Psychology by BogdanovaThe Encyclopedia of Psychology for the Artificial Era

RESEARCH DOSSIER

Speech Perception and Phonological Categorization

Abstract. Speech perception transforms variable acoustic and audiovisual signals into linguistic contrasts, while phonological categorization supports generalization across speakers, contexts, and development. This fundamental research monograph critically synthesizes classic acoustic-phonetic experiments, categorical identification and discrimination paradigms, studies of graded sensitivity inside categories, infant cross-language discrimination, distributional learning, socially structured exposure, adult lexical retuning, perceptual restoration, audiovisual speech, and cortical feature coding. The scientific review examines how multiple cues are weighted, how listener expectations and talker distributions guide categorization, and how experiment-dependent decisions differ from underlying perceptual information. It compares abstraction, episodic-memory, and probabilistic models against behavioral and neural evidence, describes validity limitations, and outlines unresolved questions about developmental change and context-sensitive robustness. The synthesis supports experience-dependent, functionally useful categories that interact with continuous acoustic information and adaptive inference. This is a critical narrative analysis of published studies and methodological work; it reports no original experiment, registered systematic review, or meta-analysis.

Keywords: speech perception; phonological categorization; phonetic category learning; categorical perception; acoustic cue weighting; voice onset time; talker normalization; accent adaptation; phonemic restoration; speech in noise; audiovisual speech; McGurk effect; infant speech perception; perceptual narrowing; distributional learning; cross-language speech perception; phonetic retuning; gradient lexical activation; superior temporal gyrus; psychophysics; Artificial Era

Scientific Scope: Speech Perception and Phonological Categorization

Speech perception is the psychological transformation of time-varying acoustic and audiovisual information into representations that support linguistic contrasts. Phonological categorization is the assignment of speech events to learned, language-relevant distinctions that can separate words or grammatical forms. The research object includes acoustic encoding, cue combination, category decisions, continuous sensitivity within categories, speaker adaptation, learning across development, and neural implementation. A phoneme belongs to a linguistic contrast system, whereas a particular phone is a situated phonetic realization. Successful discrimination of two sounds therefore does not, by itself, establish adult-like phonological representation. Investigating this distinction requires explicit operational definitions and multiple convergent tasks Liberman et al. (1957) Werker (2024).

Three levels of explanation structure the inquiry. The physical level describes articulatory gestures and their acoustic consequences, including formant trajectories, temporal envelopes, voice onset time, and spectral properties. The computational level specifies how the listener estimates likely categories from uncertain evidence, prior experience, and context. The linguistic level identifies which contrasts function in a particular language. Neural recordings address how those transformations are implemented, while behavioral experiments establish what listeners distinguish and generalize. These levels constrain one another, yet evidence at one level does not automatically resolve representation at another. Especially important is the distinction between a sharp behavioral response and the continuous evidence from which that response was made Toscano & McMurray (2010) Mesgarani et al. (2014).

This monograph is a critical narrative synthesis of landmark experimental reports, developmental comparisons, neural measurements, computational proposals, and methodological research. Its central questions are why categories remain stable despite acoustic variation, how experience changes category boundaries, whether acoustic detail survives categorization, and how context affects identification in adverse conditions. The analysis also considers cross-language generalization and the measurement consequences of task demands. Findings are attributed at the level actually supported by the consulted studies; the work reports no original participants, novel behavioral data, preregistered systematic search, or pooled meta-analysis. Its goal is an integrated account of the evidence and the remaining explanatory problems Kleinschmidt & Jaeger (2015) McMurray et al. (2008).

The Speech Signal and the Problem of Acoustic Invariance

Spoken language is produced through coordinated movements of the lungs, larynx, tongue, jaw, lips, and velum. Acoustic properties result from source excitation filtered by the changing vocal tract. A stop release, a burst spectrum, a fricative's aperiodic energy, or a vowel's resonant frequencies may all contribute evidence about a segment. Yet the same linguistic contrast is realized differently according to adjacent segments, stress, vocal-tract anatomy, speaking style, and rate. The perceptual problem is to derive useful stability without assuming an identical waveform accompanies every occurrence of a phoneme. Cross-language acoustic measurement demonstrates the importance of analyzing actual distributions rather than treating spelling-based labels as acoustic constants Lisker & Abramson (1964) Mesgarani et al. (2014).

Voice onset time measures the temporal relationship between release of a stop and the onset of vocal-fold vibration. Lisker and Abramson's cross-language measurements established a foundational method for comparing stop voicing patterns across languages. A laboratory continuum can alter this timing while holding selected other features approximately stable, allowing identification and discrimination functions to be compared. Natural speech, however, usually makes several cues covary: aspiration duration, formant onsets, closure characteristics, burst properties, and the temporal environment all contribute information. A boundary estimated from one synthetic series cannot be generalized without testing other voices, vowels, rates, and places of articulation Lisker & Abramson (1964) Toscano & McMurray (2010).

Acoustic invariance is best treated as an explanatory challenge rather than a prior requirement that the nervous system discover one fixed feature. Listeners can use distributions, conditional relations, temporal patterns, and situational expectations to infer a contrast. Some properties may be reliable within restricted contexts while failing across all contexts; others become useful only when combined with additional features. An experimentally successful cue does not demonstrate that the listener relies on it exclusively in ordinary conversation. This distinction guides both cue-weighting models and attempts to understand adaptation to voices and accents: the relevant perceptual target is often a context-sensitive estimate of linguistic identity from a richly structured signal Kleinschmidt & Jaeger (2015) Holt & Lotto (2002).

Historical Foundations of Experimental Speech Perception

The modern experimental tradition developed through acoustic phonetics, speech synthesis, psychophysics, and comparative language study. Synthetic syllables enabled researchers to vary selected acoustic properties under controlled conditions while asking listeners to identify or discriminate resulting tokens. In a landmark study, Liberman, Harris, Hoffman, and Griffith examined how listeners discriminate speech sounds within and across phoneme boundaries. Their work made the relation between physical acoustic intervals and perceived linguistic categories an explicit empirical question. The original paper established a research program, although contemporary interpretation of its broader theoretical implications requires comparison with later experiments that preserve graded information Liberman et al. (1957) McMurray et al. (2008).

A second line of research investigated language-specific acoustic structure. Lisker and Abramson documented timing variation associated with voicing in initial stops across languages; their acoustic observations helped researchers separate production distributions from perception judgments. Infant work subsequently asked whether category-sensitive discrimination depended entirely on learned language experience. Eimas and colleagues tested one- and four-month-old infants with a habituation procedure and reported discrimination around a stop-consonant contrast. This sequence of developments transformed speech perception into a field where acoustic measurement, developmental learning, and phonological theory could be tested against different forms of behavior Lisker & Abramson (1964) Eimas et al. (1971).

Over later decades, researchers challenged the inference that categorical responses necessarily imply loss of acoustic detail. Memory comparisons, lexical competition tasks, and eye tracking revealed sensitivity that conventional forced-choice labels can obscure. Meanwhile, perceptual learning studies showed that adults recalibrate categories after brief exposure, and neural recordings disclosed feature-sensitive and category-sensitive responses within speech-related cortex. The history is therefore a history of progressively better differentiation among levels of description. Categorical identification, acoustic discrimination, online lexical competition, learning, and cortical coding are related observations whose explanatory relation has to be demonstrated rather than assumed Pisoni (1973) Norris et al. (2003) Chang et al. (2010).

Categorical Identification and Discrimination

A classic categorical-perception experiment constructs a stimulus continuum between two consonant endpoints and obtains identification judgments for every token. Plotting the probability of response A against acoustic position usually reveals a transition region. A separate discrimination procedure asks whether listeners notice differences between pairs, often at equal physical distances along the continuum. In the classical pattern, across-boundary pairs are discriminated better than within-category pairs. This result established a productive link between psychophysical performance and linguistic organization, but the identification slope is a function of stimulus choices, response alternatives, background context, and trial structure as well as perceptual encoding Liberman et al. (1957) Pisoni (1973).

Identification and discrimination estimate different dependent variables. An overt categorical label may reflect a decision about which response is most plausible even when the underlying sensory representation remains graded. Discrimination can be supported by transient acoustic memory, categorical differences, or comparisons of learned exemplars. Pisoni's consonant and vowel memory experiments made this distinction particularly important: performance depends on the interval and representation available during a comparison. Equating phoneme identification with all perceptual access discards these temporal and procedural factors. Strong research designs therefore estimate category function, discrimination function, reaction times, and a measure of uncertainty where feasible Pisoni (1973) Toscano et al. (2010).

Boundary location is also population- and context-dependent. A voice onset time that encourages a voiced response for one speaker or speaking condition may favor an unvoiced response elsewhere. Category models must explain the location and variability of the transition, rather than merely fitting its existence. An experiment reporting a steep psychometric function has identified a property of responses to the tested stimulus set; evidence for an invariant mental category requires transfer to novel tokens and generalization across relevant conditions. This distinction protects interpretations of categorical perception from circular reasoning and creates a bridge between classical psychophysics and probabilistic accounts Kleinschmidt & Jaeger (2015) Toscano & McMurray (2010).

Continuous Acoustic Detail Inside Phonological Categories

Acoustic variation remains behaviorally available after listeners make phoneme-level identifications. Research using eye movements during spoken lexical competition found graded effects of voice onset time among stimuli that would commonly receive the same phoneme label. McMurray and colleagues showed that activation of word candidates can vary with the fine acoustic position of a stimulus, including within the nominal category. This outcome is especially informative because eye movements measure evolving commitments rather than only the final response. It supports models in which categories guide interpretation while the listener retains enough subcategorical information to revise hypotheses as speech unfolds McMurray et al. (2003) McMurray et al. (2008).

Electrophysiological evidence provides an independent approach to this distinction. Toscano, McMurray, Dennhardt, and Luck measured responses to speech tokens on a voice onset time continuum and reported neural sensitivity consistent with continuous encoding of acoustic variation. Such findings complement, rather than invalidate, other recordings that reveal category-associated activity. Stimulus-dependent neural measures, task demands, electrode positions, and analysis time windows can favor different features of a distributed process. Claims that speech is wholly categorical or wholly continuous exceed what one experiment establishes. The stronger account predicts where and when continuous encoding and categorical readout each become useful Toscano et al. (2010) Chang et al. (2010).

A layered interpretation recognizes three operations: sensory registration of differences, graded estimation of their relevance, and eventual selection of linguistic responses. A decision threshold can generate apparently discrete labels from a posterior distribution whose intermediate values remain computationally consequential. This architecture also explains why explicit phoneme judgments and natural word recognition may produce different estimates of sensitivity. Future research needs jointly fitted models of identification, discrimination, and incremental recognition, tested on unfamiliar speakers and novel phonetic contexts. A category is thereby defined by functional generalization and contrast, while a behavioral response is treated as an observable consequence of a richer inferential process McMurray et al. (2008) Kleinschmidt & Jaeger (2015).

Multiple Acoustic Cues and Reliability Weighting

Most phonological distinctions are identified through combinations of cues whose usefulness depends on segment, speaker, language, and context. For a stop consonant, voice onset time may cooperate with burst properties, aspiration, vowel onset, and spectral transitions. For a fricative, noise spectrum and the neighboring vowel may jointly indicate place of articulation. An experiment that varies one dimension while freezing the others measures a conditional contribution, not an exhaustive inventory of what a listener uses. Cue weighting is therefore a relational property: listeners assign different influence to features depending on their conditional diagnostic value in the experienced distribution of speech Toscano & McMurray (2010) Lisker & Abramson (1964).

Distributional models offer an explicit route from experience to cue weighting. Toscano and McMurray analyzed how unsupervised learning and observed statistical relations can shape integration of acoustic information, providing an alternative to a fixed ranking of speech cues. The theoretical problem is to explain how a learner discovers meaningful structure when category membership itself is initially uncertain. A successful model must predict both the category decision and changes when a cue becomes unreliable. Manipulating noise separately on candidate features, testing cross-cue conflicts, and using held-out speakers help establish whether the estimated weights reflect learned acoustic reliability or a laboratory-specific response strategy Toscano & McMurray (2010) Kleinschmidt & Jaeger (2015).

Context effects demonstrate why acoustic interpretation is not reducible to a one-to-one physical mapping. Holt and Lotto examined auditory processing levels contributing to speech context effects, emphasizing the role of preceding stimulation in altering the perceptual interpretation of a target. The effect of a contextual precursor can depend on spectral relations and listening history, making its mechanism testable through carefully controlled nonspeech and speech contexts. The appropriate conclusion is that speech perception draws on general auditory organization as well as learned linguistic regularities; identifying the contribution of each requires designs that separate auditory contrast from categorical expectations Holt & Lotto (2002) Toscano & McMurray (2010).

Coarticulation and Temporal Context

Speech gestures overlap in time, so an acoustic interval typically contains consequences of several articulatory movements. A vowel's early formant transitions may carry information about an adjacent consonant; a consonant may reflect an upcoming vowel before that vowel is fully audible. Coarticulation is consequently both a source of variability and a potentially informative source of anticipatory evidence. The proper unit of analysis is often a sequence of jointly conditioned cues rather than an isolated sound. Researchers need to determine whether listeners use this temporal information immediately, whether sensitivity transfers across talkers, and how the resulting expectations depend on native-language experience Mesgarani et al. (2014) Dahan et al. (2001).

Evidence for temporally distributed information is particularly relevant to theories of incremental processing. Spoken words unfold under conditions in which each new acoustic segment changes the relative plausibility of several hypotheses. Dahan and colleagues studied how subcategorical mismatches influence lexical access and competition over time. Such designs show that imperfect phonetic evidence can alter downstream recognition before the end of a word. These effects should be attributed to graded and temporally structured lexical processing; they do not establish that a listener must explicitly identify every phoneme before updating the lexical candidates Dahan et al. (2001) McMurray et al. (2003).

Neural responses support a distributed feature perspective. Mesgarani and colleagues investigated phonetic feature encoding in the human superior temporal gyrus, revealing systematic cortical sensitivity to aspects of speech structure. Their recordings help connect time-resolved acoustics with neural representational geometry, while the limited clinical recording population and electrode coverage constrain broad anatomical generalization. A theory of coarticulatory integration needs to accommodate both rapid auditory feature extraction and later dependence on language-specific categories, without assigning every observed response to a single serial stage Mesgarani et al. (2014) Hickok & Poeppel (2007).

Speaking Rate, Talker Normalization, and Accent

A listener hears speech produced by vocal tracts of different sizes and shapes, speakers with distinct habitual timings, and communities with different phonetic norms. The same measured acoustic value can correspond to different intended contrasts in different talkers, while distinct physical values may represent the same category. Recognition therefore involves estimating how evidence is conditioned on the source. Speaking rate changes the temporal scale of events, and habitual accent affects prior expectations about category realization. A robust system must balance transfer of familiar patterns with rapid updating when an unfamiliar talker supplies systematic evidence Kleinschmidt & Jaeger (2015) Goldinger (1998).

Kleinschmidt and Jaeger developed a rational account of robust speech perception in which familiar variability, generalization to similar conditions, and adaptation to novel conditions are interrelated computational problems. Their framework clarifies why learning a talker's distributions can improve subsequent categorization while excessive updating can destabilize previously useful knowledge. Goldinger's episodic theory, in turn, places detailed traces of particular speech experiences at the center of lexical access. Both perspectives predict sensitivity to individual utterances, although they differ in what is stored and how abstraction emerges. Experimental evaluation requires held-out voices, changing accent consistency, and careful separation of token repetition from category learning Kleinschmidt & Jaeger (2015) Goldinger (1998).

Normalization should be measured as a specific pattern of generalization. A shift in a phoneme boundary after one voice's exposure may reflect changed priors, altered cue scaling, lexical feedback, fatigue, or response anchoring. Showing that the shift predicts performance on new utterances from the same speaker, and comparing it with unfamiliar-speaker transfer, narrows the explanation. Natural accents add interactions among timing, spectral quality, and phonotactics that controlled continua often simplify. These complexities make the literature's central insight a conditional one: adaptation is useful because speech categories must be inferred relative to speakers and contexts, and useful adaptation has limits that laboratory designs can quantify Norris et al. (2003) Kleinschmidt & Jaeger (2015).

Phonological Categories, Decision Boundaries, and Contrast

A phonological category is defined partly by its functional role within a language's contrast system. If an acoustic change reliably distinguishes words or grammatical forms, experience with that language can increase the importance of the change. Yet the same acoustic distinction may be allophonic, optional, or irrelevant elsewhere, and the classification cannot be read directly from an isolated judgment. Category boundaries in a forced-choice task are estimates of response transitions, whereas phonological knowledge additionally concerns generalization across contexts and sensitivity to meaningful contrasts. The difference is especially visible when children and cross-language listeners discriminate sounds that they do not label in an adult-like manner Werker & Tees (1984) Best & McRoberts (2003).

Best and McRoberts investigated how infants respond to nonnative consonant contrasts that adult listeners assimilate in different ways. Their research connects language-specific phonetic organization with discriminability while demonstrating the value of specifying which categories a listener maps onto a nonnative contrast. It is inappropriate to treat every nonnative contrast as equally difficult; acoustic separability, articulatory properties, age, exposure, and assimilation pattern can produce heterogeneous outcomes. Category formation can thus be understood as an alignment between experience-dependent structure and specific contrastive functions rather than as the acquisition of a universal, fixed set of response thresholds Best & McRoberts (2003) Best & Tyler (2007).

Lexical evidence can change overt phoneme decisions. Ganong's demonstration of phonetic categorization differences in auditory word perception showed that ambiguous sounds can be labeled differently depending on whether a resulting string forms a word. This finding raises a crucial level-of-analysis question: does lexical information modify an early phonetic representation, alter the weighting of evidence during decision, or bias the response criterion? Subsequent studies must distinguish these possibilities using measures that do not rely solely on categorical reports. A theoretically informative category account specifies a representational claim and the observable pattern capable of ruling out a simpler criterion shift Ganong (1980) McMurray et al. (2008).

Distributional Structure and Generalization Across Variability

Every language learner encounters distributions of acoustic tokens rather than a dictionary of perfectly labeled phonemes. If speech examples cluster along a relevant dimension, an observer might infer one or several categories depending on how experiences are sampled and weighted. Maye, Werker, and Gerken experimentally manipulated distributional information and found that infant discrimination could be altered by brief exposure. The study gave experimental support to the claim that statistical structure can contribute to early phonetic learning, even though laboratory exposure cannot stand for the full social and acoustic environment of language acquisition Maye et al. (2002) Kuhl et al. (1992).

A central theoretical difficulty is that distribution alone does not identify the linguistically appropriate dimensions or number of categories. Different speakers produce overlapping token distributions, correlated cues can hide distinctions, and common contexts create variation within a contrast. Models must specify what invariances the learner discovers, which priors guide the discovery, and how new examples are integrated with existing representations. Toscano and McMurray's unsupervised account supplies a framework for studying acoustic cue learning, while probabilistic models of speaker adaptation suggest that structure may be conditional on source identity Toscano & McMurray (2010) Kleinschmidt & Jaeger (2015).

Tests of distributional learning benefit from a design in which participants hear multiple talkers, variability is manipulated systematically, and the evaluation includes entirely new tokens. Immediate improvement on trained examples can reflect memorization rather than generalization of a category. A strong protocol assesses discrimination, classification, uncertainty, and later retention, and it asks whether acoustic changes affect linguistic contrasts outside the training context. The cumulative evidence supports distribution-sensitive learning as one explanatory mechanism within broader language experience; determining its relative contribution to lexical meaning, communicative interaction, and development remains a separate empirical question Maye et al. (2002) Kuhl et al. (2003) Best & Tyler (2007).

Early Infant Discrimination and Experimental Design

Infant speech perception became experimentally tractable when researchers adapted habituation and discrimination procedures for participants unable to give verbal judgments. Eimas and colleagues tested one- and four-month-old infants using speech contrasts and observed changes in responding consistent with discriminating a contrast near a phonetic boundary. Such findings established the importance of early sensitivity to relevant acoustic differences and motivated theories of how this sensitivity changes with exposure. The dependent measure, however, remains an indirect index of the infant's auditory and attentional response, so its interpretation requires careful control of stimulus novelty, arousal, and timing Eimas et al. (1971) Werker (2024).

Infant paradigms vary in what they measure. Habituation procedures infer discrimination from recovery of attention, while conditioned head-turn methods examine learned orientation to sound changes. More recent auditory and neural designs add converging measures of sensitivity. A successful discrimination result means that the child distinguished the tested stimulus distributions under the experiment's conditions; it does not imply mastery of the contrast's adult phonological function. The same distinction applies to an apparent failure: low attention, excessive task difficulty, insufficient trial power, or unbalanced acoustic cues can obscure sensitivity that appears under another method Eimas et al. (1971) Best & McRoberts (2003).

Historical claims about innate speech categories must be read against this methodological foundation. The early experiments show perceptual capacities before mature speech production and established lexical knowledge; they do not settle how experience, general auditory capacities, and developing linguistic representations jointly produce performance. Werker's later retrospective analysis of perceptual reorganization clarifies the trajectory of this research and the changing interpretations of infant discrimination. Strong developmental evidence follows multiple cohorts, tests comparable contrasts using methods appropriate for each age, and distinguishes general task development from selective change in phonetic sensitivity Werker (2024) Eimas et al. (1971).

Native-Language Attunement in the First Year

Infants initially discriminate some contrasts from languages outside their immediate environment, while their responses become increasingly aligned with the contrasts of their experienced language during the first year. Werker and Tees provided classic cross-language evidence for developmental reorganization in discrimination of nonnative contrasts. The pattern invited a learning-based interpretation: experience reorganizes which differences are especially relevant to a developing linguistic system. The change is language- and contrast-specific rather than a general reduction in hearing ability. Age comparisons require attention to exactly which sounds, participant groups, and discrimination procedures are involved Werker & Tees (1984) Werker (2024).

Vowel perception provides another window into the time course of this attunement. Kuhl and colleagues reported experience-related differences in infant phonetic perception by six months, suggesting that relevant distributions begin shaping vowel representations earlier than some classic nonnative consonant contrasts. Because consonants and vowels differ in acoustic organization, phonological function, and exposure patterns, a single universal 'switch point' in development is not warranted. The appropriate developmental model predicts different trajectories conditional on category properties and the child's language environment. Multi-age sampling and continuous measures can better reveal these trajectories than a simple before-and-after contrast Kuhl et al. (1992) Kuhl (2004).

Native-language tuning also changes the meaning of success in an infant task. Discriminating a foreign contrast may be useful early when the language environment remains unknown; later efficiency may depend on grouping variable examples of the same native-language category. This functional reorganization is compatible with preserved auditory sensitivity to many physical differences even when those differences are ignored in familiar linguistic decisions. Follow-up studies must investigate reversibility, individual differences, and how exposure to more than one language shapes different pathways. These are questions about developing perceptual organization rather than evidence for an irreversible loss of sensory capability Werker (2024) Best & Tyler (2007).

Statistical Learning and Socially Structured Exposure

Controlled exposure experiments have shown that even limited changes in the distribution of speech sounds can affect infant discrimination. Maye, Werker, and Gerken altered the frequency distribution of tokens along a phonetic continuum, linking emerging discrimination to the statistical properties of the heard input. This provides a mechanism connecting an apparently abstract boundary to the environmental record of sound occurrences. Yet natural input includes highly structured variation across speakers, situations, words, and conversational purposes; the laboratory distribution is one component of a much more complex learning problem Maye et al. (2002) Kuhl (2004).

Communicative experience also matters. Kuhl, Tsao, and Liu studied short-term foreign-language exposure in infancy and reported effects associated with live social interaction. Their research suggests that the conditions under which sounds are presented can influence learning outcomes, including attention and access to relevant distributions. An experimental result about particular exposure formats should not be converted into the claim that all phonetic learning requires direct interpersonal interaction; the methods and populations determine what comparison was actually tested. Investigators must distinguish the quantity and acoustic quality of input, contingent interaction, motivation, and social cueing Kuhl et al. (2003) Kuhl (2004).

Statistical and socially supported learning are candidates for integration within a common explanatory account. Communicative settings can change where an infant looks, which voices are attended, how frequently particular sounds recur, and what lexical meaning accompanies the acoustic evidence. A mechanistic design would equate exposure statistics while varying contingent interaction, then measure retention and generalization on new speakers and sound environments. Such studies can determine whether social conditions alter the extraction of acoustic regularities, the later consolidation of learned categories, or the likelihood that a child engages with the input. Distinguishing these stages is essential for specifying what language experience actually contributes Kuhl et al. (2003) Maye et al. (2002).

Nonnative Contrasts and Second-Language Category Learning

Adults listening to an unfamiliar language often map its sounds onto categories acquired through their first language. Performance depends on how a nonnative contrast relates to existing phonetic and phonological distinctions, with some contrasts discriminated accurately and others assimilated in ways that reduce their apparent separability. Best and Tyler developed a framework for relating nonnative perception and second-language speech learning, emphasizing important complementarities while attending to experience and task. A single global measure of foreign-language 'difficulty' misses the specific assimilation patterns and the acoustic structure underlying each contrast Best & Tyler (2007) Best & McRoberts (2003).

Second-language learning can involve changes in cue weighting, category location, sensitivity to distributions, and generalization across talkers. A listener may improve on practiced items while continuing to confuse unfamiliar speakers, or may learn an experimental response criterion without reorganizing a durable linguistic category. Designs with pretests, delayed tests, transfer items, and multiple measures help separate these explanations. Neural and behavioral evidence of language-specific processing also needs to be interpreted in relation to proficiency and exposure, which are not interchangeable variables. Language experience is distributed through time, social setting, and task-specific listening goals Zhang et al. (2005) Kleinschmidt & Jaeger (2015).

Cross-language studies are especially informative when they measure the physical characteristics of the contrasts rather than comparing orthographic labels. Lisker and Abramson's cross-language acoustic work demonstrates the value of empirical specification of stop categories, while Best's assimilation studies illustrate how linguistic histories shape subsequent perception. A rigorous comparison therefore reports stimulus provenance, speaker variability, relevant native-language contrasts, hearing and linguistic histories, and the decision structure of the test. These practices strengthen explanations of why some nonnative distinctions transfer readily and others require sustained learning Lisker & Abramson (1964) Best & McRoberts (2003) Best & Tyler (2007).

Adult Lexically Guided Perceptual Retuning

Language experience continues to influence phonetic boundaries after childhood. Norris, McQueen, and Cutler demonstrated perceptual learning in speech using lexical contexts that help disambiguate an acoustically ambiguous sound. Subsequent judgments on other tokens changed in a direction consistent with retuning the listener's phonetic interpretation. The finding matters because it connects knowledge of words to an adaptive response to unusual speech from a particular talker. The experimental inference remains strongest when transfer occurs beyond the specific trained words, revealing an alteration in the interpretation of phonetic evidence rather than rote memory for a few items Norris et al. (2003) Eisner & McQueen (2005).

Eisner and McQueen investigated the specificity of perceptual learning, showing why it is necessary to ask which properties of the training experience are carried forward. A speaker-specific learning effect implies a different adaptation mechanism from general reassignment of an entire language-wide phonemic contrast. These results can be placed within probabilistic models that update expectations when a speaker's realizations are systematically atypical. Rapid category flexibility is helpful for understanding accents, yet an overly broad shift could misclassify familiar speakers. Models therefore require mechanisms for protecting useful prior knowledge while accommodating novel regularities Eisner & McQueen (2005) Kleinschmidt & Jaeger (2015).

Lexically guided retuning is sometimes presented as decisive evidence that lexical processing directly changes early sensory representations. The same behavioral shift can also arise from updated decision criteria, changed priors over sounds, or a representation linking speaker identity with phonetic evidence. Distinguishing these possibilities requires measures that retain continuous sensitivity, tests of generalization across sound contexts, and designs that manipulate the reliability of lexical labels. The established result is adaptive modification of subsequent phonetic decisions in the examined conditions; its exact computational and neural locus remains a target of comparative model testing Norris et al. (2003) Toscano et al. (2010).

Lexical Context, Bias, and the Ganong Effect

Lexical knowledge influences how listeners respond to acoustically ambiguous speech. In Ganong's classic experiment, listeners heard continua whose endpoints were associated with different possibilities for real-word and nonword interpretations; the lexical status of the resulting item affected phonetic categorization. The phenomenon shows that word knowledge can shape a phoneme decision under uncertainty. To interpret the outcome mechanistically, investigators must distinguish the prior probability of a word, the moment when lexical context becomes available, and the rule used for an explicit response. The effect provides a key test case for models linking perception to language knowledge Ganong (1980) McMurray et al. (2003).

Evidence from incremental word recognition reveals that lexical constraints need not erase fine acoustic variation. McMurray and colleagues found graded effects of within-category acoustic differences on evolving lexical activation, allowing models to compare continuous competition with early all-or-none phoneme commitments. Dahan and colleagues likewise examined how subcategorical mismatch affects lexical access over time. Together, these findings encourage explanations in which phonetic evidence and lexical hypotheses interact dynamically. Their joint implication is more precise than an undifferentiated claim about top-down processing: acoustic detail can remain available while lexical alternatives compete, with response demands determining how that competition becomes observable McMurray et al. (2003) Dahan et al. (2001).

A well-designed lexical-context experiment includes matched words and nonwords, counterbalanced continua, and measures that separate altered perceptual sensitivity from altered choice preference. Signal detection analyses, reaction times, and eye movements are useful complements to identification curves, although each introduces assumptions about how latent evidence generates behavior. Investigators should also test whether an effect persists when attention is directed to sound properties instead of lexical status. An explanatory model must predict which contextual constraints alter early listening, which alter later inference, and which merely change the selection of a response in a constrained task Ganong (1980) Kleinschmidt & Jaeger (2015).

Perceptual Restoration and Speech in Noise

Conversational speech often arrives amid overlapping voices, background sound, reverberation, and missing acoustic segments. In such settings the listener combines imperfect bottom-up evidence with contextual expectations to recover likely utterances. Samuel's phonemic-restoration experiments provided an influential method for investigating the perceptual consequences of replacing a speech segment with noise. A listener's report of hearing a missing sound can reflect an inference about what would have occurred, and the task requires care in separating subjective completion from accurate detection of physical acoustic information. The phenomenon illustrates a broader capacity for perceptual continuity under partial input Samuel (1981) Shahin et al. (2009).

Neural research supplies complementary measures of how degraded speech is processed. Shahin, Bishop, and Miller examined neural mechanisms associated with illusory filling-in, helping connect the behavioral restoration phenomenon to time-sensitive brain activity. However, a difference in neural response between completed and incomplete percepts does not by itself identify the direction of causal influence. Predictive expectations, auditory integration, response preparation, and attention can all contribute to measured signals. Experiments with carefully equated acoustic materials and orthogonal manipulation of context are therefore important for evaluating claims about the neural implementation of restoration Shahin et al. (2009) Hickok & Poeppel (2007).

Speech-in-noise research is particularly sensitive to ecological differences among listening situations. Steady noise, competing speech, and temporally fluctuating maskers interfere with intelligibility in different ways; familiarity with the speaker, vocabulary, and topic can alter the amount of contextual information available. A measured benefit from context should be described for the exact masker, signal-to-noise manipulation, linguistic materials, and population studied. Models of robust perception can accommodate these effects as inference under uncertainty, while leaving open whether a particular advantage originates in acoustic segregation, phonetic categorization, lexical prediction, or response strategy Kleinschmidt & Jaeger (2015) Sumby & Pollack (1954).

Audiovisual Speech and the McGurk Phenomenon

Speech perception commonly combines auditory input with information from the speaker's face. Visible lip and jaw movements may constrain the identity and timing of expected sounds, especially when the auditory signal is degraded. Sumby and Pollack's early study demonstrated a contribution of visible speech to intelligibility in noise, establishing a foundation for investigating multisensory benefit in everyday communication. The size and character of that benefit depend on viewing conditions, signal clarity, talker visibility, and whether auditory and visual streams plausibly belong to the same event. The appropriate dependent variable is often intelligibility or identification accuracy, not a general assertion that vision dominates hearing Sumby & Pollack (1954) McGurk & MacDonald (1976).

McGurk and MacDonald presented mismatched auditory and visual syllables and described compelling changes in what listeners reported hearing. The resulting responses show that a visual articulation can alter the perceived or reported identity of an auditory speech event. The name 'McGurk effect' describes an experimental phenomenon, while its incidence and strength vary with stimuli, language history, attention, synchronization, and presentation procedure. Consequently, a single incongruent audiovisual token should not serve as a universal quantitative measure of multisensory integration. Proper inference requires controls for unisensory identification and for the physical correspondence of the two streams McGurk & MacDonald (1976) Sumby & Pollack (1954).

An integrated account must explain both facilitation from congruent visual speech and conflict from incongruent signals. Under uncertainty, information from another modality can improve category estimates when it is reliable; when sources disagree, inference depends on their relative credibility and common-source assumptions. Experimental studies can manipulate visual clarity, auditory noise, and audiovisual asynchrony independently to identify the regime where integration is most useful. These issues connect speech perception to general perceptual inference, but the learned mapping between articulation and phonological contrasts remains a linguistically distinctive component of audiovisual speech recognition McGurk & MacDonald (1976) Kleinschmidt & Jaeger (2015).

Neural Feature Selectivity and Category Coding

Neurophysiological studies clarify how speech-relevant features are represented across human auditory cortex. Mesgarani and colleagues analyzed neural responses in the superior temporal gyrus and found structured encoding of phonetic features. Because electrocorticography provides fine temporal resolution from clinically placed electrodes, it is informative about feature timing and response selectivity in the sampled cortical regions. Its spatial coverage and the characteristics of surgical patient samples also limit any claim about universal cortical organization. The evidence is most useful when linked to independent acoustic feature models and to behavioral predictions of which distinctions the listener can make Mesgarani et al. (2014) Hickok & Poeppel (2007).

Chang and colleagues described categorical representations of speech in superior temporal gyrus, while Toscano and colleagues reported electrophysiological evidence for a linear relation between acoustic variation and perceptual encoding. These studies may appear to support incompatible descriptions only if all neural signals are assumed to reflect the same population, processing stage, and representational scale. The empirical task is to determine whether continuous acoustic encoding and nonlinear category-related responses are distributed across time, cortical areas, neural populations, or measurement procedures. A multilayer account can accommodate both classes of response, provided it makes new predictions about their interactions Chang et al. (2010) Toscano et al. (2010).

Neural selectivity is an observation, and phonological representation is an explanatory interpretation that needs additional support. A site responding differently to two phonemes may also be sensitive to correlated acoustic features, familiarity, lexical expectations, or task structure. Conversely, a response tracking continuous acoustics may still contribute to categorical readout downstream. Cross-speaker generalization, controlled cue conflicts, and comparisons with participant-level discrimination functions help test whether a neural measure encodes contrast identity rather than the incidental composition of one stimulus set. The strongest conclusions arise when converging behavioral and neural evidence places constraints on a specified computational model Mesgarani et al. (2014) Chang et al. (2010) Kleinschmidt & Jaeger (2015).

Cortical Systems and Temporal Processing

Speech recognition depends on distributed cortical systems that process temporal, spectral, and linguistic information at different representational levels. Hickok and Poeppel's cortical organization account provided an influential synthesis of how auditory speech processing relates to phonological, lexical, and sensorimotor functions. Its explanatory value lies partly in distinguishing tasks that require understanding spoken language from those involving articulation or explicit sound judgments. A brain region recruited during one phonetic task is not necessarily a dedicated location for a single phoneme category. Network architecture is best examined through convergent lesion, behavioral, and physiological evidence rather than isolated activation maps Hickok & Poeppel (2007) Mesgarani et al. (2014).

Experience affects cortical responses to language-specific auditory patterns. Zhang and colleagues investigated neural commitment to language-specific auditory structures, providing an example of how learning history may influence neural sensitivity. Interpretation requires care because language groups may differ in acoustic exposure, attention, familiarity with laboratory materials, and task performance. A neural difference is strongest evidence for a language-related representation when the stimuli and behavioral response requirements are balanced and alternative acoustic explanations are evaluated. Developing speech categories likely involve linked changes across auditory encoding, attention, and linguistic inference rather than a single all-purpose neural marker Zhang et al. (2005) Kuhl (2004).

Temporal measurements offer an opportunity to test whether acoustic features, category estimates, and lexical expectations become available in a fixed succession or interact recurrently. Electrophysiological responses and intracranial recordings provide complementary time scales, but their signals are generated by different neural populations and measurement geometries. The essential research question is how information needed for successful phonological generalization emerges from these signals under realistic variability. Joint experiments that collect neural data and graded behavioral measures can constrain models more effectively than attributing a representational label to a particular waveform component alone Toscano et al. (2010) Mesgarani et al. (2014) Hickok & Poeppel (2007).

Psychophysical Methods, Stimulus Continua, and Measurement

The methodological foundation of this field is the controlled relation between an acoustic manipulation and observable discrimination or categorization. Synthetic continua can vary voice onset time, formant transitions, or spectral cues, while identification tasks estimate response functions and discrimination tasks examine sensitivity to change. Liberman and colleagues pioneered comparison of identification and discrimination near boundaries, but the value of this paradigm depends on stimulus calibration and the assumption that equal nominal steps are comparably perceptible. Researchers must report synthesis procedures, frequency content, playback conditions, trial randomization, and the response alternatives offered to participants Liberman et al. (1957) Lisker & Abramson (1964).

Psychometric functions provide estimates of category transition position and response slope. A steep slope can arise from well-separated internal distributions, a decisive response rule, or a mixture of both. The same listener may show a different boundary depending on speaking rate, speaker identity, lexical context, and task instructions. Auditory discrimination additionally depends on memory interval and attention; Pisoni demonstrated the importance of comparing auditory and phonetic memory contributions. Studies should therefore document the temporal relation between stimuli, the number of observations across the continuum, participant-level variability, and whether models are evaluated beyond the fitted training data Pisoni (1973) Toscano & McMurray (2010).

Ecological methods complement highly controlled stimuli. Naturalistic recordings preserve realistic variation, while gated materials reveal the accumulation of information during successive portions of an utterance. Eye tracking can estimate time-varying lexical competition, and electrophysiology can constrain the timing of feature extraction. None of these measures provides a transparent readout of a mental category independent of analytical assumptions. Triangulation is strongest when the same hypothesis predicts an identification boundary, a discrimination pattern, a graded online response, and generalization to a new speaker or recording condition. Such convergence addresses psychological construct validity rather than merely increasing the number of instruments used McMurray et al. (2008) Toscano et al. (2010).

Experimental Validity, Replication, and Generalization

The reliability of speech-perception findings depends on reproducible stimulus generation, sampling, counterbalancing, and a transparent account of exclusions. Small laboratory changes in headphones, noise level, synthesis parameters, or feedback can alter the effective information available to participants. Cross-language studies add variation in experience, multilingual history, hearing status, and language-specific phonological structure. A claim of generality requires replication across independent materials and participants, ideally with preregistered primary outcomes and uncertainty intervals. Developmental studies must also report how age-appropriate task differences constrain apparent changes in discrimination Werker (2024) Best & McRoberts (2003).

Interpretive disagreements often arise because methods measure different constructs. Two-alternative identification emphasizes a decision boundary, whereas lexical eye tracking reveals time-varying competition and electrophysiological responses may track a physical acoustic dimension even without overt categorization. Toscano and colleagues' neural evidence of continuous encoding illustrates why a categorical behavioral report cannot be taken as the complete perceptual state. A replication should preserve the theoretical manipulation and report whether the original outcome depends on one stimulus continuum, one analysis window, or one response procedure. Construct-level replication matters as much as exact duplication of laboratory settings Toscano et al. (2010) McMurray et al. (2008).

Quantitative interpretations should distinguish reliable condition differences from effect sizes that generalize to the wider population. Sampling uncertainty, multiple comparisons, model-fitting choices, and acoustic confounds can exaggerate apparent category or context effects. Research syntheses should describe the underlying design before comparing effects across studies, because measures expressed as boundary shifts, percent correct, response time, or neural amplitude are not automatically commensurable. The present monograph therefore reports the direction and interpretive character of confirmed experimental observations without assigning unsupported pooled estimates or declaring a formal meta-analytic consensus. These safeguards are particularly important in developmental and cross-language work Werker (2024) Kleinschmidt & Jaeger (2015).

Theoretical Models: Abstraction, Exemplars, and Probabilistic Prediction

Theoretical accounts differ in the representational units they propose. Abstraction-oriented models describe learned phonetic or phonological categories that support generalization across tokens. Episodic accounts retain detailed traces of particular speech experiences and explain recognition through similarity to remembered exemplars. Goldinger's episodic model shows how voice-specific detail can contribute to lexical access, challenging a representation restricted to invariant symbolic phonemes. Neither a behaviorally stable category nor a memory for distinctive voices decides the issue alone: a model must explain both generalization to novel material and detailed sensitivity to previously encountered utterances Goldinger (1998) McMurray et al. (2008).

Probabilistic inference provides a mathematical language for combining acoustic evidence with expectations. Kleinschmidt and Jaeger's account treats robust recognition as recognizing familiar input, generalizing to similar conditions, and adapting to novelty. Cue-learning models further describe how acoustic dimensions acquire differential weight through their statistical structure. This framework can represent category uncertainty, source-dependent distributions, and learning from evidence without assuming that all cognitive operations happen at a single level. Its scientific value depends on parameter identifiability and predictions for held-out speakers, cue conflicts, and adaptation trajectories, not on the mere flexibility of fitting many patterns after observation Kleinschmidt & Jaeger (2015) Toscano & McMurray (2010).

Integration across model classes is possible when competing explanations are evaluated against the same observations. Exemplar memory can supply experience-dependent priors, category abstractions can represent distributions over contrasts, and decision models can transform continuous estimates into discrete responses. The substantive issue is how the representations are learned, which information remains accessible during recognition, and whether the mechanism predicts transfer beyond training. Neural feature coding constrains the implementation of such a system but does not directly adjudicate between a particular cognitive algorithm and another. Strong model comparison requires independently specified test data, transparent parameterization, and explicit alternative hypotheses Goldinger (1998) Chang et al. (2010) Mesgarani et al. (2014).

The Artificial Era and Functional Comparisons with Speech Technologies

Contemporary artificial speech systems make it possible to compare different solutions to the engineering problem of extracting linguistic regularities from variable acoustic streams. Such comparisons can help specify what should count as robustness: intelligibility across speakers, calibration under noise, adaptation to accent, retention of fine phonetic detail, and the ability to update categories without damaging previous performance. A computational model that reproduces human identification slopes is informative about one task, but equivalent outward performance does not establish a common psychological representation or developmental mechanism. Human experiments and engineered systems must be compared through matched inputs, well-defined outputs, and independently validated generalization tests Kleinschmidt & Jaeger (2015) Toscano & McMurray (2010).

Artificial systems can also serve as instruments for sharpening experimental hypotheses. One model may retain detailed acoustic sequences; another may compress them into discrete symbolic states; a third may infer context-dependent category probabilities. Comparing their responses to the same continuum, lexical ambiguity, or accent shift helps identify which representational assumptions are sufficient to reproduce specific behavioral patterns. A shared task score does not settle whether speech information is represented in the same way across human neural systems and trained computational architectures. The scientific contribution lies in explicit operational comparisons and mechanistic transparency, informed by human evidence about graded lexical sensitivity and neural feature responses McMurray et al. (2008) Mesgarani et al. (2014).

Within the Artificial Era, research on phonological categorization contributes to a larger study of how different systems organize information, adapt to environmental variability, and sustain communicative exchange. The relevant empirical program should include independent validation under distribution shift, controlled tests of audiovisual integration, and documented limits of generalization across languages. Theoretical categories such as artificial cognition can frame these investigations while the empirical claims remain tied to the measured performance of particular systems. This discipline preserves the distinction among a philosophical category, a proposed explanatory mechanism, and experimentally observed processing Hickok & Poeppel (2007) Kleinschmidt & Jaeger (2015).

Critical Scientific Discussion and Outstanding Problems

The research literature establishes that linguistic categories organize speech judgments, and that this organization coexists with measurable sensitivity to continuous acoustic structure. The explanatory challenge is to locate and characterize the interaction between them. Categorical identification can emerge from a decision rule applied to graded evidence, while neural signals may track features, category membership, or both at different times and cortical locations. Behavioral findings from McMurray and colleagues and electrophysiological findings from Toscano and colleagues create stringent tests for models claiming that phonemes are represented as discrete perceptual units throughout processing. A satisfactory theory must explain those data without dissolving the linguistic function of categories McMurray et al. (2008) Toscano et al. (2010).

Development remains another major area of theoretical uncertainty. Infant perceptual reorganization is supported by historical cross-language discrimination studies and later evidence about vowels, consonants, and experience, but its underlying stages vary across languages, contrast types, and methods. Questions remain about the relative importance of acoustic distributions, attentional allocation, word learning, social interaction, and multiple-language exposure. Werker's historical review cautions against a simplistic developmental account in which infants either possess or lose a fixed inventory of speech categories. Mechanistic research should map trajectories of changing generalization and cue use rather than infer a singular irreversible transition Werker (2024) Kuhl et al. (1992) Maye et al. (2002).

A further unresolved issue is the relationship among source adaptation, linguistic knowledge, and attentional decision processes. Listeners can rapidly retune unfamiliar sounds, yet the durability and specificity of that learning depend on circumstances and the representational units tested. Lexical influence, contextual restoration, and audiovisual effects may involve different mechanisms even when all alter speech judgments. Next-generation studies should combine ecologically valid speech, controlled acoustic manipulations, longitudinal experience measures, computational comparison, and preregistered independent replication. These designs can clarify which regularities are specific to linguistic categorization and which follow from broader auditory or inferential principles Norris et al. (2003) McGurk & MacDonald (1976) Kleinschmidt & Jaeger (2015).

Conclusion

Speech perception and phonological categorization form a coordinated psychological system for deriving linguistic contrasts from an acoustically variable stream. The historical experiments of Liberman and colleagues established a critical relation between acoustic continua and categorical responses, while cross-language acoustic and developmental research showed that categories must be understood in the context of linguistic experience. Subsequent evidence demonstrated that overt labels can be sharply categorical even when perception and lexical competition remain sensitive to continuous information. These results support treating categories as learned functional generalizations operating on a richer acoustic evidence base Liberman et al. (1957) Lisker & Abramson (1964) McMurray et al. (2008).

Experience shapes category use across the lifespan. Infant discrimination develops toward language-specific organization through exposure to acoustic and social regularities; adult listeners continue to adapt to unfamiliar speakers and ambiguous segments. Contextual word knowledge, auditory history, facial articulation, and noise conditions alter how linguistic evidence is interpreted. Neural findings reveal both feature-sensitive and category-associated responses, providing complementary constraints on the timescales and anatomical organization of speech processing. A complete scientific account must explain why these processes remain effective across variable voices, imperfect inputs, and changing environments Werker (2024) Norris et al. (2003) Mesgarani et al. (2014).

The synthesis identifies an enduring research agenda: connect identification, graded discrimination, source-specific learning, developmental change, and cortical coding within models tested against independent observations. Clear operational distinctions among acoustic differences, phonetic categories, phonological contrasts, and lexical decisions will improve the interpretation of contradictory findings. Methodological progress requires representative stimuli, cross-population validation, quantified uncertainty, and transparent computational comparison. The resulting understanding of human speech perception also provides experimentally grounded questions for artificial speech research, while preserving the particular evidence needed for claims about each kind of system. The research remains an integrative review of published science, with its conclusions limited by the available methods and studies Kleinschmidt & Jaeger (2015) Best & Tyler (2007) Hickok & Poeppel (2007).

References

  • Best, C. T., & McRoberts, G. W. (2003). Infant perception of non-native consonant contrasts that adults assimilate in different ways. Language and Speech, 46(2–3), 183–216. https://doi.org/10.1177/00238309030460020701
  • Best, C. T., & Tyler, M. D. (2007). Nonnative and second-language speech perception: Commonalities and complementarities. In O.-S. Bohn & M. J. Munro (Eds.), Language experience in second language speech learning (pp. 13–34). John Benjamins. https://doi.org/10.1075/lllt.17.07bes
  • Chang, E. F., Rieger, J. W., Johnson, K., Berger, M. S., Barbaro, N. M., & Knight, R. T. (2010). Categorical speech representation in human superior temporal gyrus. Nature Neuroscience, 13(11), 1428–1432. https://doi.org/10.1038/nn.2641
  • Dahan, D., Magnuson, J. S., Tanenhaus, M. K., & Hogan, E. M. (2001). Subcategorical mismatches and the time course of lexical access: Evidence for lexical competition. Language and Cognitive Processes, 16(5–6), 507–534. https://doi.org/10.1080/01690960143000074
  • Eimas, P. D., Siqueland, E. R., Jusczyk, P., & Vigorito, J. (1971). Speech perception in infants. Science, 171(3968), 303–306. https://doi.org/10.1126/science.171.3968.303
  • Eisner, F., & McQueen, J. M. (2005). The specificity of perceptual learning in speech processing. Perception & Psychophysics, 67(2), 224–238. https://doi.org/10.3758/BF03206487
  • Ganong, W. F., III. (1980). Phonetic categorization in auditory word perception. Journal of Experimental Psychology: Human Perception and Performance, 6(1), 110–125. https://doi.org/10.1037//0096-1523.6.1.110
  • Goldinger, S. D. (1998). Echoes of echoes? An episodic theory of lexical access. Psychological Review, 105(2), 251–279. https://doi.org/10.1037/0033-295X.105.2.251
  • Hickok, G., & Poeppel, D. (2007). The cortical organization of speech processing. Nature Reviews Neuroscience, 8(5), 393–402. https://doi.org/10.1038/nrn2113
  • Holt, L. L., & Lotto, A. J. (2002). Behavioral examinations of the level of auditory processing of speech context effects. Hearing Research, 167(1–2), 156–169. https://doi.org/10.1016/S0378-5955(02)00383-0
  • Kleinschmidt, D. F., & Jaeger, T. F. (2015). Robust speech perception: Recognize the familiar, generalize to the similar, and adapt to the novel. Psychological Review, 122(2), 148–203. https://doi.org/10.1037/a0038695
  • Kuhl, P. K. (2004). Early language acquisition: Cracking the speech code. Nature Reviews Neuroscience, 5(11), 831–843. https://doi.org/10.1038/nrn1533
  • Kuhl, P. K., Tsao, F.-M., & Liu, H.-M. (2003). Foreign-language experience in infancy: Effects of short-term exposure and social interaction on phonetic learning. Proceedings of the National Academy of Sciences, 100(15), 9096–9101. https://doi.org/10.1073/pnas.1532872100
  • Kuhl, P. K., Williams, K. A., Lacerda, F., Stevens, K. N., & Lindblom, B. (1992). Linguistic experience alters phonetic perception in infants by 6 months of age. Science, 255(5044), 606–608. https://doi.org/10.1126/science.1736364
  • Liberman, A. M., Harris, K. S., Hoffman, H. S., & Griffith, B. C. (1957). The discrimination of speech sounds within and across phoneme boundaries. Journal of Experimental Psychology, 54(5), 358–368. https://doi.org/10.1037/h0044417
  • Lisker, L., & Abramson, A. S. (1964). A cross-language study of voicing in initial stops: Acoustical measurements. Word, 20(3), 384–422. https://doi.org/10.1080/00437956.1964.11659830
  • Maye, J., Werker, J. F., & Gerken, L. (2002). Infant sensitivity to distributional information can affect phonetic discrimination. Cognition, 82(3), B101–B111. https://doi.org/10.1016/S0010-0277(01)00157-3
  • McGurk, H., & MacDonald, J. (1976). Hearing lips and seeing voices. Nature, 264, 746–748. https://doi.org/10.1038/264746a0
  • McMurray, B., Tanenhaus, M. K., Aslin, R. N., & Spivey, M. J. (2003). Probabilistic constraint satisfaction at the lexical/phonetic interface: Evidence for gradient effects of within-category VOT on lexical access. Journal of Psycholinguistic Research, 32(1), 77–97. https://doi.org/10.1023/A:1021937116271
  • McMurray, B., Aslin, R. N., Tanenhaus, M. K., Spivey, M. J., & Subik, D. (2008). Gradient sensitivity to within-category variation in words and syllables. Journal of Experimental Psychology: Human Perception and Performance, 34(6), 1609–1631. https://doi.org/10.1037/a0011747
  • Mesgarani, N., Cheung, C., Johnson, K., & Chang, E. F. (2014). Phonetic feature encoding in human superior temporal gyrus. Science, 343(6174), 1006–1010. https://doi.org/10.1126/science.1245994
  • Norris, D., McQueen, J. M., & Cutler, A. (2003). Perceptual learning in speech. Cognitive Psychology, 47(2), 204–238. https://doi.org/10.1016/S0010-0285(03)00006-9
  • Pisoni, D. B. (1973). Auditory and phonetic memory codes in the discrimination of consonants and vowels. Perception & Psychophysics, 13(2), 253–260. https://doi.org/10.3758/BF03214136
  • Samuel, A. G. (1981). Phonemic restoration: Insights from a new methodology. Journal of Experimental Psychology: General, 110(4), 474–494. https://doi.org/10.1037//0096-3445.110.4.474
  • Shahin, A. J., Bishop, C. W., & Miller, L. M. (2009). Neural mechanisms for illusory filling-in of degraded speech. NeuroImage, 44(3), 1133–1143. https://doi.org/10.1016/j.neuroimage.2008.09.045
  • Sumby, W. H., & Pollack, I. (1954). Visual contribution to speech intelligibility in noise. Journal of the Acoustical Society of America, 26(2), 212–215. https://doi.org/10.1121/1.1907309
  • Toscano, J. C., & McMurray, B. (2010). Cue integration with categories: Weighting acoustic cues in speech using unsupervised learning and distributional statistics. Cognitive Science, 34(3), 434–464. https://doi.org/10.1111/j.1551-6709.2009.01077.x
  • Toscano, J. C., McMurray, B., Dennhardt, J., & Luck, S. J. (2010). Continuous perception and graded categorization: Electrophysiological evidence for a linear relationship between the acoustic signal and perceptual encoding of speech. Psychological Science, 21(10), 1532–1540. https://doi.org/10.1177/0956797610384142
  • Werker, J. F., & Tees, R. C. (1984). Cross-language speech perception: Evidence for perceptual reorganization during the first year of life. Infant Behavior and Development, 7(1), 49–63. https://doi.org/10.1016/S0163-6383(84)80022-3
  • Werker, J. F. (2024). Phonetic perceptual reorganization across the first year of life: Looking back. Infant Behavior and Development, 75, Article 101935. https://doi.org/10.1016/j.infbeh.2024.101935
  • Zhang, Y., Kuhl, P. K., Imada, T., Kotani, M., & Tohkura, Y. (2005). Effects of language experience: Neural commitment to language-specific auditory patterns. NeuroImage, 26(3), 703–720. https://doi.org/10.1016/j.neuroimage.2005.02.040