LLMs Encode Relevance as a Layer-Wise Cross-Lingual Signal

• URL: http://arxiv.org/abs/2607.15555v1
• Date: Fri, 17 Jul 2026 01:54:45 GMT
• Title: LLMs Encode Relevance as a Layer-Wise Cross-Lingual Signal
• Abstract summary: We study whether query-document relevance is linearly decodable from residual-stream activations in instruction-tuned language models.<n>Our results suggest that q-d relevance is encoded as a depth-dependent signal.<n>In several models, validation-selected probes match or outperform generated judgments and better preserve system rankings.
• Score: 8.33491432828288
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large language models (LLMs) are increasingly used in information retrieval (IR) pipelines as relevance judges and re-rankers. Yet most analyses remain output-centric, evaluating generated labels or scores while offering limited insight into how relevance is represented inside the model. In this work, we study whether query-document (q-d) relevance is linearly decodable from residual-stream activations in instruction-tuned LLMs, how this signal compares with generated relevance judgments, and whether it transfers across languages. Using the TREC DL20 and MIRACL evaluation collections, we guide medium-scale LLMs (4-9B parameters) with UMBRELA-style relevance judgment prompts, extract last-token activations from every transformer layer, and train linear probes to predict relevance labels. We compare probe predictions with generated judgments and use TREC DL20 to test whether probe-derived pseudo-labels preserve system rankings against human judgments. Our results suggest that q-d relevance is encoded as a depth-dependent signal: probe performance is weak in early layers and strongest in middle-to-late layers, indicating that relevance becomes more linearly accessible after contextual integration. Most importantly, in several models, validation-selected probes match or outperform generated judgments and better preserve system rankings, revealing a separation between internal relevance representation and external expression. Multilingual experiments suggest partial cross-language portability, although transfer remains weaker than within-language decoding. Overall, this work provides a representation-level perspective on LLM-based relevance assessment. Layer-wise probing can help diagnose where relevance emerges, when generated judgments fail to reflect internally available evidence, and how relevance representations vary across languages, datasets, and model families.

Related papers

• RSRank: Learning Relevance from Representational Shifts [2.996596877662786]
  We introduce a lightweight training framework that learns projections RS to calibrated relevance scores.<n>Our training objectives naturally filter irrelevant content at a zero threshold, reducing dependence tuning.
  arXiv  Detail & Related papers  (2026-06-16T03:29:23Z)
• Align Documents to Questions: Question-Oriented Document Rewriting for Retrieval-Augmented Generation [51.55755193937205]
  We propose QREAM, a style-controlled rewriter that aligns retrieved documents with a question-oriented style while preserving facts.<n>Our framework consists of two stages: (1) QREAM-ICL, which uses stylistic seeds to guide iterative rewriting exploration; and (2) QREAM-FT, a lightweight student model distilled from denoised ICL outputs.
  arXiv  Detail & Related papers  (2026-04-19T08:39:21Z)
• The Overlooked Role of Graded Relevance Thresholds in Multilingual Dense Retrieval [7.238707143014165]
  We analyze how graded relevance scores and the threshold used to convert them into binary labels affect dense retrieval.<n>Using a multilingual dataset with LLM-annotated relevance scores, we examine monolingual, multilingual mixture, and cross-lingual retrieval scenarios.
  arXiv  Detail & Related papers  (2026-01-07T21:14:48Z)
• Query-Document Dense Vectors for LLM Relevance Judgment Bias Analysis [4.719505127252616]
  Large Language Models (LLMs) have been used as relevance assessors for Information Retrieval (IR) evaluation collection creation.<n>We aim to understand if LLMs make systematic mistakes when judging relevance, rather than just understanding how good they are on average.<n>We introduce a clustering-based framework that embeds query-document (Q-D) pairs into a joint semantic space.
  arXiv  Detail & Related papers  (2026-01-05T03:02:33Z)
• Can LLM Annotations Replace User Clicks for Learning to Rank? [112.2254432364736]
  Large-scale supervised data is essential for training modern ranking models, but obtaining high-quality human annotations is costly.<n>Click data has been widely used as a low-cost alternative, and with recent advances in large language models (LLMs), LLM-based relevance annotation has emerged as another promising annotation.<n> Experiments on both a public dataset, TianGong-ST, and an industrial dataset, Baidu-Click, show that click-supervised models perform better on high-frequency queries.<n>We explore two training strategies -- data scheduling and frequency-aware multi-objective learning -- that integrate both supervision signals.
  arXiv  Detail & Related papers  (2025-11-10T02:26:14Z)
• When LLMs Disagree: Diagnosing Relevance Filtering Bias and Retrieval Divergence in SDG Search [0.0]
  Large language models (LLMs) are increasingly used to assign document relevance labels in information retrieval pipelines.<n>LLMs often disagree on borderline cases, raising concerns about how such disagreement affects downstream retrieval.<n>We show that model disagreement is systematic, not random.<n>We propose using classification disagreement as an object of analysis in retrieval evaluation, particularly in policy-relevant or thematic search tasks.
  arXiv  Detail & Related papers  (2025-07-02T20:53:51Z)
• Limitations of Automatic Relevance Assessments with Large Language Models for Fair and Reliable Retrieval Evaluation [2.9180406633632523]
  Large language models (LLMs) are gaining much attention as tools for automatic relevance assessment.<n>Recent research has shown that LLM-based assessments yield high systems ranking correlation with human-made judgements.<n>We look at how LLM-generated judgements preserve ranking differences among top-performing systems and also how they preserve pairwise significance evaluation as human judgements.
  arXiv  Detail & Related papers  (2024-11-20T11:19:35Z)
• ELCoRec: Enhance Language Understanding with Co-Propagation of Numerical and Categorical Features for Recommendation [38.64175351885443]
  Large language models have been flourishing in the natural language processing (NLP) domain. Despite the intelligence shown by the recommendation-oriented finetuned models, LLMs struggle to fully understand the user behavior patterns. Existing works only fine-tune a sole LLM on given text data without introducing that important information to it.
  arXiv  Detail & Related papers  (2024-06-27T01:37:57Z)
• DARG: Dynamic Evaluation of Large Language Models via Adaptive Reasoning Graph [70.79413606968814]
  We introduce Dynamic Evaluation of LLMs via Adaptive Reasoning Graph Evolvement (DARG) to dynamically extend current benchmarks with controlled complexity and diversity. Specifically, we first extract the reasoning graphs of data points in current benchmarks and then perturb the reasoning graphs to generate novel testing data. Such newly generated test samples can have different levels of complexity while maintaining linguistic diversity similar to the original benchmarks.
  arXiv  Detail & Related papers  (2024-06-25T04:27:53Z)
• Evaluating Generative Language Models in Information Extraction as Subjective Question Correction [49.729908337372436]
  We propose a new evaluation method, SQC-Score. Inspired by the principles in subjective question correction, we propose a new evaluation method, SQC-Score. Results on three information extraction tasks show that SQC-Score is more preferred by human annotators than the baseline metrics.
  arXiv  Detail & Related papers  (2024-04-04T15:36:53Z)
• "Knowing When You Don't Know": A Multilingual Relevance Assessment Dataset for Robust Retrieval-Augmented Generation [90.09260023184932]
  Retrieval-Augmented Generation (RAG) grounds Large Language Model (LLM) output by leveraging external knowledge sources to reduce factual hallucinations. NoMIRACL is a human-annotated dataset for evaluating LLM robustness in RAG across 18 typologically diverse languages. We measure relevance assessment using: (i) hallucination rate, measuring model tendency to hallucinate, when the answer is not present in passages in the non-relevant subset, and (ii) error rate, measuring model inaccuracy to recognize relevant passages in the relevant subset.
  arXiv  Detail & Related papers  (2023-12-18T17:18:04Z)
• Using Natural Language Explanations to Rescale Human Judgments [81.66697572357477]
  We propose a method to rescale ordinal annotations and explanations using large language models (LLMs)<n>We feed annotators' Likert ratings and corresponding explanations into an LLM and prompt it to produce a numeric score anchored in a scoring rubric.<n>Our method rescales the raw judgments without impacting agreement and brings the scores closer to human judgments grounded in the same scoring rubric.
  arXiv  Detail & Related papers  (2023-05-24T06:19:14Z)
• UnifieR: A Unified Retriever for Large-Scale Retrieval [84.61239936314597]
  Large-scale retrieval is to recall relevant documents from a huge collection given a query. Recent retrieval methods based on pre-trained language models (PLM) can be coarsely categorized into either dense-vector or lexicon-based paradigms. We propose a new learning framework, UnifieR which unifies dense-vector and lexicon-based retrieval in one model with a dual-representing capability.
  arXiv  Detail & Related papers  (2022-05-23T11:01:59Z)

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.