pith. sign in

arxiv: 2606.00432 · v1 · pith:3TDQQXNBnew · submitted 2026-05-29 · 💻 cs.LG

Grounded Decoding: Retrieval-Anchored Probability Fusion for Faithful RAG

Ibne Farabi Shihab , Fariya Afrin , Sanjeda Akter , Anuj Sharma This is my paper

Pith reviewed 2026-06-28 22:42 UTC · model grok-4.3

classification 💻 cs.LG
keywords retrieval-augmented generationfactual consistencydecoding strategyKL barycenterprobability fusionRAGgrounded generation
0
0 comments X

The pith

Grounded Decoding obtains the next-token distribution as the unique solution to a KL-barycenter objective that geometrically fuses the full RAG distribution with a retrieval-only distribution.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper constructs two distributions at every generation step, one conditioned on the full RAG prompt and one conditioned only on retrieved evidence, then derives their fusion as the minimizer of a weighted sum of KL divergences. This yields a normalized geometric mean that recovers standard RAG when the grounding weight is zero and shifts probability mass toward evidence as the weight increases. An adaptive scheme further modulates the weight according to distributional disagreement and retriever confidence. The approach is training-free and is shown to raise factual accuracy and citation quality on ALCE, Natural Questions, and FActScore while preserving fluency. A sympathetic reader would care because it offers a lightweight way to counteract the tendency of LLMs to favor parametric knowledge over retrieved facts.

Core claim

The central claim is that the optimal next-token distribution at each step is the unique minimizer of a KL-barycenter objective over the probability simplex, which produces the normalized geometric fusion of the full RAG distribution and the retrieval-only distribution. This formulation recovers standard RAG decoding exactly when the grounding weight is zero and smoothly increases the influence of retrieved evidence as the weight grows. A conflict-aware adaptive weighting scheme is introduced that adjusts the grounding strength dynamically on the basis of distributional disagreement and retriever confidence.

What carries the argument

KL-barycenter objective over the probability simplex, which yields the normalized geometric fusion of the full RAG distribution and the retrieval-only distribution

If this is right

  • Standard RAG decoding is recovered exactly when the grounding weight is set to zero.
  • Factual accuracy and citation quality rise on the ALCE, Natural Questions, and FActScore benchmarks.
  • Fluency metrics remain comparable to standard RAG and other decoding baselines.
  • The adaptive weighting scheme modulates grounding strength according to distributional disagreement and retriever confidence.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same barycenter construction could be applied when multiple retrieval sources are available by including additional distributions in the objective.
  • If retrieval contains systematic errors, the adaptive scheme would need to detect and down-weight those cases to avoid amplifying mistakes.
  • Probability-level fusion offers a potential lightweight substitute for logit-level interventions in other constrained decoding settings.

Load-bearing premise

The retrieval-only distribution acts as a faithful factual anchor whose geometric fusion with the full RAG distribution increases consistency without introducing new factual errors or harming fluency.

What would settle it

A controlled test on inputs where the retrieved documents contain clear factual errors, measuring whether the fused outputs produce more factual mistakes than standard RAG outputs.

Figures

Figures reproduced from arXiv: 2606.00432 by Anuj Sharma, Fariya Afrin, Ibne Farabi Shihab, Sanjeda Akter.

Figure 1
Figure 1. Figure 1: Grounded Decoding maintains two matched decoding streams. The full stream conditions on the query and [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Computation of the adaptive grounding weight [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Prompt templates used for the full stream and retrieval stream. Both streams share identical system message, [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Geometric interpretation of Grounded Decoding as KL-barycenter probability fusion between the full [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Geometric interpretation of Grounded Decoding on the probability simplex. Each vertex corresponds to a [PITH_FULL_IMAGE:figures/full_fig_p015_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of factuality and QA performance [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Citation quality evaluation across different [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 12
Figure 12. Figure 12: Token-level adaptive grounding dynamics for a representative FActScore example. The grounding strength ηt remains low for function words such as “The” and “was”, while it increases sharply at factual spans in￾cluding “novelist”, “1947”, and “Trieste”. This behavior indicates that adaptive grounding selectively activates on content-bearing tokens that are more prone to factual errors, while remaining inact… view at source ↗
Figure 13
Figure 13. Figure 13: Performance–Latency Comparison across decoding methods. We compare Standard RAG, CAD, AdaCAD, and Grounded Decoding in terms of decoding latency (ms/token) versus FActScore. While CAD and AdaCAD incur higher latency without substantial gains in factual accuracy, Grounded Decoding achieves the best trade-off, significantly improving FActScore while maintaining comparable computational cost to other advance… view at source ↗
Figure 14
Figure 14. Figure 14: Token-level grounding analysis of adaptive decoding. (a) The heatmap shows token-wise grounding [PITH_FULL_IMAGE:figures/full_fig_p018_14.png] view at source ↗
read the original abstract

As retrieval-augmented generation (RAG) systems scale, it becomes increasingly challenging to ensure faithful grounding in external evidence. Large language models may still prioritize parametric knowledge over retrieved information when conflicts arise. We propose a novel training-free decoding framework, \emph{Grounded Decoding}, designed to improve factual consistency in RAG without modifying model parameters. Unlike standard approaches that rely on a single conditional distribution, our method constructs two matched-prompt distributions at every generation step: (1) a full RAG distribution conditioned on the query, retrieved documents, and generated prefix, and (2) a retrieval-only distribution conditioned solely on retrieved evidence and the same prefix. The final next-token distribution is derived as the unique solution to a KL-barycenter objective over the probability simplex, yielding a normalized geometric fusion of the two distributions.This formulation naturally recovers standard RAG when the grounding weight is zero and smoothly shifts probability mass toward retrieved evidence as grounding strength increases. We further introduce a conflict-aware adaptive weighting scheme that dynamically adjusts grounding based on distributional disagreement and retriever confidence. Experiments on ALCE, Natural Questions, and FActScore demonstrate consistent improvements in factual accuracy and citation quality over standard RAG and competitive decoding-time baselines, while maintaining fluency. Our results indicate that probability-level fusion provides a strong and efficient alternative to logit-level intervention methods for faithful RAG decoding.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 3 minor

Summary. The manuscript proposes Grounded Decoding, a training-free decoding framework for retrieval-augmented generation. At each step it constructs a full RAG distribution (conditioned on query, documents, and prefix) and a retrieval-only distribution (conditioned only on evidence and prefix), then takes their normalized geometric fusion as the unique minimizer of a weighted KL-barycenter objective over the simplex. An adaptive weighting scheme adjusts the grounding strength according to distributional disagreement and retriever . Experiments on ALCE, Natural Questions, and FActScore report gains in factual accuracy and citation quality while preserving fluency; the method recovers standard RAG when the grounding weight is zero.

Significance. If the claimed uniqueness and empirical improvements hold, the work supplies a simple, training-free probability-level fusion technique that anchors generation to retrieved evidence via the geometric-mean property of the KL barycenter. The fact that the formulation is parameter-light, recovers the baseline, and is supported by standard convex optimization arguments is a strength; reproducible gains on established factual-consistency benchmarks would make the method a practical addition to decoding-time interventions for RAG.

minor comments (3)
  1. [Abstract] The abstract asserts that the fused distribution is the 'unique solution' to the KL-barycenter objective and yields a normalized geometric fusion, yet supplies neither a short derivation nor a reference to the standard variational characterization (min_R ∑ w_i KL(R || P_i) yields R ∝ ∏ P_i^{w_i}). Adding one paragraph or an appendix equation would make the central claim self-contained.
  2. No ablation isolates the contribution of the barycenter fusion itself from the adaptive weighting scheme. Reporting results with fixed versus adaptive weights (and with the retrieval-only distribution ablated) would clarify which component drives the reported gains on ALCE, NQ, and FActScore.
  3. The adaptive weighting depends on distributional disagreement and retriever scores; a brief discussion of how these quantities are computed and any sensitivity to their estimation would improve reproducibility.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their thoughtful summary and positive evaluation of Grounded Decoding. We are pleased that the significance of the training-free KL-barycenter fusion, its recovery of standard RAG, and the reported gains on factual-consistency benchmarks were recognized. The recommendation of minor revision is appreciated; we will incorporate any editorial or presentational suggestions in the revised manuscript.

Circularity Check

0 steps flagged

No significant circularity; derivation is standard math

full rationale

The central claim—that the fused distribution is the unique minimizer of the weighted KL-barycenter objective and equals the normalized geometric mean—follows directly from the known variational characterization of the right KL barycenter (min_R ∑ w_i KL(R || P_i) yields R ∝ ∏ P_i^{w_i} by strict convexity). This is external mathematics, not a self-definition, fitted input renamed as prediction, or self-citation chain. The adaptive weighting scheme is a separate heuristic using disagreement and retriever scores; it does not make the fusion equation reduce to its inputs by construction. Experiments on ALCE/NQ/FActScore provide external validation rather than tautological confirmation. No load-bearing step matches any enumerated circularity pattern.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The central claim rests on the existence of a unique KL-barycenter solution and on the premise that the retrieval-only distribution supplies an independent factual signal; one explicit free parameter (grounding weight) and an adaptive weighting rule whose internal coefficients are unspecified are introduced to control the fusion.

free parameters (2)
  • grounding weight
    Scalar that interpolates between standard RAG (weight zero) and stronger retrieval anchoring; its value is chosen per experiment or via the adaptive rule.
  • adaptive weighting coefficients
    Parameters inside the conflict-aware rule that map distributional disagreement and retriever confidence into the effective grounding strength.
axioms (1)
  • standard math Existence and uniqueness of the solution to the KL-barycenter objective on the probability simplex
    Invoked to guarantee that the fused distribution is well-defined and unique.

pith-pipeline@v0.9.1-grok · 5784 in / 1519 out tokens · 25798 ms · 2026-06-28T22:42:18.022321+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

57 extracted references · 12 canonical work pages · 1 internal anchor

  1. [1]

    Retrieval-Augmented Generation for Knowledge-Intensive

    Lewis, Patrick and Perez, Ethan and Piktus, Aleksandra and Petroni, Fabio and Karpukhin, Vladimir and Goyal, Naman and K. Retrieval-Augmented Generation for Knowledge-Intensive. Advances in Neural Information Processing Systems , volume =

  2. [2]

    2020 , publisher =

    Guu, Kelvin and Lee, Kenton and Tung, Zora and Pasupat, Panupong and Chang, Ming-Wei , booktitle =. 2020 , publisher =

    2020

  3. [3]

    Proceedings of the 39th International Conference on Machine Learning , pages =

    Improving Language Models by Retrieving from Trillions of Tokens , author =. Proceedings of the 39th International Conference on Machine Learning , pages =. 2022 , publisher =

    2022

  4. [4]

    ACM Computing Surveys , volume =

    Survey of Hallucination in Natural Language Generation , author =. ACM Computing Surveys , volume =. 2023 , doi =

    2023

  5. [5]

    ACM Transactions on Information Systems , volume =

    A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions , author =. ACM Transactions on Information Systems , volume =. 2025 , doi =

    2025

  6. [6]

    Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing , pages =

    Enabling Large Language Models to Generate Text with Citations , author =. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing , pages =. 2023 , publisher =

    2023

  7. [7]

    2023 , publisher =

    Min, Sewon and Krishna, Kalpesh and Lyu, Xinxi and Lewis, Mike and Yih, Wen-tau and Koh, Pang Wei and Iyyer, Mohit and Zettlemoyer, Luke and Hajishirzi, Hannaneh , booktitle =. 2023 , publisher =

    2023

  8. [8]

    Transactions of the Association for Computational Linguistics , volume =

    Natural Questions: A Benchmark for Question Answering Research , author =. Transactions of the Association for Computational Linguistics , volume =. 2019 , doi =

    2019

  9. [9]

    Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 2: Short Papers) , pages =

    Trusting Your Evidence: Hallucinate Less with Context-aware Decoding , author =. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 2: Short Papers) , pages =. 2024 , publisher =

    2024

  10. [10]

    2025 , publisher =

    Wang, Han and Prasad, Archiki and Stengel-Eskin, Elias and Bansal, Mohit , booktitle =. 2025 , publisher =

    2025

  11. [11]

    Findings of the Association for Computational Linguistics: ACL 2024 , pages =

    Discerning and Resolving Knowledge Conflicts through Adaptive Decoding with Contextual Information-Entropy Constraint , author =. Findings of the Association for Computational Linguistics: ACL 2024 , pages =. 2024 , publisher =

    2024

  12. [12]

    2025 , publisher =

    Khandelwal, Anant and Gupta, Manish and Agrawal, Puneet , booktitle =. 2025 , publisher =

    2025

  13. [13]

    Decoding as Optimisation on the Probability Simplex: From Top-

    Ji, Xiaotong and Tutunov, Rasul and Zimmer, Matthieu and Bou-Ammar, Haitham , journal =. Decoding as Optimisation on the Probability Simplex: From Top-

  14. [14]

    Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages =

    Contrastive Decoding: Open-ended Text Generation as Optimization , author =. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages =. 2023 , publisher =

    2023

  15. [15]

    Chuang, Yung-Sung and Xie, Yujia and Luo, Hongyin and Kim, Yoon and Glass, James and He, Pengcheng , booktitle =

  16. [16]

    International Conference on Learning Representations , year =

    Generalization through Memorization: Nearest Neighbor Language Models , author =. International Conference on Learning Representations , year =

  17. [17]

    International Conference on Learning Representations , year =

    The Curious Case of Neural Text Degeneration , author =. International Conference on Learning Representations , year =

  18. [18]

    Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages =

    Hierarchical Neural Story Generation , author =. Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages =. 2018 , publisher =

    2018

  19. [19]

    Turning Up the Heat: Min-

    Nguyen, Minh and Baker, Andrew and Neo, Clement and Roush, Allen and Kirsch, Andreas and Shwartz-Ziv, Ravid , booktitle =. Turning Up the Heat: Min-. 2025 , note =

    2025

  20. [20]

    Advances in Neural Information Processing Systems , volume =

    Training Language Models to Follow Instructions with Human Feedback , author =. Advances in Neural Information Processing Systems , volume =

  21. [21]

    Advances in Neural Information Processing Systems , volume =

    Direct Preference Optimization: Your Language Model is Secretly a Reward Model , author =. Advances in Neural Information Processing Systems , volume =

  22. [22]

    Transactions on Machine Learning Research , year =

    Unsupervised Dense Information Retrieval with Contrastive Learning , author =. Transactions on Machine Learning Research , year =

  23. [23]

    Asai, Akari and Wu, Zeqiu and Wang, Yizhong and Sil, Avirup and Hajishirzi, Hannaneh , booktitle =. Self-. 2024 , url =

    2024

  24. [24]

    Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing , pages =

    Entity-Based Knowledge Conflicts in Question Answering , author =. Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing , pages =. 2021 , publisher =

    2021

  25. [25]

    2024 , howpublished =

    Meta. 2024 , howpublished =

    2024

  26. [26]

    Jiang, Albert Q. and Sablayrolles, Alexandre and Mensch, Arthur and Bamford, Chris and Chaplot, Devendra Singh and de las Casas, Diego and Bressand, Florian and Lengyel, Gianna and Lample, Guillaume and Saulnier, Lucile and Lavaud, L. arXiv preprint arXiv:2310.06825 , year =

    Pith/arXiv arXiv

  27. [27]

    2021 , publisher =

    Yang, Kevin and Klein, Dan , booktitle =. 2021 , publisher =

    2021

  28. [28]

    International Conference on Learning Representations , year =

    Plug and Play Language Models: A Simple Approach to Controlled Text Generation , author =. International Conference on Learning Representations , year =

  29. [29]

    2021 , publisher =

    Krause, Ben and Gotmare, Akhilesh Deepak and McCann, Bryan and Keskar, Nitish Shirish and Joty, Shafiq and Socher, Richard and Rajani, Nazneen Fatema , booktitle =. 2021 , publisher =

    2021

  30. [30]

    Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi. 2024. https://arxiv.org/abs/2310.11511 Self- RAG : Learning to retrieve, generate, and critique through self-reflection . In International Conference on Learning Representations

    Pith/arXiv arXiv 2024

  31. [31]

    Sebastian Borgeaud, Arthur Mensch, Jordan Hoffmann, Trevor Cai, Eliza Rutherford, Katie Millican, George van den Driessche, Jean-Baptiste Lespiau, Bogdan Damoc, Aidan Clark, Diego de Las Casas, Aurelia Guy, Jacob Menick, Roman Ring, Tom Hennigan, Saffron Huang, Loren Maggiore, Chris Jones, Albin Cassirer, and 9 others. 2022. https://arxiv.org/abs/2112.044...

    Pith/arXiv arXiv 2022

  32. [32]

    Yung-Sung Chuang, Yujia Xie, Hongyin Luo, Yoon Kim, James Glass, and Pengcheng He. 2024. DoLa : Decoding by contrasting layers improves factuality in large language models. In International Conference on Learning Representations

    2024

  33. [33]

    Sumanth Dathathri, Andrea Madotto, Janice Lan, Jane Hung, Eric Frank, Piero Molino, Jason Yosinski, and Rosanne Liu. 2020. Plug and play language models: A simple approach to controlled text generation. In International Conference on Learning Representations

    2020

  34. [34]

    Angela Fan, Mike Lewis, and Yann Dauphin. 2018. https://doi.org/10.18653/v1/P18-1082 Hierarchical neural story generation . In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 889--898. Association for Computational Linguistics

    work page doi:10.18653/v1/p18-1082 2018

  35. [35]

    Tianyu Gao, Howard Yen, Jiatong Yu, and Danqi Chen. 2023. https://doi.org/10.18653/v1/2023.emnlp-main.398 Enabling large language models to generate text with citations . In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 6465--6488. Association for Computational Linguistics

    work page doi:10.18653/v1/2023.emnlp-main.398 2023

  36. [36]

    Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat, and Ming-Wei Chang. 2020. REALM : Retrieval-augmented language model pre-training. In Proceedings of the 37th International Conference on Machine Learning, pages 3929--3938. PMLR

    2020

  37. [37]

    Ari Holtzman, Jan Buys, Li Du, Maxwell Forbes, and Yejin Choi. 2020. The curious case of neural text degeneration. In International Conference on Learning Representations

    2020

  38. [38]

    Lei Huang, Weijiang Yu, Weitao Ma, Weihong Zhong, Zhangyin Feng, Haotian Wang, Qianglong Chen, Weihua Peng, Xiaocheng Feng, Bing Qin, and Ting Liu. 2025. https://doi.org/10.1145/3703155 A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions . ACM Transactions on Information Systems, 43(2):1--55. Also avail...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1145/3703155 2025

  39. [39]

    Gautier Izacard, Mathilde Caron, Lucas Hosseini, Sebastian Riedel, Piotr Bojanowski, Armand Joulin, and Edouard Grave. 2022. Unsupervised dense information retrieval with contrastive learning. Transactions on Machine Learning Research

    2022

  40. [40]

    Xiaotong Ji, Rasul Tutunov, Matthieu Zimmer, and Haitham Bou-Ammar. 2026. Decoding as optimisation on the probability simplex: From top- K to top- P (nucleus) to best-of- K samplers. arXiv preprint arXiv:2602.18292

    arXiv 2026

  41. [41]

    Ziwei Ji, Nayeon Lee, Rita Frieske, Tiezheng Yu, Dan Su, Yan Xu, Etsuko Ishii, Ye Jin Bang, Andrea Madotto, and Pascale Fung. 2023. https://doi.org/10.1145/3571730 Survey of hallucination in natural language generation . ACM Computing Surveys, 55(12):1--38

    work page doi:10.1145/3571730 2023

  42. [42]

    Anant Khandelwal, Manish Gupta, and Puneet Agrawal. 2025. CoCoA : Confidence- and context-aware adaptive decoding for resolving knowledge conflicts in large language models. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 6835--6855. Association for Computational Linguistics

    2025

  43. [43]

    Urvashi Khandelwal, Omer Levy, Dan Jurafsky, Luke Zettlemoyer, and Mike Lewis. 2020. Generalization through memorization: Nearest neighbor language models. In International Conference on Learning Representations

    2020

  44. [44]

    Ben Krause, Akhilesh Deepak Gotmare, Bryan McCann, Nitish Shirish Keskar, Shafiq Joty, Richard Socher, and Nazneen Fatema Rajani. 2021. https://doi.org/10.18653/v1/2021.findings-emnlp.424 GeDi : Generative discriminator guided sequence generation . In Findings of the Association for Computational Linguistics: EMNLP 2021, pages 4929--4952. Association for ...

    work page doi:10.18653/v1/2021.findings-emnlp.424 2021

  45. [45]

    Dai, Jakob Uszkoreit, Quoc Le, and Slav Petrov

    Tom Kwiatkowski, Jennimaria Palomaki, Olivia Redfield, Michael Collins, Ankur Parikh, Chris Alberti, Danielle Epstein, Illia Polosukhin, Jacob Devlin, Kenton Lee, Kristina Toutanova, Llion Jones, Matthew Kelcey, Ming-Wei Chang, Andrew M. Dai, Jakob Uszkoreit, Quoc Le, and Slav Petrov. 2019. https://doi.org/10.1162/tacl_a_00276 Natural questions: A benchma...

    work page doi:10.1162/tacl_a_00276 2019

  46. [46]

    u ttler, Mike Lewis, Wen-tau Yih, Tim Rockt \

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich K \"u ttler, Mike Lewis, Wen-tau Yih, Tim Rockt \"a schel, Sebastian Riedel, and Douwe Kiela. 2020. Retrieval-augmented generation for knowledge-intensive NLP tasks. In Advances in Neural Information Processing Systems, volume 33, pages 9459--9474

    2020

  47. [47]

    doi: 10.18653/v1/2023.acl-long.687

    Xiang Lisa Li, Ari Holtzman, Daniel Fried, Percy Liang, Jason Eisner, Tatsunori B. Hashimoto, Luke Zettlemoyer, and Mike Lewis. 2023. https://doi.org/10.18653/v1/2023.acl-long.687 Contrastive decoding: Open-ended text generation as optimization . In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Pap...

    work page doi:10.18653/v1/2023.acl-long.687 2023

  48. [48]

    Shayne Longpre, Kartik Perisetla, Anthony Chen, Nikhil Ramesh, Chris DuBois, and Sameer Singh. 2021. https://doi.org/10.18653/v1/2021.emnlp-main.565 Entity-based knowledge conflicts in question answering . In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 7052--7063. Association for Computational Linguistics

    work page doi:10.18653/v1/2021.emnlp-main.565 2021

  49. [49]

    Meta AI . 2024. Meta Llama 3 model card. https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md. Model card, accessed 2026-05-15

    2024

  50. [50]

    Sewon Min, Kalpesh Krishna, Xinxi Lyu, Mike Lewis, Wen-tau Yih, Pang Wei Koh, Mohit Iyyer, Luke Zettlemoyer, and Hannaneh Hajishirzi. 2023. https://doi.org/10.18653/v1/2023.emnlp-main.741 FActScore : Fine-grained atomic evaluation of factual precision in long form text generation . In Proceedings of the 2023 Conference on Empirical Methods in Natural Lang...

    work page doi:10.18653/v1/2023.emnlp-main.741 2023

  51. [51]

    Minh Nguyen, Andrew Baker, Clement Neo, Allen Roush, Andreas Kirsch, and Ravid Shwartz-Ziv. 2025. Turning up the heat: Min- P sampling for creative and coherent LLM outputs. In International Conference on Learning Representations. Oral; arXiv:2407.01082

    arXiv 2025

  52. [52]

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul F. Christiano, Jan Leike, and Ryan Lowe. 2022. Training language models to follow instructions with hum...

    2022

  53. [53]

    Manning, Stefano Ermon, and Chelsea Finn

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D. Manning, Stefano Ermon, and Chelsea Finn. 2023. Direct preference optimization: Your language model is secretly a reward model. In Advances in Neural Information Processing Systems, volume 36

    2023

  54. [54]

    Weijia Shi, Xiaochuang Han, Mike Lewis, Yulia Tsvetkov, Luke Zettlemoyer, and Wen-tau Yih. 2024. https://doi.org/10.18653/v1/2024.naacl-short.69 Trusting your evidence: Hallucinate less with context-aware decoding . In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technolo...

    work page doi:10.18653/v1/2024.naacl-short.69 2024

  55. [55]

    Han Wang, Archiki Prasad, Elias Stengel-Eskin, and Mohit Bansal. 2025. AdaCAD : Adaptively decoding to balance conflicts between contextual and parametric knowledge. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 11636...

    2025

  56. [56]

    Kevin Yang and Dan Klein. 2021. https://doi.org/10.18653/v1/2021.naacl-main.276 FUDGE : Controlled text generation with future discriminators . In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 3511--3535. Association for Computational Linguistics

    work page doi:10.18653/v1/2021.naacl-main.276 2021

  57. [57]

    Xiaowei Yuan, Zhao Yang, Yequan Wang, Shengping Liu, Jun Zhao, and Kang Liu. 2024. https://doi.org/10.18653/v1/2024.findings-acl.234 Discerning and resolving knowledge conflicts through adaptive decoding with contextual information-entropy constraint . In Findings of the Association for Computational Linguistics: ACL 2024, pages 3903--3922. Association fo...

    work page doi:10.18653/v1/2024.findings-acl.234 2024