This paper introduces a novel method for modeling the local geometry of sentence embeddings, focusing on how paraphrased sentences are organized in embedding space. The researchers developed a geometric modeling scheme using affine, quadratic, and cubic fitted models, along with a surface-based latent probing procedure. Experiments demonstrated that nonlinear models better represent embedding clouds than affine models, and they introduced CoPaGE-300K, a dataset of semantically similar sentence variants. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Introduces a new method for analyzing sentence embedding geometry and a dataset that could aid in understanding representation spaces.
RANK_REASON The cluster contains an academic paper detailing a new method for analyzing sentence embeddings and a new dataset. [lever_c_demoted from research: ic=1 ai=1.0]