@inproceedings{ding-etal-2026-hierarchical,
    title = "Hierarchical Token Prepending: Enhancing Information Flow in Decoder-based {LLM} Embeddings",
    author = "Ding, Xueying  and
      Huang, Xingyue  and
      Ju, Mingxuan  and
      Collins, Liam  and
      Liu, Yozen  and
      Akoglu, Leman  and
      Shah, Neil  and
      Zhao, Tong",
    editor = "Liakata, Maria  and
      Moreira, Viviane P.  and
      Zhang, Jiajun  and
      Jurgens, David",
    booktitle = "Proceedings of the 64th Annual Meeting of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl/2026.acl-long.870/",
    pages = "19047--19066",
    ISBN = "979-8-89176-390-6",
    abstract = "Large language models produce powerful text embeddings, but their causal attention mechanism restricts the flow of information from later to earlier tokens, degrading representation quality. While recent methods attempt to solve this by prepending a single summary token, they over-compress information, hence harming performance on long documents. We propose \textit{Hierarchical Token Prepending} (HTP), a method that resolves two critical bottlenecks. To mitigate attention-level compression, HTP partitions the input into blocks and prepends block-level summary tokens to subsequent blocks, creating multiple pathways for backward information flow. To address readout-level over-squashing, we replace last-token pooling with mean-pooling, a choice supported by theoretical analysis. HTP achieves consistent performance gains across 11 retrieval datasets and 30 general embedding benchmarks, especially in long-context settings. As a simple, architecture-agnostic method, HTP effectively enhances both zero-shot and fine-tuned models, offering a scalable route to superior long-document embeddings."
}