arXiv：cs.CL· Juri Opitz, Andrianos Michail·· 4 小时前
研究发现嵌入模型对物理量测量建模存在奇特偏差
Embedding Models Measure in Peculiar Ways
AI 导读
研究考察嵌入空间能否反映质量、距离、时间和体积的物理测量，发现物理测量在嵌入空间中仅被弱建模，并呈现出相当奇特的测量模式。进一步分析表明，物理测量的嵌入表示强烈受表层字符串相似性影响，而重新校准相似度并未显著改善对齐效果。
正文
Abstract:Embedding spaces define notions of semantic similarity and distance. We study whether those embeddings reflect physical measurements of mass, distance, time and volume, which admit a unique, objective notion of semantic equivalence and distance. We find that physical measurement is only weakly modeled in the embedding space, and that instead quite peculiar measurement patterns can be observed. Further analysis indicates that embedding representations of physical measurements are strongly influenced by superficial string similarity, and recalibration of similarity does not substantially improve the alignment.
| Subjects: | Computation and Language (cs.CL); Machine Learning (cs.LG) |
| Cite as: | arXiv:2609.20821 [cs.CL] |
| (or arXiv:2609.20821v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2609.20821 arXiv-issued DOI via DataCite |
Submission history
From: Juri Opitz [view email]
[v1]
Thu, 17 Sep 2026 17:59:54 UTC (1,235 KB)
[v2]
Thu, 8 Oct 2026 15:31:40 UTC (1,235 KB)
来源：arXiv：cs.CL · arxiv.org