mlsanityData to 7 Oct 2026

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

arXiv 2407.03005 ↗First posted 3 Jul 2024cs.CL cs.AI cs.SDcited by 19

Not on the map. Why?

Abstract (arXiv)

What do deep neural speech models know about phonology? Existing work has examined the encoding of individual linguistic units such as phonemes in these models. Here we investigate interactions between units. Inspired by classic experiments on human speech perception, we study how Wav2Vec2 resolves phonotactic constraints. We synthesize sounds on an acoustic continuum between /l/ and /r/ and embed them in controlled contexts where only /l/, only /r/, or neither occur in English. Like humans, Wav2Vec2 models show a bias towards the phonotactically admissable category in processing such ambiguous sounds. Using simple measures to analyze model internals on the level of individual stimuli, we find that this bias emerges in early layers of the model's Transformer module. This effect is amplified by ASR finetuning but also present in fully self-supervised models. Our approach demonstrates how controlled stimulus designs can help localize specific linguistic knowledge in neural speech models.

Key facts. Cited by 19 papers in the map (data through 7 Oct 2026). Builds on 10 of its 26 references that are in the map. First posted on arXiv 3 Jul 2024.

Builds on

10 of 26 references.

References · 10

  1. 2020wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representationscited by 3,853
  2. 2023What Do Self-Supervised Speech Models Know About Words?cited by 62
  3. 2023Probing self-supervised speech models for phonetic and phonemic information: a case study in aspirationcited by 35
  4. 2022Do self-supervised speech models develop human-like perception biases?cited by 28
  5. 2017Encoding of phonology in a recurrent neural model of grounded speechcited by 27
  6. 2023DecoderLens: Layerwise Interpretation of Encoder-Decoder Transformerscited by 22
  7. 2023Wave to Syntax: Probing spoken language models for syntaxcited by 18
  8. 2020Analyzing analytical methods: The case of phonology in neural models of spoken languagecited by 16
  9. 2023Homophone Disambiguation Reveals Patterns of Context Mixing in Speech Transformerscited by 13
  10. 2022Successes and critical failures of neural networks in capturing human-like speech recognitioncited by 7

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 1; 2025: 10; 2026: 8.

19 papers.

Sort:

By area:

  1. 2025How Linguistics Learned to Stop Worrying and Love the Language Modelscited by 50
  2. 2026Self-Supervised Speech Models Encode Phonetic Context via Position-dependent Orthogonal Subspacescited by 7
  3. 2024Exploring bat song syllable representations in self-supervised audio encoderscited by 5
  4. 2025What do self-supervised speech models know about Dutch? Analyzing advantages of language-specific pre-trainingcited by 4
  5. 2025Trick or Neat: Adversarial Ambiguity and Language Model Evaluationcited by 3
  6. 2026Emergence of Phonemic, Syntactic, and Semantic Representations in Artificial Neural Networkscited by 2
  7. 2025Automatic classification of stop realisation with wav2vec2.0cited by 1
  8. 2025From Words to Waves: Analyzing Concept Formation in Speech and Text-Based Foundation Modelscited by 1
  9. 2025The Curious Case of Visual Grounding: Different Effects for Speech- and Text-based Language Encoderscited by 1
  10. 2025Language models as tools for investigating the distinction between possible and impossible natural languagescited by 0
  11. 2025Large language models have learned to use languagecited by 0
  12. 2025Large language models are not about natural languagecited by 0
7 more
  1. 2025Linguists should learn to love speech-based deep learning modelscited by 0
  2. 2026Are Language Models Models?cited by 0
  3. 2026Tracking the emergence of linguistic structure in self-supervised models learning from speechcited by 0
  4. 2026Across the Levels of Analysis: Explaining Predictive Processing in Humans Requires More Than Machine-Estimated Probabilitiescited by 0
  5. 2026You Can't Fight in Here! This is BBS!cited by 0
  6. 2026Not-So-Strange Love: Language Models and Generative Linguistic Theories are More Compatible than They Appearcited by 0
  7. 2026Perceptual compensation for tonal context in self-supervised speech modelscited by 0