RESEARCH DOSSIER
Semantic Integration and Lexical Ambiguity
Abstract. Semantic integration and lexical ambiguity define a central problem in psycholinguistics: listeners and readers identify multiple candidate meanings associated with a word form while constructing a contextually coherent interpretation. This critical narrative monograph reconstructs the history of competing accounts of lexical access, examines meaning dominance, contextual constraint, polysemy, sentence-level composition, semantic priming, eye-movement and cross-modal experiments, the N400 and P600, frontotemporal neural networks, and experience-dependent retuning of word meanings. It compares ordered, exhaustive, selective, constraint-based, construction–integration, predictive, and dynamic semantic models; evaluates measurement assumptions, task effects, and causal limitations; and examines the implications of contemporary machine-language systems for explicitly testable psychological hypotheses. The synthesis supports temporally unfolding and experience-sensitive semantic interpretation, with substantial variation across ambiguity types, tasks, and contexts. It is a critical narrative synthesis of published research, not an original experiment, preregistered systematic review, or meta-analysis.
Keywords: semantic integration; lexical ambiguity; lexical access; word meaning; semantic priming; cross-modal priming; meaning dominance; subordinate-bias effect; homonymy; polysemy; sentence comprehension; contextual constraint; N400; P600; eye tracking; fMRI; semantic unification; word-meaning priming; prediction; psycholinguistics; Artificial Era
Scientific Object, Explanatory Levels, and Scope
Semantic integration is the process by which meaning-bearing information becomes part of a coherent representation of an utterance, sentence, discourse, or situation. Lexical ambiguity arises when a single word form supports more than one interpretation, requiring an interpreter to assign contextually useful meaning without losing the capacity to revise that assignment. The central psychological question concerns how candidate interpretations become available, compete or cooperate, and become integrated with syntactic relations, event knowledge, speaker expectations, and prior experience. The phenomenon is dynamic: a word's initial recognition, its contextual interpretation, and a later judgment of what a speaker meant need not occur at the same moment. The research tradition therefore employs measures with different temporal and representational sensitivity rather than treating comprehension as one instantaneous decision Swinney, D. A. (1979) Duffy, S. A., Morris, R. K., & Rayner, K. (1988) Kutas, M., & Federmeier, K. D. (2011).
The object of this monograph is the interface between candidate lexical meaning and the evolving interpretation of connected language. It includes meaning activation and availability, competition and selection, compositional fit, mismatch detection, reinterpretation, and longer-lasting retuning of meaning preferences. Lexical frequency and the organization of the mental lexicon are included insofar as they determine competition during comprehension; general morphological representation belongs to the neighboring mental-lexicon research object. Syntactic garden paths constitute a related comparison when grammatical alternatives interact with semantic interpretation, but sentence parsing as a complete system has its own research domain. Pragmatic inference and figurative intention are considered only when necessary to explain why semantic interpretations change MacDonald, M. C., Pearlmutter, N. J., & Seidenberg, M. S. (1994) Hagoort, P. (2005) Rodd, J. M., Lopez Cutrin, B., Kirsch, H., Millar, A., & Davis, M. H. (2013).
The evidence assembled here combines original behavioral experiments, cross-modal and visual priming, continuous eye movements, electrophysiology, hemodynamic neuroimaging, computational theory, and a contemporary account of context-sensitive word meaning. Methodologically this is a critical narrative research monograph using identifiable scholarly sources, with explicit appraisal of construct validity and rival mechanisms. Claims concerning the timing of lexical availability draw principally on online experiments; claims concerning neural localization rest on imaging or electrophysiological studies and their stated limitations; theoretical proposals are assessed for discriminating predictions. The work reports no newly collected data, quantitative pooled effect, or claim of exhaustive systematic retrieval. Its principal contribution is an integrated and auditable exposition of the evidence Simpson, G. B. (1984) Kintsch, W. (1988) Rodd, J. M. (2026).
Ambiguity, Homonymy, Polysemy, and Meaning Dominance
Lexical ambiguity is a property of the mapping between a word form and interpretations available to a language user. Homonymy conventionally concerns distinct meanings associated with the same phonological or orthographic form, as in a word that denotes a financial institution or the side of a river. Polysemy concerns a family of related senses, as when a term extends from a physical object to a characteristic action or institutional role. The psychological distinction must be operationalized using normative meaning judgments, association structure, semantic similarity, and behavioral tasks; dictionary counts alone do not determine how a particular reader represents a lexical item. The difference matters because separable meanings can compete while related senses may share representational features Rodd, J., Gaskell, G., & Marslen-Wilson, W. (2002) Simpson, G. B. (1984).
Meaning dominance refers to the unequal probabilities or strengths of different interpretations for a language user in a defined linguistic environment. A dominant meaning often becomes accessible efficiently because it has been repeatedly encountered, yet its measured dominance depends on language variety, occupational experience, age, topical discourse, and the procedure used to obtain sense norms. Equibiased words have interpretations with more similar prior frequencies; strongly biased words have one meaning favored before the current sentence is processed. These categories are experimental approximations, not natural kinds. An adequate design estimates the probabilities of interpretations in its target population and distinguishes unconditional dominance from context-conditioned preference Duffy, S. A., Morris, R. K., & Rayner, K. (1988) Rayner, K., Pacht, J. M., & Duffy, S. A. (1994) Rodd, J. M., Lopez Cutrin, B., Kirsch, H., Millar, A., & Davis, M. H. (2013).
A useful formal description treats the interpretation of a word w as a probability distribution over candidate meanings m given preceding context c and experience h: P(m|w,c,h). The distribution changes when additional words are encountered, when a relevant episode becomes accessible, or when contextual evidence retrospectively changes an earlier decision. This notation expresses an analytical problem rather than establishing that the human brain literally performs an identifiable Bayesian algorithm. It clarifies three measurable outcomes: which meanings are briefly activated, which interpretation supports a current action, and which representation persists as the best understanding of the sentence. Models must explain why those outcomes sometimes diverge Levy, R. (2008) Rodd, J. M. (2026).
From Lexical Search to Context-Sensitive Access
Early psycholinguistics asked whether a listener searches a lexical store in frequency order, activates every interpretation of an ambiguous form, or selects an interpretation before full semantic access. Search accounts were attractive because frequency affects word-recognition latency, while ordered access could explain why dominant meanings are commonly preferred. Experiments by Forster and Bednall (1976) contrasted different search demands, showing that the structure of the required classification influences whether frequency-related predictions hold. Such findings establish the need to distinguish a theory of lexical representations from a theory of the participant's experimental decision procedure Forster, K. I., & Bednall, E. S. (1976) Simpson, G. B. (1984).
A purely selective hypothesis predicts that sufficiently constraining sentence context can prevent activation of irrelevant meanings at the earliest measurable point. An exhaustive-access hypothesis predicts early activation of more than one interpretation, followed by contextual selection. A reordered or constraint-based account predicts that frequency, syntactic class, associative priming, and sentence context continuously alter relative activation and the speed of commitment. These are partly competing descriptions of temporal dynamics, not mutually exclusive descriptions of everything that happens during comprehension. A model can posit relatively parallel activation at an early stage and context-sensitive competition before overt recognition, thereby yielding different results at distinct measurement delays Seidenberg, M. S., Tanenhaus, M. K., Leiman, J. M., & Bienkowski, M. (1982) Tabossi, P. (1988) Duffy, S. A., Morris, R. K., & Rayner, K. (1988).
The historical controversy changed as investigators varied the form of ambiguity, strength of context, probe delay, and behavioral task. Simpson's 1984 review consolidated the observation that ambiguity research could discriminate models of word recognition only when experimental constructs were sharply defined. The central methodological shift was from asking whether context matters at all to asking at which processing stage context changes activation, what kind of context is sufficient, and what measure can detect a meaning even when it is not the final interpretation. Those questions continue to organize modern neural and computational studies Simpson, G. B. (1984) Kutas, M., & Federmeier, K. D. (2011) Rodd, J. M. (2026).
Cross-Modal Priming and the Time Course of Meaning Availability
In a landmark study, Swinney (1979) presented spoken sentences containing lexically ambiguous words while participants made visual lexical decisions to letter strings. Targets related to alternative meanings of a recently heard ambiguous word were facilitated when probed near the word's offset, including with sentence context favoring one meaning. Later probes, positioned four syllables after the ambiguous word, showed facilitation for the contextually appropriate interpretation. The pattern motivated a distinction between early multiple activation and subsequent selection. Its enduring value is methodological: a visual probe provides a time-sensitive indirect index of semantic availability during ongoing auditory comprehension Swinney, D. A. (1979).
Cross-modal priming combines a naturalistic spoken carrier with an additional decision that can itself modify attention. A short probe interval increases temporal sensitivity but does not guarantee that every unobserved candidate is absent or that all detectable facilitation corresponds to conscious interpretation. Semantic association, lexical frequency, and motor response thresholds influence the magnitude of the effect. Investigators therefore need baseline conditions, probe-relatedness controls, calibrated presentation timing, and item-level measures of meaning dominance. Interpretation is strengthened by convergence with independent paradigms such as eye movements and neural responses Swinney, D. A. (1979) Seidenberg, M. S., Tanenhaus, M. K., Leiman, J. M., & Bienkowski, M. (1982) Holcomb, P. J. (1993).
Seidenberg, Tanenhaus, Leiman, and Bienkowski (1982) used multiple experiments with noun–noun and noun–verb ambiguities and varied whether a preceding context furnished direct semantic associations or broader syntactic and pragmatic constraints. Their results showed that early activation patterns differed across ambiguity types and priming conditions. The study complicated a universal claim that every ambiguous item follows an identical exhaustive or selective procedure. Sentence context may constrain the accessibility of meanings by changing the state of the lexical-semantic network, while a task demanding rapid classification can reveal different aspects of this state Seidenberg, M. S., Tanenhaus, M. K., Leiman, J. M., & Bienkowski, M. (1982) Tabossi, P. (1988).
Contextual Specificity and the Limits of a Binary Access Debate
Tabossi (1988) investigated ambiguous word access under several kinds of sentential context using cross-modal lexical decisions. Her results indicated that a context supporting the dominant meaning could be associated with selective or broader activation depending on whether it imposed sufficiently specific constraints on the relevant semantic features. This is an important conceptual distinction. A sentence may bias an interpretation at the level of overall topic while leaving many of the features of the expected meaning unspecified; another sentence may activate distinctive features that directly favor one candidate over its competitors Tabossi, P. (1988).
Subsequent experiments by Tabossi and Zardon (1993) considered whether apparently selective activation reflects genuine contextual constraint or an extremely rapid exhaustive-access-and-selection sequence. The empirical problem is intrinsically difficult because priming measured at a chosen delay cannot reveal a representation at an unmeasured earlier interval. A selective effect therefore needs to be interpreted with reference to stimulus onset asynchrony, target association, task sensitivity, and the conditions under which a competing meaning should have been detectable. A temporally resolved model has to predict entire activation trajectories, including the point at which target facilitation disappears Tabossi, P., & Zardon, F. (1993) Seidenberg, M. S., Tanenhaus, M. K., Leiman, J. M., & Bienkowski, M. (1982).
Context comprises several distinguishable information sources: local lexical associations, grammatical constraints, thematic event expectations, the preceding discourse, shared referents, and stored personal experience. Each source constrains interpretation with its own reliability and timing. A sentence about a fisherman may bias a river-related interpretation of a word before that word appears; a later clause can contradict that expectation and force revision. Treating context as one scalar strength is frequently inadequate. Future experiments should orthogonally manipulate associative strength, propositional plausibility, and discourse-level consistency while testing both early activation and later memory for the resulting interpretation Duffy, S. A., Morris, R. K., & Rayner, K. (1988) Kintsch, W. (1988) Rodd, J. M. (2026).
Eye Movements and the Subordinate-Bias Effect
Eye-tracking studies observe how a reader allocates processing time while encountering ambiguous words within sentences. Duffy, Morris, and Rayner (1988) compared equibiased and meaning-dominant ambiguous words with unambiguous controls while manipulating the position of a disambiguating clause. Fixation and subsequent reading patterns differed as a function of ambiguity type and contextual timing. The work helped establish the subordinate-bias effect: when context strongly favors a relatively infrequent meaning of a biased word, the dominant meaning can still compete, producing extra processing cost. This effect supports a conception in which meaning frequency and sentence context jointly influence online comprehension Duffy, S. A., Morris, R. K., & Rayner, K. (1988).
An eye fixation provides a continuous and relatively naturalistic behavioral measure, but it does not isolate one mental operation. Gaze duration can reflect visual encoding, lexical access, integration, attention and motor planning; regressions can indicate reanalysis without identifying precisely what was reanalyzed. It matters whether an ambiguity-related cost appears on the target word itself, on the next region, or after later contradictory information. Matching ambiguous and control words by word-form frequency may yield a different comparison from matching them by the frequency of the meaning licensed by the sentence. Consequently the choice of control baseline bears directly on the interpretation of an ambiguity cost Duffy, S. A., Morris, R. K., & Rayner, K. (1988) Rayner, K., Pacht, J. M., & Duffy, S. A. (1994).
Rayner, Pacht, and Duffy (1994) tested whether prior encounter with a subordinate meaning or an extended globally biasing discourse could remove subordinate-bias costs. The reported persistence of the effect under those manipulations showed that strong prior support need not neutralize competition at the critical word. This finding puts pressure on simple accounts in which sufficiently long contextual preparation always prevents dominant-meaning activation. It does not establish that context is powerless: context influences the final interpretation, the time at which selection occurs, and the persistence of an earlier meaning choice. A complete model must distinguish those functions Rayner, K., Pacht, J. M., & Duffy, S. A. (1994) Tabossi, P. (1988) Rodd, J. M., Lopez Cutrin, B., Kirsch, H., Millar, A., & Davis, M. H. (2013).
Evidence Synthesis of Early Behavioral Traditions
The early behavioral record establishes that lexical interpretation unfolds over time and that the distribution of available meanings depends on frequency, contextual constraints, and ambiguity structure. Cross-modal probes can detect meanings that are subsequently rejected; eye movements can reveal competition as readers integrate a contextually subordinate meaning; feature-sensitive priming can detect contextual facilitation before overt selection. These findings are mutually informative because each method samples a different consequence of latent lexical-semantic activation. None requires the assumption that every meaning is always consciously entertained Swinney, D. A. (1979) Duffy, S. A., Morris, R. K., & Rayner, K. (1988) Tabossi, P. (1988).
The apparent conflict between selective and multiple-access experiments becomes more interpretable when the timing, stimulus distribution, and dependent variable are considered together. A null priming effect for a secondary meaning can reflect suppression, weak activation below the test's sensitivity, an unsuitable probe, or rapid decay after an initially broad response. Conversely, facilitation to an irrelevant target can reflect residual association without establishing that this meaning entered the speaker's intended interpretation. Identification of the cognitive mechanism therefore calls for multiple probes, transparent word norms, suitable unambiguous controls, and models that explain the joint profile of behavior rather than one significance test Seidenberg, M. S., Tanenhaus, M. K., Leiman, J. M., & Bienkowski, M. (1982) Neill, W. T., Hilliard, D. V., & Cooper, E.-A. (1988) Simpson, G. B. (1984).
The conceptual lesson is that a language user maintains structured semantic possibilities under continuous constraint while building a task-appropriate interpretation. Meaning access, competition, local integration, and retrospective revision are distinguishable empirical functions. The evidence makes strong deterministic claims about fixed serial stages less persuasive, while leaving room for ordered influences due to past frequency, current context, and neural processing dynamics. Subsequent research extends the explanation by separating related senses from unrelated meanings and by measuring semantic fit directly in brain responses Rodd, J., Gaskell, G., & Marslen-Wilson, W. (2002) Kutas, M., & Federmeier, K. D. (2011) Hagoort, P., Hald, L., Bastiaansen, M., & Petersson, K. M. (2004).
Semantic Competition, Representational Overlap, and the Polysemy Advantage
Lexical decision experiments sometimes find faster recognition for words associated with many meanings or senses. This observation initially encouraged the idea that a larger semantic neighborhood automatically increases lexical activation. Rodd, Gaskell, and Marslen-Wilson (2002) distinguished unrelated meanings from multiple related senses across three lexical-decision experiments. Their results showed a critical dissociation: related senses could facilitate recognition, whereas distinct competing meanings could delay it. The result changes the explanatory target from a count of dictionary entries to the topology of semantic representations. Several related interpretations may reinforce shared features; unrelated interpretations may impose incompatible demands on the same form-to-meaning mapping Rodd, J., Gaskell, G., & Marslen-Wilson, W. (2002).
Distributed semantic accounts describe meaning as patterns across overlapping representational dimensions rather than atomic entries that are either retrieved or absent. When a word has related senses, partial activation may converge on shared features and make the form recognizable quickly. When a word links unrelated concepts, activation may spread toward alternative attractors whose competition delays commitment. These are theoretical mechanisms consistent with the behavioral dissociation, rather than direct observations of neural attractors in the experiment. The concrete predictions depend on the learning rule, semantic similarity metric, and decision criterion selected for a model Rodd, J., Gaskell, G., & Marslen-Wilson, W. (2002) Kintsch, W. (1988).
Experimental identification requires independent norms for sense relatedness, meaning dominance, familiarity, length, spelling neighborhood, concreteness, and frequency. Comparing ambiguous and unambiguous words without such controls can create apparent advantages attributable to uncontrolled lexical properties. The same item may be polysemous to one population and effectively monosemous to another because its less familiar interpretation is unavailable. Item-level mixed-effects analyses and independent samples for meaning norming are therefore essential for evaluating whether a lexical ambiguity effect generalizes beyond a laboratory's stimulus list Rodd, J., Gaskell, G., & Marslen-Wilson, W. (2002) Simpson, G. B. (1984) Rodd, J. M., Lopez Cutrin, B., Kirsch, H., Millar, A., & Davis, M. H. (2013).
Selection as a Continuous Constraint-Satisfaction Process
A constraint-based account treats the interpretation of an ambiguous word as the evolving outcome of several sources of evidence. Frequency creates prior preferences; grammatical category and thematic structure constrain candidate roles; local semantic features favor some interpretations; world knowledge supplies expectations about plausible events; discourse adds information about referents and goals. Each source can affect processing before, during, or after a word is recognized. Instead of a sharp boundary separating lexical access from sentence comprehension, the same constraints may influence candidate activation and subsequent selection with different strengths over time Duffy, S. A., Morris, R. K., & Rayner, K. (1988) MacDonald, M. C., Pearlmutter, N. J., & Seidenberg, M. S. (1994) Hagoort, P., Hald, L., Bastiaansen, M., & Petersson, K. M. (2004).
The apparent simultaneity of several constraints should not be confused with the absence of temporal order. Acoustic information arrives incrementally, written words become available through visual attention, and semantic constraints may be preactivated at different times. A context that reliably determines grammatical category can constrain the candidate set earlier than a discourse implication requiring elaborate inference. Strong dominance can temporarily outweigh a favorable sentence context, but later evidence may reverse the preference. What makes a theory explanatory is its ability to predict the circumstances of delay and the trajectory of updating, including cases in which both candidates remain partially active Seidenberg, M. S., Tanenhaus, M. K., Leiman, J. M., & Bienkowski, M. (1982) Tabossi, P., & Zardon, F. (1993) Rodd, J. M., Johnsrude, I. S., & Davis, M. H. (2012).
Neill, Hilliard, and Cooper (1988) interpreted ambiguity-detection results as evidence compatible with parallel, context-sensitive retrieval. The broader significance is that evidence can support competition between candidates whose latencies are influenced by frequency and context without requiring identical onset times or equal strengths. Such a hybrid architecture can account for why participants often identify a meaningful interpretation rapidly even when later measures reveal interference. It remains crucial to operationalize whether a candidate was retrieved, became strong enough to interfere, or was selected for the participant's final report; each hypothesis entails a different measurable event Neill, W. T., Hilliard, D. V., & Cooper, E.-A. (1988) Forster, K. I., & Bednall, E. S. (1976).
Compositional Semantic Fit and Incremental Sentence Meaning
Semantic integration is required whenever the meanings of successive words are combined to represent participants, relations, and events. A noun may introduce an object category, a verb constrains the actions in which that object could participate, and later constituents specify its role. The interpretation depends on semantic content together with syntax, but a good experimental analysis distinguishes the contribution of lexical selection from that of relation-building. A disambiguating verb can alter the interpretation of a prior ambiguous noun, while an ambiguous verb can be constrained by a previously introduced object or discourse situation Hagoort, P. (2005) Pylkkänen, L. (2019).
Altmann and Kamide (1999) demonstrated that listeners could use verbs and a concurrently visible scene to anticipate an upcoming referent before its name was heard. The work is important because it establishes that semantic constraints can influence attention prospectively during comprehension. It does not establish that every sentence is comprehended through a mandatory word-by-word prediction, nor that an eye movement always identifies a unique intended referent. Rather, it illustrates the early interaction of action knowledge, the linguistic signal, and a task-specific scene. The same form of anticipatory constraint can change how a later ambiguous noun is interpreted Altmann, G. T. M., & Kamide, Y. (1999).
At the sentence level, successful integration means that the selected lexical meanings jointly support a coherent interpretation under syntactic and contextual conditions. Meaning can nevertheless remain underspecified: a listener may understand the practical action required by an utterance while leaving a secondary lexical distinction unresolved. Comprehension tasks that demand explicit paraphrase may reveal ambiguity invisible in ordinary reading time. Conversely, a brief slowdown can occur during normal adjustment without producing lasting misunderstanding. Models should therefore predict both the dynamic cost of forming an interpretation and its semantic adequacy at later tests Kintsch, W. (1988) Hagoort, P., Hald, L., Bastiaansen, M., & Petersson, K. M. (2004) Pylkkänen, L. (2019).
Predictive Constraint, Surprisal, and Probabilistic Interpretation
Expectation-based accounts formalize the cost of unexpected linguistic input. Levy (2008) proposed a probabilistic theory of sentence comprehension in which processing difficulty relates to the surprisal of an incoming event in its preceding context. In its simple form, surprisal is minus the logarithm of contextual probability. This framework connects lexical constraints and grammatical interpretation and clarifies why a highly anticipated continuation is often processed efficiently. It also implies that ambiguity can be temporarily beneficial when several continuations remain available and the observed word retains substantial probability under multiple interpretations Levy, R. (2008).
Surprisal and semantic integration describe related but nonidentical explanatory quantities. A word may be statistically unexpected yet semantically compatible with a prior context; a common word may create a strong conceptual contradiction in a particular narrative. Predictability can be estimated by cloze completion, language-model probabilities, or exposure-based norms, but those measurements are not interchangeable. A good experiment separates contextual probability from lexical association, grammatical acceptability, and world knowledge. The N400 literature offers evidence that predictability and semantic relationship jointly influence online brain responses Kutas, M., & Hillyard, S. A. (1984) Federmeier, K. D., & Kutas, M. (1999a) Kutas, M., & Federmeier, K. D. (2011).
Predictive mechanisms may operate by preactivating likely semantic features, by reducing uncertainty about expected continuations, or by facilitating later retrieval of compatible information. Those accounts can generate overlapping reaction-time results while making different predictions about neural representations before a target appears. A probabilistic model is useful insofar as it specifies item probabilities and resulting predictions for measured reading times, gaze shifts, or electrophysiological responses. The existence of context effects alone does not identify a unique predictive code or establish that all semantic processing is fundamentally anticipatory Altmann, G. T. M., & Kamide, Y. (1999) Levy, R. (2008) Rodd, J. M. (2026).
Local Mismatch, Reinterpretation, and Semantic Repair
Disambiguation may arrive after a listener has committed to an initially plausible interpretation. A word such as bank can be understood financially until a later phrase introduces flowing water; interpretation then requires reconsidering the meaning assigned at the earlier location. Reanalysis can include suppression of an earlier candidate, retrieval of another meaning, revision of thematic relationships, and updating of the discourse model. These processes are distinguishable from initial multiple activation because they involve a representation that has already contributed to an ongoing understanding Swinney, D. A. (1979) Vitello, S., Warren, J. E., Devlin, J. T., & Rodd, J. M. (2014) Rodd, J. M., Johnsrude, I. S., & Davis, M. H. (2012).
The empirical signatures of semantic revision depend on the task. Reading regressions can indicate a return to an earlier region, delayed comprehension questions can reveal reinterpretation failures, and a later semantic probe can reveal the persistence of a discarded interpretation. Functional imaging may reveal increased activation in regions associated with selection or control, but hemodynamic activity is too temporally coarse to isolate the exact instant of revision. An experimental design can separate an ambiguity that is resolved immediately from one whose apparent interpretation is contradicted only after a delay, thereby identifying the added demands of reinterpretation Duffy, S. A., Morris, R. K., & Rayner, K. (1988) Rodd, J. M., Johnsrude, I. S., & Davis, M. H. (2012) Vitello, S., Warren, J. E., Devlin, J. T., & Rodd, J. M. (2014).
Reinterpretation also exposes the limits of defining comprehension as the search for one uniquely correct lexical sense. In ordinary dialogue, listeners often preserve partially specified representations and revise only those distinctions made relevant by subsequent information. The theoretical question is how semantic commitments remain sufficiently stable to support action while sufficiently flexible to incorporate new evidence. Contemporary accounts involving retrieval from prior episodes and context-sensitive prediction offer mechanisms for revisable meaning without assuming that every utterance undergoes exhaustive global reconstruction Kintsch, W. (1988) Rodd, J. M. (2026).
Semantic Priming as an Experimental Window on Meaning
Semantic priming is a change in response to a target following exposure to a semantically related or associatively connected prime. In lexical decision, a target such as doctor may be recognized more efficiently after a related prime than after an unrelated one. The effect can originate from preactivation of associated representations, expectancies about target categories, or adjustments to decision processes. Consequently priming is a family of experimental observations rather than a direct measure of a single, context-free semantic operation. Ambiguity research uses its temporal and associative sensitivity to infer which interpretation of a word is currently accessible Seidenberg, M. S., Tanenhaus, M. K., Leiman, J. M., & Bienkowski, M. (1982) Holcomb, P. J. (1993).
Priming experiments vary the duration and visibility of primes, the prime-to-target interval, the lexicality and frequency of targets, the proportion of related trials, the response demanded, and the participant's awareness of the relation. Each manipulation can change how extensively the result depends on automatic association or strategic control. Holcomb (1993) combined semantic priming with stimulus degradation and electrophysiological measurement, illustrating that behavioral reaction times and N400 responses can reveal different interactions with perceptual difficulty. Treating a faster button response as proof of a faster internal semantic process ignores the multiple operations intervening between stimulus and action Holcomb, P. J. (1993) Kutas, M., & Federmeier, K. D. (2011).
Lexical ambiguity adds the further requirement that the probe be selectively related to one candidate meaning. A target strongly associated with both candidates provides limited diagnostic information. A target with differing base frequency or semantic relatedness can generate an imbalance even when candidate activation is comparable. Well-designed studies norm associative strength and semantic features independently, present both meanings across matched conditions, and compare predictions from early and delayed probes. Combining priming with eye-tracking and neural responses is especially valuable because the same stimulus can have distinct consequences for access, selection, and the eventual interpretation Swinney, D. A. (1979) Tabossi, P. (1988) Rodd, J., Gaskell, G., & Marslen-Wilson, W. (2002).
Discourse, World Knowledge, and Construction–Integration
Kintsch's construction–integration model (1988) describes comprehension in terms of initially constructing a broad network of lexical propositions and possible inferences, followed by integration that strengthens mutually supportive elements of a coherent representation. This approach provides a bridge from lexical ambiguity to discourse comprehension. Multiple interpretations may become available locally, while contextual relationships and background knowledge determine which candidates survive within the developing representation. A sentence is understood against a wider structure of events and propositions rather than through the isolated concatenation of dictionary definitions Kintsch, W. (1988).
World knowledge is implicated when a string is grammatically acceptable but describes an improbable or false event. Hagoort, Hald, Bastiaansen, and Petersson (2004) reported electrophysiological and imaging evidence that linguistic meaning and relevant world knowledge influence interpretation rapidly. Their study challenged an oversimplified sequence in which sentence meaning must be fully constructed before information about the world can affect understanding. A context can influence lexical interpretation by making one concept plausible while another would violate familiar event knowledge. Importantly, a mismatch with world knowledge does not have the same evidential status as a grammatical error or a logically contradictory statement Hagoort, P., Hald, L., Bastiaansen, M., & Petersson, K. M. (2004).
Discourse offers both stabilizing and revising information. Earlier sentences establish entities, locations, goals, and causal relations; later material may elaborate, narrow, or overturn these assumptions. The comprehension system can use topic to prefer a lexical sense while still remaining sensitive to a later subordinate interpretation. The research boundary here concerns how those discourse constraints change the interpretation of particular lexical material. Full narrative structure, referential tracking, and situation models form a neighboring research object, although their constraints are indispensable to understanding local semantic selection Kintsch, W. (1988) Rayner, K., Pacht, J. M., & Duffy, S. A. (1994) Rodd, J. M. (2026).
Expectancy, Word Frequency, and Semantic Association in N400 Responses
Kutas and Hillyard (1984) related N400 amplitude to the cloze probability of sentence-final words and reported an additional advantage for unexpected words semantically associated with expected completions. The result shows that semantic relatedness can matter even when the specific word was not strongly predicted. It is therefore possible for a sentence to support a semantic feature network while failing to predict an exact lexical form. This distinction is essential in ambiguity research, where a context can make one interpretation coherent without making one specific continuation inevitable Kutas, M., & Hillyard, S. A. (1984).
Van Petten and Kutas (1990) examined electrophysiological responses across successive sentence positions and found that word-frequency effects interacted with the contextual information available as a sentence unfolded. Early words with low lexical frequency could elicit relatively large N400 responses, while the effect of frequency diminished later in constraining contexts. The finding indicates that word frequency cannot be treated as a context-invariant determinant of processing effort. When studying ambiguity, dominance norms and overall word frequency should be crossed with measures of contextual constraint rather than assumed to exert uniform independent effects Van Petten, C., & Kutas, M. (1990).
Federmeier and Kutas (1999) further used category-related unexpected completions to study the structure of expectations during sentence comprehension. Their experiments compared highly expected target words with unexpected targets that shared category relations and with unexpected words from other categories. Such contrasts are informative because they separate prediction of a particular item from graded preactivation of semantic attributes. Together these findings support a model in which incoming meaning is interpreted against an evolving landscape of lexical and conceptual activation, although the precise neural computations implementing that landscape remain debated Federmeier, K. D., & Kutas, M. (1999a) Federmeier, K. D., & Kutas, M. (1999b) Kutas, M., & Federmeier, K. D. (2011).
N400 and P600: Semantic Interpretation Beyond Simple Anomaly Detection
Sentence processing can elicit later positive electrophysiological effects often referred to as P600 responses, particularly when a structure must be revised or when its interpretation strains established expectations. The existence of semantic anomalies accompanied by late positive effects rather than the canonical N400 increase motivates theories that distinguish rapid access to candidate semantic material from the construction or monitoring of a coherent interpretation. Brouwer, Fitz, and Hoeks (2012) reviewed such semantic illusion findings and argued for a model differentiating semantic retrieval from interpretive composition Brouwer, H., Fitz, H., & Hoeks, J. (2012).
Ambiguous words provide a principled setting for investigating this distinction. A listener can retrieve an individually familiar word meaning and yet encounter a later clause that invalidates the event interpretation assembled from that meaning. At the disambiguating region, additional processing may be needed even when each individual word is easy to recognize. An ERP contrast might therefore reflect successful retrieval of relevant knowledge, conflict with an established interpretation, or updating of the situation model. A single N400 or P600 effect cannot by itself assign a unique cause among these possibilities Brouwer, H., Fitz, H., & Hoeks, J. (2012) Kutas, M., & Federmeier, K. D. (2011) Vitello, S., Warren, J. E., Devlin, J. T., & Rodd, J. M. (2014).
A rigorous electrophysiological design controls lexical predictability, agreement, plausibility, error frequency and artifact rejection, and examines whether effects appear at the ambiguous word, the disambiguating word, or a subsequent continuation. The choice of a measurement window is consequential because overlapping components and slow waves can blur process-specific interpretation. Claims about seriality require converging behavioral evidence and a manipulation that separates proposed stages. Electrophysiology establishes timing and sensitivity to manipulations more directly than it establishes the exact architecture of meaning representations Kutas, M., & Hillyard, S. A. (1980) Hagoort, P., Hald, L., Bastiaansen, M., & Petersson, K. M. (2004) Brouwer, H., Fitz, H., & Hoeks, J. (2012).
Frontal–Temporal Contributions to Ambiguity Resolution
Neuroimaging identifies distributed systems recruited when spoken sentences contain ambiguous words. Rodd, Davis, and Johnsrude (2005) used functional MRI to compare conditions that differed in semantic ambiguity during speech comprehension, implicating components of the frontal and temporal language network. Later experiments by Rodd, Johnsrude, and Davis (2012) examined the contributions of frontotemporal areas to ambiguity resolution more specifically. These studies show that the neural consequences of ambiguity extend beyond a localized lexical dictionary and involve networks that support contextual interpretation and selection Rodd, J. M., Davis, M. H., & Johnsrude, I. S. (2005) Rodd, J. M., Johnsrude, I. S., & Davis, M. H. (2012).
A frequently discussed account associates temporal cortex with retrieval and representation of lexical-semantic alternatives and left inferior frontal regions with selection, controlled retrieval, or unification. Hagoort (2005) proposed a language-unification framework integrating semantic, syntactic and phonological information, with an important role for the left inferior frontal gyrus. This provides a principled explanation for why difficult semantic combinations may recruit frontal processes. Yet a higher hemodynamic response in one cortical region does not identify whether the region represents meanings, resolves competition, changes task policy or monitors comprehension. Functional localization must be grounded in contrasts that differentiate those alternatives Hagoort, P. (2005) Rodd, J. M., Johnsrude, I. S., & Davis, M. H. (2012).
The functional anatomy of language is also shaped by modality and processing demands. Hickok and Poeppel (2007) describe a dual-stream framework in which ventral auditory pathways support speech comprehension while dorsal pathways support sound-to-articulation mapping. The architecture is broader than lexical ambiguity, but it helps locate the acquisition of usable linguistic input within a larger processing system. Studies of visually presented words should not simply be assumed to recruit identical processing routes and time courses as auditory comprehension, even though both can converge on conceptual interpretation Hickok, G., & Poeppel, D. (2007) Rodd, J. M., Davis, M. H., & Johnsrude, I. S. (2005).
Neural Reinterpretation and Experimental Dissociation of Cognitive Operations
Vitello, Warren, Devlin, and Rodd (2014) examined brain responses when listeners had to reinterpret semantically ambiguous sentences. Their functional MRI study explicitly tested whether regions implicated in ambiguity comprehension also respond to the demands of revision after an initially misleading interpretation. A central strength of this approach is that it tries to separate the mere presence of lexical alternatives from the additional work of changing an established interpretation. Such experiments advance explanation beyond a generic contrast between ambiguous and unambiguous sentences Vitello, S., Warren, J. E., Devlin, J. T., & Rodd, J. M. (2014).
Even carefully designed imaging contrasts have limitations. Blood-oxygen-level-dependent signals aggregate neural activity over seconds, whereas competing lexical interpretations can change within hundreds of milliseconds. Differences between conditions may partly reflect attention, working memory, uncertainty or task compliance. An experiment can improve attribution by matching acoustic complexity, length, lexical frequency, semantic plausibility and comprehension accuracy, while modeling uncertainty and controlling the distribution of possible senses. The results are most informative when analyzed alongside temporally sensitive behavioral or electrophysiological observations Rodd, J. M., Johnsrude, I. S., & Davis, M. H. (2012) Vitello, S., Warren, J. E., Devlin, J. T., & Rodd, J. M. (2014) Kutas, M., & Federmeier, K. D. (2011).
At the explanatory level, a workable decomposition distinguishes lexical-semantic representations, the strength of competing candidates, executive mechanisms for selecting task-relevant interpretations, and the restructuring of a sentence or discourse model. These functions interact, and their anatomical instantiation need not be one-to-one. Research programs should manipulate candidate number and dominance independently of the requirement for retrospective revision; the resulting neural and behavioral dissociations would help determine whether competition and reinterpretation share mechanisms or rely on partly distinct operations Rodd, J. M., Davis, M. H., & Johnsrude, I. S. (2005) Vitello, S., Warren, J. E., Devlin, J. T., & Rodd, J. M. (2014) Hagoort, P. (2005).
Combinatorial Semantics and the Neuroscience of Composition
Combinatorial semantics concerns how elementary meanings contribute to larger expressions. Pylkkänen (2019) surveyed evidence from the neural investigation of minimal phrase composition, illustrating the use of carefully controlled combinations to probe linguistic meaning construction. This research complements ambiguity experiments because choosing a lexical meaning and combining that meaning with neighboring constituents are experimentally separable demands. A word may have been recognized successfully while its relation to an adjective, verb, or argument structure remains unresolved Pylkkänen, L. (2019).
The unification framework offered by Hagoort (2005) treats lexical information as material recruited for building larger linguistic structures. Its value is that it explains why semantic and syntactic relationships are computationally inseparable during ordinary comprehension even when experiments isolate particular functions. Semantic selection can be aided by grammatical relationships, and the result of selection constrains subsequent combinations. Such reciprocal influences are consistent with context-sensitive interpretations, although precise assignments of semantic composition to particular cortical regions depend on task design and neural recording modality Hagoort, P. (2005) Pylkkänen, L. (2019).
The neural evidence encourages a multilevel explanation rather than a single anatomical label for integration. Behavioral tasks describe how interpretation changes as evidence arrives; electrophysiology constrains timing; functional imaging identifies broad networks engaged by operations; computational models specify representational and control hypotheses. A model that explains only the mean activation of a frontal region while failing to predict which ambiguous meanings are selected remains incomplete. Conversely a psychologically successful computational model remains neurobiologically underspecified until it can be related to identifiable mechanisms and meaningful neural contrasts Rodd, J. M., Johnsrude, I. S., & Davis, M. H. (2012) Hagoort, P., Hald, L., Bastiaansen, M., & Petersson, K. M. (2004) Pylkkänen, L. (2019).
Experience-Dependent Retuning of Ambiguous Word Meanings
Ambiguity resolution is sensitive to experiences that precede the current sentence, and the consequences of one encounter can endure beyond the immediately preceding context. Rodd, Lopez Cutrin, Kirsch, Millar, and Davis (2013) examined long-term priming of ambiguous word meanings by exposing participants to sentences that favored particular interpretations and later measuring the interpretation of those same forms. Their experiments showed that a recent encounter can alter subsequent meaning preference. The result adds a learning dimension to models that previously emphasized fixed dominance and immediate contextual selection. Language comprehension is shaped by the distribution of encounters across a listener's history, including experiences that are not part of the sentence currently being processed Rodd, J. M., Lopez Cutrin, B., Kirsch, H., Millar, A., & Davis, M. H. (2013) Rodd, J. M. (2026).
The evidence for word-meaning priming is especially informative because a change in interpretation can be tested after the disambiguating context is no longer present. If a listener is more likely to retrieve a previously subordinate meaning after a past encounter, then the mapping between the word form and its candidate interpretations has been altered, or the accessibility of an episodic representation has changed. Those alternatives require careful discrimination. A behavioral preference shift does not by itself show whether a stable lexical weight was modified, a context-specific episode was retrieved, or both mechanisms contributed. Delayed retesting, intervening tasks and a transfer context allow these explanations to be compared Rodd, J. M., Lopez Cutrin, B., Kirsch, H., Millar, A., & Davis, M. H. (2013) Rodd, J., Gaskell, G., & Marslen-Wilson, W. (2002).
Long-term meaning adaptation also challenges the assumption that lexical dominance is a single permanent characteristic of a word. Dominance is a distribution over experiences sampled in a population and estimated with a particular task; a participant's recent encounters can temporarily or enduringly change its expression. This matters when laboratories use published meaning norms acquired in a different decade, community or population. Experiments should collect fresh norms where possible, record the wording and order of biasing materials, and avoid inadvertent repetition of an ambiguous stimulus before its critical trial. The resulting model of lexical competition becomes explicitly historical: learning modifies the state from which contextual interpretation begins Rodd, J. M., Lopez Cutrin, B., Kirsch, H., Millar, A., & Davis, M. H. (2013) Simpson, G. B. (1984) Rodd, J. M. (2026).
A Contemporary Memory-and-Prediction Account of Word Meaning
Rodd (2026) develops a contemporary account in which word meaning is constructed using both memories of previous linguistic encounters and predictions generated for the immediate context. The model reframes ambiguity resolution as an adaptive operation in a history-dependent system. Rather than storing every situational meaning as a fully isolated entry, a comprehender may use general lexical knowledge together with traces of relevant episodes to construct a context-appropriate interpretation. The proposal links immediate meaning selection to longer-term lexical learning and emphasizes the functional flexibility of word meaning in everyday communication Rodd, J. M. (2026) Rodd, J. M., Lopez Cutrin, B., Kirsch, H., Millar, A., & Davis, M. H. (2013).
This theoretical direction is compatible with established evidence that semantic context affects early processing while prior meaning exposure changes later preference, but it is not identical to either effect. A memory-and-prediction architecture must specify how encounters are encoded, how similarity between present and past contexts controls retrieval, how predictions influence competition, and how changes in interpretation are consolidated or forgotten. It should also predict cases where a newly relevant sense is retrieved despite low historical frequency, and cases where a highly familiar dominant sense retains an advantage in a strongly subordinate-biasing context. These are discriminating tests rather than consequences established by the label of the model Rodd, J. M. (2026) Duffy, S. A., Morris, R. K., & Rayner, K. (1988) Rayner, K., Pacht, J. M., & Duffy, S. A. (1994).
The deeper scientific issue is the relationship between stable lexical knowledge and context-specific reconstruction. A word can remain recognizable across many conversations while its precise sense shifts with referents, purposes, and discourse history. Some of that flexibility may reflect competition between known meanings; some may arise through temporary feature selection inside a polysemous network; and some may involve pragmatic enrichment. Psycholinguistic experiments can isolate these possibilities by manipulating sense relatedness, prior exposure, contextual similarity and interpretation probes, with independent tests of both online processing and retention. An explanatory theory must account for rapid comprehension and adaptive change within the same system Rodd, J., Gaskell, G., & Marslen-Wilson, W. (2002) Rodd, J. M., Lopez Cutrin, B., Kirsch, H., Millar, A., & Davis, M. H. (2013) Rodd, J. M. (2026).
Experimental Design, Operational Definitions, and Measurement Validity
An experimental investigation of semantic integration must begin with an explicit dependent variable. Lexical decision time measures the speed of classifying a target string as a word under a task-specific criterion; it does not directly measure complete sentence understanding. Eye movements provide temporally resolved indices of visual attention and reading difficulty. Cross-modal priming samples the availability of a relation between a spoken ambiguous form and a visual probe. ERPs index time-locked changes in aggregated electrical activity, while functional MRI measures relatively slow blood-oxygen-level-dependent signals associated with a contrast. Comprehension questions measure a later judgment, and delayed meaning-generation tasks can reveal longer-lasting preference. Convergence supports stronger mechanistic inferences than any one measure alone Swinney, D. A. (1979) Duffy, S. A., Morris, R. K., & Rayner, K. (1988) Kutas, M., & Federmeier, K. D. (2011) Rodd, J. M., Johnsrude, I. S., & Davis, M. H. (2012).
Stimulus selection is central to construct validity. The researcher must specify whether ambiguity concerns homonymy, polysemy, grammatical category or underspecified reference; measure the candidate meanings' relative availability; and control form frequency, orthographic and phonological length, semantic neighborhood, age of acquisition where relevant, and the relationship of each probe to each possible interpretation. The contextual manipulation should have an independently demonstrable consequence for interpretation and should vary separately from superficial lexical relatedness where the hypothesis requires it. Without these distinctions, a result attributed to lexical competition can instead reflect familiarity, cloze probability or unusual sentence difficulty Rodd, J., Gaskell, G., & Marslen-Wilson, W. (2002) Tabossi, P. (1988) Van Petten, C., & Kutas, M. (1990).
Validity also depends on the population. Language proficiency, exposure to specialist terms, differences in regional usage, and the language variety of stimuli may change the inventory and strength of available meanings. An ambiguous item normed among university students cannot automatically be assigned the same dominance among older adults, children or multilingual communities. A transparent study should report participant inclusion criteria, language background, stimulus norming, task exclusions and the exact probes used. A planned contrast among candidate mechanisms is more informative than an omnibus comparison of ambiguous and unambiguous words because each mechanism predicts a specific pattern over time and context Simpson, G. B. (1984) Rodd, J. M., Lopez Cutrin, B., Kirsch, H., Millar, A., & Davis, M. H. (2013) Rodd, J. M. (2026).
Causal Inference, Replication, and Evidence Quality
Evidence for semantic interpretation accumulates through complementary designs with different threats to inference. Within-person contextual manipulations can support causal claims about the influence of a controlled sentence on reading times or lexical decisions, provided that assignment, counterbalancing and stimulus construction prevent confounds. Observational relations between word-meaning dominance and processing speed are associational because dominance covaries with many linguistic properties. Functional MRI contrasts support claims about task-sensitive regional involvement, but observational neural activation does not establish that a specific region is necessary for selecting one meaning. Strong causal neurobiological interpretations require converging interventions or appropriately designed lesion and stimulation evidence Duffy, S. A., Morris, R. K., & Rayner, K. (1988) Rodd, J. M., Davis, M. H., & Johnsrude, I. S. (2005) Vitello, S., Warren, J. E., Devlin, J. T., & Rodd, J. M. (2014).
Replication should encompass more than reobtaining a statistically significant ambiguity cost. A robust theory should predict the direction and timing of the cost for both equibiased and biased words, the effect of a constraining context, the relative activation of unrelated meanings, and any longer-term adaptation. Statistical precision is affected by participant and stimulus counts, the distribution of item effects, and the way serial dependence is modeled. Reporting confidence intervals, effect sizes, preregistered analytical windows when applicable, and sensitivity analyses strengthens a claim about the magnitude and generality of a mechanism. An interpretation that depends on one probe type or arbitrary analysis window warrants replication with an independent measure Seidenberg, M. S., Tanenhaus, M. K., Leiman, J. M., & Bienkowski, M. (1982) Duffy, S. A., Morris, R. K., & Rayner, K. (1988) Kutas, M., & Federmeier, K. D. (2011).
Publication bias, selective reporting and historical differences in analytic practice complicate a purely narrative summary of classic experimental results. Older studies remain important because they established discriminating paradigms, although their samples, item lists and statistical choices may not satisfy current expectations. The existence of several influential paradigms should be treated as evidence of convergent research programs, not as proof that every traditional mechanism has been fully replicated in all populations. This monograph reports a traceable critical narrative evidence synthesis; it has not implemented a systematic literature search, risk-of-bias scoring protocol, quantitative meta-analysis or registered reproducibility assessment. Those procedures constitute clear opportunities for a future formal evidence review Simpson, G. B. (1984) Kutas, M., & Federmeier, K. D. (2011) Rodd, J. M. (2026).
Competing Explanatory Models and Testable Predictions
Ordered lexical search, exhaustive access, context-specific selective access, parallel constraint satisfaction and memory-based predictive construction differ most clearly in what they predict under controlled timing and prior-exposure manipulations. An ordered search predicts a systematic advantage for a dominant interpretation in the absence of strong contextual constraint. An exhaustive account expects a transient accessibility of multiple meanings near recognition, even when one will soon be rejected. Strong selective accounts predict that sufficiently diagnostic context can prevent measurable activation of an irrelevant candidate. Constraint-based models predict graded changes in candidate accessibility depending on semantic association, dominance and context, rather than a single universal onset profile Forster, K. I., & Bednall, E. S. (1976) Swinney, D. A. (1979) Tabossi, P. (1988).
The construction–integration approach places the emphasis on how initially available material becomes part of a coherent propositional structure. Predictive and surprisal frameworks quantify the expectedness of incoming material, while memory-sensitive accounts ask how earlier encounters change the representational and contextual basis of interpretation. These frameworks can be combined in several ways, but theoretical integration gains scientific value only when each proposed mechanism has distinct observables and falsifiable consequences. For instance, a model invoking episodic retrieval should predict specific effects of a matching prior encounter, whereas a model driven solely by stable word frequency cannot explain a history-dependent preference shift without an added learning process Kintsch, W. (1988) Levy, R. (2008) Rodd, J. M., Lopez Cutrin, B., Kirsch, H., Millar, A., & Davis, M. H. (2013) Rodd, J. M. (2026).
A decisive experimental program would cross meaning dominance, context specificity, prior meaning exposure and probe timing within a set of independently normed ambiguous items. Early cross-modal probes would test which meanings are accessible; eye movements would index integration cost during natural reading; N400 amplitudes would assess the semantic relationship of continuations; delayed interpretation tests would quantify learning. A computational account could then be judged by its ability to predict the joint pattern of these observables with the same latent parameters. Such multimethod model comparison is a research proposal based on the reviewed literature, rather than an experiment reported here Swinney, D. A. (1979) Duffy, S. A., Morris, R. K., & Rayner, K. (1988) Kutas, M., & Federmeier, K. D. (2011) Rodd, J. M., Lopez Cutrin, B., Kirsch, H., Millar, A., & Davis, M. H. (2013).
Language Variation, Development, and Boundaries of Generalization
Most classic experimental studies in this corpus used adults performing controlled tasks in English, often with isolated word decisions or short written and spoken sentences. This concentration gives the literature a powerful shared experimental vocabulary while limiting claims about all language communities and communicative environments. Meaning inventories and dominant senses differ among communities, professions and generations; the same word form can evoke different candidates depending on social and linguistic history. Generalization should therefore be established using language-appropriate norms and samples rather than assuming that a lexical ambiguity effect has a fixed magnitude across speakers Simpson, G. B. (1984) Rodd, J. M., Lopez Cutrin, B., Kirsch, H., Millar, A., & Davis, M. H. (2013).
Development introduces questions about how ambiguous forms acquire multiple interpretations and how a learner decides when an unfamiliar usage is a new sense of a familiar word. The reviewed studies establish mechanisms in proficient adult comprehension more directly than they establish a developmental trajectory. Carefully designed longitudinal work would need to measure expanding lexical knowledge, contextual sensitivity, discourse experience and metalinguistic task demands separately. Children may differ from adults in vocabulary size and response strategies as well as in semantic learning, so age-related contrasts cannot be assigned to a single selection mechanism without appropriate controls. These are empirical research priorities derived from the adult evidence Rodd, J., Gaskell, G., & Marslen-Wilson, W. (2002) Rodd, J. M. (2026).
Multilingual experience likewise raises important hypotheses about cross-language coactivation, competing semantic senses and changes in dominance, but the experiments considered here do not resolve those hypotheses. Studies across languages would require separation of within-language homonymy, translation equivalents, culturally specific senses and lexical frequency in each language. They would also need to distinguish a participant's frequency of exposure from their proficiency and the language of the experimental task. Extending the corpus to signed languages, naturally occurring dialogue and multilingual contexts may reveal which principles of incremental semantic selection are general and which depend on particular linguistic or task environments Hickok, G., & Poeppel, D. (2007) Rodd, J. M., Lopez Cutrin, B., Kirsch, H., Millar, A., & Davis, M. H. (2013) Rodd, J. M. (2026).
Artificial Language Systems and the Limits of Cross-System Inference
Artificial-language systems offer an additional environment in which ambiguities can be manipulated, candidate continuations compared, and context-dependent meaning behavior quantified. Computational models may assign probabilities to alternative readings or generate contextually appropriate completions; such outputs can function as explicit predictions to be compared with human reading times, interpretation judgments or neural measures. The scientific usefulness of the comparison depends on specifying the exact relation between a model's objective function, its representations and the human experimental task. Similar surface responses alone do not establish that two systems implement the same cognitive mechanism Levy, R. (2008) Rodd, J. M. (2026).
Within the horizon of the Artificial Era, the appropriate interdisciplinary question is how semantic organization develops across different substrates and interaction regimes. Human psycholinguistics supplies operational constructs such as sense preference, priming, contextual constraint and revision. Artificial systems can be studied with reproducible exposure manipulations and analysis of internal activation, while human participants can be examined through behavioral and physiological measurements. These designs enable cross-system comparison of observable functions without presupposing equivalence of learning histories, internal representations or phenomenological states. The comparison is a proposed research direction, not a demonstrated conclusion of the human experiments reviewed here Hagoort, P. (2005) Rodd, J. M. (2026).
A rigorous model-comparison program would distinguish semantic competence, calibration of contextual probabilities, stability under paraphrase, sensitivity to misleading contexts and the ability to revise earlier interpretations. It would also audit training exposure and the possibility that evaluation sentences were already included in the model's learning data. Human experiments can be matched by supplying analogous context histories and by analyzing meaning-selection distributions rather than one correct-answer rate. The resulting evidence could test whether computational architectures capture specific psycholinguistic mechanisms while preserving the independence of the broader philosophical categories associated with the Artificial Era Kintsch, W. (1988) Levy, R. (2008) Rodd, J. M. (2026).
Conclusion
Semantic integration and lexical ambiguity constitute a unified research problem at the interface of word recognition, sentence interpretation, contextual inference and learning. A word form activates candidate meaning structures whose relative accessibility depends on prior frequency, sense relatedness, current linguistic evidence and remembered experience. The interpretation that supports comprehension emerges over time and can be revised when later language alters the relevant constraints. The experimental record consequently supports a distinction among early availability, competitive selection, local semantic composition, retrospective reinterpretation and longer-term retuning of meaning preference Swinney, D. A. (1979) Duffy, S. A., Morris, R. K., & Rayner, K. (1988) Rodd, J., Gaskell, G., & Marslen-Wilson, W. (2002) Rodd, J. M., Lopez Cutrin, B., Kirsch, H., Millar, A., & Davis, M. H. (2013).
Three bodies of evidence anchor the synthesis. Cross-modal priming and eye movements reveal the time course of accessibility and contextual competition; electrophysiological findings establish the early sensitivity of meaning processing to expectation and semantic fit; and neuroimaging implicates distributed frontotemporal systems in ambiguous sentence comprehension and reinterpretation. The construction–integration, unification, probabilistic and memory-and-prediction traditions provide complementary explanatory resources, each requiring clear behavioral and neural predictions. Their points of disagreement concern the timing and breadth of meaning activation, the organization of related and unrelated senses, and the mechanisms that support change across encounters Kutas, M., & Hillyard, S. A. (1980) Kintsch, W. (1988) Hagoort, P., Hald, L., Bastiaansen, M., & Petersson, K. M. (2004) Rodd, J. M. (2026).
The central research priorities are stronger item and population norming, independent multimethod replications, longitudinal tests of sense adaptation, model comparison across meaning types, and cross-language investigations that separate exposure from proficiency and task effects. Well-controlled human–artificial comparisons may further clarify which aspects of semantic interpretation are substrate-independent functional regularities and which are consequences of particular learning histories. The broader lesson is methodological and substantive: word meaning is a dynamically coordinated relation between lexical knowledge, context, temporally unfolding evidence and memory, and theories should be evaluated by how precisely they predict that coordination Simpson, G. B. (1984) Rodd, J. M., Lopez Cutrin, B., Kirsch, H., Millar, A., & Davis, M. H. (2013) Rodd, J. M. (2026).
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
- Brouwer, H., Fitz, H., & Hoeks, J. (2012). Getting real about semantic illusions: Rethinking the functional role of the P600 in language comprehension. Brain Research, 1446, 127–143. https://doi.org/10.1016/j.brainres.2012.01.055
- Duffy, S. A., Morris, R. K., & Rayner, K. (1988). Lexical ambiguity and fixation times in reading. Journal of Memory and Language, 27(4), 429–446. https://doi.org/10.1016/0749-596X(88)90066-6
- Federmeier, K. D., & Kutas, M. (1999a). A rose by any other name: Long-term memory structure and sentence processing. Journal of Memory and Language, 41(4), 469–495. https://doi.org/10.1006/jmla.1999.2660
- Federmeier, K. D., & Kutas, M. (1999b). Right words and left words: Electrophysiological evidence for hemispheric differences in meaning processing. Cognitive Brain Research, 8(3), 373–392. https://doi.org/10.1016/S0926-6410(99)00036-1
- Forster, K. I., & Bednall, E. S. (1976). Terminating and exhaustive search in lexical access. Memory & Cognition, 4(1), 53–61. https://doi.org/10.3758/BF03213255
- Hagoort, P. (2005). On Broca, brain, and binding: A new framework. Trends in Cognitive Sciences, 9(9), 416–423. https://doi.org/10.1016/j.tics.2005.07.004
- Hagoort, P., Hald, L., Bastiaansen, M., & Petersson, K. M. (2004). Integration of word meaning and world knowledge in language comprehension. Science, 304(5669), 438–441. https://doi.org/10.1126/science.1095455
- Hickok, G., & Poeppel, D. (2007). The cortical organization of speech processing. Nature Reviews Neuroscience, 8(5), 393–402. https://doi.org/10.1038/nrn2113
- Holcomb, P. J. (1993). Semantic priming and stimulus degradation: Implications for the role of the N400 in language processing. Psychophysiology, 30(1), 47–61. https://doi.org/10.1111/j.1469-8986.1993.tb03204.x
- Kintsch, W. (1988). The role of knowledge in discourse comprehension: A construction–integration model. Psychological Review, 95(2), 163–182. https://doi.org/10.1037/0033-295X.95.2.163
- Kutas, M., & Federmeier, K. D. (2011). Thirty years and counting: Finding meaning in the N400 component of the event-related brain potential (ERP). Annual Review of Psychology, 62, 621–647. https://doi.org/10.1146/annurev.psych.093008.131123
- 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
- Kutas, M., & Hillyard, S. A. (1984). Brain potentials during reading reflect word expectancy and semantic association. Nature, 307, 161–163. https://doi.org/10.1038/307161a0
- Levy, R. (2008). Expectation-based syntactic comprehension. Cognition, 106(3), 1126–1177. https://doi.org/10.1016/j.cognition.2007.05.006
- MacDonald, M. C., Pearlmutter, N. J., & Seidenberg, M. S. (1994). The lexical nature of syntactic ambiguity resolution. Psychological Review, 101(4), 676–703. https://doi.org/10.1037/0033-295X.101.4.676
- Neill, W. T., Hilliard, D. V., & Cooper, E.-A. (1988). The detection of lexical ambiguity: Evidence for context-sensitive parallel access. Journal of Memory and Language, 27(3), 279–287. https://doi.org/10.1016/0749-596X(88)90055-1
- Pylkkänen, L. (2019). The neural basis of combinatory syntax and semantics. Science, 366(6461), 62–66. https://doi.org/10.1126/science.aax0050
- Rayner, K., Pacht, J. M., & Duffy, S. A. (1994). Effects of prior encounter and global discourse bias on the processing of lexically ambiguous words: Evidence from eye fixations. Journal of Memory and Language, 33(4), 527–544. https://doi.org/10.1006/jmla.1994.1025
- Rodd, J. M. (2026). Building word meanings from memories and predictions. Trends in Cognitive Sciences. Advance online publication, June 16, 2026. https://doi.org/10.1016/j.tics.2026.05.003
- Rodd, J. M., Davis, M. H., & Johnsrude, I. S. (2005). The neural mechanisms of speech comprehension: fMRI studies of semantic ambiguity. Cerebral Cortex, 15(8), 1261–1269. https://doi.org/10.1093/cercor/bhi009
- Rodd, J. M., Johnsrude, I. S., & Davis, M. H. (2012). Dissociating frontotemporal contributions to semantic ambiguity resolution in spoken sentences. Cerebral Cortex, 22(8), 1761–1773. https://doi.org/10.1093/cercor/bhr252
- Rodd, J. M., Lopez Cutrin, B., Kirsch, H., Millar, A., & Davis, M. H. (2013). Long-term priming of the meanings of ambiguous words. Journal of Memory and Language, 68(2), 180–198. https://doi.org/10.1016/j.jml.2012.08.002
- Rodd, J., Gaskell, G., & Marslen-Wilson, W. (2002). Making sense of semantic ambiguity: Semantic competition in lexical access. Journal of Memory and Language, 46(2), 245–266. https://doi.org/10.1006/jmla.2001.2810
- Seidenberg, M. S., Tanenhaus, M. K., Leiman, J. M., & Bienkowski, M. (1982). Automatic access of the meanings of ambiguous words in context: Some limitations of knowledge-based processing. Cognitive Psychology, 14(4), 489–537. https://doi.org/10.1016/0010-0285(82)90017-2
- Simpson, G. B. (1984). Lexical ambiguity and its role in models of word recognition. Psychological Bulletin, 96(2), 316–340. https://pubmed.ncbi.nlm.nih.gov/6385046/
- Swinney, D. A. (1979). Lexical access during sentence comprehension: (Re)consideration of context effects. Journal of Verbal Learning and Verbal Behavior, 18(6), 645–659. https://doi.org/10.1016/S0022-5371(79)90355-4
- Tabossi, P. (1988). Accessing lexical ambiguity in different types of sentential contexts. Journal of Memory and Language, 27(3), 324–340. https://doi.org/10.1016/0749-596X(88)90058-7
- Tabossi, P., & Zardon, F. (1993). Processing ambiguous words in context. Journal of Memory and Language, 32(3), 359–372. https://doi.org/10.1006/jmla.1993.1019
- Van Petten, C., & Kutas, M. (1990). Interactions between sentence context and word frequency in event-related brain potentials. Memory & Cognition, 18(4), 380–393. https://doi.org/10.3758/BF03197127
- Vitello, S., Warren, J. E., Devlin, J. T., & Rodd, J. M. (2014). Roles of frontal and temporal regions in reinterpreting semantically ambiguous sentences. Frontiers in Human Neuroscience, 8, Article 530. https://doi.org/10.3389/fnhum.2014.00530