CS Paper Radar

Semantically Rich, Metrically Affect-Blind: A Geometric Blind Spot in Modern Text Embedders

Mahdi Naser Moghadasi

cs.CL cs.LG ·submitted 2026-10-23 03:50 UTC · v1
code/data

Abstract

We report a systematic dissociation in modern text embedders: sentiment is almost perfectly decodable from their embeddings, yet almost entirely absent from their metric. Across eleven frozen encoders spanning 2021 to 2026, including 2026-generation models and the 4B-parameter LLM embedder Qwen3-Embedding-4B, a linear probe recovers sentence polarity at up to 91.3% accuracy while cosine similarity separates polarity by a margin of only 0.003-0.064. Qwen3-Embedding-4B, the strongest probe in the panel, reaches 91.3% probe accuracy but only 64.3 retrieval MAP. Because retrieval, clustering, deduplication, and retrieval-augmented generation operate on the metric rather than on a trained probe, these systems are effectively blind to information their encoder demonstrably contains. A task-conditioning control suggests this reflects the pretraining objective rather than a prompting artifact: conditioning Jina-v5 on its classification task rather than its retrieval task recovers only about half of the MAP gain that injection achieves on the same weights (63.2 to 73.8, versus 85.1), and leaves the alignment gap essentially unchanged. Sentiment-Injected Sentence Embeddings (SISE) train a small (0.13M-0.69M-parameter) supervised-contrastive head on a frozen encoder in seconds on a CPU, using the labelled training set. The head improves every encoder released since 2022 on full-data probe accuracy (up to +5.8 points), retrieval MAP (up to +26.0) and polarity margin (7-61x). This is a specialization, not a free gain: the compact 128-d head loses 38-41 Spearman points on STS-B, while a fused variant that concatenates it with the base embedding loses 2.4-12.0. With the head, a 0.6B encoder at 128 dimensions matches a 4B encoder given the same head (91.8 vs. 90.3 MAP). We release a CPU-reproducible pipeline and a reusable diagnostic, the Attribute Alignment Gap.

Comment
6 pages, 2 figures, 3 tables. Code: https://github.com/mahdinaser/sent2vec-sise

Closest papers

Nearest neighbours by meaning across the whole index.

Reducing Representational Interference in Retrieval Embeddings with Trainable Prefix Tokens

cs.IR ·2026-11-05 ·Seongtae Hong, Youngjoon Jang, Jungseob Lee, Chanjun Park, Heuiseok Lim
top venue similarity 0.65
Abstract · CIKM 2026

Embedding models compress variable-length text into fixed-size vector representations, but this compression can entangle semantic content with residual token-level information, such as artifacts arising from attention sink behavior. We propose a simple method that prepends trainable [DUMP] tokens to the input during embedding model training. The method keeps the model architecture, pooling strategy, and training loss unchanged, while optimizing the added tokens through the standard contrastive retrieval objective. By introducing trainable prefix positions that interact with regular tokens during encoding, [DUMP] tokens help reduce the influence of residual or uninformative token-level signals on the final retrieval representation. Experiments on standard retrieval and long-context benchmarks show that our method consistently improves over finetuned baselines across encoder and decoder embedding models. Further analysis shows that [DUMP] tokens alter token-level activation and attention patterns associated with low-information tokens. These results suggest that trainable prefix tokens provide a lightweight way to improve retrieval representations.

0.39

MoganColBERT-TR: A Late-Interaction Multi-Vector Retrieval Model for Turkish

cs.CL ·2026-08-26 ·Furkan Yilmaz, Habibe Aleyna Tasdemir, Muhammed Faruk Gozay
benchmarkcode/data similarity 0.65
Abstract · 12 pages, 6 tables. Model weights: https://huggingface.co/moganai

We previously reported a ModernBERT encoder trained from scratch for Turkish (MoganBERT-TR) and a single-vector embedding model built on top of it (MoganBERT-embed). This work introduces the third model in that lineage: MoganColBERT-TR, a multi-vector retrieval model that, instead of compressing a query or a document into a single vector, represents it at the token level through a 768->128 projection and scores it with MaxSim late interaction. The model is not trained from scratch: the embedding model's encoder is taken as the starting point and adapted to the ColBERT objective with a single-epoch distillation phase. Training data is produced from two sources - title-to-passage pairs carved out of our own pretraining corpus in the character domain and at sentence boundaries, and two Turkish question-based retrieval sets - and is distilled from the soft scores of a cross-encoder teacher (bge-reranker-v2-m3) over one positive and seven mined negatives. We show that in hard negative mining, rank-based skipping alone is insufficient and must be combined with a group mask and a cosine ceiling. Evaluation is carried out with the official pipeline of TurkColBERT, a benchmark built for Turkish late-interaction retrieval (PLAID index, exact MaxSim), on five Turkish BEIR datasets; none of them appears in our training pool, so all five results are clean zero-shot. With 148.9M parameters, MoganColBERT-TR reaches an overall score of 37.36 (35.53 nDCG@100, 31.81 nDCG@10) averaged over the five datasets and finishes second among the five models compared: it outperforms the twice-as-large ColmmBERT-base-TR on four of five datasets and by +3.05 overall, and the benchmark's largest model by +12.30. The gap to the leading model (mLateOn) is concentrated on ArguAna-TR, the dataset with by far the longest queries.

0.41

Do Language Models Feel the Sentiment? Investigating Emotion Vectors

cs.CL ·2026-11-10 ·Ali Hassan Khan, Saad Bin Ather, Muhammad Saif, Maryam Bashir, Hajra Waheed
similarity 0.65
Abstract · 16 pages, 3 figures, 11 tables. Accepted at ExplAI-2026: 2nd Workshop on Advanced AI in Explainability and Ethics for the Sustainable Develo…

Emotion vectors are directions in a language model's internal representations that are associated with emotional concepts. Prior work shows that these vectors can influence model behavior, but their role in sentiment classification remains unclear. This study investigates emotion directions in Qwen2.5-0.5B-Instruct and Qwen2.5-3B-Instruct. Eight directions extracted from GoEmotions are evaluated through sentiment probing, activation interventions, and supervised correction. Emotion-only probes achieve 88.66% and 90.70% accuracy on binary TweetEval. In exploratory matched-ablation experiments, raw emotion-space removal causes 2.39 and 1.10 percentage points more accuracy loss than matched isotropic random-subspace removal for 0.5B and 3B, respectively. All four TweetEval raw and denoised effects are significant after Holm adjustment across eight comparisons. An exploratory SST-2 extension shows smaller raw excess losses of 0.66 and 0.57 points, with no effects surviving Holm adjustment. Frozen sentiment-correction classifiers are also tested on 20,000 Amazon Polarity reviews. Emotion-based correction improves accuracy over uncorrected Qwen, but does not consistently outperform margin-only baselines. These results distinguish sentiment information available in model representations, the effects of changing those representations, and their practical value for classification. They do not establish a uniquely emotional mechanism.

0.18

When Does Label-Free Deconfounding of Document Embeddings Work?

cs.CL ·2026-11-14 ·Anna Kiepura, Yingqiang Gao, Elliott Ash, Yu Fan
similarity 0.65
Abstract

Document embeddings encode spurious attributes that can distort semantic comparisons; yet supervised erasure requires metadata that is often unavailable. We systematically investigate label-free deconfounding across corpora and embedding models. Our results show that label-free methods improve semantic utility over unprocessed embeddings, while simple geometric approaches recover much of the supervised benefit, sometimes matching metadata-supervised erasure in linear confounder recoverability and semantic utility. We identify two distinct causes of ineffective erasure: poor proxy-confounder alignment and insufficient intervention coverage. Crucially, confounder recoverability and semantic utility are distinct dimensions of deconfounding success: utility can improve despite residual confounder information, while stronger suppression does not necessarily improve utility. Moreover, reduced linear recoverability can coexist with substantial nonlinear recoverability. Together, these findings distinguish confounder discovery from intervention effectiveness and highlight the importance of evaluating nuisance suppression alongside semantic preservation.

0.18

Embedding Cosine Measures Text Length Before It Measures Homogenization

cs.CL cs.CYcs.SI ·2026-11-16 ·Fengming Liu
similarity 0.64
Abstract · 11 pages, 10 figures, 2 tables. Re-measurement of arXiv:2608.13674

Mean pairwise cosine similarity between transformer embeddings is a standard measure of whether online discourse is converging under generative AI. When the embeddings are mean-pooled outputs of an encoder never trained for sentence similarity, this measure mostly tracks text length. In 6.06 million Reddit comments, an uncentered DeBERTa-v3 measure had suggested that only conservative discourse broke toward homogeneity after late 2022. Truncating the same comments from 10 to 150 words raises their mean cosine by eight times the reported break. A placebo that swaps every comment for a random one of equal length reproduces 81% of the asymmetry, and a fixed-content simulation reproduces 78%. The leading principal direction carries 99.9% of raw DeBERTa homogeneity, and once length is handled the asymmetry disappears under every encoder. On 575,000 Stack Exchange answers the measure reports a post-ChatGPT homogenization of technical sites that the placebo fully reproduces, although answers grew longer. A census of 31 studies locates the risk where anisotropic encoders meet time-series or length-unequal designs. A reverse control on a well-known experiment shows that centering with renormalization can reverse a genuine effect. I propose a low-cost diagnostic protocol. Corrected measures show that the two political camps converged between 2019 and 2022 and stopped converging before ChatGPT. Cross-camp replies engage their targets more closely than same-camp replies.

0.30

One Geometry, Different Outcomes: Readout-Dependent Effects of the Modality Gap in Vision-Language Models

cs.CV cs.AIcs.LG ·2026-09-28 ·Aditya Sharma, Divya Saxena
similarity 0.64
Abstract · 9 pages, 3 figures, 3 tables

Contrastive vision-language models learn shared embedding spaces by aligning matched image-text pairs, yet their representations remain separated by a modality gap. Prior work reports divergent effects of modifying this gap: reducing it can improve zero-shot classification and cross-modal alignment, whereas removing gap-related structure can degrade image-text retrieval. In this paper, we provide a unified geometric explanation for these task-dependent effects. Across CLIP and SigLIP encoders, we find that a single dominant direction captures 94.4-99.9% of the squared norm of the image-text mean separation, revealing that the mean-separation component is approximately rank-one. A decomposition of the similarity score then identifies three task-specific roles. In zero-shot classification, query-side fixed gap-offset subtraction is exactly equivalent to an additive class bias. In standard cross-modal retrieval, projecting out the gap direction and renormalising residuals discards candidate-specific norm information, inducing a multiplicative ranking distortion; a geometry-derived exponent tracks the grid-search optimum (Spearman rho = 0.93) and restores performance in some settings, although the gains transfer unevenly. In mixed-modal retrieval, the gap direction sorts candidates by modality; its removal can improve cross-modal ranking, unlike random or non-gap controls. Residual semantic structure after removal defines the limits of the rank-one account. Together, these results explain why gap modification can improve, degrade, or restore performance across downstream settings. By clarifying when and why gap modification changes model behavior, this account provides a principled basis for selecting gap interventions in similarity-based vision-language systems across evaluated downstream tasks.

0.22

ISBC: Interpretable Sparse Binary Coding of Pretrained Text Embeddings

cs.LG ·2026-10-31 ·Stefano Maria Iacus, Haodong Qi
similarity 0.64
Abstract · updated version for smoother reading of the results

Pretrained text embeddings encode semantic structure in high-dimensional continuous spaces whose coordinates are difficult to interpret. We introduce Interpretable Sparse Binary Coding (ISBC), which learns a small set of binary dimensions from frozen embeddings without taxonomy labels or downstream outcomes. A trainable encoder and affine decoder combine reconstruction with controls on activity, factor utilization, dependence, and activation uncertainty. Individual bits and their conjunctions select overlapping document groups; complete activation profiles provide finer contexts for conditional comparisons. We evaluate these representations across corpora, backbones, and code configurations through reconstruction, geometry, predictive probes, and document inspection. Empirical cases identify both distinctions within existing categories and shared content across their boundaries. In a mixed-domain corpus, the most frequent profiles predominantly identify source collections, while selected within-source comparisons reveal sentiment and migration contrasts. Selected associations persist in additional documents encoded without refitting. In an engagement application, bits alone improve prediction of log-transformed outcomes over a constant reference, while their combinations identify within-event content contrasts associated with engagement. The results support using binary coordinates and their combinations for empirical interpretation, with complete profiles providing a fine, partly initialization-dependent description rather than a unique discrete taxonomy.

0.30

Search-by-Answer Bias in Retrieval Models

cs.CL cs.IR ·2026-11-08 ·Adiel Avigad, Shauli Ravfogel, Yoav Goldberg
similarity 0.63
Abstract · 17 pages, 9 figures

Dense retrievers are meant to select passages that match a query's semantic specification. We show they suffer from search-by-answer bias: they favor passages that mention answer objects the model knows (scores as likely), over passages that support valid alternative answers. We test this with controlled factual contrasts from LAMA/TREx. For each retriever, we measure how strongly the model scores the canonical object over an alternative object using a token-prediction head, and ask whether this score predicts passage rankings. Across eight retrievers, including DPR, Spider variants, Contriever, E5, and Qwen3-Embedding, larger token-prediction scores consistently predict higher retrieval scores for passages containing the corresponding object. This holds when passages differ only in the answer object, and even when the passage that mentions the high-scoring object expresses a wrong relation. Dense retrieval thus exhibits search-by-answer bias during passage selection, shaping which evidence is retrieved.

0.17