TETHER: Training-Free, Elastic, ToC-Anchored Hierarchy Extraction and Recovery in Design-Driven Long Documents
Published: November 11, 2026 — Accepted to EMNLP 2026 (Main)
Authors: Yu Zou, Yan Chen, Jialei Zeng, Xiaorui Zhou, Aixi Zhong, Qi Zhou, Haoran You, Ke Jiang, Changyu Hu
Subjects: Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
Converting complex documents into structured knowledge has become critical infrastructure for deploying Large Language Models (LLMs) in downstream research—and with retrieval-augmented and long-context pipelines now routine in 2026, the demand for reliable document structuring has only intensified. Existing bottom-up paradigms, however, remain blind to the Table of Contents (ToC) as a macro-structure. Relying on localized visual cues, they exhibit heuristic fragility in high-entropy layouts where visual prominence decouples from logical depth. Meanwhile, direct LLM extraction lacks spatial grounding, triggering hallucinations in long contexts.
To address this, we introduce TETHER, a training-free framework that elevates the ToC from discarded noise to a vital structural prior, enabling robust adaptation across design-driven long documents. TETHER operates via a two-stage neuro-symbolic mechanism:
- Skeleton construction — It builds a document skeleton from the raw ToC using a Large Multimodal Model (LMM) empowered by spatio-semantic constraints.
- Anchor-based alignment — It then tethers content blocks to this skeleton, effectively resolving boundary ambiguities.
Results
Evaluated across a multilingual benchmark of ESG reports, product manuals, and annual reports, TETHER delivers highly precise, state-of-the-art (SOTA) zero-shot hierarchy recovery, outperforming deep hierarchical recovery models, dedicated parsers, and general-purpose LMMs. Stress tests further highlight TETHER's resilience to sparse omissions and textual noise, demonstrating that its precise semantic alignment is fundamentally anchored in topological and depth fidelity. Crucially, ablations confirm the framework's architectural irreducibility, while achieving 90% token reduction and drastic speedups over end-to-end baselines—establishing a compelling equilibrium between structural precision and deployment scalability.
via ArXiv CL+LG
