Exploring phonological complexity in statistical learning of artificial words

Akshay R. Maggu; Tobias Overath

doi:10.1371/journal.pone.0341771

Abstract

Purpose

This study examined whether phonological complexity enhances auditory word learning within a statistical learning framework. Specifically, we tested if exposure to phonologically complex speech patterns (i.e., marked consonant clusters) facilitates the segmentation and generalization of both complex and simple artificial words.

Method

Seventy-eight adults were randomly assigned to either a complex or simple pattern induction group and exposed to bisyllabic artificial words varying in onset complexity. Participants then heard an auditory stream containing both complex and simple words, followed by a wordlikeness rating task assessing both previously heard and novel items.

Results

Exposure to complex speech patterns did not enhance wordlikeness ratings for either previously heard (stream) or novel (generalization) artificial words. Wordlikeness ratings were significantly higher for stream items than for generalization items, indicating sensitivity to exposure, but no reliable effects of induction condition or stimulus complexity were observed.

Conclusions

Findings suggest that passive exposure to complex patterns does not enhance generalization within a statistical learning paradigm.

Citation: Maggu AR, Overath T (2026) Exploring phonological complexity in statistical learning of artificial words. PLoS One 21(2): e0341771. https://doi.org/10.1371/journal.pone.0341771

Editor: Nicola Molinaro, Basque Center on Cognition Brain and Language, SPAIN

Received: October 10, 2025; Accepted: January 12, 2026; Published: February 3, 2026

Copyright: © 2026 Maggu, Overath. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: All data underlying the findings reported in this study, including behavioral data, stimuli, experimental scripts, and analysis code, are publicly available from the OSF repository (https://doi.org/10.17605/OSF.IO/7Q5UX).

Funding: This study was financially supported by the Institute for the Brain and Cognitive Sciences, University of Connecticut (https://braincognitivesciences.institute.uconn.edu/) in the form of seed funding received by AM. No additional external funding was received for this study.

Competing interests: The authors have declared that no competing interests exist.

Introduction

Speech sound acquisition is central to spoken language and remains a topic of debate across theoretical frameworks. One open question concerns the role of input complexity in driving acquisition. Complexity can be defined in various ways—phonologically (e.g., markedness), perceptually (e.g., stimulability), or statistically (e.g., transitional probabilities)—and may interact with learner-specific factors such as developmental stage or prior exposure. Some theories emphasize that acquisition proceeds from simple to complex structures based on perceptual salience, articulatory ease, or early developmental accessibility (e.g., behaviorist, connectionist, and dynamic systems theories), while others argue that exposure to complex, marked forms facilitates system-wide generalization by engaging abstract grammatical representations, e.g., [1]. A broad range of theoretical accounts address these issues, including usage-based [2], connectionist [3], dynamic systems [4], Bayesian [5], and statistical learning frameworks [6]. While these models differ in their assumptions, the present study focuses specifically on two well-defined and empirically grounded accounts—linguistic complexity, operationalized through phonological markedness, and statistical learning, defined as sensitivity to the distributional structure of speech. These frameworks offer contrasting yet complementary predictions regarding how input complexity influences learning. Our goal, in the current study, is to examine how phonological complexity may modulate the statistical learning of word-like units in an artificial language paradigm. Understanding how these two learning mechanisms jointly influence acquisition has implications for both theoretical models of language development and clinical approaches to speech sound disorders.

Complexity in speech and language acquisition

Complexity-based theories of language acquisition, in line with generative grammar [7], natural phonology [8], and optimality theory (OT) [9], propose that exposure to complex linguistic input facilitates acquisition by restructuring the internal grammar and triggering generalization to less complex forms. Linguistic complexity in speech sound acquisition can be defined along several dimensions, including phonological markedness, lexical frequency, stimulability, and developmental consistency. Gierut [10] outlined that complexity may arise from linguistic, psycholinguistic, articulatory-phonetic, and conventional clinical factors, all of which may influence the efficacy of intervention and generalization patterns.

Phonological complexity, in particular, has been the focus of numerous speech therapy-based treatment studies [11] focused on speech production in children (3–7-year-olds) with phonological disorders (i.e., speech sound difficulties characterized by deficits in phonological representations and rule-based pattern learning, such as systematic error patterns, rather than motor articulation impairments; see [12,13]). Gierut et al. [14] conducted a seminal study comparing two approaches: training with sounds representing the child’s “most knowledge” (i.e., partially acquired) versus “least knowledge” (i.e., completely unknown). Children trained on least knowledge, or more complex, sounds demonstrated broader generalization to untreated sounds (i.e., the sounds that were not used as training stimuli), supporting the idea that complex targets result in system-wide change. Extending this framework, Gierut et al. [1] selected marked speech sounds based on implicational relationships derived from OT hierarchies. Their findings showed that speech therapy with marked sounds, such as fricatives or clusters, resulted in improvement in production of both marked and unmarked sounds, while speech therapy with unmarked sounds yielded limited generalization. Similarly, Powell et al. [15] trained children with either fricatives (more marked) or stops (less marked) and found that those trained with fricatives improved in producing both fricatives and stops, whereas those trained on stops showed limited improvement beyond the trained sounds. Further evidence comes from studies that directly manipulated the number or novelty of phonemes in treatment. Gierut [16] found that training with two new phonemes led to greater generalization than training with one new phoneme. Gierut and Neumann [17] compared homonymous and non-homonymous minimal pair therapy approaches and found that non-homonymous pairs, which were fully outside the child’s system, induced greater phonological change. Dinnsen [18] examined therapy outcomes in children trained on either marked onset clusters or less marked affricates and found that the cluster group showed more robust generalization. These findings reinforce the notion that marked or complex input engages representational mechanisms that are not triggered by simple input, leading to reorganization of the phonological system. In addition to phonological complexity, other factors such as lexical frequency and stimulability have been shown to modulate speech production outcomes. Gierut et al. [19] reported that high-frequency words, which are more complex at the sublexical level, led to greater generalization than low-frequency words. Powell et al. [20] demonstrated that treatment involving conceptual and linguistic engagement was more effective than simple motor-based articulation training. Gierut [21] showed that minimal pairs using unknown, and therefore more complex, sounds were more beneficial than those using familiar contrasts. Across these studies, complexity, whether defined by knowledge status, markedness, frequency, stimulability, or methodological strategy, has consistently been associated with greater system-wide generalization.

The effect of complexity is not limited to speech production. Maggu et al. [22] found that exposure to more marked speech contrasts in a short (15–20 minutes in total) pseudoword learning task facilitated perceptual generalization. In this study, Cantonese listeners were trained to associate Hindi prevoiced dental-retroflex contrasts (more marked) with pictures to form novel words. Those trained on prevoiced contrasts generalized to voiceless contrasts (less marked), but the reverse was not true. This finding extended the complexity effect from production to perceptual learning in typical adults. Similar complexity effects have also been observed in syntax. Thompson and Shapiro [23] demonstrated that training with syntactically complex sentence structures, such as object relatives, led to generalization to simpler structures, such as actives, while the reverse pattern did not hold. Collectively, the aforementioned findings provide strong empirical support for the complexity hypothesis. Exposure to complex structures facilitates acquisition by engaging deeper grammatical or cognitive operations and enabling generalization across untreated forms. While most of this evidence has emerged from treatment/training-based studies, the current study builds on this work by embedding linguistic complexity within a statistical learning framework to examine how these mechanisms jointly influence word learning.

On the other hand, it should be noted that not all findings support the complexity hypothesis. Rvachew and Nowak [24] found that children with phonological disorders demonstrated greater improvement when treatment targeted stimulable, early-acquired sounds compared to non-stimulable, complex ones. These results align with principles from dynamic systems theory, which emphasizes that stability in underlying components is essential for development to proceed. In the domain of speech acquisition, stimulability of simple sounds is thought to provide the foundation for acquiring more complex forms [4,25]. For example, teaching marked consonant clusters may be more successful if the child is first stimulable for unmarked plosives. Moreover, dynamic systems theory, like behaviorist accounts [26], highlights the role of reinforcement and graded exposure in shaping accurate speech patterns. Together, these findings underscore that complexity effects may be moderated by individual learner profiles and developmental readiness.

Statistical learning in speech and language acquisition

In general, statistical learning suggests that acquisition of speech and language is a result of exposure to patterns that repeat with high probability. More specifically, it has been found that infants and children, even with minimal exposure to an artificial language, can exploit the transitional probability (TP) within a stream of sounds to find patterns that can be construed as “words” [6,27,28]. TPs are calculated as the conditional probability of one syllable following another. For example, in the four-syllable stream “prettybaby,” the TP between “pre” and “tty” (P(tty|pre)) and between “ba” and “by” (P(by|ba)) is relatively high because these syllables co-occur within real words [29]. In contrast, the TP between “tty” and “ba” (P(ba|tty)) is low, as this syllable pair straddles a word boundary. Listeners, including infants, are sensitive to these differences and are more likely to segment high-TP syllable pairs as cohesive word-like units. Indeed, infants as young as 8 months have been shown to extract trisyllabic words from continuous speech streams with only two minutes of exposure and perform above chance on segmentation tasks [30]. Further, statistical learning is a domain-general phenomenon, shown to operate not only in speech and language acquisition, but also in the domains of vision and music, and even in non-human species [30–32]. Besides the studies in typical speech and language acquisition, the statistical learning paradigm has also been used to study language acquisition in disorders such as specific language impairment [33].

Further, it has been found that statistical learning can be modulated by even prior brief exposure to phonological patterns. Along these lines, the statistical learning paradigm has been used in conjunction with pattern induction [34], i.e., it has been found that exposure to simple nonsense consonant (C) vowel (V) speech patterns facilitates parsing out the “words” from an auditory stream. Saffran and Thiessen [34] investigated how prior exposure to simple speech sound patterns facilitates SL in infants. In their experiments, 9-month-old infants were first exposed to different types of “pattern induction” stimuli—CVCV (e.g., “pada”) or CVCCVC (e.g., “padtak”) nonsense words—for 2 minutes. Then, the infants heard an artificial language stream for 1 minute containing new CVCV or CVCCVC items with statistically defined word boundaries based on transitional probabilities, following which word segmentation was evaluated using a head-turn procedure. The key finding was that infants exposed to simple CVCV patterns in the induction phase showed stronger statistical learning (i.e., better word segmentation) than those exposed to complex CVCCVC patterns. This suggested that even brief exposure to simple structural patterns can tune infants’ expectations about phonological regularities, enhancing their ability to extract words from fluent speech. Notably, learning generalized from simple to complex forms but not the reverse—posing a challenge to the complexity hypothesis, which predicts broader generalization from complex input. These results highlight the importance of input structure in scaffolding learning. However, the complexity manipulation in this study was based on syllabic structure (i.e., number of consonants), not phonological markedness. As such, while these findings show how structural simplicity can support statistical learning, they do not directly address complexity effects as defined by phonological markedness.

Combination of linguistic complexity and statistical learning

Evidence from syntax-based studies suggests that a combination of innate rules/complexity and statistical learning might be needed to drive language acquisition [29]. Evidence from computational studies in phonology suggests that when a phonological constraint hierarchy is applied to an auditory stream, this leads to better word segmentation performance as compared to statistical learning alone [35]. However, to date, no behavioral studies have directly examined how phonological complexity, defined by markedness constraint hierarchy, interacts with statistical learning in shaping word segmentation.

To fill this gap, the current study combined statistical learning and phonological complexity (based on OT constraint hierarchy) to examine whether this combination enhanced learning and generalization beyond statistical learning alone. Specifically, we adapted the pattern induction paradigm of Saffran and Thiessen [34] and manipulated phonological complexity based on the markedness implicational hierarchy (i.e., clusters > plosives [36]). In the first stage (Pattern Induction), one group of participants was exposed to bisyllabic artificial words beginning with clusters (i.e., complex patterns), while the other group was exposed to bisyllabic artificial words starting with plosives (i.e., simple patterns). In the second stage (Word Segmentation), both groups were exposed to an auditory stream containing two new cluster-initial words (e.g.,/blɑfi/,/ʃɹubɑ/) and two new plosive-initial words (e.g.,/bɑfi/,/ʃubɑ/). In the third stage (Wordlikeness Testing), participants rated both previously heard words (to assess learning) and novel words (to assess generalization) on a 7-point Likert scale based on perceived wordlikeness. We hypothesized that if phonological complexity, as defined by OT constraint hierarchy (i.e., clusters > plosives) interacted with statistical learning, the participants induced with complex patterns (i.e., cluster-initial words), after listening to the auditory stream with both complex and simple words, would rate the words starting with both clusters and plosives as more word-like. On the other hand, if the phonological complexity did not interact with statistical learning, the participants induced with complex patterns (i.e., starting with clusters), after listening to the auditory stream with both complex and simple words, would rate the words starting with clusters as more word-like, but they would not rate the words starting with plosives as word-like.

Materials and methods

Participants

We recruited a total of 83 participants from the University of Connecticut community. They were paid at the rate of $15/hour for their participation in the study. Five participants were excluded due to being outside the normative range of hearing abilities leading to the final included sample of 78 participants (mean age: 20.47 years; SD: 3.21; 15 males, 63 females). All participants self-reported as monolingual speakers of American English, with no history of speech or language disorders, special education services, or accessibility accommodations. Participants represented a range of academic majors, including but not limited to psychology, engineering, communication disorders, biology, nursing, education, and business. The study was approved by the Institutional Review Board of the University of Connecticut (IRB # H23-0786). All participants provided written informed consent prior to participation. The recruitment start and end dates were 04/09/2024 and 27/10/2024, respectively.

A priori power analysis: A power analysis was conducted using pilot data from 10 participants exposed to complex patterns and 10 to simple patterns. The analysis determined that, to achieve a power of 0.8 with an α of 0.5 for an effect size of 0.25, a sample of 36 participants per group would be required. To account for potential attrition, the current study recruited slightly more participants than the suggested sample size.

Hearing screening

All participants were screened for their hearing abilities for frequencies 250 Hz, 500 Hz, 1000 Hz, 2000 Hz, and 4000 Hz using a Maico MA 27 screening audiometer. Pure tones were routed via DD45 headphones. Participants with > 20 dB HL on any tested frequency were excluded from the study. Five participants did not meet this criterion and were excluded; all remaining participants met the hearing screening criterion at all tested frequencies.

Material

Stimuli consisted of bisyllabic artificial words, either as C_mVCV (with marked, or complex initials) or as C_uVCV (with unmarked, or simple initials) combinations. C_mVCV sound combinations started with clusters (e.g.,/tr/ as in “trim”) while C_uVCV sound combinations started with plosives (e.g.,/t/ as in “tim”). Stimuli were recorded by a 25-year-old phonetically-trained male speaker using a Shure SM10A microphone and Praat [37] with a 44,100 Hz sampling rate and 16-bit resolution. Each syllable in the sound stimuli was further duration-normalized to 300 ms, intensity-normalized to 70 dB and pitch-matched to 120 Hz flat pitch contour. Table 1 lists the speech stimuli used across the three stages of the study.

Download:

Table 1. Stimuli (listed in International Phonetic Alphabet) used across the three stages of the study: (A) Pattern Induction; (B) Word Segmentation; and (C) Wordlikeness Testing.

https://doi.org/10.1371/journal.pone.0341771.t001

Procedure

The study consisted of three stages: Pattern Induction, Word Segmentation, and Wordlikeness Testing (Fig 1). The study was run using E-prime 3.0 (E-Studio Build 3.0.3.219) on a Dell laptop connected to a Roland Rubix 22 sound card and stimuli were delivered binaurally at a comfortable intensity level of 70 dB SPL via Sennheiser HD 280 Pro headphones. Participants were instructed at the start of the study to pay attention throughout the multiple stages of the study, with a special emphasis on the use of information from the first two stages for the tasks in the third stage. Fig 1 shows a flowchart of the study. The whole procedure took 5–6 minutes to complete.

Download:

Fig 1. Flowchart describing the procedure of the study which entailed using the knowledge from the (A) Pattern Induction stage to (B) segment the words from the auditory stream to (C) rate the wordlikeness on a 7-point Likert scale.

https://doi.org/10.1371/journal.pone.0341771.g001

Pattern induction.

The participants were randomly allocated to one of the two groups – Complex pattern group (n = 39; mean age: 19.94; SD: 2.4; 30 females) and Simple pattern group (n = 39; mean age: 21; SD: 3.8; 33 females). Participants in the Complex pattern group were exposed to the 30 bisyllabic speech stimuli with clusters (i.e., complex stimuli) in the initial position, while the Simple pattern group were exposed to 30 bisyllabic speech stimuli with plosives or fricatives (i.e., simple stimuli) in the initial position (Fig 1 (A)). All participants listened to a random order of presentation of the 30 stimuli. Each stimulus was 600 ms long and separated by an inter-stimulus interval of 2 s. Participants were instructed to attentively listen to the artificial words, as they would use them in subsequent stages of the experiment. However, they were not informed that the task involved identifying word boundaries. This phase was conducted without any feedback, consistent with procedures in seminal pattern induction-based statistical learning studies, e.g., [34]. Following the presentation of the 30 artificial words, the experiment moved to the second stage – Word Segmentation.

Word segmentation.

All participants, regardless of the patterns that they were exposed to (i.e., Complex or Simple) in the first stage, were asked to listen to an auditory stream that contained four artificial words, of which two words started with complex clusters (e.g.,/blɑfi/,/ʃɹubɑ/) and the other two started with simple plosives or fricatives (e.g.,/bɑfi/,/ʃubɑ/). These four stimuli were randomly repeated 26 times each, leading to a total of 104 stimuli presentations over a duration of 1 min and 12 secs (Fig 1 (B)). Each participant listened to one of three auditory streams with different speech stimuli (see Table 1), and within each auditory stream, one of three separate randomizations of the stimuli. Auditory streams and randomizations were counterbalanced across participants; because preliminary analyses found no differences between streams and their versions, data from the three streams were pooled for subsequent analyses. As in the Pattern Induction phase, participants were instructed only to listen attentively. As this phase was designed to promote passive exposure, no mention was made of word boundaries or statistical regularities, and no feedback was provided. Following the presentation of the 104 stimuli, the experiment moved to the third stage – Wordlikeness Testing.

Wordlikeness testing.

All participants, regardless of the type of patterns that they were exposed to (Complex or Simple) in the first stage, were given the Wordlikeness Testing that contained complex and simple artificial words, both old (from the Word Segmentation stage) and new (not presented in Pattern Induction and Word Segmentation stages). Specifically, stimuli in this stage consisted of 8 artificial words, with four artificial words (two simple, two complex) from the word segmentation stage (i.e., learned stimuli), and four novel artificial words (two simple, two complex) that did not appear in the word segmentation stage (i.e., generalization stimuli) (see Table 1 for details). Based on the knowledge of the stimuli that they were exposed to in the pattern induction stage and in the word segmentation stage, participants were asked to rate the stimuli in this stage on a 7-point Likert scale in terms of their wordlikeness (based on their exposure so far), with 1 being “least likely” and 7 being “most likely” a ‘word’ (Fig 1 (C)). Each word was played twice in a randomized order. The use of a rating scale rather than forced-choice recognition was intended to elicit gradient judgments of wordlikeness and reduce reliance on explicit recall or binary recognition. This measure has been used in prior work to index perceptual sensitivity to statistical structure without requiring overt segmentation judgments [38,39].

Statistical analysis

Wordlikeness ratings were analyzed using linear mixed-effects models implemented in the lme4 package in R. Fixed effects included Induction Group (Simple vs. Complex), Stimulus Complexity (Simple vs. Complex), Novelty (Stream vs. Generalization), and all interactions. Participant was included as a random intercept to account for repeated measurements within individuals. This modeling approach allowed all conditions to be analyzed within a single framework, rather than conducting separate analyses for stream and generalization items.

Model assumptions were assessed by visual inspection of residual distributions and residuals-versus-fitted plots. Post-hoc comparisons were conducted using estimated marginal means (emmeans) with Holm correction for multiple comparisons.

Results

To evaluate whether pattern induction (Simple vs. Complex) modulated wordlikeness ratings as a function of stimulus complexity (Simple vs. Complex) and novelty (Word Segmentation Stimuli vs. Generalization Stimuli), wordlikeness ratings were analyzed using a linear mixed-effects model with fixed effects of Induction Group, Stimulus Complexity, Novelty, and their interactions, and a random intercept for participant. The analysis revealed a significant main effect of Novelty, such that Word Segmentation items were rated as more wordlike than Generalization items, F(1, 234) = 9.42, p = .002 (Fig 2). No other main effects or interactions reached statistical significance (all ps ≥ .106). Planned contrasts indicated that induction-group differences were not significant within the Word Segmentation Stimuli condition for either simple stimuli (Simple induction: M = 4.20; Complex induction: M = 3.58; p = .105) or complex stimuli (Simple induction: M = 3.84; Complex induction: M = 3.83; p = .973), nor within the Generalization Stimuli condition (all ps ≥ .471). Although the Word Segmentation/Simple condition showed a numerically higher mean following simple induction, this difference did not reach statistical significance after correction.

Download:

Fig 2. Comparison of the group exposed to Complex versus Simple patterns on the wordlikeness of the complex and simple stimuli for the (A) Word Segmentation Stimuli and for the (B) Generalization Stimuli.

Error bars = ± 1 SEM.

https://doi.org/10.1371/journal.pone.0341771.g002

Discussion

In the current study, we investigated whether phonological complexity interacts with statistical learning to maximize word learning. More specifically, we examined whether exposing young adults to complex speech patterns that rank higher in the markedness hierarchy (i.e., clusters) leads to increased segmentation of both complex (i.e., C_mVCV) and simple words (i.e., C_uVCV) such that wordlikeness ratings would be enhanced for both complex and simple words. Overall, we found no effect of exposure of complex stimuli on wordlikeness ratings for complex and simple words, both for the words they were exposed to (i.e., stream condition) and novel words (i.e., generalization condition). Overall, wordlikeness ratings were reliably higher for items encountered in the word segmentation stage than for novel generalization items.

Contrary to predictions derived from the complexity hypothesis, exposure to complex patterns did not lead to enhanced ratings of either complex or simple stream items. Linear mixed-effects modeling revealed no significant main effects or interactions involving induction group or stimulus complexity. Although a numerical difference was observed for simple word segmentation items, this effect did not reach statistical significance after correction and did not extend to complex stimuli or generalization items. These findings suggest that, within a passive statistical learning paradigm, exposure to complex phonological structures alone is insufficient to drive system-wide generalization.

These results diverge from prior treatment-based and explicit learning studies [1,11,14,40,41] that suggest that exposure to more complex or more marked structures leads to the acquisition of both marked and unmarked structures. The main point of distinction between the current and the previous studies that showed the generalization effects of exposure to complex speech sounds on both complex and simple sounds is that the previous studies, e.g., Maggu et al. [22] used an explicit word-picture association paradigm, while the methodology of the current study involved embedding complex structures along with simple structures within a statistical learning paradigm. This difference in findings could suggest that the principles of complexity theories [42–44] are valid for explicit learning of complex structures that rank higher in the markedness hierarchy to promote acquisition of structures that are less or equivalently complex, which might not be the case when using an implicit statistical learning paradigm as in the current study.

In fact, our results are more aligned with findings from Saffran and Thiessen [34], who demonstrated that infants exposed to simpler phonotactic patterns (CVCV) were more successful in segmenting novel words than those exposed to more complex patterns (CVCCVC). In their study, learning generalized from simple to complex patterns but not the reverse, an outcome that challenges the complexity hypothesis and supports the idea that structural simplicity can scaffold early statistical learning. Similarly, in the current study, participants exposed to simple patterns exhibited numerically higher wordlikeness ratings for simple stream items and comparable ratings for complex stream items relative to participants exposed to complex patterns. Although these differences did not reach statistical significance after correction, the pattern is consistent with the idea that prior exposure to simpler structures may support the processing of more complex forms during passive exposure in an artificial language learning context.

Further, the current findings are also different from computational data on maximizing speech sound acquisition by embedding linguistic constraints within a statistical learning paradigm [35], probably due to inherent differences in the research designs of existing computational findings and the current behavioral study. The present findings can also be interpreted in the context of research on substantive bias in phonology, which suggests that learners tend to favor patterns that are more perceptually and articulatorily familiar or typologically less marked over those that are more marked [45,46]. Consequently, due to substantive bias in artificial language learning, exposure to complex or highly marked structures may not have generalized to less marked ones. Consistent with this view, the current study revealed limited evidence for complexity-driven advantages and no reliable generalization to novel items, suggesting that statistical learning in this context may be guided primarily by surface-level familiarity and distributional regularities rather than by abstract markedness relationships. This interpretation aligns with distributional learning models [47] and usage-based theories, e.g., [48], where learners extract patterns from frequent, perceptually salient input.

The current study has a few limitations: First, the participants did not have any active task to perform during the auditory stream (Word Segmentation) stage; as a result, their levels of attention could not be directly monitored. Second, no feedback was provided to the participants throughout the study, with the first two stages (i.e., Pattern Induction and Word Segmentation) not having a specific active task. Although this design was intended to promote passive statistical learning, the absence of explicit goals or reinforcement may have reduced learning strength even among attentive participants. Third, the use of wordlikeness ratings as the sole dependent measure provides only an indirect proxy for segmentation and learning, and may have been influenced by factors unrelated to statistical learning. Fourth, while the task was designed to minimize explicit awareness, we cannot fully rule out the possibility that participants developed some conscious insight into the structure of the stimuli. Finally, all stimuli were produced by a single talker; greater variability in talkers might have facilitated more robust generalization [49].

In addition, prior work using EEG and event-related potentials has demonstrated neural signatures of statistical learning that may not always be fully captured by overt behavioral measures [50–52]. These findings highlight the sensitivity of neural measures to implicit learning processes and suggest that future studies combining behavioral and electrophysiological approaches may provide a more complete characterization of how phonological complexity interacts with statistical learning.

In summary, the present results suggest that phonological complexity does not enhance generalization within a passive statistical learning paradigm. Instead, statistical learning appears to be driven primarily by exposure to encountered items, with limited transfer to novel forms. These findings refine existing accounts of complexity-based learning by delineating the conditions under which complexity effects are likely to emerge and highlight the importance of task demands and learning context in shaping phonological acquisition.

Acknowledgments

We would like to thank Arinjoy Bhattacharjee, Emma Kindl, and Sophia Dmitriyeva, for their assistance with the project.

References

1. Gierut JA, Morrisette ML, Hughes MT, Rowland S. Phonological Treatment Efficacy and Developmental Norms. LSHSS. 1996;27(3):215–30.
- View Article
- Google Scholar
2. Tomasello M. Constructing a language: A usage-based theory of language acquisition. Harvard University Press. 2009.
3. Elman JL. Learning and development in neural networks: the importance of starting small. Cognition. 1993;48(1):71–99. pmid:8403835
- View Article
- PubMed/NCBI
- Google Scholar
4. Rvachew S, Bernhardt BM. Clinical implications of dynamic systems theory for phonological development. Am J Speech Lang Pathol. 2010;19(1):34–50. pmid:19644125
- View Article
- PubMed/NCBI
- Google Scholar
5. Xu F, Tenenbaum JB. Word learning as Bayesian inference. Psychol Rev. 2007;114(2):245–72. pmid:17500627
- View Article
- PubMed/NCBI
- Google Scholar
6. Saffran JR, Newport EL, Aslin RN. Word Segmentation: The Role of Distributional Cues. Journal of Memory and Language. 1996;35(4):606–21.
- View Article
- Google Scholar
7. Chomsky N, Halle M. The sound pattern of English. Haper Row. 1968.
8. Stampe D. Dissertation on natural phonology. Taylor & Francis. 1979.
9. Prince A, Smolensky P. Optimality Theory: Constraint Interaction in Generative Grammar. Wiley-Blackwell. 2004.
10. Gierut JA. Complexity in Phonological Treatment: Clinical Factors. Lang Speech Hear Serv Sch. 2001;32(4):229–41. pmid:27764450
- View Article
- PubMed/NCBI
- Google Scholar
11. Maggu AR, Kager R, To CKS, Kwan JSK, Wong PCM. Effect of Complexity on Speech Sound Development: Evidence From Meta-Analysis Review of Treatment-Based Studies. Front Psychol. 2021;12:651900. pmid:33995208
- View Article
- PubMed/NCBI
- Google Scholar
12. A S H A. Speech sound disorders: Articulation and phonology. American Speech-Language-Hearing Association. American Speech-Language-Hearing Association. 2020.
13. Dodd B. Differential Diagnosis of Pediatric Speech Sound Disorder. Curr Dev Disord Rep. 2014;1(3):189–96.
- View Article
- Google Scholar
14. Gierut JA, Elbert M, Dinnsen DA. A functional analysis of phonological knowledge and generalization learning in misarticulating children. J Speech Hear Res. 1987;30(4):462–79. pmid:3695441
- View Article
- PubMed/NCBI
- Google Scholar
15. Powell TW, Elbert M, Dinnsen DA. Stimulability as a factor in the phonological generalization of misarticulating preschool children. J Speech Hear Res. 1991;34(6):1318–28. pmid:1787714
- View Article
- PubMed/NCBI
- Google Scholar
16. Gierut JA. The conditions and course of clinically induced phonological change. J Speech Hear Res. 1992;35(5):1049–63. pmid:1447917
- View Article
- PubMed/NCBI
- Google Scholar
17. Gierut JA, Neumann HJ. Teaching and learning /θ/: A non-confound. Clin Linguist Phon. 1992;6(3):191–200. pmid:21269158
- View Article
- PubMed/NCBI
- Google Scholar
18. Dinnsen DA. Fundamentals of Optimality Theory. Optimality Theory, Phonological Acquisition and Disorders. Equinox. 2008. 3–36.
19. Gierut JA, Morrisette ML, Champion AH. Lexical constraints in phonological acquisition. J Child Lang. 1999;26(2):261–94. pmid:11706466
- View Article
- PubMed/NCBI
- Google Scholar
20. Powell TW, Elbert M, Miccio AW, Strike-Roussos C, Brasseur J. Facilitating [s] production in young children: an experimental evaluation of motoric and conceptual treatment approaches. Clin Linguist Phon. 1998;12(2):127–46. pmid:21434786
- View Article
- PubMed/NCBI
- Google Scholar
21. Gierut JA. Homonymy in phonological change. Clin Linguist Phon. 1991;5(2):119–37. pmid:23682602
- View Article
- PubMed/NCBI
- Google Scholar
22. Maggu AR, Kager R, Xu S, Wong PCM. Complexity drives speech sound development: Evidence from artificial language training. J Exp Psychol Hum Percept Perform. 2019;45(5):628–44. pmid:30945908
- View Article
- PubMed/NCBI
- Google Scholar
23. Thompson CK, Shapiro LP. Complexity in treatment of syntactic deficits. Am J Speech Lang Pathol. 2007;16(1):30–42. pmid:17329673
- View Article
- PubMed/NCBI
- Google Scholar
24. Rvachew S, Nowak M. The effect of target-selection strategy on phonological learning. J Speech Lang Hear Res. 2001;44(3):610–23. pmid:11407566
- View Article
- PubMed/NCBI
- Google Scholar
25. Bernhardt B. Developmental implications of nonlinear phonological theory. Clin Linguist Phon. 1992;6(4):259–81. pmid:20670203
- View Article
- PubMed/NCBI
- Google Scholar
26. Skinner BF. Verbal Behavior. Acton, Mass.: Copley Publishing Group. 1957.
27. Jusczyk PW, Friederici AD, Wessels JMI, Svenkerud VY, Jusczyk AM. Infants′ Sensitivity to the Sound Patterns of Native Language Words. Journal of Memory and Language. 1993;32(3):402–20.
- View Article
- Google Scholar
28. Thiessen ED, Saffran JR. Learning to Learn: Infants’ Acquisition of Stress-Based Strategies for Word Segmentation. Language Learning and Development. 2007;3(1):73–100.
- View Article
- Google Scholar
29. Yang CD. Universal Grammar, statistics or both?. Trends Cogn Sci. 2004;8(10):451–6. pmid:15450509
- View Article
- PubMed/NCBI
- Google Scholar
30. Saffran JR, Aslin RN, Newport EL. Statistical learning by 8-month-old infants. Science. 1996;274(5294):1926–8. pmid:8943209
- View Article
- PubMed/NCBI
- Google Scholar
31. Fiser J, Aslin RN. Statistical learning of new visual feature combinations by infants. Proc Natl Acad Sci U S A. 2002;99(24):15822–6. pmid:12429858
- View Article
- PubMed/NCBI
- Google Scholar
32. Gómez R, Gerken L. Infant artificial language learning and language acquisition. Trends Cogn Sci. 2000;4(5):178–86. pmid:10782103
- View Article
- PubMed/NCBI
- Google Scholar
33. Evans JL, Saffran JR, Robe-Torres K. Statistical learning in children with specific language impairment. J Speech Lang Hear Res. 2009;52(2):321–35. pmid:19339700
- View Article
- PubMed/NCBI
- Google Scholar
34. Saffran JR, Thiessen ED. Pattern induction by infant language learners. Dev Psychol. 2003;39(3):484–94. pmid:12760517
- View Article
- PubMed/NCBI
- Google Scholar
35. Adriaans F, Kager R. Adding generalization to statistical learning: The induction of phonotactics from continuous speech. Journal of Memory and Language. 2010;62(3):311–31.
- View Article
- Google Scholar
36. Greenberg JH. Some generalizations concerning initial and final consonant clusters. Universals of human language. Stanford, CA: Stanford University Press. 1978. 243–79.
37. Boersma P, Weenink D. Praat: doing phonetics by computer. 2010.
38. Endress AD, Mehler J. The surprising power of statistical learning: When fragment knowledge leads to false memories of unheard words. Journal of Memory and Language. 2009;60(3):351–67.
- View Article
- Google Scholar
39. Siegelman N, Bogaerts L, Elazar A, Arciuli J, Frost R. Linguistic entrenchment: Prior knowledge impacts statistical learning performance. Cognition. 2018;177:198–213. pmid:29705523
- View Article
- PubMed/NCBI
- Google Scholar
40. Morrisette ML, Dinnsen DA, Gierut JA. Markedness and Context Effects in the Acquisition of Place Features. Can J Linguist. 2003;48(3–4):329–55.
- View Article
- Google Scholar
41. Tyler AA, Figurski GR. Phonetic inventory changes after treating distinctions along an implicational hierarchy. Clinical Linguistics & Phonetics. 1994;8(2):91–107.
- View Article
- Google Scholar
42. Chomsky N, Skinner BF. Verbal behavior. Language. 1959;35(1):26.
- View Article
- Google Scholar
43. Ingram D. Jakobson revisited: Some evidence from the acquisition of Polish. Lingua. 1988;75(1):55–82.
- View Article
- Google Scholar
44. Jakobson R. Child language, aphasia and phonological universals. Walter de Gruyter GmbH & Co KG. 1941.
45. Myers S, Padgett J. Domain generalisation in artificial language learning. Phonology. 2014;31(3):399–433.
- View Article
- Google Scholar
46. Zheng S, Do Y. Substantive Bias in Artificial Phonology Learning. Language and Linguist Compass. 2024;19(1).
- View Article
- Google Scholar
47. Maye J, Werker JF, Gerken L. Infant sensitivity to distributional information can affect phonetic discrimination. Cognition. 2002;82(3):B101-11. pmid:11747867
- View Article
- PubMed/NCBI
- Google Scholar
48. Bybee J. Phonology and Language Use. Cambridge University Press. 2003.
49. Van Hedger SC, Winspear M, Batterink L. Talker variability facilitates the statistical learning of speech sounds. Center for Open Science. 2021.
- View Article
- Google Scholar
50. Francois C, Schön D. Learning of musical and linguistic structures: comparing event-related potentials and behavior. Neuroreport. 2010;21(14):928–32. pmid:20697301
- View Article
- PubMed/NCBI
- Google Scholar
51. Francois C, Schön D. Musical expertise boosts implicit learning of both musical and linguistic structures. Cereb Cortex. 2011;21(10):2357–65. pmid:21383236
- View Article
- PubMed/NCBI
- Google Scholar
52. Schön D, François C. Musical expertise and statistical learning of musical and linguistic structures. Front Psychol. 2011;2:167. pmid:21811482
- View Article
- PubMed/NCBI
- Google Scholar

[ref1] 1. Gierut JA, Morrisette ML, Hughes MT, Rowland S. Phonological Treatment Efficacy and Developmental Norms. LSHSS. 1996;27(3):215–30.
View Article
Google Scholar

[2] View Article

[3] Google Scholar

[ref2] 2. Tomasello M. Constructing a language: A usage-based theory of language acquisition. Harvard University Press. 2009.

[ref3] 3. Elman JL. Learning and development in neural networks: the importance of starting small. Cognition. 1993;48(1):71–99. pmid:8403835
View Article
PubMed/NCBI
Google Scholar

[6] View Article

[7] PubMed/NCBI

[8] Google Scholar

[ref4] 4. Rvachew S, Bernhardt BM. Clinical implications of dynamic systems theory for phonological development. Am J Speech Lang Pathol. 2010;19(1):34–50. pmid:19644125
View Article
PubMed/NCBI
Google Scholar

[10] View Article

[11] PubMed/NCBI

[12] Google Scholar

[ref5] 5. Xu F, Tenenbaum JB. Word learning as Bayesian inference. Psychol Rev. 2007;114(2):245–72. pmid:17500627
View Article
PubMed/NCBI
Google Scholar

[14] View Article

[15] PubMed/NCBI

[16] Google Scholar

[ref6] 6. Saffran JR, Newport EL, Aslin RN. Word Segmentation: The Role of Distributional Cues. Journal of Memory and Language. 1996;35(4):606–21.
View Article
Google Scholar

[18] View Article

[19] Google Scholar

[ref7] 7. Chomsky N, Halle M. The sound pattern of English. Haper Row. 1968.

[ref8] 8. Stampe D. Dissertation on natural phonology. Taylor & Francis. 1979.

[ref9] 9. Prince A, Smolensky P. Optimality Theory: Constraint Interaction in Generative Grammar. Wiley-Blackwell. 2004.

[ref10] 10. Gierut JA. Complexity in Phonological Treatment: Clinical Factors. Lang Speech Hear Serv Sch. 2001;32(4):229–41. pmid:27764450
View Article
PubMed/NCBI
Google Scholar

[24] View Article

[25] PubMed/NCBI

[26] Google Scholar

[ref11] 11. Maggu AR, Kager R, To CKS, Kwan JSK, Wong PCM. Effect of Complexity on Speech Sound Development: Evidence From Meta-Analysis Review of Treatment-Based Studies. Front Psychol. 2021;12:651900. pmid:33995208
View Article
PubMed/NCBI
Google Scholar

[28] View Article

[29] PubMed/NCBI

[30] Google Scholar

[ref12] 12. A S H A. Speech sound disorders: Articulation and phonology. American Speech-Language-Hearing Association. American Speech-Language-Hearing Association. 2020.

[ref13] 13. Dodd B. Differential Diagnosis of Pediatric Speech Sound Disorder. Curr Dev Disord Rep. 2014;1(3):189–96.
View Article
Google Scholar

[33] View Article

[34] Google Scholar

[ref14] 14. Gierut JA, Elbert M, Dinnsen DA. A functional analysis of phonological knowledge and generalization learning in misarticulating children. J Speech Hear Res. 1987;30(4):462–79. pmid:3695441
View Article
PubMed/NCBI
Google Scholar

[36] View Article

[37] PubMed/NCBI

[38] Google Scholar

[ref15] 15. Powell TW, Elbert M, Dinnsen DA. Stimulability as a factor in the phonological generalization of misarticulating preschool children. J Speech Hear Res. 1991;34(6):1318–28. pmid:1787714
View Article
PubMed/NCBI
Google Scholar

[40] View Article

[41] PubMed/NCBI

[42] Google Scholar

[ref16] 16. Gierut JA. The conditions and course of clinically induced phonological change. J Speech Hear Res. 1992;35(5):1049–63. pmid:1447917
View Article
PubMed/NCBI
Google Scholar

[44] View Article

[45] PubMed/NCBI

[46] Google Scholar

[ref17] 17. Gierut JA, Neumann HJ. Teaching and learning /θ/: A non-confound. Clin Linguist Phon. 1992;6(3):191–200. pmid:21269158
View Article
PubMed/NCBI
Google Scholar

[48] View Article

[49] PubMed/NCBI

[50] Google Scholar

[ref18] 18. Dinnsen DA. Fundamentals of Optimality Theory. Optimality Theory, Phonological Acquisition and Disorders. Equinox. 2008. 3–36.

[ref19] 19. Gierut JA, Morrisette ML, Champion AH. Lexical constraints in phonological acquisition. J Child Lang. 1999;26(2):261–94. pmid:11706466
View Article
PubMed/NCBI
Google Scholar

[53] View Article

[54] PubMed/NCBI

[55] Google Scholar

[ref20] 20. Powell TW, Elbert M, Miccio AW, Strike-Roussos C, Brasseur J. Facilitating [s] production in young children: an experimental evaluation of motoric and conceptual treatment approaches. Clin Linguist Phon. 1998;12(2):127–46. pmid:21434786
View Article
PubMed/NCBI
Google Scholar

[57] View Article

[58] PubMed/NCBI

[59] Google Scholar

[ref21] 21. Gierut JA. Homonymy in phonological change. Clin Linguist Phon. 1991;5(2):119–37. pmid:23682602
View Article
PubMed/NCBI
Google Scholar

[61] View Article

[62] PubMed/NCBI

[63] Google Scholar

[ref22] 22. Maggu AR, Kager R, Xu S, Wong PCM. Complexity drives speech sound development: Evidence from artificial language training. J Exp Psychol Hum Percept Perform. 2019;45(5):628–44. pmid:30945908
View Article
PubMed/NCBI
Google Scholar

[65] View Article

[66] PubMed/NCBI

[67] Google Scholar

[ref23] 23. Thompson CK, Shapiro LP. Complexity in treatment of syntactic deficits. Am J Speech Lang Pathol. 2007;16(1):30–42. pmid:17329673
View Article
PubMed/NCBI
Google Scholar

[69] View Article

[70] PubMed/NCBI

[71] Google Scholar

[ref24] 24. Rvachew S, Nowak M. The effect of target-selection strategy on phonological learning. J Speech Lang Hear Res. 2001;44(3):610–23. pmid:11407566
View Article
PubMed/NCBI
Google Scholar

[73] View Article

[74] PubMed/NCBI

[75] Google Scholar

[ref25] 25. Bernhardt B. Developmental implications of nonlinear phonological theory. Clin Linguist Phon. 1992;6(4):259–81. pmid:20670203
View Article
PubMed/NCBI
Google Scholar

[77] View Article

[78] PubMed/NCBI

[79] Google Scholar

[ref26] 26. Skinner BF. Verbal Behavior. Acton, Mass.: Copley Publishing Group. 1957.

[ref27] 27. Jusczyk PW, Friederici AD, Wessels JMI, Svenkerud VY, Jusczyk AM. Infants′ Sensitivity to the Sound Patterns of Native Language Words. Journal of Memory and Language. 1993;32(3):402–20.
View Article
Google Scholar

[82] View Article

[83] Google Scholar

[ref28] 28. Thiessen ED, Saffran JR. Learning to Learn: Infants’ Acquisition of Stress-Based Strategies for Word Segmentation. Language Learning and Development. 2007;3(1):73–100.
View Article
Google Scholar

[85] View Article

[86] Google Scholar

[ref29] 29. Yang CD. Universal Grammar, statistics or both?. Trends Cogn Sci. 2004;8(10):451–6. pmid:15450509
View Article
PubMed/NCBI
Google Scholar

[88] View Article

[89] PubMed/NCBI

[90] Google Scholar

[ref30] 30. Saffran JR, Aslin RN, Newport EL. Statistical learning by 8-month-old infants. Science. 1996;274(5294):1926–8. pmid:8943209
View Article
PubMed/NCBI
Google Scholar

[92] View Article

[93] PubMed/NCBI

[94] Google Scholar

[ref31] 31. Fiser J, Aslin RN. Statistical learning of new visual feature combinations by infants. Proc Natl Acad Sci U S A. 2002;99(24):15822–6. pmid:12429858
View Article
PubMed/NCBI
Google Scholar

[96] View Article

[97] PubMed/NCBI

[98] Google Scholar

[ref32] 32. Gómez R, Gerken L. Infant artificial language learning and language acquisition. Trends Cogn Sci. 2000;4(5):178–86. pmid:10782103
View Article
PubMed/NCBI
Google Scholar

[100] View Article

[101] PubMed/NCBI

[102] Google Scholar

[ref33] 33. Evans JL, Saffran JR, Robe-Torres K. Statistical learning in children with specific language impairment. J Speech Lang Hear Res. 2009;52(2):321–35. pmid:19339700
View Article
PubMed/NCBI
Google Scholar

[104] View Article

[105] PubMed/NCBI

[106] Google Scholar

[ref34] 34. Saffran JR, Thiessen ED. Pattern induction by infant language learners. Dev Psychol. 2003;39(3):484–94. pmid:12760517
View Article
PubMed/NCBI
Google Scholar

[108] View Article

[109] PubMed/NCBI

[110] Google Scholar

[ref35] 35. Adriaans F, Kager R. Adding generalization to statistical learning: The induction of phonotactics from continuous speech. Journal of Memory and Language. 2010;62(3):311–31.
View Article
Google Scholar

[112] View Article

[113] Google Scholar

[ref36] 36. Greenberg JH. Some generalizations concerning initial and final consonant clusters. Universals of human language. Stanford, CA: Stanford University Press. 1978. 243–79.

[ref37] 37. Boersma P, Weenink D. Praat: doing phonetics by computer. 2010.

[ref38] 38. Endress AD, Mehler J. The surprising power of statistical learning: When fragment knowledge leads to false memories of unheard words. Journal of Memory and Language. 2009;60(3):351–67.
View Article
Google Scholar

[117] View Article

[118] Google Scholar

[ref39] 39. Siegelman N, Bogaerts L, Elazar A, Arciuli J, Frost R. Linguistic entrenchment: Prior knowledge impacts statistical learning performance. Cognition. 2018;177:198–213. pmid:29705523
View Article
PubMed/NCBI
Google Scholar

[120] View Article

[121] PubMed/NCBI

[122] Google Scholar

[ref40] 40. Morrisette ML, Dinnsen DA, Gierut JA. Markedness and Context Effects in the Acquisition of Place Features. Can J Linguist. 2003;48(3–4):329–55.
View Article
Google Scholar

[124] View Article

[125] Google Scholar

[ref41] 41. Tyler AA, Figurski GR. Phonetic inventory changes after treating distinctions along an implicational hierarchy. Clinical Linguistics & Phonetics. 1994;8(2):91–107.
View Article
Google Scholar

[127] View Article

[128] Google Scholar

[ref42] 42. Chomsky N, Skinner BF. Verbal behavior. Language. 1959;35(1):26.
View Article
Google Scholar

[130] View Article

[131] Google Scholar

[ref43] 43. Ingram D. Jakobson revisited: Some evidence from the acquisition of Polish. Lingua. 1988;75(1):55–82.
View Article
Google Scholar

[133] View Article

[134] Google Scholar

[ref44] 44. Jakobson R. Child language, aphasia and phonological universals. Walter de Gruyter GmbH & Co KG. 1941.

[ref45] 45. Myers S, Padgett J. Domain generalisation in artificial language learning. Phonology. 2014;31(3):399–433.
View Article
Google Scholar

[137] View Article

[138] Google Scholar

[ref46] 46. Zheng S, Do Y. Substantive Bias in Artificial Phonology Learning. Language and Linguist Compass. 2024;19(1).
View Article
Google Scholar

[140] View Article

[141] Google Scholar

[ref47] 47. Maye J, Werker JF, Gerken L. Infant sensitivity to distributional information can affect phonetic discrimination. Cognition. 2002;82(3):B101-11. pmid:11747867
View Article
PubMed/NCBI
Google Scholar

[143] View Article

[144] PubMed/NCBI

[145] Google Scholar

[ref48] 48. Bybee J. Phonology and Language Use. Cambridge University Press. 2003.

[ref49] 49. Van Hedger SC, Winspear M, Batterink L. Talker variability facilitates the statistical learning of speech sounds. Center for Open Science. 2021.
View Article
Google Scholar

[148] View Article

[149] Google Scholar

[ref50] 50. Francois C, Schön D. Learning of musical and linguistic structures: comparing event-related potentials and behavior. Neuroreport. 2010;21(14):928–32. pmid:20697301
View Article
PubMed/NCBI
Google Scholar

[151] View Article

[152] PubMed/NCBI

[153] Google Scholar

[ref51] 51. Francois C, Schön D. Musical expertise boosts implicit learning of both musical and linguistic structures. Cereb Cortex. 2011;21(10):2357–65. pmid:21383236
View Article
PubMed/NCBI
Google Scholar

[155] View Article

[156] PubMed/NCBI

[157] Google Scholar

[ref52] 52. Schön D, François C. Musical expertise and statistical learning of musical and linguistic structures. Front Psychol. 2011;2:167. pmid:21811482
View Article
PubMed/NCBI
Google Scholar

[159] View Article

[160] PubMed/NCBI

[161] Google Scholar

Figures

Abstract

Purpose

Method

Results

Conclusions

Introduction

Complexity in speech and language acquisition

Statistical learning in speech and language acquisition

Combination of linguistic complexity and statistical learning

Materials and methods

Participants

Hearing screening

Material

Procedure

Pattern induction.

Word segmentation.

Wordlikeness testing.

Statistical analysis

Results

Discussion

Acknowledgments

References