Pith. sign in

REVIEW 3 major objections 4 minor 45 references

What do self-supervised speech models know about Dutch? Analyzing advantages of language-specific pre-training

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Pre-training a Wav2Vec2 model exclusively on Dutch improves how it encodes Dutch phonetic and lexical information, compared with English or multilingual pre-training.

desk verdict Solid Dutch SSL resource paper, but the headline claim is undercut by corpus/domain overlap; needs matched controls or a qualified abstract. read the letter →

arxiv 2506.00981 v2 pith:DU2SASNM submitted 2025-06-01 cs.CL cs.AIcs.SDeess.AS

Marianne de Heer Kloots , Hosein Mohebbi , Charlotte Pouw , Gaofei Shen , Willem Zuidema , Martijn Bentum This is my paper
classification cs.CLcs.AIcs.SDeess.AS
keywords self-supervisedspeechWav2Vec2language-specificpre-trainingDutchrepresentationsprobingautomaticrecognitionrepresentationalsimilarityanalysis
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

This paper asks whether the internal representations of a self-supervised speech model become better at a specific language when the model is pre-trained on that language alone. The authors pre-train a Wav2Vec2 model on 831 hours of Dutch and compare it with an English-only model of similar size and a much larger multilingual model, testing how well Dutch phones and words can be decoded from each model's hidden layers. On nearly all of their phonetic and lexical measures the Dutch model comes out ahead, and it also reaches the lowest word error rate when all models are fine-tuned for Dutch speech recognition. The authors read this as evidence that language-specific pre-training sharpens the encoding of that language's phonetic and lexical structure, while noting that part of the Dutch advantage may reflect the conversational character of its pre-training data rather than the Dutch language itself.

What carries the argument

The argument is carried by a controlled comparison of four Wav2Vec2 base models with identical architecture but different pre-training data: w2v2-nl (831 hours of Dutch from CGN, Multilingual LibriSpeech, and CommonVoice), the original English base model, a multilingual model trained on 100k hours from VoxPopuli, and a nonspeech acoustic baseline. The evaluation rests on the newly curated SSL-NL set, which contains phone- and word-level forced alignments for read speech (MLS) and conversational speech (IFADV). The analysis methods fall into two groups: trained linear transforms (phone identity probes and LDA silhouette scores) and zero-shot metrics (ABX discrimination, PCA silhouette scores, and RSA against Dutch Fasttext vectors). The key observation is that the language-specific advantage is clearly detected by the trained probes but only partially by zero-shot distances, leading the authors to conclude that language-specific phonetic information occupies a small decodable subspace of the representations.

What would settle it

Train a Dutch-only model on read speech only and an English-only model on comparable conversational speech, then rerun the IFADV comparisons; if the English conversational model matches the Dutch model's advantage on IFADV phones and words, the claimed language-specific benefit is largely a domain effect.

Watch

Extended reading notes

Core claim

The central claim is that pre-training a Wav2Vec2 model exclusively on Dutch improves its representation of Dutch phonetic and lexical information compared with pre-training on a similar amount of English or on a much larger amount of multilingual data. The paper shows that the advantage appears in trained phone-identity probes, in LDA-based clustering of phones and words, and in representational similarity to Dutch word vectors, and that it is only partially visible in zero-shot ABX and PCA measures. The authors argue that the language-specific information is therefore real but concentrated in a subspace that linear transformations can expose. They also report that the same ranking holds for downstream ASR: fine-tuned on the same Dutch read-speech data, the Dutch model has the lowest word error rate on every test set, followed by the multilingual model and then the English model.

Load-bearing premise

The observed Dutch advantage is assumed to come from the Dutch content of the pre-training data rather than from its domain, even though the Dutch model's data included conversational and interview speech while the English and multilingual models were trained on read audiobooks and parliament recordings.

Editorial extensions

If this is right

  • If the central claim is right, monolingual pre-training of modest size (under a thousand hours) can beat both same-sized English and much larger multilingual pre-training for representing that language's phones and words.
  • The divergence between trained-probe and zero-shot measures implies that studies relying only on ABX-style distances may systematically underestimate language-specific structure in high-dimensional speech representations.
  • The alignment between probe performance and downstream ASR word error rates suggests that representational quality measured by linear probes is a meaningful predictor of fine-tuned transcription performance for a language.
  • The larger gaps on conversational data indicate that matching the domain of the pre-training data matters for representing phones and words, not just for conversational-level patterns.

Reading between the lines

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

  • A direct test that would separate the language and domain explanations is to pre-train a Dutch model on read speech only and an English model on conversational speech only, then compare on IFADV; if the domain, not the language, drives the gap, the English conversational model should close it.
  • The subspace interpretation predicts that the Dutch model's advantage should be removable by projecting out a few principal components aligned with Dutch-specific phone contrasts, a manipulation the paper does not perform.
  • For other languages, the size of the language-specific benefit should track the phonetic distance from English and from the languages in the multilingual model; the paper's small but consistent effect for Dutch, a language close to English, suggests larger effects for more distant languages.
  • If language-specific information is concentrated in a low-dimensional subspace, then multilingual models may already contain it; a probe trained on a small amount of target-language data could extract it, which would be a cheaper route than monolingual pre-training.
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

3 major / 4 minor

Summary. The paper introduces SSL-NL, a Dutch evaluation set of phone- and word-level alignments from MLS read speech and IFADV conversational speech, and releases Wav2Vec2-NL, a Dutch-only Wav2Vec2 model trained on 831 hours of CGN, MLS, and CommonVoice. It compares this model with an English-only Wav2Vec2 base, a multilingual VoxPopuli model, and a non-speech AudioSet model. Using phone identity probes, phone ABX, PCA/LDA silhouette clustering, word clustering, and FastText RSA, the paper finds that the Dutch model best encodes Dutch phonetic and lexical features, with the advantage clearest for trained linear probes and for the conversational IFADV corpus. It then fine-tunes all models on Dutch CGN-o read speech and reports lower WER for the Dutch model across CGN-o, IFADV, MLS, CommonVoice, and N-Best test sets. The authors conclude that Dutch-specific pre-training improves Dutch linguistic representations and downstream ASR, while noting that analysis-method choice and data domain affect the size of the observed advantage.

Significance. If the main claim is accepted, the paper makes a useful empirical contribution to the interpretability of self-supervised speech models: it provides a new Dutch model and evaluation resource, compares multiple analysis methodologies on the same representations, and includes zero-shot measures that are not optimized on the test speakers. The release of the SSL-NL set and Wav2Vec2-NL with training manifests is a concrete resource for future work on language-specific SSL representations. The paper also honestly reports that the advantage varies across methods and datasets, and it explicitly acknowledges a domain confound for the IFADV results. These strengths are substantial, but the headline causal claim is currently broader than the experimental design supports because the Dutch model's pre-training data overlaps in corpus and domain with the evaluation data in ways that the baselines do not share.

major comments (3)
  1. [§2, §3, §5] The central comparison is confounded by training/evaluation overlap. w2v2-nl was pre-trained on 211 hours of MLS and 537 hours of CGN (Section 2). The phonetic and lexical analyses use a 'held-out' MLS subset whose speaker- and book-level disjointness from pre-training is not stated (Section 3), and the ASR experiment fine-tunes on 78 hours of CGN component o and evaluates on a further 10 hours of CGN-o, without stating that component o was excluded from the 537 hours of CGN used for pre-training (Sections 2 and 5). Since fb-en and fb-voxp-100k were pre-trained on neither MLS nor CGN, the observed advances could reflect pre-training on the same corpus (MLS), pre-training on the same recordings or domain (CGN-o for ASR, conversational speech for IFADV), rather than exclusively Dutch language-specific pre-training. Section 6 acknowledges a domain effect for IFADV, but it does not address the MLS corpus overlap or the CGN-o overlap in the ASR comparison. Please provide explicit speaker/book/recording disjointness guarantees, add an evaluation condition on a corpus absent from all pre-training data, or restrict the Abstract's causal wording accordingly.
  2. [Abstract and §6] The Abstract's first claim, that 'pre-training exclusively on Dutch improves the representation of Dutch linguistic features,' is not supported in that unqualified form by the paper's own discussion. Section 6 attributes the larger IFADV differences to 'an effect of the pre-training data domain beyond its language-specificity,' and the word-level advantages are especially prominent on IFADV in Figure 2. The reported advantage is therefore a joint effect of language and training-domain match, not a pure language effect. The claim should either be restricted to the conditions where the language variable is not entangled with domain (e.g., read speech from MLS, if overlap is controlled), or the experiments should add a control model trained on Dutch read speech only, or on conversational non-Dutch speech, to disentangle language from domain.
  3. [§2] The Abstract compares 'similar amounts of English or larger amounts of multilingual data,' but this is only a match on corpus hours. w2v2-nl was trained for 100k steps with a modified fairseq configuration, while the paper does not report the number of training steps or exact optimization/masking schedule for fb-en and fb-voxp-100k. The models therefore also differ in training compute and recipe, which is an additional uncontrolled variable in a three-model comparison. Reporting the training configurations of all models, and where possible matching training steps or at least documenting them, is necessary to support the attribution of the advantage to language-specific pre-training rather than to differences in optimization.
minor comments (4)
  1. [§4.1] The phrase 'the high-font vowels' should be 'the high-front vowels'.
  2. [Figure 2, §4.1] The claim that the English model shows 'significantly higher' scores than the nonspeech baseline is not supported by any reported statistical test; the 95% confidence intervals shown in Figure 2 are informative but do not by themselves establish significance, especially with many layers and analysis variants compared.
  3. [§4.2] The subspace explanation ('language-specific phonetic information may be encoded in a small subspace') is plausible but is not directly tested; it would be strengthened by an explicit subspace-alignment or canonical-correlation analysis between the models' representation spaces.
  4. [§2] The text says CGN segments are limited to 2–15 seconds, but then states that 'across the full training set, audio samples range between 2 and 20 seconds'; please clarify whether this refers to the other sources or to a later revision of the sampling procedure.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: representational and ASR claims are evaluated on held-out data against external references; the corpus/domain overlap is an experimental confound, not a circular derivation.

full rationale

The paper's central claim is that a Dutch-only Wav2Vec2 model encodes Dutch phonetic and lexical features better than English-only or multilingual models. This is an empirical comparison, not a derivation from definitions. Phone identity probes are fit on training speakers and evaluated on held-out speakers; ABX and RSA use zero-shot distances against external phonetic categories and Dutch Fasttext embeddings; word clustering uses a held-out speaker split. The comparison models (fb-en, fb-voxp-100k, nonspeech) are external. Although the abstract's causal wording is stronger than the evidence, the paper itself acknowledges in Section 6 that part of the IFADV advantage may reflect 'an effect of the pre-training data domain beyond its language-specificity.' A separate concern is that Section 5 does not state whether the CGN-o ASR test split was excluded from the 537 hours of CGN used to pre-train w2v2-nl, which would be data leakage rather than circularity. No fitted parameter is renamed as a prediction, and no load-bearing claim rests on a self-citation or on a uniqueness theorem. The evaluation is therefore not equivalent to the model's training objective by construction.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The central empirical comparison rests on the quality of the underlying alignments and the representativeness of the linguistic categories, and on the assumption that architectural identity isolates the effect of pre-training data. No free parameters are fitted in support of the claim; the analysis probes are validation tools, not parts of the claim itself.

assumptions (3)
  • domain assumption WebMAUS forced alignments correctly identify phone and word boundaries in Dutch audio.
    All phone- and word-level representations are mean-pooled over these alignments (Section 3). If boundaries are inaccurate, the measured representations and differences between models would be distorted.
  • domain assumption The 37 phone categories and the BAK preschooler word list are representative of Dutch phonetics and lexicon.
    The evaluation set SSL-NL is built from these categories and words (Sections 3.1 and 3.2). A different choice of categories or words could change which models show an advantage.
  • domain assumption The three speech models have identical architectures and differ only in pre-training data.
    Section 2 states 'identical architectures (7 CNN + 12 Transformer layers)'. The comparison attributes score differences to pre-training data, but training steps and data hours also differ across models, which is not controlled.

how reviews work

0 comments
Cite this review

Pith. "Pith review of What do self-supervised speech models know about Dutch? Analyzing advantages of language-specific pre-training." pith.science (2026). https://pith.science/paper/DU2SASNM

@misc{pith2026250600981,
  author       = {Pith},
  title        = {Pith review of: What do self-supervised speech models know about Dutch? Analyzing advantages of language-specific pre-training},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DU2SASNM}},
  note         = {Machine review of arXiv:2506.00981}
}
read the original abstract

How language-specific are speech representations learned by self-supervised models? Existing work has shown that a range of linguistic features can be successfully decoded from end-to-end models trained only on speech recordings. However, it's less clear to what extent pre-training on specific languages improves language-specific linguistic information. Here we test the encoding of Dutch phonetic and lexical information in internal representations of self-supervised Wav2Vec2 models. Pre-training exclusively on Dutch improves the representation of Dutch linguistic features as compared to pre-training on similar amounts of English or larger amounts of multilingual data. This language-specific advantage is well-detected by trained clustering or classification probes, and partially observable using zero-shot metrics. Furthermore, the language-specific benefit on linguistic feature encoding aligns with downstream performance on Automatic Speech Recognition.

Figures

Figures reproduced from arXiv: 2506.00981 by the authors.

Figure 1
Figure 1. A set of high front vowels (phonemically distinctive in Dutch) is most discriminably represented in a model trained ex￾clusively on Dutch. Data points are vowel occurrences sampled across 3 speakers in Multilingual LibriSpeech. contrasts [10]. However, the same models do show distinct pat￾terns in predicting brain activity for their pre-training language vs. a different language [15], and crosslingual comparisons of… view at source ↗
Figure 2
Figure 2. Layerwise phonetic and lexical analyses, across a read speech (MLS, top row) and a dialogue (IFADV, bottom row) dataset of spoken Dutch. Measures marked * involve optimized linear transforms, whereas others are computed zero-shot; shading indicates 95% confidence intervals. The Dutch w2v2-nl model achieves highest scores across most analyses of Dutch phone and word encoding, though the size of this language-specific… view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

45 extracted references · 42 canonical work pages

  1. [1]

    Introduction In recent years, self-supervised learning (SSL) algorithms have been demonstrated to learn powerful representations of spoken language, in terms of both their downstream task performance and the richness of their embedding spaces. Despite being trained only on unlabeled speech recordings, neural SSL models vastly outperform acoustic baselines...

    work page

  2. [2]

    Models To investigate the effect of language-specific pre-training on lin- guistic feature encoding, we compare three Wav2Vec2 models with identical architectures (7 CNN + 12 Transformer layers), but with varying amounts of Dutch and other languages in their pre-training data. We train w2v2-nl (Wav2Vec2-NL) on 831 hours of spo- ken Dutch, combining data f...

    work page

  3. [3]

    and lexical [4] features. Understanding the linguistic infor- mation these models encode is relevant not only for engineers interpreting model behaviour, but also for cognitive scientists studying language acquisition from spoken input [5, 6]. Despite the wide interest for interpreting SSL model func- tioning, research has so far primarily focused on Engl...

    work page Pith review arXiv

  4. [4]

    We use a sep- arate subset of MLS audiobook segments (held-out from w2v2- nl training data), as well as the IFADV corpus [23] of face- to-face conversational speech

    Representational analysis methods We create theSSL-NL evaluation set by sampling speech record- ings from two different datasets of spoken Dutch. We use a sep- arate subset of MLS audiobook segments (held-out from w2v2- nl training data), as well as the IFADV corpus [23] of face- to-face conversational speech. Based on the speech recordings and orthograph...

    work page 2022

  5. [5]

    Results of representational analyses 4.1. Phonetic and lexical representations improve with language-specific pre-training Across most of our phonetic and lexical analyses, we observe a moderate to substantial advantage of the w2v2-nl model over its English and multilingual counterparts (Figure 2). This sug- gests that Dutch-specific pre-training enhances...

    work page

  6. [6]

    Each SSL model is fine-tuned on Dutch read-aloud speech from the CGN (component o), using 78 hours of training data while reserving 10 hours each for de- velopment and testing

    Downstream ASR performance We fine-tune our SSL models for speech-to-text transcription to examine whether the language-specific advantages in self- supervised representations also lead to improved performance on downstream ASR tasks. Each SSL model is fine-tuned on Dutch read-aloud speech from the CGN (component o), using 78 hours of training data while ...

    work page

  7. [7]

    Discussion & Conclusions We introduced the SSL-NL evaluation set, and used it to com- pare a new monolingual Dutch Wav2Vec2 model against exist- ing English and multilingual models. What do these models represent about the phonetic and lexical structure of Dutch? We find that linguistic information at both levels can be accurately decoded from each model’...

    work page

  8. [8]

    This work used the Dutch national e-infrastructure with the support of the SURF Cooperative using grant no

    Acknowledgements We would like to thank David van Leeuwen and Nik Vaessen for sharing the N-Best evaluation dataset. This work used the Dutch national e-infrastructure with the support of the SURF Cooperative using grant no. EINF-8324

    work page

Show all 45 references
  1. [9]

    Phonemic competition in end-to-end ASR models,

    L. ten Bosch, M. Bentum, and L. Boves, “Phonemic competition in end-to-end ASR models,” in Interspeech 2023, pp. 586–590

    2023

  2. [10]

    The Processing of Stress in End-to-End Automatic Speech Recognition Models,

    M. Bentum, L. ten Bosch, and T. Lentz, “The Processing of Stress in End-to-End Automatic Speech Recognition Models,” in Inter- speech 2023, pp. 2350–2354

    2023

  3. [11]

    Human-like Linguistic Biases in Neural Speech Models: Phonetic Categorization and Phonotactic Constraints in Wav2Vec2.0,

    M. de Heer Kloots and W. Zuidema, “Human-like Linguistic Biases in Neural Speech Models: Phonetic Categorization and Phonotactic Constraints in Wav2Vec2.0,” inInterspeech 2024, pp. 4593–4597

    2024

  4. [12]

    What Do Self- Supervised Speech Models Know About Words?

    A. Pasad, C.-M. Chien, S. Settle, and K. Livescu, “What Do Self- Supervised Speech Models Know About Words?” TACL, vol. 12, pp. 372–391, 2024

    2024

  5. [13]

    Simulating Early Phonetic and Word Learning Without Linguistic Categories,

    M. Lavechin, M. de Seyssel, H. Titeux, G. Wisniewski, H. Bredin, A. Cristia, and E. Dupoux, “Simulating Early Phonetic and Word Learning Without Linguistic Categories,”Developmental Science, vol. 28, no. 2, p. e13606, 2025

    2025

  6. [14]

    Cognitive science in the era of artificial intelligence: A roadmap for reverse-engineering the infant language-learner,

    E. Dupoux, “Cognitive science in the era of artificial intelligence: A roadmap for reverse-engineering the infant language-learner,” Cognition, vol. 173, pp. 43–59, 2018

    2018

  7. [15]

    Encoding of lexical tone in self-supervised models of spoken lan- guage,

    G. Shen, M. Watkins, A. Alishahi, A. Bisazza, and G. Chrupała, “Encoding of lexical tone in self-supervised models of spoken lan- guage,” in NAACL 2024, Mexico City, Mexico, pp. 4250–4261

    2024

  8. [16]

    A layer-wise analysis of Man- darin and English suprasegmentals in SSL speech models,

    A. de la Fuente and D. Jurafsky, “A layer-wise analysis of Man- darin and English suprasegmentals in SSL speech models,” in In- terspeech 2024, pp. 1290–1294

    2024

  9. [17]

    What Has LeBenchmark Learnt about French Syntax?

    Z. Dugonji ´c, A. Pupier, B. Lecouteux, and M. Coavoux, “What Has LeBenchmark Learnt about French Syntax?” in LREC- COLING 2024 , N. Calzolari, M.-Y . Kan, V . Hoste, A. Lenci, S. Sakti, and N. Xue, Eds., pp. 17 493–17 499

    2024

  10. [18]

    Do self-supervised speech models de- velop human-like perception biases?

    J. Millet and E. Dunbar, “Do self-supervised speech models de- velop human-like perception biases?” in ACL 2022, S. Muresan, P. Nakov, and A. Villavicencio, Eds., pp. 7591–7605

    2022

  11. [19]

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

    A. Baevski, H. Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A framework for self-supervised learning of speech representa- tions,” CoRR, vol. abs/2006.11477, 2020

    2006 arXiv

  12. [20]

    Unsuper- vised Pretraining Transfers Well Across Languages,

    M. Rivi `ere, A. Joulin, P.-E. Mazar ´e, and E. Dupoux, “Unsuper- vised Pretraining Transfers Well Across Languages,” in ICASSP 2020, pp. 7414–7418, iSSN: 2379-190X

    2020

  13. [21]

    Xls-r: Self-supervised cross-lingual speech representation learning at scale,

    A. Babu, C. Wang, A. Tjandra, K. Lakhotia, Q. Xu, N. Goyal, K. Singh, P. von Platen, Y . Saraf, J. Pino, A. Baevski, A. Con- neau, and M. Auli, “Xls-r: Self-supervised cross-lingual speech representation learning at scale,” in Interspeech 2022, pp. 2278– 2282

    2022

  14. [22]

    Wav2vec behind the scenes: How end2end models learn phonet- ics,

    T. tom Dieck, P. A. P´erez-Toro, T. Arias, E. Noeth, and P. Klumpp, “Wav2vec behind the scenes: How end2end models learn phonet- ics,” in Interspeech 2022, pp. 5130–5134

    2022

  15. [23]

    Toward a realistic model of speech processing in the brain with self-supervised learning,

    J. Millet, C. Caucheteux, P. Orhan, Y . Boubenec, A. Gramfort, E. Dunbar, C. Pallier, and J.-R. King, “Toward a realistic model of speech processing in the brain with self-supervised learning,” NeurIPS 2022, vol. 35, pp. 33 428–33 443

    2022

  16. [24]

    CGN, an annotated corpus of spoken Dutch,

    I. Schuurman, M. Schouppe, H. Hoekstra, and T. van der Wouden, “CGN, an annotated corpus of spoken Dutch,” in LINC-03 at EACL 2003

    2003

  17. [25]

    MLS: A Large-Scale Multilingual Dataset for Speech Research,

    V . Pratap, Q. Xu, A. Sriram, G. Synnaeve, and R. Collobert, “MLS: A Large-Scale Multilingual Dataset for Speech Research,” in Interspeech 2020, pp. 2757–2761

    2020

  18. [26]

    Common V oice: A Massively-Multilingual Speech Corpus,

    R. Ardila, M. Branson, K. Davis, M. Henretty, M. Kohler, J. Meyer, R. Morais, L. Saunders, F. M. Tyers, and G. We- ber, “Common V oice: A Massively-Multilingual Speech Corpus,” 2020

    2020

  19. [27]

    wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Represen- tations,

    A. Baevski, Y . Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Represen- tations,” in NeurIPS 2020, vol. 33, pp. 12 449–12 460

    2020

  20. [28]

    Fairseq: A Fast, Extensible Toolkit for Se- quence Modeling

    M. Ott, S. Edunov, A. Baevski, A. Fan, S. Gross, N. Ng, D. Grang- ier, and M. Auli, “Fairseq: A Fast, Extensible Toolkit for Se- quence Modeling.” ACL, 2019, pp. 48–53

    2019

  21. [29]

    V oxPopuli: A Large- Scale Multilingual Speech Corpus for Representation Learning, Semi-Supervised Learning and Interpretation,

    C. Wang, M. Riviere, A. Lee, A. Wu, C. Talnikar, D. Haziza, M. Williamson, J. Pino, and E. Dupoux, “V oxPopuli: A Large- Scale Multilingual Speech Corpus for Representation Learning, Semi-Supervised Learning and Interpretation,” in ACL-IJCNLP 2021, C. Zong, F. Xia, W. Li, and ...

    2021

  22. [30]

    HuggingFace’s Trans- formers: State-of-the-art Natural Language Processing,

    T. Wolf, L. Debut, V . Sanh, J. Chaumond, C. Delangue, A. Moi, P. Cistac, T. Rault, R. Louf, M. Funtowicz, J. Davison, S. Shleifer, P. v. Platen, C. Ma, Y . Jernite, J. Plu, C. Xu, T. L. Scao, S. Gugger, M. Drame, Q. Lhoest, and A. M. Rush, “HuggingFace’s Trans- formers: State...

    2020

  23. [31]

    The IFADV Corpus: a Free Dialog Video Corpus,

    R. van Son, W. Wesseling, E. Sanders, and H. van den Heuvel, “The IFADV Corpus: a Free Dialog Video Corpus,” in LREC 2008, N. Calzolari, K. Choukri, B. Maegaard, J. Mariani, J. Odijk, S. Piperidis, and D. Tapias, Eds. ELRA

    2008

  24. [32]

    Domain- informed probing of wav2vec 2.0 embeddings for phonetic fea- tures,

    P. C. English, J. Kelleher, and J. Carson-Berndsen, “Domain- informed probing of wav2vec 2.0 embeddings for phonetic fea- tures,” in SIGMORPHON 2022, pp. 83–91

    2022

  25. [33]

    Modeling the initial state of early phonetic learning in infants,

    M. Poli, T. Schatz, E. Dupoux, and M. Lavechin, “Modeling the initial state of early phonetic learning in infants,” Language De- velopment Research, vol. 5, no. 1, 2024

    2024

  26. [34]

    Silhouettes: A graphical aid to the interpretation and validation of cluster analysis,

    P. J. Rousseeuw, “Silhouettes: A graphical aid to the interpretation and validation of cluster analysis,” Journal of Computational and Applied Mathematics, vol. 20, pp. 53–65, 1987

    1987

  27. [35]

    Scikit-learn: Machine learning in Python,

    F. Pedregosa, G. Varoquaux, A. Gramfort, V . Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V . Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay, “Scikit-learn: Machine learning in Python,” JMLR, vol. 12, pp. 2825...

    2011

  28. [36]

    Mulder, Y

    F. Mulder, Y . Timman, and S. Verhallen, Handreiking bij de Ba- siswoordenlijst Amsterdamse Kleuters (BAK). ITTA, 2009

    2009

  29. [37]

    Representational similarity analysis - connecting the branches of systems neuro- science,

    N. Kriegeskorte, M. Mur, and P. A. Bandettini, “Representational similarity analysis - connecting the branches of systems neuro- science,” Frontiers in Systems Neuroscience, vol. 2, no. 4, pp. 1– 28, 2008

    2008

  30. [38]

    Enriching Word Vectors with Subword Information,

    P. Bojanowski, E. Grave, A. Joulin, and T. Mikolov, “Enriching Word Vectors with Subword Information,”TACL, vol. 5, pp. 135– 146, 2017

    2017

  31. [39]

    N-best: the northern- and southern-dutch benchmark evaluation of speech recognition technology

    J. M. Kessens and D. A. van Leeuwen, “N-best: the northern- and southern-dutch benchmark evaluation of speech recognition technology.” inInterspeech 2007, 2007, pp. 1354–1357

    2007

  32. [40]

    An information-theoretic analysis of self-supervised discrete repre- sentations of speech,

    B. M. Abdullah, M. M. Shaik, B. M ¨obius, and D. Klakow, “An information-theoretic analysis of self-supervised discrete repre- sentations of speech,” in Interspeech 2023, pp. 2883–2887

    2023

  33. [41]

    Do speech registers differ in the predictability of words?

    M. Bentum, L. Ten Bosch, A. Van den Bosch, and M. Ernestus, “Do speech registers differ in the predictability of words?” Inter- national Journal of Corpus Linguistics, vol. 24, no. 1, pp. 98–130, 2019

    2019

  34. [42]

    From text to talk: Harnessing conversational corpora for humane and diversity-aware language technology,

    M. Dingemanse and A. Liesenfeld, “From text to talk: Harnessing conversational corpora for humane and diversity-aware language technology,” in ACL 2022, S. Muresan, P. Nakov, and A. Villavi- cencio, Eds., Dublin, Ireland, pp. 5614–5633

    2022

  35. [43]

    Speech self-supervised representation benchmarking: Are we doing it right?

    S. Zaiem, Y . Kemiche, T. Parcollet, S. Essid, and M. Ravanelli, “Speech self-supervised representation benchmarking: Are we doing it right?” in Interspeech 2023, pp. 2873–2877

    2023

  36. [44]

    LeBenchmark 2.0: A standardized, replicable and enhanced framework for self-supervised represen- tations of French speech,

    T. Parcollet, H. Nguyen, S. Evain, M. Zanon Boito, A. Pupier, S. Mdhaffar, H. Le, S. Alisamir, N. Tomashenko, M. Dinarelli, S. Zhang, A. Allauzen, M. Coavoux, Y . Est `eve, M. Rouvier, J. Goulian, B. Lecouteux, F. Portet, S. Rossato, F. Ringeval, D. Schwab, and L. Besacier, “L...

    2024

  37. [45]

    On the encoding of gender in transformer-based asr representations,

    A. Krishnan, B. M. Abdullah, and D. Klakow, “On the encoding of gender in transformer-based asr representations,” in Interspeech 2024, pp. 3090–3094

    2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.