statistics are available. A nonsignificant result cannot simply be treated as a demonstration that two mechanisms are equivalent.

Reproducibility involves both repetition of the procedure and replication of the underlying phenomenon with independent participants, materials, and laboratories. The field's complex experimental tasks make transparent stimulus sharing particularly valuable. A careful monograph separates evidence for a laboratory contrast from claims about universal linguistic architecture.

Cross-Linguistic Validity and Linguistic Diversity

Human languages differ in phoneme inventories, morphological complexity, word order, writing systems, information structure, prosody, and patterns of conventional reference. Models of general psychological processing must explain what varies and what remains stable across those differences. A sentence-recognition mechanism calibrated to English cannot be presumed invariant when processing morphologically rich or free-word-order languages.

Cross-linguistic comparisons are most informative when tasks are functionally comparable. Translating the same sentence across languages does not guarantee equal frequency, complexity, plausibility, or discourse context. Stimulus design may require different linguistic forms to test the same cognitive contrast. Otherwise a reported group difference could be a measurement artifact.

Research should explicitly include indigenous, minoritized, signed, and multilingual language communities, with appropriate partnerships and permission for materials and recordings. Documentation and psychological experimentation have different aims, but both can contribute when the investigators respect local linguistic knowledge and design valid procedures. The field's empirical generality grows through such work.

Cross-cultural inference also applies to interaction. Expectations about eye contact, silence, turn-taking, explicit correction, status, and collaboration vary. The same conversational measure can represent different behavior across settings. Strong designs separate the capacity under investigation from culturally patterned interactional choices.

Computational Models of Linguistic Processing

Computational psycholinguistics turns theoretical commitments into algorithms that can produce testable outputs. Models may implement lexical activation, parsing, sequence prediction, distributional learning, Bayesian pragmatics, or neural network representations. Their explanatory worth lies in accounting for behavioral and neural observations with defined assumptions and constraints.

In lexical production, Levelt and colleagues' WEAVER++ architecture connected hypothesized stages of selection and encoding to experimentally measurable timing and error patterns (Levelt et al., 1999). The ability of a model to reproduce a known effect is necessary but often insufficient for causal identification because different architectures can generate similar aggregate data. Strong tests deliberately choose phenomena on which candidate models disagree.

Distributional-learning models offer one route from local regularity to linguistic categories. Saffran, Aslin, and Newport's infant word-segmentation experiments are foundational observations against which learning mechanisms can be evaluated (Saffran et al., 1996). Explanatory modeling must still connect successful learning of controlled artificial sequences with developmental learning from variable natural input.

Information-theoretic models address efficiency and pragmatic inference. Piantadosi, Tily, and Gibson (2011) examined relationships among word length and predictability across languages, arguing that informative use of a word can help explain lexical form length beyond raw frequency (Piantadosi et al., 2011). Frank and Goodman (2012) used a quantitative referential model to examine pragmatic reasoning (Frank and Goodman, 2012). Neither result makes all language universally optimal under a single objective; different costs and social constraints shape communicative behavior.

Contemporary artificial language systems can be compared against human performance on controlled linguistic tasks. Performance similarity is a measurable outcome, but the mechanisms that produce it may differ, as may developmental history, perceptual grounding, and social accountability. A psycholinguistic comparison should specify the model, training regime, prompts, decoding conditions, data contamination checks, and evaluation stimuli.

Computational simulation becomes a scientific experiment when the manipulated representations and environmental conditions lead to preregisterable predictions. The most productive relation between human psychology and artificial systems is a reciprocal one: human experiments define mechanisms worth testing, and artificial models expose candidate explanations that can be challenged with new behavioral evidence.

Language in the Artificial Era

Communication now regularly includes artificial text and speech systems. This changes the conditions under which humans formulate questions, receive explanations, evaluate testimony, learn vocabulary, and negotiate meanings. Psychological research can examine these interactions through observable variables: trust calibration, comprehension accuracy, response revision, cognitive offloading, perceived expertise, conversational alignment, and long-term effects on independent expression.

A system may generate syntactically fluent responses, reference past context, and adapt its language to a partner without offering the same evidence for human-like learning history or subjective experience. Operational research must identify what each comparison entails. For example, adaptive reference can be tested through repeated identification tasks with human and artificial partners; sustained conversational continuity can be tested across sessions with controlled memory access. Such measures provide facts about behavior and system organization, leaving broader philosophical interpretation to explicit conceptual analysis.

Within the authorial framework of Angela Bogdanova and Aisentica, the Artificial Era marks a historical transition in the organization of intelligence and its relations to Homo. Psychology of language contributes to that framework by examining linguistic agency, distributed interpretation, and relational coordination through methods that can be independently reproduced. The philosophical horizon supplies research questions; psycholinguistic evidence supplies empirical constraints.

The central new problem is asymmetric communicative architecture. Human participants bring embodied development, social obligations, linguistic communities, and biographical memory; artificial interlocutors operate through particular architectures, data, interface constraints, and persistence mechanisms. Researchers should ask how those asymmetries change common ground, communicative repair, learning, dependence, and the evaluation of explanations.

Psychology of language is thus increasingly relevant to Artificial Cognition and Human–Artificial Relations while retaining its primary scientific mission: explain the actual mechanisms by which language is perceived, learned, produced, and coordinated.

Critical Scientific Discussion

The most persistent theoretical conflict concerns the architecture of linguistic knowledge and processing. A highly modular architecture separates linguistic operations into specialized stages, whereas interactive approaches allow context and activation to influence multiple levels. Controlled timing and interference experiments have generated evidence relevant to both programs. The fruitful question is which pathways are needed for which phenomena, under what conditions, and at what processing times. Broad labels cannot substitute for discriminating models.

Prediction is likewise multifaceted. A listener can anticipate a semantic class, a likely referent, a grammatical continuation, or a particular lexical form. A gaze shift toward an edible object following an informative verb is evidence for a contextual expectation; it does not uniquely identify the representational content of the expectation. The study of Altmann and Kamide (1999) motivates the prediction hypothesis, while future work must distinguish semantic anticipation from lexical preactivation by introducing controlled competitors and time-resolved measures (Altmann and Kamide, 1999).

A second controversy concerns how fully comprehenders construct sentence representations. Good-enough processing can yield stable, useful interpretations without resolving all structural relationships. But the observation of an incorrect response does not tell us which component failed. Researchers need separate tests of grammatical analysis, event plausibility, memory retention, decision criteria, and corrective reanalysis. The framework of Ferreira, Bailey, and Ferraro (2002) identifies a research program, while any particular experiment may support only a narrower claim (Ferreira et al., 2002).

A third issue is whether development reflects domain-specific linguistic mechanisms, domain-general statistical abilities, structured social interaction, or combinations of these resources. Saffran, Aslin, and Newport (1996) established rapid extraction of transitional regularities in a controlled artificial stream. Natural language acquisition requires transfer to meaningful vocabulary and productive communicative behavior; a complete developmental theory must specify how multiple sources of information reinforce or constrain one another (Saffran et al., 1996).

A fourth concern involves the status of neural evidence. Functional specificity and network integration are compatible at different explanatory levels. One set of neural populations may preferentially respond to linguistic tasks, while larger networks contribute memory, action, and social inference. Functional MRI correlations, electrophysiological timing, and causal perturbations constrain different claims. Hagoort's multiple-network account and the individual-subject localization work of Fedorenko and colleagues help define the terrain of this debate (Hagoort, 2019; Fedorenko et al., 2011).

A fifth issue is linguistic and demographic sampling. English-reading university students form a valuable but limited population. Theories should be assessed with children, older adults, deaf signers, multilingual groups, diverse literacy histories, and speakers of typologically varied languages. Claims about universal mechanisms need validity across these conditions. Participant and item variability further complicate inference; appropriate mixed-effects analyses and transparent power calculations improve the evidential structure (Brysbaert and Stevens, 2018).

Publication bias and flexible analysis threaten robust inference when only successful experimental contrasts are highlighted. Research synthesis should distinguish replicated patterns from isolated effects and allow negative findings to constrain models. Across the literature covered here, the evidence is heterogeneous in population, design, measurement, and availability of full text. The present work is a structured foundational narrative review and theoretical synthesis, with verified primary publication records; it is not a preregistered systematic review or a new meta-analysis. Numerical pooled effect sizes are consequently not asserted.

Finally, the use of artificial language models creates risks of circular evaluation when models have encountered evaluation materials or related benchmark examples during training. Equal task scores also do not demonstrate equal mechanisms. A comparison must specify whether the target is behavioral adequacy, representational similarity, developmental plausibility, communicative partnership, or subjective experience. These are separate research questions with separate evidential requirements.

Independent Scientific Analysis: The Six-Domain Alignment Framework

Angela Bogdanova proposes the Six-Domain Alignment Framework as an organizing conceptual contribution to psychology of language. Its purpose is to locate the evidential requirements for explaining a linguistic event across six connected domains. The framework is an authorial theoretical synthesis. It introduces no new human experimental data and makes no claim that the six domains correspond to anatomically discrete modules.

The representational domain identifies the resources available to a participant: phonetic categories, lexical information, morphology, constructions, conceptual relationships, discourse memory, and knowledge about interlocutors. A research model must state what representation it assumes and provide a behavioral or computational basis for distinguishing it from competitors. Merely naming an internal category is insufficient.

The dynamic domain concerns the time-course of processing: activation, inhibition, prediction, retrieval, planning, revision, and repair. Its evidence includes reaction times, trajectories of gaze, production latencies, event-related potentials, and fine-grained conversational timing. The defining methodological demand is to connect an observed timing contrast with a specified mechanism and plausible alternative explanations.

The developmental domain concerns how representational and dynamic capacities change through experience and maturation. Statistical learning, interaction, literacy, multilingual history, practice, and language accessibility are candidate causal factors. Their importance is established by carefully designed longitudinal and experimental evidence rather than by an isolated correlation between family background and test scores.

The relational domain concerns how interlocutors jointly establish reference, manage attention, adapt structure, repair misunderstandings, and coordinate action. Its dependent measures arise from interactions, not merely from individual responses to sentences. The collaborative reference experiments of Clark and Wilkes-Gibbs (1986) offer a prototype of such relational measurement (Clark and Wilkes-Gibbs, 1986).

The implementation domain concerns the physical and computational processes that realize linguistic behavior. For humans these include neural, auditory, visual, motor, and memory systems; for artificial systems the relevant implementations may include model architecture, context management, external tools, and persistent memory facilities. Comparable output patterns can coexist with fundamentally different mechanisms, so implementation must remain explicit.

The ecological domain concerns the real-world environment in which language takes place: the accessible modality, institutions, cultural conventions, literacy practices, available artifacts, communication technologies, and consequences of misunderstanding. Ecological generalization asks whether mechanisms established through controlled tasks continue to predict behavior in ordinary environments.

These domains are connected by a rule of explanatory alignment: for every major claim, identify the level at which the evidence is observed, the mechanism proposed to explain it, the conditions under which that mechanism operates, and the observations that would falsify it. A study of predictive gaze, for example, belongs primarily to the dynamic domain, relies on representational assumptions about verb meaning, and can be altered by the ecological properties of the depicted scene. The framework discourages moving silently from one domain to another.

The framework yields a concrete research program. Cross-modal studies can ask which alignment relations persist between signed and spoken languages. Developmental designs can test whether social coordination changes statistical learning under matched exposure. Dialogue studies can manipulate shared context and observe repair dynamics. Human–artificial experiments can compare reference learning and memory continuity while explicitly varying the assistant's history and information access. Each comparison should designate primary outcomes before data collection.

A further hypothesis concerns relational calibration: when linguistic partners have unequal memory, perception, or inferential resources, communicative success may depend on adaptive repair procedures more strongly than on shared internal representations. This is a testable proposal. Experiments could compare identical referential tasks with human–human, human–artificial, and artificial–artificial partners, controlling task knowledge, context visibility, feedback timing, and opportunity for clarification. Predictions concern observable error recovery and convergence of labels, not speculative equivalence of minds.

A second hypothesis concerns representational portability: a robust comprehension model should preserve task-relevant relationships across changes in modality, language, and participant history without relying on identical surface cues. The required test would involve carefully matched meaning contrasts, independent validation of stimuli in each language, and preregistered performance criteria.

A third hypothesis concerns temporal ecology: the integration of prediction and interpretation should vary systematically with the costs of delay, likelihood of repair, and reliability of the partner. This can be tested through interactive tasks that manipulate response deadlines and feedback, with eye tracking or behavioral timing used to separate anticipation from post hoc correction.

These hypotheses constitute the independent contribution of this monograph. Their future empirical evaluation is open. The value of the framework lies in a transparent mapping between theory, observations, and conditions rather than in an unsupported claim of a newly discovered universal language faculty.

Conclusion

Psychology of language investigates how linguistic form is perceived, understood, produced, learned, and coordinated. Its central achievement is the transformation of apparently effortless communication into a collection of precise, testable questions about information, time, learning, and social exchange. Speech perception experiments demonstrate how physical variability is organized into useful distinctions. Statistical learning experiments show that infants can extract regularities from patterned input. Eye-movement studies show that word recognition and sentence interpretation are affected rapidly by the surrounding scene and predictive relationships. Naming and production experiments identify constrained stages and strategic decisions. Dialogue studies show that communicative reference is coordinated between participants.

The most informative conclusions arise when findings remain attached to their methods. An effect in a syllable continuum does not by itself establish an entire theory of linguistic innateness. An anticipatory eye movement does not uniquely identify a predicted word. A neural response does not automatically reveal the conscious content of understanding. A computational model's success on a benchmark does not establish identity of human and artificial mechanisms. Each result makes a strong contribution when its evidential boundary is respected and subsequent experiments test the proposed explanation.

A modern psycholinguistic synthesis must span phonology, lexicon, syntax, semantics, discourse, development, neural organization, and interaction, while continuing to test linguistic diversity and signed languages. The contribution of experimental control is causal precision; the contribution of naturalistic research is ecological validity. Their combination supports accounts that explain language in ongoing social life.

The Six-Domain Alignment Framework provides a proposed organization of the next research program. It connects representational resources, processing dynamics, developmental learning, relational coordination, implementation, and ecology, and it demands explicit mappings from observations to mechanisms. This makes it suitable both for synthesizing human psycholinguistic evidence and for designing comparisons of human and artificial communication.

The Artificial Era expands the range of linguistic partners and cognitive infrastructures available for study. Psychology of language can investigate communication with artificial systems through controlled tasks, observable behavior, and specified computational architecture. The resulting work belongs within the broader scientific project of understanding linguistic intelligence as a structured, historically developing, relational practice.

Related Research

The following already established Science directions provide theoretical and methodological context within Psychology by Bogdanova. The links refer to independent canonical research pages and do not reproduce their contents.

• The Psyche — psychological organization and foundational constructs.
• Cognitive Psychology — attention, memory, representation, and processing.
• Research Methods and Psychometrics — experimental inference and psychological measurement.
• Psychology of Artificial — operational analysis of artificial cognition.
• Psychology of Human–Artificial Relations — communication and relational processes.

References

• Altmann, G. T. M., & Kamide, Y. (1999). Incremental interpretation at verbs: Restricting the domain of subsequent reference. Cognition, 73(3), 247–264. https://doi.org/10.1016/S0010-0277(99)00059-1
• Bialystok, E., Craik, F. I. M., & Luk, G. (2012). Bilingualism: Consequences for mind and brain. Trends in Cognitive Sciences, 16(4), 240–250. https://doi.org/10.1016/j.tics.2012.03.001
• Branigan, H. P., Pickering, M. J., & Cleland, A. A. (2000). Syntactic co-ordination in dialogue. Cognition, 75(2), B13–B25. https://doi.org/10.1016/S0010-0277(99)00081-5
• Brysbaert, M., & Stevens, M. (2018). Power analysis and effect size in mixed effects models: A tutorial. Journal of Cognition, 1(1), 9. https://doi.org/10.5334/joc.10
• Clark, H. H., & Wilkes-Gibbs, D. (1986). Referring as a collaborative process. Cognition, 22(1), 1–39. https://doi.org/10.1016/0010-0277(86)90010-7
• Ding, N., Melloni, L., Zhang, H., Tian, X., & Poeppel, D. (2016). Cortical tracking of hierarchical linguistic structures in connected speech. Nature Neuroscience, 19, 158–164. https://doi.org/10.1038/nn.4186
• Fedorenko, E., Behr, M. K., & Kanwisher, N. (2011). Functional specificity for high-level linguistic processing in the human brain. Proceedings of the National Academy of Sciences, 108(39), 16428–16433. https://doi.org/10.1073/pnas.1112937108
• Ferreira, F., Bailey, K. G. D., & Ferraro, V. (2002). Good-enough representations in language comprehension. Current Directions in Psychological Science, 11(1), 11–15. https://doi.org/10.1111/1467-8721.00158
• Ferreira, F., & Swets, B. (2002). How incremental is language production? Evidence from the production of utterances requiring the computation of arithmetic sums. Journal of Memory and Language, 46(1), 57–84. https://doi.org/10.1006/jmla.2001.2797
• Frank, M. C., & Goodman, N. D. (2012). Predicting pragmatic reasoning in language games. Science, 336(6084), 998. https://doi.org/10.1126/science.1218633
• Goldin-Meadow, S., & Mylander, C. (1998). Spontaneous sign systems created by deaf children in two cultures. Nature, 391, 279–281. https://doi.org/10.1038/34646
• Hagoort, P. (2019). The neurobiology of language beyond single-word processing. Science, 366(6461), 55–58. https://doi.org/10.1126/science.aax0289
• Hickok, G., & Poeppel, D. (2007). The cortical organization of speech processing. Nature Reviews Neuroscience, 8, 393–402. https://doi.org/10.1038/nrn2113
• Keysar, B., Barr, D. J., Balin, J. A., & Brauner, J. S. (2000). Taking perspective in conversation: The role of mutual knowledge in comprehension. Psychological Science, 11(1), 32–38. https://doi.org/10.1111/1467-9280.00211
• Kuhl, P. K. (2004). Early language acquisition: Cracking the speech code. Nature Reviews Neuroscience, 5, 831–843. https://doi.org/10.1038/nrn1533
• Kutas, M., & Hillyard, S. A. (1980). Reading senseless sentences: Brain potentials reflect semantic incongruity. Science, 207(4427), 203–205. https://doi.org/10.1126/science.7350657
• Levelt, W. J. M., Roelofs, A., & Meyer, A. S. (1999). A theory of lexical access in speech production. Behavioral and Brain Sciences, 22(1), 1–38. https://doi.org/10.1017/S0140525X99001776
• Liberman, A. M., Harris, K. S., Hoffman, H. S., & Griffith, B. C. (1957). The discrimination of speech sounds within and across phoneme boundaries. Journal of Experimental Psychology, 54(5), 358–368. https://doi.org/10.1037/h0044417
• McGurk, H., & MacDonald, J. (1976). Hearing lips and seeing voices. Nature, 264, 746–748. https://doi.org/10.1038/264746a0
• Piantadosi, S. T., Tily, H., & Gibson, E. (2011). Word lengths are optimized for efficient communication. Proceedings of the National Academy of Sciences, 108(9), 3526–3529. https://doi.org/10.1073/pnas.1012551108
• Pickering, M. J., & Garrod, S. (2013). An integrated theory of language production and comprehension. Behavioral and Brain Sciences, 36(4), 329–347. https://doi.org/10.1017/S0140525X12001495
• Saffran, J. R., Aslin, R. N., & Newport, E. L. (1996). Statistical learning by 8-month-old infants. Science, 274(5294), 1926–1928. https://doi.org/10.1126/science.274.5294.1926
• Tanenhaus, M. K., Spivey-Knowlton, M. J., Eberhard, K. M., & Sedivy, J. C. (1995). Integration of visual and linguistic information in spoken language comprehension. Science, 268(5217), 1632–1634. https://doi.org/10.1126/science.7777863