Localizing Prompt Ambiguity in Large Language Models with Probe-Targeted Attribution

• URL: http://arxiv.org/abs/2606.05486v1
• Date: Wed, 03 Jun 2026 22:15:43 GMT
• Title: Localizing Prompt Ambiguity in Large Language Models with Probe-Targeted Attribution
• Abstract summary: We introduce PRIG, a gradient attribution method that uses a probe logit to attribute latent ambiguity to token positions.<n>Specifically, PRIG trains a linear probe to distinguish clear prompts from ambiguous prompts and attributes the probe score to earlier token representations in the residual stream.<n>In this setting, PRIG localizes ambiguous spans substantially better than gradient attribution baselines, achieving 0.840 AUROC on the combined synthetic benchmark and 0.891 AUROC on the gold set.
• Score: 9.857349182313092
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Prompt ambiguity is a common source of failure in large language models, but is difficult to localize because it is a latent property of the prompt, while existing attribution methods are designed to explain observable outputs such as logits or generated tokens. We introduce PRIG, a gradient attribution method that uses a probe logit to attribute latent ambiguity to token positions. Specifically, PRIG trains a linear probe to distinguish clear prompts from ambiguous prompts and attributes the probe score to earlier token representations in the residual stream. To enable token-level evaluation, we construct synthetic ambiguity datasets across coding, math, and writing by rewriting one task-critical sentence per prompt, and complement them with a human-written gold benchmark. In this setting, PRIG localizes ambiguous spans substantially better than gradient attribution baselines, achieving 0.840 AUROC on the combined synthetic benchmark and 0.891 AUROC on the gold set. It also outperforms GPT-5.4 on sentence-level ambiguity identification and retains useful signal out-of-domain. These results establish PRIG as a practical tool for identifying which parts of a prompt are ambiguous. More broadly, they suggest that latent prompt properties can be localized through intermediate representations, rather than through output-level attribution.

Related papers

• Optimal Watermark Localization in Mixed-Source Large Language Model Texts [41.92651100461471]
  Watermarking provides a principled way to authenticate text generated by large language models (LLMs)<n>In practice, the final text may be mixed-source, with watermark evidence surviving at only a subset of token positions after rewriting, insertion, deletion, or paraphrasing.<n>We formulate localization as a token-level multiple-testing problem based on pivotal statistics, with a latent indicator recording whether watermark dependence survives at each position.
  arXiv  Detail & Related papers  (2026-08-14T21:41:50Z)
• GradCuit: Credit-Assigned Gradient Flow Enables Robust and Interpretable Test-Time Latent Reasoning [61.67411833252235]
  We introduce GradCuit, which inserts optimizable latent states at a selected Transformer layer between the hidden representations of the prompt and the generated continuation.<n>Across five instruction-tuned backbones, three reasoning benchmarks, and two answer formats, GradCuit achieves an average accuracy of 64.5%.<n>For interpretability, token-level gradient attribution reveals that latent influence concentrates on reasoning-connector tokens, while layer analysis identifies early-to-middle Transformer layers as the most effective optimization space.
  arXiv  Detail & Related papers  (2026-08-03T17:55:24Z)
• Detecting LLM-Generated Tokens in Human--LLM Coauthored Text [11.447769496973137]
  Existing methods for detecting LLM-generated text mainly focus on document-level classification and cannot identify which parts of the text are generated by LLMs.<n>This paper introduces a new method to address this urgent need.<n>Our method operates at the token level, the natural unit of modern language models, and builds on existing token-level detection scores.
  arXiv  Detail & Related papers  (2026-07-23T16:01:16Z)
• Human Label Variation as Stable Signal: Learning Annotator-Specific Explanation Behavior via Cross-Annotator Preference Optimization [51.93456979139756]
  Free-text explanations extend human label variation beyond label disagreement.<n>We study whether large language models can learn and reproduce such annotator-specific label-explanation behavior.
  arXiv  Detail & Related papers  (2026-05-27T17:55:00Z)
• Interpretable Discriminative Text Representations via Agreement and Label Disentanglement [4.2788570168072395]
  We propose an operational criterion for interpretable discriminative text representations.<n>Each coordinate should satisfy conceptual clarity, measured by chance of agreement.<n>We instantiate this criterion in LLM-assisted Feature Discovery.
  arXiv  Detail & Related papers  (2026-05-20T04:41:44Z)
• GUIDED: Granular Understanding via Identification, Detection, and Discrimination for Fine-Grained Open-Vocabulary Object Detection [54.19989440021701]
  Fine-grained open-vocabulary object detection (FG-OVD) aims to detect novel object categories described by attribute-rich texts.<n>FG-OVD achieves new state-of-the-art results, demonstrating the benefits of disentangled modeling and modular optimization.
  arXiv  Detail & Related papers  (2026-03-27T22:08:11Z)
• Log Probabilities Are a Reliable Estimate of Semantic Plausibility in Base and Instruction-Tuned Language Models [50.15455336684986]
  We evaluate the effectiveness of LogProbs and basic prompting to measure semantic plausibility. We find that LogProbs offers a more reliable measure of semantic plausibility than direct zero-shot prompting. We conclude that, even in the era of prompt-based evaluations, LogProbs constitute a useful metric of semantic plausibility.
  arXiv  Detail & Related papers  (2024-03-21T22:08:44Z)
• Zero and Few-shot Semantic Parsing with Ambiguous Inputs [45.285508941560295]
  We introduce AmP, a framework, dataset, and challenge for translating ambiguous natural language to formal representations like logic and code. Using AmP, we investigate how several few-shot text-to-code systems handle ambiguity, introducing three new metrics. We find that large pre-trained models perform poorly at capturing the distribution of possible meanings without deliberate instruction.
  arXiv  Detail & Related papers  (2023-06-01T15:46:36Z)
• BERM: Training the Balanced and Extractable Representation for Matching to Improve Generalization Ability of Dense Retrieval [54.66399120084227]
  We propose a novel method to improve the generalization of dense retrieval via capturing matching signal called BERM. Dense retrieval has shown promise in the first-stage retrieval process when trained on in-domain labeled datasets.
  arXiv  Detail & Related papers  (2023-05-18T15:43:09Z)
• Ambiguity-Resistant Semi-Supervised Learning for Dense Object Detection [98.66771688028426]
  We propose a Ambiguity-Resistant Semi-supervised Learning (ARSL) for one-stage detectors. Joint-Confidence Estimation (JCE) is proposed to quantifies the classification and localization quality of pseudo labels. ARSL effectively mitigates the ambiguities and achieves state-of-the-art SSOD performance on MS COCO and PASCAL VOC.
  arXiv  Detail & Related papers  (2023-03-27T07:46:58Z)
• Locally Aggregated Feature Attribution on Natural Language Model Understanding [12.233103741197334]
  Locally Aggregated Feature Attribution (LAFA) is a novel gradient-based feature attribution method for NLP models. Instead of relying on obscure reference tokens, it smooths gradients by aggregating similar reference texts derived from language model embeddings. For evaluation purpose, we also design experiments on different NLP tasks including Entity Recognition and Sentiment Analysis on public datasets.
  arXiv  Detail & Related papers  (2022-04-22T18:59:27Z)
• Exploring the Role of BERT Token Representations to Explain Sentence Probing Results [15.652077779677091]
  We show that BERT tends to encode meaningful knowledge in specific token representations. This allows the model to detect syntactic and semantic abnormalities and to distinctively separate grammatical number and tense subspaces.
  arXiv  Detail & Related papers  (2021-04-03T20:40:42Z)
• Dive into Ambiguity: Latent Distribution Mining and Pairwise Uncertainty Estimation for Facial Expression Recognition [59.52434325897716]
  We propose a solution, named DMUE, to address the problem of annotation ambiguity from two perspectives. For the former, an auxiliary multi-branch learning framework is introduced to better mine and describe the latent distribution in the label space. For the latter, the pairwise relationship of semantic feature between instances are fully exploited to estimate the ambiguity extent in the instance space.
  arXiv  Detail & Related papers  (2021-04-01T03:21:57Z)

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.