Why Mean Pooling Works: Quantifying Second-Order Collapse in Text Embeddings

• URL: http://arxiv.org/abs/2604.27398v1
• Date: Thu, 30 Apr 2026 04:09:26 GMT
• Title: Why Mean Pooling Works: Quantifying Second-Order Collapse in Text Embeddings
• Abstract summary: Mean pooling, which averages token embeddings, is the standard approach to constructing text embeddings.<n>This paper examines whether mean pooling actually works well in real models.
• Score: 30.943998879066857
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: 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.

Related papers

• Does Text Actually Help? Uncovering and Resolving Text Collapse in Multimodal Time Series Forecasting [7.885089945243776]
  We argue that text collapse is a consequence of a fundamental asymmetry in time series forecasting.<n>We propose textbfREST-TS (textbfResidual-textbfExclusive textbfSupervision for textbfText in textbfTime textbfSeries), which turns the asymmetry into a design principle.
  arXiv  Detail & Related papers  (2026-06-17T15:47:26Z)
• Hacking Generative Perplexity: Why Unconditional Text Evaluation Needs Distributional Metrics [49.443264461057645]
  Diffusion and continuous flow-based language models have emerged as the leading non-autoregressive alternatives to language modeling.<n>Progress in both paradigms is overwhelmingly tracked by generative perplexity (gen-PPL): the per-token negative log-likelihood of samples under a frozen autoregressive (AR) scorer.<n>We argue that this metric is unsound. By construction, gen-PPL measures only predictability under the scoring AR, not grammaticality or semantic coherence.
  arXiv  Detail & Related papers  (2026-06-07T02:35:56Z)
• LRANet++: Low-Rank Approximation Network for Accurate and Efficient Text Spotting [118.93173826110815]
  We propose a novel parameterized text shape method based on low-rank approximation for precise detection.<n>By exploiting the inherent shape correlation among different text contours, our method achieves consistency and compactness in shape representation.<n>We integrate the enhanced detection module with a lightweight recognition branch to form an end-to-end text spotting framework, termed LRANet++.
  arXiv  Detail & Related papers  (2025-11-08T03:08:03Z)
• The Medium Is Not the Message: Deconfounding Document Embeddings via Linear Concept Erasure [98.71456610527598]
  Embedding-based similarity metrics can be influenced by spurious attributes like the text's source or language.<n>This paper shows that a debiasing algorithm that removes information about observed confounders from the encoder representations substantially reduces these biases at a minimal computational cost.
  arXiv  Detail & Related papers  (2025-07-01T23:17:12Z)
• Dense Retrievers Can Fail on Simple Queries: Revealing The Granularity Dilemma of Embeddings [65.31723739561151]
  This work stems from an observed limitation of text encoders: embeddings may not be able to recognize fine-grained entities or events within encoded semantics.<n>We introduce a new evaluation dataset, CapRetrieval, in which passages are image captions and queries are phrases targeting entity or event concepts in diverse forms.<n>We finetune encoders with our proposed data generation strategies, enabling a small 0.1B encoder to outperform the state-of-the-art 7B model.
  arXiv  Detail & Related papers  (2025-06-10T09:00:33Z)
• What Are You Token About? Dense Retrieval as Distributions Over the Vocabulary [68.77983831618685]
  We propose to interpret the vector representations produced by dual encoders by projecting them into the model's vocabulary space. We show that the resulting projections contain rich semantic information, and draw connection between them and sparse retrieval.
  arXiv  Detail & Related papers  (2022-12-20T16:03:25Z)
• Toward the Understanding of Deep Text Matching Models for Information Retrieval [72.72380690535766]
  This paper aims at testing whether existing deep text matching methods satisfy some fundamental gradients in information retrieval. Specifically, four attributions are used in our study, i.e., term frequency constraint, term discrimination constraint, length normalization constraints, and TF-length constraint. Experimental results on LETOR 4.0 and MS Marco show that all the investigated deep text matching methods satisfy the above constraints with high probabilities in statistics.
  arXiv  Detail & Related papers  (2021-08-16T13:33:15Z)
• Statistically significant detection of semantic shifts using contextual word embeddings [7.439525715543974]
  We propose an approach to estimate semantic shifts by combining contextual word embeddings with permutation-based statistical tests. We demonstrate the performance of this approach in simulation, achieving consistently high precision by suppressing false positives. We additionally analyzed real-world data from SemEval-2020 Task 1 and the Liverpool FC subreddit corpus.
  arXiv  Detail & Related papers  (2021-04-08T13:58:54Z)
• Language Modeling with Reduced Densities [0.0]
  We show that sequences of symbols from a finite alphabet, such as those found in a corpus of text, form a category enriched over probabilities. We then address a second fundamental question: How can this information be stored and modeled in a way that preserves the categorical structure?
  arXiv  Detail & Related papers  (2020-07-08T00:41:53Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.