mlsanityData to 7 Oct 2026

[b] = [d] - [t] + [p]: Self-supervised Speech Models Discover Phonological Vector Arithmetic

arXiv 2602.18899 ↗First posted 21 Feb 2026eess.AS cs.CL cs.LGcited by 11

Not on the map. Why?

Abstract (arXiv)

Self-supervised speech models (S3Ms) are known to encode rich phonetic information, yet how this information is structured remains underexplored. We conduct a comprehensive study across 96 languages to analyze the underlying structure of S3M representations, with particular attention to phonological vectors. We first show that there exist linear directions within the model's representation space that correspond to phonological features. We further demonstrate that the scale of these phonological vectors correlate to the degree of acoustic realization of their corresponding phonological features in a continuous manner. For example, the difference between [d] and [t] yields a voicing vector: adding this vector to [p] produces [b], while scaling it results in a continuum of voicing. Together, these findings indicate that S3Ms encode speech using phonologically interpretable and compositional vectors, demonstrating phonological vector arithmetic. All code and interactive demos are available at https://github.com/juice500ml/phonetic-arithmetic .

Key facts. Cited by 11 papers in the map (data through 7 Oct 2026). Builds on 38 of its 84 references that are in the map. First posted on arXiv 21 Feb 2026.

Builds on

38 of 84 references.

References · 38

  1. 2013Efficient Estimation of Word Representations in Vector Spacecited by 4,346
  2. 2020wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representationscited by 3,853
  3. 2021HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Unitscited by 2,301
  4. 2021WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processingcited by 1,609
  5. 2019Common Voice: A Massively-Multilingual Speech Corpuscited by 1,358
  6. 2023Steering Language Models With Activation Engineeringcited by 824
  7. 2019How Contextual are Contextualized Word Representations? Comparing the Geometry of BERT, ELMo, and GPT-2 Embeddingscited by 770
  8. 2019LibriTTS: A Corpus Derived from LibriSpeech for Text-to-Speechcited by 639
  9. 2020Unsupervised Cross-lingual Representation Learning for Speech Recognitioncited by 496
  10. 2020MLS: A Large-Scale Multilingual Dataset for Speech Researchcited by 467
  11. 2019Libri-Light: A Benchmark for ASR with Limited or No Supervisioncited by 462
  12. 2021VoxPopuli: A Large-Scale Multilingual Speech Corpus for Representation Learning, Semi-Supervised Learning and Interpretationcited by 455
26 more
  1. 2021GigaSpeech: An Evolving, Multi-domain ASR Corpus with 10,000 Hours of Transcribed Audiocited by 391
  2. 2021Layer-wise Analysis of a Self-supervised Speech Representation Modelcited by 293
  3. 2022Comparative layer-wise analysis of self-supervised speech modelscited by 155
  4. 2023Vocos: Closing the gap between time-domain and Fourier-based neural vocoders for high-quality audio synthesiscited by 150
  5. 2021Simple and Effective Zero-shot Cross-lingual Phoneme Recognitioncited by 79
  6. 2024Self-Supervised Speech Representations are More Phonetic than Semanticcited by 70
  7. 2023What Do Self-Supervised Speech Models Know About Words?cited by 62
  8. 2018Towards Understanding Linear Word Analogiescited by 57
  9. 2022Word Discovery in Visually Grounded, Self-Supervised Speech Modelscited by 37
  10. 2022Evidence of Vocal Tract Articulation in Self-Supervised Learning of Speechcited by 30
  11. 2024SyllableLM: Learning Coarse Semantic Units for Speech Language Modelscited by 24
  12. 2020Phonological Features for 0-shot Multilingual Speech Synthesiscited by 20
  13. 2022Opening the Black Box of wav2vec Feature Encodercited by 18
  14. 2023Wave to Syntax: Probing spoken language models for syntaxcited by 18
  15. 2024Coding Speech through Vocal Tract Kinematicscited by 18
  16. 2020Analogies minus analogy test: measuring regularities in word embeddingscited by 17
  17. 2023Universal Automatic Phonetic Transcription into the International Phonetic Alphabetcited by 16
  18. 2023Self-Supervised Models of Speech Infer Universal Articulatory Kinematicscited by 16
  19. 2023An Information-Theoretic Analysis of Self-supervised Discrete Representations of Speechcited by 15
  20. 2025Leveraging Allophony in Self-Supervised Speech Models for Atypical Pronunciation Assessmentcited by 15
  21. 2024FLEURS-R: A Restored Multilingual Speech Corpus for Generation Taskscited by 13
  22. 2023Understanding Probe Behaviors through Variational Bounds of Mutual Informationcited by 11
  23. 2025Emergent morpho-phonological representations in self-supervised speech modelscited by 11
  24. 2024Phonetic Segmentation of the UCLA Phonetics Lab Archivecited by 9
  25. 2023PWESuite: Phonetic Word Embeddings and Tasks They Facilitatecited by 6
  26. 2017Learning weakly supervised multimodal phoneme embeddingscited by 5

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 11.

11 papers.

  1. 2026Self-Supervised Speech Models Encode Phonetic Context via Position-dependent Orthogonal Subspacescited by 7
  2. 2026An Empirical Recipe for Universal Phone Recognitioncited by 5
  3. 2026Adapting Self-Supervised Speech Representations for Cross-lingual Dysarthria Detection in Parkinson's Diseasecited by 4
  4. 2026Speech Playground: An Interactive Tool for Speech Analysis and Comparisoncited by 1
  5. 2026Language Orthogonalization for Zero-Shot Cross-Lingual Audio Deepfake Detectioncited by 1
  6. 2026A Native-Reference Coordinate Geometry for L2 Pronunciation Deviation Using Self-Supervised Speech Modelscited by 1
  7. 2026Phone Segmentation and Recognition through Phonological Activation Mappingcited by 0
  8. 2026Quantifying Consonant Contributions to Word Intelligibility via Acoustic Maskingcited by 0
  9. 2026A Native-Reference Phone-Class Geometry for Second-Language Pronunciation Analysiscited by 0
  10. 2026Do Audio Language Models Hear and Read Distinctive Features Alike?cited by 0
  11. 2026Child-Adapted Structured Phonological Representations for Interpretable Speech Sound Analysiscited by 0