Pith. sign in

REVIEW 5 major objections 5 minor 36 references

Do Self-Supervised Speech Models Exhibit the Critical Period Effects in Language Acquisition?

T0 review · 5 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A speech model trained like a child shows no critical-period effects

desk verdict First systematic test of critical-period effects in a speech model, but the timing manipulation is confounded with total training steps, so the strong innateness conclusion is not yet supported. read the letter →

arxiv 2508.21210 v1 pith:4SOJ6RHL submitted 2025-08-28 cs.CL

Yurie Koga , Shunsuke Kando , Yusuke Miyao This is my paper
classification cs.CL
keywords criticalperiodself-supervisedspeechmodelsHuBERTphonediscriminationchild-directedsecondlanguageacquisitionL1attritionABXtask
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Human language learning has a critical period: first-language attrition is worse if exposure ends early, and second-language mastery declines the later exposure begins. The paper asks whether these effects appear in a self-supervised speech model that learns purely by statistical patterns in child-directed speech. It trains HuBERT sequentially on pairs of languages, varying when the second language starts and when the first stops, then measures phone discrimination. The human pattern does not appear: later L2 onset improves L2 performance, and later L1 offset increases L1 forgetting. The authors conclude that critical-period phenomena are not an inevitable product of statistical learning and may depend on innate neurobiological mechanisms.

What carries the argument

HuBERT: a self-supervised speech representation model trained to predict masked hidden units from surrounding speech. The training schedule is the timing instrument: checkpoints saved after 5k to 120k steps of L1 training in the second training iteration serve as 'ages of L2 onset,' and each checkpoint is then continued on L2 for a fixed 120k steps. The outcome measure is ABX phone-discrimination accuracy, which tests whether the model can tell which of two triphone contexts matches a third stimulus, using embeddings from the ninth Transformer layer.

What would settle it

Train a matched control set in which total training time before evaluation is held constant by inserting filler training (on a third language or non-speech audio) between L1 and L2; if the delayed-L2-onset model still outperforms, timing is what matters, but if the advantage tracks total steps, the reported absence of critical-period effects is confounded.

Watch

Extended reading notes

Core claim

The paper establishes that HuBERT, a self-supervised speech model trained by masked prediction, does not reproduce the human critical-period pattern for phonological acquisition. In the L2-acquisition condition, models whose English training began after more L1 exposure (Japanese or French) achieved higher English ABX phone-discrimination accuracy, not lower. In the L1-attrition condition, models with a later L1 offset lost more L1 accuracy after L2-only training, while models with an early offset improved. The authors interpret these results, together with earlier text-language-model findings, as evidence that critical-period effects are not a necessary consequence of statistical learning a

Load-bearing premise

The load-bearing premise is that varying the number of L1 training steps before a fixed 120k-step L2 block changes 'age of L2 onset' rather than simply total training amount.

Editorial extensions

If this is right

  • If the result holds, delaying second-language exposure in a self-supervised speech learner makes it better, not worse, at L2 phone discrimination—the opposite of the human age-of-acquisition curve.
  • It extends the earlier text-model null result to the speech modality, where children actually receive their early language input, so the negative evidence for pure statistical-learning accounts gets stronger.
  • More training before L2 helps L2 performance, suggesting the observed pattern is governed by total training budget rather than a sensitive window.
  • When L2 arrives late, keeping L1 in the training mixture protects L1 accuracy; this matches continual-learning rehearsal and suggests where to look for attrition-like behavior.
  • The paper's own next step is to test syntax and semantics, where CP effects are also reported; the current null is phonological only.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A decisive control would equalize total training steps (e.g., insert filler training between L1 and L2); if the delayed-onset advantage disappears, the paper's null result reflects training amount, not absence of a critical period.
  • The ABX triphone task may miss the perceptual loss that defines human CP effects; a non-native contrast test such as English /r/-/l/ for Japanese L1 might show the human-like decline even if ABX does not.
  • The early-offset L1 'improvement' may be an artifact of the pseudo-label regime: L2 training generates labels from an early L1 checkpoint, and the gain could reflect continued phonological learning under new labels rather than resistance to attrition.
  • Using a closer L1-L2 pair (French-English) already boosts L2 accuracy; systematically varying phonological distance could reveal whether a 'useful, similar' first language acts like early L2 exposure.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. Using HuBERT, the paper asks whether self-supervised speech models reproduce two critical-period effects: worse L2 phone discrimination with delayed L2 onset, and stronger L1 retention with delayed L1 offset. It trains HuBERT on child-directed speech (English, Japanese, French) in two directions, varying the number of L1 training steps before a fixed 120k-step L2 training stage, under L2-only and L1+L2 settings. Phone discrimination is measured on the English ABX ZeroSpeech benchmark. The paper reports that later L2 onset improves L2 performance and later L1 offset leads to L1 forgetting, opposite to human CP effects, and concludes that CP phenomena are not inevitable consequences of statistical learning alone.

Significance. The question is well-motivated and the move from text LMs to speech models is timely. Strengths include the use of naturalistic child-directed corpora, multiple language pairs, a standard ABX evaluation, and comparison against monolingual/bilingual baselines. However, the experimental design does not isolate exposure timing from training amount or pseudo-label quality, and the evaluation covers only English as the tested language. Because these confounds affect both directions of the central null result, the paper's current evidence does not support its conclusion. If the confounds are addressed and the null result persists, this would be a notable negative result for the innateness debate.

major comments (5)
  1. [Sec. III-B.2, Fig. 3, Sec. IV-A] The L2-onset manipulation varies L1 training length while holding L2 training at 120k steps; later-onset models receive up to 115k more total steps (L1-5k-L2-120k vs L1-120k-L2-120k). The better L2 ABX performance with later onset therefore may reflect more total optimization, not later onset. The paper itself says 'phone discrimination performance in one language improves with longer training regardless of the language used for training.' Match total training steps across onset conditions, or otherwise control for total step count, before interpreting the L2 curves as evidence about timing.
  2. [Sec. III-B.2] Pseudo-labels for the L2 training stage are generated from each L1-trained checkpoint: 'pseudo-labels in it3 were generated using representations from each L1-trained checkpoint.' Thus early-onset models receive L2 labels produced by undertrained encoders, while late-onset models receive higher-quality labels. This is a second confound: improved L2 performance may be an artifact of better target labels rather than of later onset. Use a fixed label source or demonstrate that label quality does not vary systematically with onset.
  3. [Sec. IV-B, Fig. 5(a)] The L1-attrition gain is final L1 accuracy minus pre-L2 L1 accuracy at the checkpoint. Early-offset models are far from converged on L1 when L2 begins; their low pre-L2 baseline makes any later improvement look like 'gain,' while late-offset models have already converged and any interference appears as loss. This makes the attrition pattern attributable to training state rather than offset timing. A control that continues training without switching languages would separate continued convergence from actual L1 retention.
  4. [Table I, Sec. III-C, Sec. IV-A] All evaluations use English ABX. Consequently, L2 acquisition is tested only when English is the L2 (Japanese-L1 and French-L1 conditions). For the (EN,JA) and (EN,FR) conditions, Japanese and French L2 performance is never evaluated, so the paper cannot support claims about CP effects for L2 phonological acquisition in those languages. The conclusion should be restricted to English-L2 acquisition, or the missing L2 evaluations should be added.
  5. [Sec. V] Even if the above controls were added, the conclusion that CP phenomena 'are not inevitable outcomes of statistical learning alone' goes beyond the evidence: this is one model, one task, and one training schedule. At most, the results would show that this instantiation of statistical learning did not produce CP effects under the tested conditions. The wording should be softened to match the scope.
minor comments (5)
  1. [Sec. IV] Figures 4 and 5 report averages over two seeds but show no error bars or individual values; add variability information or statistical comparisons.
  2. [Sec. III-B] Training hyperparameters (learning rate, batch size, optimizer, masking) are not reported; include them for reproducibility.
  3. [Fig. 4] The axis label 'L1-5k-(L1)L2-120k' is hard to parse; define the notation in the caption or in the text.
  4. [Sec. III-B.1] The relationship between it1 (80k steps) and the checkpoints in it2 should be clarified; the statement that it1 can be ignored because it2 starts from scratch is non-obvious.
  5. [Abstract / Sec. IV] The abstract says delayed L2 onset 'tend to perform better,' but no significance testing is reported; clarify whether the trends are descriptive or supported by statistical tests.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the study is empirical, uses external benchmarks, and contains no fitted-parameter or self-citation reductions.

full rationale

The paper is an empirical investigation that trains HuBERT models under different L1/L2 exposure schedules and evaluates them with an external benchmark (ZeroSpeech ABX). There are no fitted parameters that are then renamed as predictions, and no derivation chain in which the target result is equivalent by construction to the input. The central claim—that HuBERT does not exhibit the human CP pattern—is an observed outcome, not a consequence of an equation or a self-citation. The paper's reliance on prior work (e.g., HuBERT, the ABX metric, and the text-LM study [11]) is external and not authored by the present authors. The most notable weakness is the confound between L2 onset and total training steps: models with later L2 onset receive more total training (Fig. 3). The paper itself acknowledges this in Section IV-A: 'Rather, this indicates that phone discrimination performance in one language improves with longer training regardless of the language used for training.' This is a validity/soundness concern about whether the manipulation isolates onset, not a circularity: the outcome is not defined in terms of the input, nor is it forced by a self-referential argument. Therefore the circularity score is 0.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The experimental design, not a derivation, carries the load. The central claim rests on the representativeness of HuBERT for statistical learning, the validity of ABX as a measure of phonological acquisition, and the mapping from training steps to exposure timing. The main hand-chosen numbers that affect the result are the training-step schedules and the evaluation layer; the key problem is the uncontrolled total training amount rather than fitted parameters.

free parameters (4)
  • L2 training duration = 120k it3 steps
    Fixed across all onset conditions to hold L2 exposure equal, but because L1 training varies, later-onset models receive much more total training; this choice creates the central confound.
  • L1 checkpoint schedule = 5k, 10k, 15k, 30k, 45k, 60k, 75k, 90k, 105k, 120k during it2
    Hand-chosen sampling points define the operationalization of L2 onset and L1 offset; early checkpoints are not converged on L1, confounding the attrition gain measure.
  • Evaluation layer = 9th Transformer layer
    Chosen based on prior reports that this layer encodes phoneme information; phoneme-level conclusions may depend on this choice.
  • Number of random seeds = 2
    Averages over two seeds with no variance reported; trend claims lack statistical support.
assumptions (5)
  • domain assumption HuBERT self-supervised training on child-directed speech is a representative implementation of statistical learning from speech, so a null result licenses inference about 'statistical learning alone'.
    Invoked in Sections I and IV-B where the absence of CP in HuBERT is used to argue that CP requires innate mechanisms; this equates the model with statistical learning generally.
  • domain assumption ABX phone discrimination accuracy (ZeroSpeech) measures phonological acquisition ability in a way comparable to human phonological CP.
    Section III-C uses ABX as the only evaluation; the paper draws phonological-acquisition conclusions from this single task.
  • domain assumption Training steps are a valid proxy for exposure time, and varying L1 steps before fixed L2 steps corresponds to manipulating L2 onset and L1 offset.
    Section III-B and Fig. 3; this mapping is necessary to connect model training to the CP timeline, and is threatened by the total-step confound.
  • domain assumption Child-directed speech corpora (Providence, MiiPro, Lyon) approximate the input conditions of human L1 and L2 acquisition.
    Section III-A; the CDS choice is motivated by faithfulness to human conditions, but no validation is provided for cross-linguistic comparability after preprocessing and subsampling.
  • domain assumption English ABX evaluation is valid for measuring L1 retention in models whose L2 is Japanese or French; continued training in other languages does not alter evaluation-language alignment.
    Section III-C and IV-B; all retention scores are computed on English triplets regardless of the L2 training language.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Do Self-Supervised Speech Models Exhibit the Critical Period Effects in Language Acquisition?." pith.science (2026). https://pith.science/paper/4SOJ6RHL

@misc{pith2026250821210,
  author       = {Pith},
  title        = {Pith review of: Do Self-Supervised Speech Models Exhibit the Critical Period Effects in Language Acquisition?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4SOJ6RHL}},
  note         = {Machine review of arXiv:2508.21210}
}
read the original abstract

This paper investigates whether the Critical Period (CP) effects in human language acquisition are observed in self-supervised speech models (S3Ms). CP effects refer to greater difficulty in acquiring a second language (L2) with delayed L2 exposure onset, and greater retention of their first language (L1) with delayed L1 exposure offset. While previous work has studied these effects using textual language models, their presence in speech models remains underexplored despite the central role of spoken language in human language acquisition. We train S3Ms with varying L2 training onsets and L1 training offsets on child-directed speech and evaluate their phone discrimination performance. We find that S3Ms do not exhibit clear evidence of either CP effects in terms of phonological acquisition. Notably, models with delayed L2 exposure onset tend to perform better on L2 and delayed L1 exposure offset leads to L1 forgetting.

Figures

Figures reproduced from arXiv: 2508.21210 by the authors.

Figure 1
Figure 1. Overview of the Critical Period (CP) effects in human language [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of the HuBERT training scheme. We train HuBERT for [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Number of training steps for each model. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: (a): L2 (EN) accuracies on the phone discrimination ABX test for models with various L1 training durations. We plot results for two L1s (Japanese and French), and for two L2 training settings with each L1: L2-only and L1+L2. The x-axis corresponds to each model (i.e., …
Figure 5
Figure 5. Figure 5: (a): L1 (EN) accuracy gains due to L2 training, computed by subtracting the L1 accuracy before L2 training from the one after L2 training. The x-axis can be regarded as the L1 exposure offset for L2-only models. (b): L1 (EN) ABX accuracy advantage of training with Fren…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

36 extracted references · 29 canonical work pages

  1. [11]

    Investigating critical period effects in language acquisition through neural language models,

    I. Constantinescu, T. Pimentel, R. Cotterell, and A. Warstadt, “Investigating critical period effects in language acquisition through neural language models,” 2024. [Online]. Available: https://arxiv.org/abs/2407.19325

    arXiv 2024

  2. [1]

    A large-scale evaluation of speech foundation models,

    S.-w. Yang, H.-J. Chang, Z. Huang, A. T. Liu, C.-I. Lai, H. Wu, J. Shi, X. Chang, H.-S. Tsai, W.-C. Huang, T.-h. Feng, P.-H. Chi, Y . Y . Lin, Y .-S. Chuang, T.-H. Huang, W.-C. Tseng, K. Lakhotia, S.-W. Li, A. Mohamed, S. Watanabe, and H.-y. Lee, “A large-scale evaluation of speech foundation models,” IEEE/ACM Transactions on Audio, Speech, and Language P...

    work page 2024

  3. [2]

    A phonetic model of non-native spoken word processing,

    Y . Matusevych, H. Kamper, T. Schatz, N. Feldman, and S. Goldwater, “A phonetic model of non-native spoken word processing,” in Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main V olume , P. Merlo, J. Tiedemann, and R. Tsarfaty, Eds. Online: Association for Computational Linguistics, Apr. 202...

    work page 2021

  4. [3]

    Mapping phonology to semantics: A computational model of cross-lingual spoken- word recognition,

    I. Zaitova, B. Abdullah, and D. Klakow, “Mapping phonology to semantics: A computational model of cross-lingual spoken- word recognition,” in Proceedings of the Ninth Workshop on NLP for Similar Languages, V arieties and Dialects , Y . Scherrer, T. Jauhiainen, N. Ljube ˇsi´c, P. Nakov, J. Tiedemann, and M. Zampieri, Eds. Gyeongju, Republic of Korea: Assoc...

    work page 2022

  5. [4]

    Modeling early phonetic acquisition from child-centered audio data,

    M. Lavechin, M. de Seyssel, M. M ´etais, F. Metze, A. Mohamed, H. Bredin, E. Dupoux, and A. Cristia, “Modeling early phonetic acquisition from child-centered audio data,” Cognition, vol. 245, p. 105734, 2024. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0010027724000209

    work page 2024

  6. [5]

    Do self-supervised speech models develop human-like perception biases?

    J. Millet and E. Dunbar, “Do self-supervised speech models develop human-like perception biases?” in Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (V olume 1: Long Papers) , S. Muresan, P. Nakov, and A. Villavicencio, Eds. Dublin, Ireland: Association for Computational Linguistics, May 2022, pp. 7591–7605. [Online...

    work page 2022

  7. [6]

    The biological foundations of language,

    E. H. Lenneberg, “The biological foundations of language,” Hospital Practice, vol. 2, no. 12, pp. 59–67, 1967. [Online]. Available: https://doi.org/10.1080/21548331.1967.11707799

    work page arXiv 1967

  8. [7]

    The critical period hypothesis: A coat of many colours,

    D. Singleton, “The critical period hypothesis: A coat of many colours,” International Review of Applied Linguistics in Language Teaching, vol. 43, no. 4, pp. 269–285, 2005. [Online]. Available: https://doi.org/10.1515/iral.2005.43.4.269

    work page doi:10.1515/iral.2005.43.4.269 2005

Show all 36 references
  1. [8]

    Pinker, The Language Instinct: How the Mind Creates Language

    S. Pinker, The Language Instinct: How the Mind Creates Language . New York: William Morrow and Company, 1994

    1994

  2. [9]

    State-dependent factors influencing neural plasticity: A partial account of the critical period,

    P. W. Munro, “State-dependent factors influencing neural plasticity: A partial account of the critical period,” in Parallel Distributed Processing, V olume 2: Explorations in the Microstructure of Cognition: Psychological and Biological Models . The MIT Press, 07 1986. [Online...

    1986 doi

  3. [10]

    Elman, E

    J. Elman, E. Bates, M. H. Johnson, A. Karmiloff-Smith, D. Parisi, and K. Plunkett, Rethinking Innateness: A Connectionist Perspective on Development . The MIT Press, 10 1996. [Online]. Available: https://doi.org/10.7551/mitpress/5929.001.0001

    1996 doi

  4. [12]

    Hubert: Self-supervised speech representation learning by masked prediction of hidden units,

    W.-N. Hsu, B. Bolte, Y .-H. H. Tsai, K. Lakhotia, R. Salakhutdinov, and A. Mohamed, “Hubert: Self-supervised speech representation learning by masked prediction of hidden units,” IEEE/ACM Transactions on Audio, Speech, and Language Processing , vol. 29, pp. 3451–3460, 2021

    2021

  5. [13]

    Maturational constraints on language learning,

    E. L. Newport, “Maturational constraints on language learning,” Cognitive Science, vol. 14, no. 1, pp. 11–28, 1990. [Online]. Available: https://www.sciencedirect.com/science/article/pii/036402139090024Q

    1990

  6. [14]

    Pallier, Critical periods in language acquisition and language attrition

    C. Pallier, Critical periods in language acquisition and language attrition. John Benjamins Publishing Company, 2007, pp. 155–168. [Online]. Available: https://doi.org/10.1075/sibil.33.11pal

    2007 doi

  7. [15]

    The optimal age to learn a foreign language,

    J. J. Asher and R. Garc ´ıa, “The optimal age to learn a foreign language,” The Modern Language Journal , vol. 53, no. 5, pp. 334–341, 1969. [On- line]. Available: https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1540- 4781.1969.tb04603.x

    1969

  8. [16]

    A critical period for second language acquisition: Evidence from 2/3 million english speakers,

    J. K. Hartshorne, J. B. Tenenbaum, and S. Pinker, “A critical period for second language acquisition: Evidence from 2/3 million english speakers,” Cognition, vol. 177, pp. 263–277, 2018. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0010027718300994

    2018

  9. [17]

    Maturational constraints in the acquisition of a native-like accent in second language learning,

    H. Seliger, S. Krashen, and P. Ladefoged, “Maturational constraints in the acquisition of a native-like accent in second language learning,” Language Sciences , vol. 36, pp. 20–22, 1975

    1975

  10. [18]

    A sensitive period for the acquisition of a nonnative phonological system,

    S. Oyama, “A sensitive period for the acquisition of a nonnative phonological system,” Journal of Psycholinguistic Research , vol. 5, no. 3, pp. 261–285, 1976

    1976

  11. [19]

    Critical period effects in second language learning: The influence of maturational state on the acquisition of english as a second language,

    J. S. Johnson and E. L. Newport, “Critical period effects in second language learning: The influence of maturational state on the acquisition of english as a second language,” Cognitive Psychology, vol. 21, no. 1, pp. 60–99, 1989. [Online]. Available: https://www.sciencedirect...

    1989

  12. [20]

    Developmentally-plausible working memory shapes a critical period for language acquisition,

    M. Mita, R. Yoshida, and Y . Oseki, “Developmentally-plausible working memory shapes a critical period for language acquisition,”

  13. [21]

    Representation learning with contrastive predictive coding,

    A. van den Oord, Y . Li, and O. Vinyals, “Representation learning with contrastive predictive coding,” 2019. [Online]. Available: https://arxiv.org/abs/1807.03748

    2019 arXiv

  14. [22]

    wav2vec 2.0: A framework for self-supervised learning of speech representations,

    A. Baevski, Y . Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A framework for self-supervised learning of speech representations,” in Advances in Neural Information Processing Systems , H. Larochelle, M. Ranzato, R. Hadsell, M. Balcan, and H. Lin, Eds., vol. 33. Curran Associat...

    2020

  15. [23]

    The childes project: Tools for analyzing talk (third edition): V olume i: Transcription format and programs, volume ii: The database,

    B. MacWhinney, “The childes project: Tools for analyzing talk (third edition): V olume i: Transcription format and programs, volume ii: The database,” Computational Linguistics , vol. 26, no. 4, pp. 657–657, 12

  16. [24]

    A joint model of word segmentation and phonological variation for English word-final /t/-deletion,

    B. B ¨orschinger, M. Johnson, and K. Demuth, “A joint model of word segmentation and phonological variation for English word-final /t/-deletion,” in Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (V olume 1: Long Papers) , H. Schuetze, ...

    2013

  17. [25]

    Babyslm: language-acquisition- friendly benchmark of self-supervised spoken language models,

    M. Lavechin, Y . Sy, H. Titeux, M. A. C. Bland ´on, O. R ¨as¨anen, H. Bredin, E. Dupoux, and A. Cristia, “Babyslm: language-acquisition- friendly benchmark of self-supervised spoken language models,” in INTERSPEECH 2023 , 2023, pp. 4588–4592

    2023

  18. [26]

    Guideline for japanese mlu — how to compute mluw and mlum—,

    M. Susanne, “Guideline for japanese mlu — how to compute mluw and mlum—,” Journal of Health and Medical Science , no. 2, pp. 1–15, 2012

    2012

  19. [27]

    Prosodically-conditioned variability in children’s production of french determiners,

    K. Demuth and A. Tremblay, “Prosodically-conditioned variability in children’s production of french determiners,”Journal of Child Language, vol. 35, no. 1, pp. 99–127, Feb. 2008

    2008

  20. [28]

    An open-source voice type classifier for child-centered daylong recordings,

    M. Lavechin, R. Bousbib, H. Bredin, E. Dupoux, and A. Cristia, “An open-source voice type classifier for child-centered daylong recordings,” in Interspeech 2020, 2020, pp. 3072–3076

    2020

  21. [29]

    fairseq: A fast, extensible toolkit for sequence modeling,

    M. Ott, S. Edunov, A. Baevski, A. Fan, S. Gross, N. Ng, D. Grangier, and M. Auli, “fairseq: A fast, extensible toolkit for sequence modeling,” in Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics (Demonstrations) ...

    2019

  22. [30]

    Layer-wise analysis of a self-supervised speech representation model,

    A. Pasad, J.-C. Chou, and K. Livescu, “Layer-wise analysis of a self-supervised speech representation model,” in 2021 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU) , 2021, pp. 914–921

    2021

  23. [31]

    Analyzing acoustic word embeddings from pre-trained self-supervised speech models,

    R. Sanabria, H. Tang, and S. Goldwater, “Analyzing acoustic word embeddings from pre-trained self-supervised speech models,” in ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , 2023, pp. 1–5

    2023

  24. [32]

    Perceptimatic: A human speech perception benchmark for unsupervised subword modelling,

    J. Millet and E. Dunbar, “Perceptimatic: A human speech perception benchmark for unsupervised subword modelling,” in Interspeech 2020 , 2020, pp. 4881–4885

    2020

  25. [33]

    The influence of initial exposure on lexical representation: Comparing early and simultaneous bilinguals,

    N. Sebasti ´an-Gall´es, S. Echeverr ´ıa, and L. Bosch, “The influence of initial exposure on lexical representation: Comparing early and simultaneous bilinguals,” Journal of Memory and Language , vol. 52, no. 2, pp. 240–255, 2005. [Online]. Available: https://www.sciencedirect...

    2005

  26. [34]

    Expe- rience replay for continual learning,

    D. Rolnick, A. Ahuja, J. Schwarz, T. Lillicrap, and G. Wayne, “Expe- rience replay for continual learning,” in Advances in Neural Informa- tion Processing Systems , H. Wallach, H. Larochelle, A. Beygelzimer, F. d'Alch ´e-Buc, E. Fox, and R. Garnett, Eds., vol. 32. Curran Assoc...

    2019

  27. [2000]

    Available: https://doi.org/10.1162/coli.2000.26.4.657

    [Online]. Available: https://doi.org/10.1162/coli.2000.26.4.657

    2000 doi

  28. [2025]

    Available: https://arxiv.org/abs/2502.04795

    [Online]. Available: https://arxiv.org/abs/2502.04795

    arXiv

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.