arXiv：cs.LG· Kwanghee Choi, Eunjung Yeo, Cheol Jun Cho, David Harwath, David R. Mortensen·· 4 小时前AI 评分50
自监督语音模型发现音系向量算术：[b] = [d] - [t] + [p]
[b] = [d] - [t] + [p]: Self-supervised Speech Models Discover Phonological Vector Arithmetic
AI 导读
自监督语音模型（S3M）的表征空间中存在对应音系特征的线性方向，其向量尺度与对应音系特征的声学实现程度呈连续相关。研究在 96 种语言上验证：用 [d] 减去 [t] 可得到清浊向量，将其加到 [p] 上得到 [b]，缩放该向量则产生清浊连续统。代码和交互 demo 已公开，论文被 ACL 2026 Findings 接收。
正文
Abstract: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 this https URL .
| Comments: | Accepted to ACL 2026 Findings |
| Subjects: | Audio and Speech Processing (eess.AS); Computation and Language (cs.CL); Machine Learning (cs.LG); Sound (cs.SD) |
| Cite as: | arXiv:2602.18899 [eess.AS] |
| (or arXiv:2602.18899v4 [eess.AS] for this version) | |
| https://doi.org/10.48550/arXiv.2602.18899 arXiv-issued DOI via DataCite |
Submission history
From: Kwanghee Choi [view email]
[v1]
Sat, 21 Feb 2026 16:43:13 UTC (1,847 KB)
[v2]
Thu, 12 Mar 2026 17:25:23 UTC (1,965 KB)
[v3]
Mon, 13 Apr 2026 18:05:11 UTC (1,959 KB)
[v4]
Tue, 6 Oct 2026 04:41:21 UTC (1,959 KB)
来源：arXiv：cs.LG · arxiv.org