mlsanityData to 7 Oct 2026

Lost in Decoding? Reproducing and Stress-Testing the Look-Ahead Prior in Generative Retrieval

arXiv 2604.23396 ↗First posted 25 Apr 2026cs.IR cs.AI cs.CLcited by 0

Not on the map. Why?

Abstract (arXiv)

Generative retrieval (GR) ranks documents by autoregressively generating document identifiers. Because many GR methods rely on trie-constrained beam search, they are vulnerable to early pruning of relevant prefixes under finite-beam decoding. Planning Ahead in Generative Retrieval (PAG) mitigates this failure mode by using simultaneous decoding to compute a document-level look-ahead prior that guides subsequent sequential decoding. We reproduce PAG at inference time and stress-test its decoding behavior. Using the authors' released checkpoint and identifier/trie artifacts under the reported decoding setup, we reproduce the main effectiveness results on MS MARCO Dev and TREC-DL 2019/2020, and corroborate the reported beam-size-latency trade-off in our hardware setting. Beyond reproduction, we introduce plan drift diagnostics that quantify how intent-preserving query variations alter the planner's top-n candidate set and highest-weight planner tokens, and how these changes affect guided decoding. We find that PAG's planning signal is brittle under lexical surface-form variation: intent-preserving typos can trigger plan collapse, where the planned candidate pool shifts enough that the look-ahead bonus provides little useful guidance, effectively reverting decoding toward weaker unguided search. We further evaluate fixed-index cross-lingual robustness using non-English mMARCO queries against an English index, and assess query-side mitigation strategies that require no re-indexing; query translation provides the strongest recovery in our setting. Overall, our results confirm PAG's reported effectiveness and the benefit of planning-guided decoding under the released inference setup, while showing that these gains depend on the stability of the planning signal under realistic query variation and query-document mismatch.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 28 of its 40 references that are in the map. First posted on arXiv 25 Apr 2026.

Builds on

28 of 40 references.

References · 28

  1. 2025Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Modelscited by 1,175
  2. 2020Beyond English-Centric Multilingual Machine Translationcited by 559
  3. 2022Transformer Memory as a Differentiable Search Indexcited by 242
  4. 2021NeuroLogic A*esque Decoding: Constrained Text Generation with Lookahead Heuristicscited by 153
  5. 2025EmbeddingGemma: Powerful and Lightweight Text Representationscited by 128
  6. 2021mMARCO: A Multilingual Version of the MS MARCO Passage Ranking Datasetcited by 116
  7. 2019On NMT Search Errors and Model Errors: Cat Got Your Tongue?cited by 114
  8. 2022Autoregressive Search Engines: Generating Substrings as Document Identifierscited by 112
  9. 2024From Matching to Generation: A Survey on Generative Information Retrievalcited by 89
  10. 2023How Does Generative Retrieval Scale to Millions of Passages?cited by 47
  11. 2023Learning to Rank in Generative Retrievalcited by 46
  12. 2022Transfer Learning Approaches for Building Cross-Language Dense Retrieval Modelscited by 36
16 more
  1. 2021Evaluating the Robustness of Retrieval Pipelines with Query Variation Generatorscited by 33
  2. 2024RetroLLM: Empowering Large Language Models to Retrieve Fine-grained Evidence within Generationcited by 20
  3. 2024Planning Ahead in Generative Retrieval: Guiding Autoregressive Generation through Simultaneous Decodingcited by 19
  4. 2024Robust Neural Information Retrieval: An Adversarial and Out-of-distribution Perspectivecited by 15
  5. 2020ProphetNet-Ads: A Looking Ahead Strategy for Generative Retrieval Models in Sponsored Search Enginecited by 11
  6. 2024Listwise Generative Retrieval Models via a Sequential Learning Processcited by 11
  7. 2025Unsupervised Corpus Poisoning Attacks in Continuous Space for Dense Retrievalcited by 11
  8. 2023A Study on FGSM Adversarial Training for Neural Retrievalcited by 10
  9. 2025Lightweight and Direct Document Relevance Optimization for Generative Information Retrievalcited by 9
  10. 2025Reproducing HotFlip for Corpus Poisoning Attacks in Dense Retrievalcited by 7
  11. 2024Constrained Decoding with Speculative Lookaheadscited by 6
  12. 2025Constrained Auto-Regressive Decoding Constrains Generative Retrievalcited by 5
  13. 2023Unlocking Anticipatory Text Generation: A Constrained Approach for Large Language Models Decodingcited by 4
  14. 2025Multilingual Generative Retrieval via Cross-lingual Semantic Compressioncited by 4
  15. 2024On the Robustness of Generative Information Retrieval Modelscited by 3
  16. 2026Multi-Step Semantic Reasoning in Generative Retrievalcited by 1

Built on by

Not cited yet.