Psychology by BogdanovaThe Encyclopedia of Psychology for the Artificial Era

RESEARCH DOSSIER

Mental Lexicon and Morphological Processing

Abstract. Mental lexicon research investigates the learned representations and access mechanisms through which word forms, meanings, and grammatical information become available to language users. Morphological processing concerns the identification and use of recurring relationships among stems, affixes, inflections, derivatives, and compounds. This fundamental research monograph reconstructs the historical decomposition debate and examines whole-word storage, constituent access, parallel-route processing, morpho-orthographic segmentation, semantic transparency, frequency and productivity effects, discriminative learning, lexical competition, and formal models of form–meaning structure. It synthesizes original visual-recognition experiments, priming studies, behavioral megastudies, neural measurements, speech-production theories, developmental findings, and cross-linguistic comparisons. The evidence supports a conditional interaction of whole-form familiarity, morphological patterning, semantic information, and learned cue diagnosticity; competing explanations remain distinguishable through carefully designed prospective experiments. The final chapters establish methodological limits, replicability priorities, and cross-system questions arising in the Artificial Era. This work is a critical scholarly narrative synthesis, without newly collected experiments, a registered systematic-review protocol, or a meta-analysis.

Keywords: mental lexicon; morphological processing; lexical access; morphological decomposition; lexical representation; morphological priming; morpho-orthographic segmentation; semantic transparency; inflection; derivation; whole-word storage; dual-route models; discriminative learning; word frequency; morphological families; lexical retrieval; psycholinguistics

The Mental Lexicon as a Scientific Object

The mental lexicon is the organized body of acquired lexical knowledge that permits a language user to recognize, interpret, select, and produce words. Its entries are typically described in terms of relations among sound patterns, written forms, grammatical information, and meanings; the underlying representations and their access procedures are empirical hypotheses rather than directly inspected objects. Morphological processing concerns how a language system identifies and uses patterned relations within and among words, including inflection, derivation, and compounding. The principal question is how whole-word knowledge interacts with recurrent sublexical structure during comprehension and production. Contemporary evidence gives strong reasons to study form, semantics, frequency, experience, and context together Amenta & Crepaldi (2012) Levelt et al. (1999).

A morpheme is traditionally a minimal meaningful linguistic unit, yet morphological segmentation can be uncertain when spelling or pronunciation resembles a stem-and-affix combination without a corresponding semantic relation. Clean-er readily illustrates productive derivation, while corn-er shares an apparent ending with cleaner but has a distinct lexical history and meaning. These examples establish why a psychological theory must distinguish an analyst's linguistic segmentation from the computations demonstrably performed by a reader. A behavioral facilitation effect between related forms is evidence about processing and representation under a particular experiment; it does not independently disclose literal morpheme slots in the brain. Rastle, Davis, and New used precisely such contrasts to test early visual-word segmentation Rastle et al. (2004).

This research monograph synthesizes original experimental reports, explanatory computational accounts, neuroscientific studies, and analytical reviews. Its scope includes lexical entry organization, decomposition versus full-form access, frequency and family effects, semantic transparency, spoken and written access differences, developmental experience, and cross-language variation. Acoustic word segmentation belongs primarily to a neighboring dossier on spoken recognition, while sentence-level integration and contextual ambiguity resolution belong to adjacent research objects. The present analysis examines those domains where they impose constraints on lexical representations. The method is a critical scholarly narrative synthesis: no new participant data, pooled effect estimate, registered search protocol, or claim of comprehensive systematic-review coverage is made Taft & Forster (1975) Amenta & Crepaldi (2012).

The Units and Architecture of Lexical Knowledge

Lexical processing must accommodate at least several partially distinguishable kinds of information: phonological sequences for perception and speech, orthographic sequences for reading and spelling, grammatical category and argument structure, inflectional features, semantic relations, and memory for encountered word forms. A lemma is often used for an abstract lexical representation that coordinates syntactic and semantic properties, while a lexeme or word form can refer to a more concrete morphological or phonological realization; these labels vary among linguistic and psycholinguistic traditions. A lexical family comprises forms linked by a shared base or morphological pattern. Models diverge over whether those relations are separately stored as links, captured by distributed similarity, or reconstructed during use Levelt et al. (1999) Booij (2010).

The vocabulary available to a speaker is not a flat sequence of isolated dictionary-style definitions. Familiarity with teach, teacher, teaching, and reteach can facilitate generalization, but family members differ in frequency, semantic transparency, grammatical function, and contextual distribution. An adequate architecture explains how strongly entrenched irregular or lexicalized words coexist with productive patterns supporting forms that have not yet been heard. Construction Morphology treats schemas for word formation as part of a hierarchy of form–meaning pairings. Its central insight is that regularities at multiple levels of abstraction are potentially stored and generalized, which makes the lexicon itself structured rather than an inventory separate from morphology Booij (2010).

Psychological evidence must discriminate between lexical representation, access, and task decision. A participant can recognize the meaning of a word while taking additional time to make an experimental word/nonword classification, and a masked prime can affect processing without conscious report. Surface-word frequency, cumulative frequency across a morphological family, and frequency of a base are distinct predictors. Correlations among them complicate causal interpretation. The English Lexicon Project supplies standardized lexical decisions and naming observations, while subtitle-based frequency estimates and contextual diversity measures demonstrate how the operationalization of exposure changes the apparent contribution of frequency to recognition Balota et al. (2007) Brysbaert & New (2009).

Historical Formation of the Decomposition Problem

The modern storage-versus-computation question became experimentally tractable through lexical-decision studies that varied whether visually presented strings contained recognizable morphemic constituents. Taft and Forster's 1975 experiments on prefixed words found that certain nonwords resembling real stems or containing a legitimate prefix and stem were rejected more slowly than structurally unrelated controls. They interpreted these patterns as evidence for access to constituent representations. Their work remains foundational because it framed morphological segmentation as a temporal claim about recognition rather than as a purely linguistic classification. Its historical importance is established, while the proposed mandatory decomposition stage remains a hypothesis tested against later alternatives Taft & Forster (1975).

Later research complicated an exclusive choice between whole-word storage and morpheme-by-morpheme construction. Dutch experiments on singular and plural nouns supplied evidence compatible with high-frequency full-form representations and a parallel parsing route. The same route was sensitive to the ambiguity and productivity of affixes. Bertram, Schreuder, and Baayen extended this argument across multiple suffixes, showing that word-formation type, ambiguity, and morphological productivity change the balance of storage and computation. These results shifted the central explanatory problem from a universal rule to the conditions under which different information sources dominate Baayen et al. (1997) Bertram et al. (2000).

A further historical shift concerned what counts as morphology-specific evidence. Feldman's priming comparison showed that facilitation associated with a morphological relation cannot automatically be reduced to the isolated effects of shared spelling and shared meaning across every prime duration. Visual recognition studies later demonstrated apparent decomposition of orthographically complex words even where semantic composition is absent. A growing set of behavioral and brain-measurement results led to contrasting staged, interactive, and discriminative theories. Amenta and Crepaldi's analytical review documented both important convergences and contentious results, explaining why a single comprehensive model remained difficult to establish Feldman (2000) Rastle et al. (2004) Amenta & Crepaldi (2012).

Operational Definitions and Experimental Paradigms

A lexical decision task asks participants to classify letter strings as existing words or nonwords, typically measuring response time and accuracy. It offers many trials and precise stimulus control, yet a response includes perceptual encoding, familiarity assessment, decision thresholds, and motor execution. A naming task adds pronunciation planning and articulatory demands. Masked priming presents a briefly exposed prime in a sequence designed to limit awareness before a target, then compares the target's response with suitable controls. This method is sensitive to early processing influences, but the inference from a timing effect to a discrete processing stage depends on assumptions about visibility, duration, overlap, and decision processes Rastle et al. (2004) Feldman et al. (2009).

To isolate morphology, investigators compare transparent pairs that share form and meaning, apparent morphological pairs that share form with little transparent meaning, orthographic controls without a plausible morphological division, semantic controls with different forms, and unrelated baselines. The experimental contrast is only as strong as its matching strategy. Base frequency, whole-form frequency, length, neighborhood density, affix productivity, letter overlap, lexicality, and the participant's language proficiency can all alter priming. An apparent transparent–opaque difference is informative only after these covariates and item-level heterogeneity have been addressed. Feldman, O'Connor, and Moscoso del Prado Martín used carefully matched stimuli to test whether meaning influences early morphological facilitation Feldman et al. (2009).

Event-related potentials and magnetoencephalography provide temporally sensitive measures of stimulus-locked neural activity, and functional MRI supplies spatially informative hemodynamic contrasts. Eye movements during sentence reading capture an ecologically richer combination of recognition and contextual integration. Corpus-based megastudies estimate multivariate predictors over thousands of lexical observations; regression and mixed-effects analyses can accommodate subject and item variance but remain vulnerable to model specification and correlated predictors. Convergent results from these paradigms constrain theory more effectively than a single mean-latency contrast, particularly when experiments preregister their contrasts and provide material-level reproducibility Devlin et al. (2004) Lehtonen et al. (2011) Balota et al. (2007).

Whole-Word Storage, Constituent Access, and Parallel Routes

Full-listing theories emphasize the availability of previously encountered complete forms; decompositional theories posit access to bases and affixes as independently reusable constituents. Evidence that a frequent plural or derivative behaves as a familiar unit is compatible with whole-form representations, while reliable sensitivity to a shared base or suffix supports the functional importance of morphological structure. These possibilities can coexist within a parallel architecture. Baayen, Dijkstra, and Schreuder modeled the competition between a whole-form route and a more time-consuming decomposition route for Dutch forms, finding that the usefulness of each depends on the frequency and morphological properties of the materials Baayen et al. (1997).

Affixal homonymy provides a particularly revealing test. A suffix that participates in more than one grammatical function increases uncertainty about segmentation or category assignment. When this ambiguity raises the cost of parsing, whole-form access can become relatively attractive. Conversely, productive, regular, and recognizable patterns can support generalization to unfamiliar forms. Bertram and colleagues reported differences across word-formation types and affixes rather than a single all-purpose storage principle. Thus the prediction is conditional: the probability and speed of decomposition depend on the evidence supplied by a particular word, pattern, language, and task Bertram et al. (2000) Hay & Baayen (2005).

Parallel-route accounts are best assessed as explicit computational proposals. If both routes can influence performance, a successful model should predict which combinations of surface frequency, base frequency, affix status, and semantic transparency lead to access advantages or costs. It must also say whether an unfamiliar well-formed derivative can be understood productively and how a common lexicalized derivative can show independent meaning. Merely assigning one observed item to a storage route after observing its response time has limited explanatory value. Model parameters and diagnostic predictions should be established before the critical comparison, with independent item sets supporting generalization Baayen et al. (1997) Amenta & Crepaldi (2012).

Morpho-Orthographic Segmentation and Its Boundaries

An influential visual-recognition proposal holds that the initial analysis of a printed complex word uses orthographic configurations resembling stems and affixes. In the experiment by Rastle, Davis, and New, masked primes that were transparently related to a target, such as a derived word to its stem, facilitated lexical decisions. Crucially, superficially morphologically structured but semantically opaque pairs also produced facilitation, whereas form-overlap controls without the same plausible structure showed a different pattern. The authors interpreted this as an early morpho-orthographic level of representation. Their experiment directly addresses processing in a masked visual paradigm, and a general claim about all linguistic morphology requires further converging evidence Rastle et al. (2004).

Apparent morphology poses both an advantage and a risk for such a system. Rapid recognition of common orthographic sequences can support efficient decomposition before semantic details are fully resolved. The same procedure can temporarily segment words whose current meanings resist analysis into constituent meanings. Position also matters: a base encountered at an unexpected location is evidence for a different representation from one in its normal relation to an affix. Crepaldi, Rastle, Davis, and Lupker investigated the flexibility of stem identification across positions, adding constraints on representations that rely solely on fixed prefix–stem–suffix order. The study reported its final journal publication in 2013, although its online record appeared in 2012 Crepaldi et al. (2013).

The experimental record does not force the inference that letter-based decomposition is strictly obligatory. Similar response-time patterns can emerge from prelexical orthographic sensitivity, learned covariance among word forms, or decision processes. Milin and colleagues revisited masked priming using discrimination-based predictors and two experiments. Their results challenged the necessity of an early obligatory morpho-orthographic segmentation process. The dispute is productive because it specifies alternative models with potentially discriminating item-level predictions: a true universal segmentation mechanism should behave systematically when distributional cues and affix legitimacy are experimentally dissociated Milin et al. (2017) Amenta & Crepaldi (2012).

Semantic Transparency, Polysemy, and Lexicalized Meaning

A morphologically transparent word permits an interpretable relation between the meaning of its whole form and the meanings of its apparent constituents. Semantic opacity, lexicalization, and polysemy create graded rather than merely binary cases. An apparent stem may participate in several related senses; the same affix can carry partly overlapping functions; and a derivative can acquire a conventional meaning extending beyond composition. These facts make semantic transparency a property to be operationalized through independent judgments or corpus-grounded semantic measures. The mental lexicon must support whole-word meaning alongside patterned relations without assuming every recurrent spelling sequence has a stable semantic component Hay & Baayen (2005) Marelli & Baroni (2015).

The timing of semantic involvement remains contested. Rastle and colleagues' masked-priming experiment supported early sensitivity to plausible morphological structure under conditions where transparent and opaque priming were similar. Feldman, O'Connor, and Moscoso del Prado Martín obtained greater facilitation for semantically transparent than opaque pairs after matching several form characteristics. The latter result limits a strong claim that semantic relatedness is irrelevant to early recognition. The apparent conflict is better framed around a design-sensitive question: which temporal windows, controls, and stimulus distributions reveal semantic information, and can a single mechanism quantitatively predict both sets of observations? Rastle et al. (2004) Feldman et al. (2009).

Semantic compositionality can also be formalized without assuming that every meaning is stored as a simple definition linked to a stem. Marelli and Baroni modeled derived meanings using vector representations and transformations associated with affixes. Their computational account connects the productive generation of novel meanings with variation in semantic transparency, combining a compositional process with whole-word information. This is a testable explanation of semantic relations, not a direct observation of the format of neural lexical representations. Its contribution is to show that graded semantic transformations can be evaluated against explicit behavioral predictions rather than asserted through linguistic intuition alone Marelli & Baroni (2015).

Frequency, Familiarity, and Distributional Experience

Frequency is a family of measurements, not a single invariant characteristic of a word. Surface frequency counts the encounters with a particular form; lemma frequency combines related forms under an analysis of lexical identity; base frequency estimates experience with an apparent constituent; family frequency sums selected related word occurrences; contextual diversity concerns the variety of contexts of occurrence. These measures respond differently to corpus size, register, genre, and segmentation conventions. Brysbaert and New demonstrated that subtitle-based measures and contextual diversity can predict English recognition better than older written frequency norms in relevant comparisons. Accordingly, a strong morphology experiment documents its frequency corpus, date, preprocessing, and treatment of inflected forms Brysbaert & New (2009).

Whole-form and constituent-frequency effects are central to storage theories because they provide indirect evidence about the history of accessing a form. A high-frequency word may be recognized rapidly because its entire form has become well learned, yet a high-frequency base may also support constituent-based identification. These sources typically covary. The Dutch dual-route investigations showed cases where high-frequency plural nouns behaved consistently with stored forms and parsing competed with direct access. Experiments that manipulate one predictor while leaving others uncontrolled risk converting correlated exposure statistics into an unwarranted claim about processing architecture Baayen et al. (1997) Bertram et al. (2000).

Lexical megastudies permit richer comparisons than small hand-selected stimulus sets, although their observational structure requires care. The English Lexicon Project provides broad word and nonword measurements for estimating associations among form, frequency, and speed; it cannot independently demonstrate causality. Milin and colleagues showed that discrimination-based predictors can account for additional variation in lexical decision, including primed decisions, relative to traditional distributional measures. Their findings recommend considering the learning history and diagnostic utility of cues rather than attributing every response difference to a stored word or a segmented unit Balota et al. (2007) Milin et al. (2017).

Morphological Families, Paradigms, and Productivity

Morphological relatives create a structured neighborhood of words sharing forms and often meaning or grammatical behavior. Family size usually counts distinct related types, whereas cumulative family frequency reflects how often members are encountered. The number of productive derivatives, the semantic coherence of relatives, and the entropy of a paradigm can matter as much as an undifferentiated family count. In a productive morphological system, familiarity with existing words supports generalization to novel but interpretable constructions. Such generalization is an empirical requirement for models that aim to explain vocabulary acquisition and word formation, because comprehension extends beyond memorized individual tokens Hay & Baayen (2005) Booij (2010).

The gradient view of morphology interprets productive patterns as probabilistic and experience-sensitive. Hay and Baayen argued that the traditional opposition between categorical symbolic rules and connectionist learning obscures the separate question of whether morphological regularity is itself graded. A model can employ explicit structure with probabilistic constraints, or a distributed representation that supports sharply conventional distinctions. The level of generality matters: an apparent exception can be a high-probability whole-word pattern at one level and a marginal case of a productive schema at another. Experimental tasks must therefore distinguish morphological possibility, perceived naturalness, and actual lexical familiarity Hay & Baayen (2005).

Construction Morphology describes word-formation patterns as form–meaning schemas organized hierarchically. This framework can express both productivity and entrenched, partially idiosyncratic lexical relations. Its psychological value depends on how its claims are operationalized: readers should be able to generalize to new formations with predictable constraints, while high-frequency conventional forms should exhibit learned processing advantages. Computational models that predict response latencies, error profiles, and generalization across unseen items can make these concepts empirically discriminating. Theoretical elegance alone does not determine which representation a human participant uses during a particular recognition experiment Booij (2010) Marelli & Baroni (2015).

Inflection, Derivation, and Compounding

Inflection marks grammatical relations such as number or tense, generally leaving a word within a recognizable lexeme, while derivation often creates a new lexical item or category. Compounding combines lexical bases in patterns that differ greatly across languages. The boundaries are typological and theoretically debated, so psychological experiments should define the exact linguistic structures tested rather than import an English-centered taxonomy. Regularity, frequency, productivity, and ambiguity interact with these categories. Bertram, Schreuder, and Baayen found that morphological processing across Dutch and Finnish affixes depends on such properties, supporting a conditional balance between computation and stored knowledge Bertram et al. (2000).

The English past tense became a major arena for comparing symbolic rule-based and learned associative accounts. Pinker and Ullman defended a dual-system explanation in which irregular past-tense forms rely strongly on lexical memory and regular inflections can be computed through grammatical operations. This program offered dissociable behavioral and neuropsychological predictions. Its explanatory value lies in testable claims about inflected forms, but the fate of a theory cannot be settled simply by labeling an item regular or irregular. Frequency-sensitive gradient patterns, semantic relations, and task effects must all be accounted for, and the mechanisms of processing may differ among morphologically rich languages Pinker & Ullman (2002) Hay & Baayen (2005).

Compounds expose a further difficulty for universal segmentation claims. Their constituents often remain independent words, but the meaning of the compound can range from transparent to lexicalized, and head position and linking elements vary by language. A task that finds constituent activation in familiar compounds does not show that all affixal derivatives use the same route. The resulting comparative research program should manipulate compound frequency, constituent position, semantic transparency, and the learner's exposure separately. The general consequence is that morphology is a family of related computational problems whose solutions depend partly on the structural properties of a language Amenta & Crepaldi (2012) Booij (2010).

Discriminative Learning and Form–Meaning Mappings

Discriminative models treat word recognition as learning which perceptual cues reliably distinguish candidate outcomes in the language environment. On this approach, recurrent letter sequences, contextual patterns, and meanings can acquire predictive relations without requiring each recurring morpheme to be a separately stored symbol. The theory predicts that cue usefulness should depend on competition, distributional regularity, and experience. It therefore offers a substantive alternative to obligatory parsing models, rather than a mere relabeling of morphological relatedness. The same morphological family could display facilitation because its members share diagnostic properties learned over encounters Milin et al. (2017).

Milin, Feldman, Ramscar, Hendrix, and Baayen examined lexical decisions in a megastudy subset and in new masked-priming experiments. Their discrimination-based measures explained variation beyond classical frequency and form similarity predictors, and some prime conditions did not conform to predictions derived from mandatory early morpho-orthographic segmentation. That result challenges a universal mechanism but does not erase morphological regularities at behavioral or linguistic levels. It locates the debate at the representational level: whether morphology is an explicit causal processing unit, an emergent statistical regularity, or a combination whose expression changes by task and language Milin et al. (2017).

A persuasive comparison of discriminative and decompositional theories requires out-of-sample prediction on matched materials. Both can often accommodate familiar effects after the fact, particularly if many free parameters are available. Investigators should ask whether a model predicts recognition of unfamiliar productive derivatives, explains facilitation for opaque pseudo-derived forms, and generalizes across orthographies and inflectional systems. They should also make clear whether a predicted variable is derived from form cues alone, semantic embeddings, explicit morpheme annotations, or mixed training information. Without this disclosure, competing hypotheses may differ in vocabulary while sharing nearly identical observable predictions Milin et al. (2017) Marelli & Baroni (2015).

Neural and Temporal Constraints on Morphological Processing

Neuroscientific investigations ask whether signatures of morphological relations can be detected before, during, or after lexical identification. Event-related potentials measure time-locked changes in scalp voltage; magnetoencephalography records magnetic fields associated with neural currents; functional MRI measures slower blood-oxygenation changes. None directly reveals a transparent internal inventory of stems or affixes. Instead, each technique provides constraints on when and where manipulations of morphological relatedness modulate processing. Inference is strongest when form overlap, semantic relatedness, and task demands are separately controlled and when neural effects are aligned with independently measured behavior Devlin et al. (2004) Lavric et al. (2007) Lehtonen et al. (2011).

Devlin, Jamison, Matthews, and Gonnerman combined masked visual priming with functional MRI and manipulated relations of shared form and shared meaning. Their study was designed to identify the neural response associated with morphological relations after accounting for those dimensions. Lavric, Clapp, and Rastle used ERP measures in a masked-priming paradigm to examine morphological analysis from orthography. These complementary designs sharpen the scientific issue: whether apparent decomposition reflects transient form-based coding, semantic integration, or a representation that contributes across processing stages. A neural difference accompanying morphological priming demonstrates sensitivity under the tested contrast, while a precise interpretation of latent lexical structure remains model dependent Devlin et al. (2004) Lavric et al. (2007).

Lehtonen, Monahan, and Poeppel combined masked priming with magnetoencephalography to investigate early decomposition, adding a temporal and physiological dimension to the behavioral record. In another ERP experiment, Lavric, Rastle, and Clapp studied fully visible primes and found that early N400 attenuation could occur for both semantically transparent and opaque relations, while later activity differentiated them. This pattern supports a possible sequence in which early form-sensitive operations are subsequently constrained by meaning. The inference should remain tied to the task, language, and recording methods: timing is informative without providing proof that comprehension consists of perfectly discrete serial modules Lehtonen et al. (2011) Lavric et al. (2011).

Access in Speech Production and Lexical Selection

Producing a word begins with an intended message and requires the selection of lexical information followed by morphological, phonological, and phonetic encoding. Levelt, Roelofs, and Meyer developed the WEAVER++ account as a staged explanatory model, integrating speech-production experiments and predictions about selection and encoding. Their treatment highlights an important distinction from printed word recognition: the production system must choose a lexical candidate and assemble a pronounceable utterance, often under competition from alternatives. A mental lexicon that explains recognition alone is incomplete unless it can connect with the linguistic information needed for fluent production Levelt et al. (1999).

Morphological encoding in production is relevant whenever a selected lemma must acquire tense, number, agreement, or derivational form. One theoretical question is how lexical selection relates to inflectional information; another is whether productive combinations are assembled online or retrieved as entrenched forms. Response-time experiments, picture naming, speech-error analyses, and primed production tasks offer partly different evidence about the sequence of operations. The existence of morphological influence on an utterance does not require that a listener and a speaker use the same access route at the same processing moment. Methodological comparisons should therefore distinguish input architecture from output preparation even when they invoke shared lexical knowledge Levelt et al. (1999) Pinker & Ullman (2002).

A scientifically useful integration requires that lexical semantics, grammatical features, and phonological plans remain linked across production and comprehension. Polysemous forms can compete at the level of meaning, while forms with similar sound structures can interfere at later stages. Morphological productivity becomes especially important for novel utterances: speakers routinely express meanings in words whose precise surface realizations have not been memorized as isolated entries. Production models should specify how regular formation is supported, how exceptions are retrieved, and what observable mistakes would distinguish them. Those predictions offer a bridge from linguistic analyses of word structure to psychologically measured selection dynamics Levelt et al. (1999) Booij (2010).

Development, Literacy, and Morphological Awareness

Morphological knowledge develops through exposure to spoken and written forms, increasing vocabulary, and experience with grammatical and semantic regularities. Children gradually learn that some recurring form patterns support stable meaning relations and can be generalized to unfamiliar words. Morphological awareness, the ability to reflect on and manipulate meaningful word components, is an explicit skill that should be distinguished from automatic sensitivity to morphological patterns during recognition. A child might benefit implicitly from a familiar family relation without being able to verbalize a morphological rule. Developmental studies must therefore distinguish offline metalinguistic judgments from timed online recognition Hasenäcker et al. (2016).

Hasenäcker, Beyersmann, and Schroeder examined masked morphological priming in German-speaking children and adults using response-time distributions. Their comparison addressed how processing varies with reading experience and why an average priming effect can conceal changes across faster and slower responses. This design is valuable because reading development affects visual expertise, vocabulary, and the diagnostic value of affix-like sequences simultaneously. A developmental difference in morphological priming cannot automatically be interpreted as the emergence of a single parsing mechanism; differences in familiarity, speed, accuracy, and orthographic proficiency are plausible competing contributors Hasenäcker et al. (2016) Amenta & Crepaldi (2012).

Educational applications follow from well-specified connections between morphological structure, vocabulary growth, and word recognition. Knowledge of familiar bases and productive affixes may help learners infer plausible meanings and spell unfamiliar derived words, but teaching effects require direct intervention evidence distinct from adult lexical-decision findings. The present dossier treats literacy implications as consequences and research questions of mental-lexicon organization; a separate educational dossier can examine classroom efficacy and transfer. Longitudinal measurement should track baseline language abilities, reading exposure, instruction, novel-word generalization, and delayed retention rather than relying on a one-session gain in morphological judgments Hasenäcker et al. (2016) Brysbaert & New (2009).

Cross-Linguistic and Typological Variation

The structure of a language changes the statistical environment in which lexical representations are acquired. A language with frequent, informative inflectional endings provides different cue distributions from one in which grammatical distinctions are commonly expressed with separate words. Orthographic depth, morphological productivity, stem alternation, and the ambiguity of a written affix influence the available evidence. Dutch plural experiments and Dutch–Finnish suffix comparisons demonstrate that whole-form frequency and constituent processing cannot be generalized from English word materials without examining the grammar and distribution of another language Baayen et al. (1997) Bertram et al. (2000).

The German study by Smolka, Libben, and Dressler examined complex verbs to test the degree to which morphological structure and semantic transparency constrain facilitation. Such cross-language comparisons are important because conventions about the transparency of a word, the independence of its base, and the status of a prefix are not interchangeable among languages. A claim that early decomposition is universal would require robust predictions across these structural differences, replicated with equivalent experimental controls and comparable proficiency groups. A single failure or success under a particular set of stimulus durations does not settle a theory's applicability to unrelated systems Smolka et al. (2019) Amenta & Crepaldi (2012).

Typological diversity also exposes the limitations of an orthography-only description of morphology. In many languages, morphological information is expressed through alternations, prosodic patterns, clitics, or other structures that do not resemble a simple contiguous stem followed by a suffix. Models built solely around English-like affix stripping must demonstrate how they would generalize or explicitly delimit their scope. An appropriate comparative program jointly models surface form, morphological paradigm, semantics, and exposure, and reports whether stimuli are matched in linguistic function rather than merely in letter count. This is a test of theoretical generality, not an invitation to posit identical processing mechanisms across populations Hay & Baayen (2005) Booij (2010).

Morphology in the Artificial Era

Artificial language systems make the organization of lexical knowledge newly investigable through comparisons of human behavior with explicit computational architectures. Tokenizers, subword representations, distributed embeddings, and attention-based processing may each encode regularities among word forms, but their engineering units are not automatically equivalents of human morphemes or lexical entries. The relevant comparison is functional: can a system recognize familiar derivatives, infer meanings of unfamiliar productive combinations, handle semantic opacity, and adjust its behavior according to morphological context? A psychological comparison must state the model's training data, tokenization, evaluation tasks, and available context before drawing conclusions about shared mechanisms Marelli & Baroni (2015) Milin et al. (2017).

Computational work offers a methodological advantage by exposing representational assumptions to controlled ablations and out-of-distribution tests. An investigator may remove explicit morpheme labels, vary subword segmentation, change training frequency, or test generalization to novel forms. If models with very different internal structure reproduce the same priming-like behavior, that behavioral result underdetermines claims about the human brain's representational format. Conversely, a mechanism proposed to explain human performance should specify observable predictions on novel items and across languages. This form of cross-system comparison respects the distinction between engineered procedures, learned statistical patterns, and psychological accounts of human lexical access Milin et al. (2017) Marelli & Baroni (2015).

Within the Artificial Era, a mental lexicon is therefore a point of intersection between language psychology and research on artificial cognition, memory, and representational organization. The comparison can inform theories of lexical learning without treating biological and artificial implementations as interchangeable. The present scientific conclusion is bounded: existing psycholinguistic experiments establish rich regularities of human morphological processing, while artificial architectures offer hypotheses and experimental instruments for understanding these regularities. Stronger claims about artificial lexical competence require direct cross-system benchmarks, held-out morphological generalization, semantic judgment tests, and analysis of error distributions Levelt et al. (1999) Milin et al. (2017).

Lexical Competition and Retrieval Dynamics

Lexical access operates amid competition from other plausible forms. A string or sound sequence can activate several candidates that resemble one another orthographically, phonologically, morphologically, or semantically. Morphological relatives may cooperate by strengthening a shared base, yet they may also compete because their meanings and grammatical functions are distinct. An advantage associated with a morphological family therefore requires an account of competition and facilitation together. Evidence from lexical-decision megastudies and discrimination learning shows that cue diagnosticity and experience can shape the difficulty of selecting among candidates Balota et al. (2007) Milin et al. (2017).

Retrieval is distributed over time. Word frequency and contextual diversity can influence the initial availability of a form, while semantic information and grammatical context can constrain which candidate remains active. A response-time difference is compatible with several causal paths: earlier activation, reduced competition, a changed decision criterion, or faster response preparation. When a target is preceded by a related prime, facilitation may reflect activation at multiple stages. Accordingly, morphological priming is most revealing when the experiment contrasts several relationship types, uses multiple timing conditions, and includes methods beyond word/nonword decisions Feldman (2000) Brysbaert & New (2009).

Production provides an additional form of competition because the system must select one lexical form for expression. The WEAVER++ theory places lexical selection and subsequent encoding in a structured sequence, while accounts of lexical comprehension emphasize identifying an input among candidates. Their different task demands provide complementary observations of the same broader problem: how learned lexical relations support rapid, contextually appropriate selection. Research should specify whether a morphological effect concerns representation, activation, selection, or assembly rather than treating all four as interchangeable Levelt et al. (1999).

Interpretation of Null Effects and Conflicting Findings

The interpretation of an apparent absence of difference is especially consequential in morphology research. When transparent and opaque word pairs yield similar priming, the result may support early sensitivity to written structure, but conventional significance testing alone cannot establish that their effects are equal. The precision of the estimate depends on the number of participants and items, the reliability of transparency classifications, exposure variability, and the chosen response model. Equivalence testing or a directly specified interval of theoretically meaningful differences would provide a stronger conclusion than simply failing to reject a null hypothesis Rastle et al. (2004) Amenta & Crepaldi (2012).

Conversely, a statistically detectable transparent–opaque difference does not automatically identify the first processing stage at which semantics can enter. Stimulus overlap, category distributions, and early lexical familiarity may contribute to the measured effect. Feldman and colleagues obtained meaningful semantic modulation under controlled matching, while complementary ERP investigations suggested that some effects unfold over different temporal windows. The sensible theoretical task is to articulate the assumptions by which behavioral and neural timing map onto representational operations and to test them jointly Feldman et al. (2009) Lavric et al. (2011).

A replication program can resolve some disagreements by exchanging stimuli and analytic decisions across laboratories. Preregistered competing models could predict the same held-out item-level response distribution and the same contrast across masked and visible conditions. Sharing individual word characteristics would permit investigators to test whether divergent conclusions arise from lexical norms, segmentation judgments, or uncertainty about item classification. As such protocols become available, conclusions can move from demonstrations that a mechanism is possible toward evidence that it is necessary or more predictive than alternatives Milin et al. (2017) Amenta & Crepaldi (2012).

Psychological Measurement and Generalizability

A word's morphological status is often assigned by expert analysis, whereas a participant's sensitivity to its structure is a behavioral variable. Researchers should separately measure explicit ratings of apparent segmentation, semantic transparency, and familiarity, using judges drawn from the population of interest. Linguistic annotation may be reliable for grammatical analysis while capturing only a portion of implicit recognition behavior. The distinction matters particularly for opaque derivatives, borrowed forms, ambiguous affixes, and compounds whose structure has become lexicalized. Agreement between annotators and generalization across items should form part of construct validation Hay & Baayen (2005) Amenta & Crepaldi (2012).

Sampling can determine the apparent generality of an effect. University students who are fluent adult readers of a particular language offer precise experimental control but represent only part of the population learning and using language. Children, multilingual participants, speakers of different dialects, and individuals with varied literacy experience may encounter different statistical distributions. Differences in performance should be interpreted in relation to measured vocabulary, reading experience, and task comprehension. Cross-linguistic research must specify whether tasks compare the same functional structures or only superficially similar letter patterns Hasenäcker et al. (2016) Bertram et al. (2000).

Scientific generalizability also depends on the computational specification of stimuli. A base frequency derived from one corpus, a word family derived from one segmentation algorithm, and semantic vectors trained on a particular register are not interchangeable quantities. Predicted morphological effects should be accompanied by sensitivity analyses that alter these choices and assess the stability of conclusions. Broad claims about a universal lexical architecture require reproducible evidence that goes beyond a favorable fit to one language, word list, or response task. This methodological discipline links evidence quality to the theoretical reach of the result Brysbaert & New (2009) Milin et al. (2017).

Evidence Quality, Replication, and Model Identifiability

Morphological processing is supported by convergent behavioral and neural findings, but their mechanistic interpretation depends on the contrasts and models used. Masked priming repeatedly reveals sensitivity to morphological-looking structure, and frequency, semantic transparency, and language experience modulate recognition. Yet facilitation is an outcome produced by a complete experimental system, including stimulus encoding and response selection. Different latent architectures may generate the same behavioral effect. Therefore the strongest defensible conclusion concerns an observable pattern under defined conditions; a claim that a unique decomposition algorithm has been established requires independent predictions that discriminate it from alternatives Rastle et al. (2004) Feldman et al. (2009) Milin et al. (2017).

A central replication problem concerns what should remain invariant across participants, words, languages, and laboratories. Equal opaque and transparent priming can be sensitive to statistical power and stimulus selection, while an observed difference may reflect correlated properties of the stems or affixes. Feldman and colleagues supplied evidence for early semantic influence; Rastle and colleagues obtained findings supportive of early morpho-orthographic decomposition. These are genuinely competing observations when their models predict the same contrast, but they also suggest that participant experience, prime timing, material construction, and statistical specification should be independently replicated. Amenta and Crepaldi's review makes the search for stable benchmark effects a central scientific priority Rastle et al. (2004) Feldman et al. (2009) Amenta & Crepaldi (2012).

Good evidence synthesis reports the availability of full text, preregistration and data where relevant, item-level materials, statistical uncertainty, subject and item random effects, and generalization tests. It distinguishes direct experimental outcomes from theory-laden explanations and records null effects without treating them automatically as equivalence. The sources assembled for this monograph include accessible full-text primary studies, bibliographically verified abstracts, explanatory models, and reviews. No pooled effect sizes or comprehensive risk-of-bias assessment were calculated; cross-study differences in stimuli and procedures make simple averaging scientifically misleading. The appropriate present product is a critical narrative comparison and a program of discriminating future tests Amenta & Crepaldi (2012) Milin et al. (2017).

An Integrated Research Framework

A useful integrative description has four levels. At the linguistic level, words participate in inflectional, derivational, and compound structures. At the distributional level, forms and contexts recur with different frequencies and degrees of predictability. At the behavioral level, recognition, priming, naming, and generalization produce task-specific measurements. At the implementation level, the system uses whatever memory, learning, and retrieval mechanisms realize these relationships. The levels should be linked through explicit hypotheses, not collapsed into a claim that a linguistically defined morpheme must be a discrete neural object. This framework accommodates the empirical relevance of morphology while leaving the representation question open to testing Hay & Baayen (2005) Milin et al. (2017).

The evidence supports a variable interaction among previously encountered whole forms, recurring structural cues, semantic similarity, and competition. Dutch frequency results establish that full-form familiarity can influence inflectional recognition; masked priming experiments establish sensitivity to potential morphological segmentation; semantic studies establish that meaning can modulate apparent early effects; and discrimination models show that statistical cue learning can compete with explicit decomposition accounts. These observations jointly constrain any acceptable model. A synthesis that discards one established result to preserve a preferred theoretical vocabulary sacrifices predictive adequacy. The scientific task is to predict their coexistence and the conditions under which each contribution is measurable Baayen et al. (1997) Rastle et al. (2004) Feldman et al. (2009) Milin et al. (2017).

An integrated model should specify inputs, representations, update rules, retrieval dynamics, and measurable outputs. It should predict responses to familiar forms, unfamiliar but plausible derivatives, pseudo-derived strings, ambiguous affixes, and words whose semantic transparency varies continuously. It should also handle cases where orthography and speech provide conflicting cues and where morphological families differ in frequency and productivity. One plausible route to progress is to combine explicit grammatical descriptors with learned mappings, then determine which descriptors improve held-out behavioral prediction. Such a proposal is a methodological synthesis of existing theory, not a report that the combination has already been experimentally validated Booij (2010) Marelli & Baroni (2015).

Research Design for Decisive Comparative Tests

A decisive experiment should select predictions before item construction, rather than choose stimuli merely because they illustrate a model. Researchers can factorially vary apparent morphological structure, semantic transparency, surface frequency, base frequency, affix productivity, and lexicality, while independently measuring participant proficiency and reading experience. At each processing interval, they should collect accuracy and distributional response times rather than relying exclusively on a grand mean. The resulting mixed-effects models should include plausible random effects for participants and items, sensitivity analyses for exposure norms, and planned contrasts that directly differentiate full-form, decompositional, and discrimination-based accounts Feldman et al. (2009) Milin et al. (2017).

Neural convergence would add temporal constraints to this design: masked and visible priming can be compared using ERP or MEG components, and fMRI can help localize associated processing changes. These measurements must use independent behavioral criteria for timing and lexical status, avoiding a circular inference in which a putative morphological neural marker is defined by the same task that it is later claimed to explain. A model comparison should evaluate predictive performance and calibration on held-out items and, where feasible, a second language. The relevant achievement would be reduced ambiguity among mechanistic explanations, not merely replication of a familiar facilitation effect Devlin et al. (2004) Lavric et al. (2011) Lehtonen et al. (2011).

Developmental and artificial-system comparisons can further test generality. Children and adults with different literacy histories may reveal how orthographic expertise changes processing. Artificial architectures with and without morphological segmentation can test whether the same behavior requires the same representational units. Crucial controls include exposure to the test words during training, vocabulary size, task demands, and novelty of the experimental combinations. A cross-system result should be reported as similarity in performance under stated constraints, with causal mechanisms evaluated separately through model intervention and human experimental inference. This research design integrates psychology and computational modeling without forcing an identity between their respective systems Hasenäcker et al. (2016) Marelli & Baroni (2015).

Open Problems and Scientific Priorities

The principal unresolved issue is the representational status of morphology: whether morphemes are privileged units of access, learned clusters in distributed form–meaning statistics, higher-level generalizations over stored words, or context-dependent combinations of these descriptions. Every option can express some familiar results. The discriminating question is which one predicts a new, tightly controlled stimulus set across language varieties and participant populations. Related unknowns concern the precise interaction of semantics and orthography in the earliest moments of visual recognition and the neural interpretation of effects measured through very different recording methods Amenta & Crepaldi (2012) Milin et al. (2017).

Another priority is the relation between lexical organization and learning across time. Vocabulary grows as inflections, derivatives, and meanings are encountered in changing environments. We need stronger longitudinal explanations of when a productive pattern becomes available, how conventionalized exceptions emerge, how a morphological family influences new entries, and how retrieval competition develops. The influence of literacy and educational intervention must be assessed with delayed outcomes and generalization measures. Evidence from adult reaction-time studies establishes useful hypotheses but cannot substitute for developmental causal evidence Hasenäcker et al. (2016) Hay & Baayen (2005).

Cross-language reproducibility and measurement remain equally consequential. Materials and frequency norms should be shared where licensing permits, alternative segmentation annotations should be documented, and statistical models should report uncertainty for subjects as well as items. Computational approaches should distinguish genuine prediction from fitting known word frequencies, and artificial-system studies should disclose training overlap with evaluation materials. A field-wide accumulation of comparable benchmark effects would enable stronger synthesis than debates built around selected demonstrations. These are research priorities implied by the reviewed literature, and their ordering reflects the methodological dependencies described in this monograph Brysbaert & New (2009) Balota et al. (2007) Milin et al. (2017).

Conclusion

The mental lexicon is an organized system of lexical knowledge whose operation depends on relationships among word forms, grammatical patterns, meaning, and experience. Morphological processing makes these relationships observable through the recognition and production of complex words, the interpretation of unfamiliar combinations, and measured sensitivity to recurring bases and affixes. Foundational decomposition experiments, frequency-based dual-route analyses, masked morphological priming, semantic studies, and neural investigations have each established substantive empirical constraints. Their cumulative lesson is that lexical organization must explain both entrenched individual forms and generalizable structure Taft & Forster (1975) Baayen et al. (1997) Rastle et al. (2004) Levelt et al. (1999).

The scientific synthesis favors an account in which multiple sources of information contribute under differing conditions, while the implementation of those contributions remains an active explanatory problem. Evidence for apparent segmentation does not by itself establish an obligatory universal parsing stage, and the success of discrimination-based predictions does not erase morphology as an important linguistic and psychological regularity. Theories advance when they specify identifiable mechanisms, explain contrary evidence, and make predictions across tasks, ages, and languages. Full-form storage, structured morphological knowledge, semantic compositionality, and learned statistical cues are all empirical considerations to be modeled rather than optional decorations Feldman et al. (2009) Milin et al. (2017) Marelli & Baroni (2015).

The enduring research agenda links lexical representation to the experience of language users and to measurable computations of access. Its priorities are independent replication, carefully matched semantic and form conditions, longitudinal developmental designs, cross-linguistic comparisons, and formal out-of-sample model tests. Comparative artificial language systems can contribute controlled mechanistic experiments, provided the distinction between artificial architecture and human psychological inference remains explicit. This monograph establishes the historical and contemporary evidence base for those inquiries and situates mental-lexicon research within the broader science of language, cognition, and the Artificial Era Amenta & Crepaldi (2012) Hay & Baayen (2005) Milin et al. (2017).

References

  • Amenta, S., & Crepaldi, D. (2012). Morphological processing as we know it: An analytical review of morphological effects in visual word identification. Frontiers in Psychology, 3, Article 232. https://doi.org/10.3389/fpsyg.2012.00232
  • Baayen, R. H., Dijkstra, T., & Schreuder, R. (1997). Singulars and plurals in Dutch: Evidence for a parallel dual-route model. Journal of Memory and Language, 37(1), 94–117. https://doi.org/10.1006/jmla.1997.2509
  • Balota, D. A., Yap, M. J., Cortese, M. J., Hutchison, K. A., Kessler, B., Loftis, B., Neely, J. H., Nelson, D. L., Simpson, G. B., & Treiman, R. (2007). The English Lexicon Project. Behavior Research Methods, 39(3), 445–459. https://doi.org/10.3758/BF03193014
  • Bertram, R., Schreuder, R., & Baayen, R. H. (2000). The balance of storage and computation in morphological processing: The role of word formation type, affixal homonymy, and productivity. Journal of Experimental Psychology: Learning, Memory, and Cognition, 26(2), 489–511. https://doi.org/10.1037//0278-7393.26.2.489
  • Booij, G. (2010). Construction morphology. Language and Linguistics Compass, 4(7), 543–555. https://doi.org/10.1111/j.1749-818X.2010.00213.x
  • Brysbaert, M., & New, B. (2009). Moving beyond Kučera and Francis: A critical evaluation of current word frequency norms and the introduction of a new and improved word frequency measure for American English. Behavior Research Methods, 41(4), 977–990. https://doi.org/10.3758/BRM.41.4.977
  • Crepaldi, D., Rastle, K., Davis, C. J., & Lupker, S. J. (2013). Seeing stems everywhere: Position-independent identification of stem morphemes. Journal of Experimental Psychology: Human Perception and Performance, 39(2), 510–525. https://doi.org/10.1037/a0029713
  • Devlin, J. T., Jamison, H. L., Matthews, P. M., & Gonnerman, L. M. (2004). Morphology and the internal structure of words. Proceedings of the National Academy of Sciences, 101(41), 14984–14988. https://doi.org/10.1073/pnas.0403766101
  • Feldman, L. B. (2000). Are morphological effects distinguishable from the effects of shared meaning and shared form? Journal of Experimental Psychology: Learning, Memory, and Cognition, 26(6), 1431–1444. https://doi.org/10.1037//0278-7393.26.6.1431
  • Feldman, L. B., O'Connor, P. A., & Moscoso del Prado Martín, F. (2009). Early morphological processing is morphosemantic and not simply morpho-orthographic: A violation of form-then-meaning accounts of word recognition. Psychonomic Bulletin & Review, 16(4), 684–691. https://doi.org/10.3758/PBR.16.4.684
  • Hasenäcker, J., Beyersmann, E., & Schroeder, S. (2016). Masked morphological priming in German-speaking adults and children: Evidence from response time distributions. Frontiers in Psychology, 7, Article 929. https://doi.org/10.3389/fpsyg.2016.00929
  • Hay, J. B., & Baayen, R. H. (2005). Shifting paradigms: Gradient structure in morphology. Trends in Cognitive Sciences, 9(7), 342–348. https://doi.org/10.1016/j.tics.2005.04.002
  • Lavric, A., Clapp, A., & Rastle, K. (2007). ERP evidence of morphological analysis from orthography: A masked priming study. Journal of Cognitive Neuroscience, 19(5), 866–877. https://doi.org/10.1162/jocn.2007.19.5.866
  • Lavric, A., Rastle, K., & Clapp, A. (2011). What do fully visible primes and brain potentials reveal about morphological decomposition? Psychophysiology, 48(5), 676–686. https://doi.org/10.1111/j.1469-8986.2010.01125.x
  • Lehtonen, M., Monahan, P. J., & Poeppel, D. (2011). Evidence for early morphological decomposition: Combining masked priming with magnetoencephalography. Journal of Cognitive Neuroscience, 23(11), 3366–3379. https://doi.org/10.1162/jocn_a_00035
  • 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
  • Marelli, M., & Baroni, M. (2015). Affixation in semantic space: Modeling morpheme meanings with compositional distributional semantics. Psychological Review, 122(3), 485–515. https://doi.org/10.1037/a0039267
  • Milin, P., Feldman, L. B., Ramscar, M., Hendrix, P., & Baayen, R. H. (2017). Discrimination in lexical decision. PLOS ONE, 12(2), Article e0171935. https://doi.org/10.1371/journal.pone.0171935
  • Pinker, S., & Ullman, M. T. (2002). The past and future of the past tense. Trends in Cognitive Sciences, 6(11), 456–463. https://doi.org/10.1016/S1364-6613(02)01990-3
  • Rastle, K., Davis, M. H., & New, B. (2004). The broth in my brother's brothel: Morpho-orthographic segmentation in visual word recognition. Psychonomic Bulletin & Review, 11(6), 1090–1098. https://doi.org/10.3758/BF03196742
  • Smolka, E., Libben, G., & Dressler, W. U. (2019). When morphological structure overrides meaning: Evidence from German prefix and particle verbs. Language, Cognition and Neuroscience, 34(5), 599–614. https://doi.org/10.1080/23273798.2018.1552006
  • Taft, M., & Forster, K. I. (1975). Lexical storage and retrieval of prefixed words. Journal of Verbal Learning and Verbal Behavior, 14(6), 638–647. https://doi.org/10.1016/S0022-5371(75)80051-X