Tomomasa Hara Tohoku University, Hiroto Kurita Tohoku University, Masaaki Imaizumi Kyoto University, RIKEN Center for Advanced Intelligence Project and The University of Tokyo, Kentaro Inui MBZUAI, RIKEN and Tohoku University, Sho Yokoi NINJAL, Tohoku University and RIKEN
原文摘要
For constructing text embeddings, mean pooling, which averages token embeddings, is the standard approach. This paper examines whether mean pooling actually works well in real models. First, we note that mean pooling can collapse information beyond the first-order statistics of the token embeddings, such as second-order statistics that capture their spatial structure, potentially mapping distinct token embedding distributions to similar text embeddings. Motivated by this concern, we propose a simple metric to quantify such a collapse induced by mean pooling. Then, using this metric, we empirically measure how often this collapse occurs in actual models and texts, and find that modern text encoders are robust to this collapse. In particular, contrastive fine-tuned text encoders tend to be less prone to the collapse than their pretrained backbone models. We also find that the robustness of these text encoders lies in the concentration of token embeddings within each text. In addition, we find that robustness to the collapse, as quantified by our proposed metric, correlates with downstream task performance. Overall, our findings offer a new perspective on why modern text encoders remain effective despite relying on seemingly coarse mean pooling.
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表