Hessian-Enhanced Token Attribution (HETA): Interpreting Autoregressive LLMs

• URL: http://arxiv.org/abs/2604.13258v1
• Date: Tue, 14 Apr 2026 19:43:43 GMT
• Title: Hessian-Enhanced Token Attribution (HETA): Interpreting Autoregressive LLMs
• Abstract summary: Hessian-Enhanced Token Attribution (HETA) is a novel attribution framework tailored for decoder-only language models.<n>HETA combines three complementary components: a semantic transition vector that captures token-to-token influence across layers, Hessian-based sensitivity scores that model second-order effects, and KL divergence to measure information loss when tokens are masked.<n> Empirical evaluations across multiple models and datasets demonstrate that HETA consistently outperforms existing methods in attribution faithfulness and alignment with human annotations.
• Score: 10.87622116540203
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Attribution methods seek to explain language model predictions by quantifying the contribution of input tokens to generated outputs. However, most existing techniques are designed for encoder-based architectures and rely on linear approximations that fail to capture the causal and semantic complexities of autoregressive generation in decoder-only models. To address these limitations, we propose Hessian-Enhanced Token Attribution (HETA), a novel attribution framework tailored for decoder-only language models. HETA combines three complementary components: a semantic transition vector that captures token-to-token influence across layers, Hessian-based sensitivity scores that model second-order effects, and KL divergence to measure information loss when tokens are masked. This unified design produces context-aware, causally faithful, and semantically grounded attributions. Additionally, we introduce a curated benchmark dataset for systematically evaluating attribution quality in generative settings. Empirical evaluations across multiple models and datasets demonstrate that HETA consistently outperforms existing methods in attribution faithfulness and alignment with human annotations, establishing a new standard for interpretability in autoregressive language models.

Related papers

• Continuous Autoregressive Language Models [56.49239051750678]
  We introduce Continuous Autoregressive Language Models (CALM)<n>CALM uses a high-fidelity autoencoder to compress a chunk of K tokens into a single continuous vector.<n>We develop a comprehensive likelihood-free framework that enables robust training, evaluation, and controllable sampling.
  arXiv  Detail & Related papers  (2025-10-31T17:58:11Z)
• Negative Matters: Multi-Granularity Hard-Negative Synthesis and Anchor-Token-Aware Pooling for Enhanced Text Embeddings [25.565372681837697]
  We introduce a Multi-Granularity Hard-negative (MGH) synthesis framework to generate diverse negative samples with varying levels of similarity with the query.<n>We also propose an Anchor Token Aware (ATA) pooling method to improve text embedding accuracy without increasing model complexity.
  arXiv  Detail & Related papers  (2025-08-31T13:24:48Z)
• SEED: A Structural Encoder for Embedding-Driven Decoding in Time Series Prediction with LLMs [3.036179638516407]
  We introduce SEED, a structural encoder for embedding-driven decoding, which integrates four stages: a token-aware encoder for patch extraction, a projection module that aligns patches with language model embeddings, and a semantic reprogramming mechanism that maps patches to task-aware prototypes.<n>This modular architecture decouples representation learning from inference, enabling efficient alignment between numerical patterns and semantic reasoning.
  arXiv  Detail & Related papers  (2025-06-25T06:40:14Z)
• Refining Sentence Embedding Model through Ranking Sentences Generation with Large Language Models [60.00178316095646]
  Sentence embedding is essential for many NLP tasks, with contrastive learning methods achieving strong performance using datasets like NLI.<n>Recent studies leverage large language models (LLMs) to generate sentence pairs, reducing annotation dependency.<n>We propose a method for controlling the generation direction of LLMs in the latent space. Unlike unconstrained generation, the controlled approach ensures meaningful semantic divergence.<n> Experiments on multiple benchmarks demonstrate that our method achieves new SOTA performance with a modest cost in ranking sentence synthesis.
  arXiv  Detail & Related papers  (2025-02-19T12:07:53Z)
• FELLE: Autoregressive Speech Synthesis with Token-Wise Coarse-to-Fine Flow Matching [56.30231216917128]
  FELLE is an autoregressive model that integrates language modeling with token-wise flow matching.<n>For each continuous-valued token, FELLE modifies the general prior distribution in flow matching by incorporating information from the previous step.<n>FELLE generates continuous-valued tokens hierarchically, conditioned on the language model's output.
  arXiv  Detail & Related papers  (2025-02-16T13:54:32Z)
• NormXLogit: The Head-on-Top Never Lies [15.215985417763472]
  We propose a novel technique for assessing the significance of individual input tokens.<n>This method operates based on the input and output representations associated with each token.<n>We reveal a significant relationship between a token's importance and the extent to which its representation can resemble the model's final prediction.
  arXiv  Detail & Related papers  (2024-11-25T10:12:27Z)
• Interpret the Internal States of Recommendation Model with Sparse Autoencoder [28.234859617081295]
  RecSAE is an automated and generalizable probing framework that interprets Recommenders with Sparse AutoEncoder.<n>It extracts interpretable latents from the internal states of recommendation models and links them to semantic concepts for interpretation.<n> RecSAE does not alter original models during interpretation and also enables targeted de-biasing to models based on interpreted results.
  arXiv  Detail & Related papers  (2024-11-09T08:22:31Z)
• Reconsidering Degeneration of Token Embeddings with Definitions for Encoder-based Pre-trained Language Models [20.107727903240065]
  We propose DefinitionEMB to re-construct isotropically distributed and semantics-related token embeddings for encoder-based language models. Our experiments demonstrate the effectiveness of leveraging definitions from Wiktionary to re-construct such embeddings.
  arXiv  Detail & Related papers  (2024-08-02T15:00:05Z)
• Scalable Learning of Latent Language Structure With Logical Offline Cycle Consistency [71.42261918225773]
  Conceptually, LOCCO can be viewed as a form of self-learning where the semantic being trained is used to generate annotations for unlabeled text. As an added bonus, the annotations produced by LOCCO can be trivially repurposed to train a neural text generation model.
  arXiv  Detail & Related papers  (2023-05-31T16:47:20Z)
• Disentanglement via Latent Quantization [60.37109712033694]
  In this work, we construct an inductive bias towards encoding to and decoding from an organized latent space. We demonstrate the broad applicability of this approach by adding it to both basic data-re (vanilla autoencoder) and latent-reconstructing (InfoGAN) generative models.
  arXiv  Detail & Related papers  (2023-05-28T06:30:29Z)
• Improve Variational Autoencoder for Text Generationwith Discrete Latent Bottleneck [52.08901549360262]
  Variational autoencoders (VAEs) are essential tools in end-to-end representation learning. VAEs tend to ignore latent variables with a strong auto-regressive decoder. We propose a principled approach to enforce an implicit latent feature matching in a more compact latent space.
  arXiv  Detail & Related papers  (2020-04-22T14:41:37Z)

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.