arxiv: 2604.06771 · v1 · submitted 2026-04-08 · 💻 cs.CL · cs.AI

Recognition: no theorem link

Multi-Faceted Self-Consistent Preference Alignment for Query Rewriting in Conversational Search

Peifeng Li, Qiaoming Zhu, Zhiyu Cao

Authors on Pith no claims yet

Pith reviewed 2026-05-10 18:07 UTC · model grok-4.3

classification 💻 cs.CL cs.AI

keywords conversational query rewritingpreference alignmentdirect preference optimizationconversational searchquery reformulationmulti-faceted optimizationself-consistent data

0 comments

The pith

Aligning query rewrites with preferences drawn from rewriting, retrieval, and response stages improves conversational search performance.

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

The paper argues that conventional conversational query rewriting treats the rewrite step in isolation and therefore misses useful signals about whether a rewrite actually helps retrieve relevant passages or generate good answers. To fix this, the authors first build preference pairs that are self-consistent across the three stages: they keep only rewrites that rank well under a rewriting metric, improve retrieval, and lead to better final responses. They then train with a prefix-guided multi-faceted version of direct preference optimization that lets the model learn distinct preference signals from each stage while sharing a common prefix. The resulting model produces more diverse rewrites that work both on topics seen during training and on new topics.

Core claim

By constructing self-consistent preference alignment data from the rewriting, retrieval, and response dimensions and applying prefix-guided multi-faceted direct preference optimization, the MSPA-CQR method learns to generate rewritten queries that are effective in both in-distribution and out-of-distribution conversational search scenarios.

What carries the argument

Self-consistent preference alignment data built from rewriting, retrieval, and response dimensions, together with prefix-guided multi-faceted direct preference optimization that trains on the three signals simultaneously.

If this is right

Rewritten queries become more diverse because each rewrite must satisfy constraints from all three stages rather than only the rewrite metric.
The learned preferences remain useful even when the test conversations cover topics absent from the training data.
The same three-stage preference construction can be applied on top of any base rewriting model without changing its architecture.
Training on aligned signals from retrieval and response reduces the chance that a rewrite looks good in isolation but harms later pipeline stages.

Where Pith is reading between the lines

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

The approach implies that isolated optimization of any single step in a multi-stage retrieval pipeline is likely to leave performance on the table.
Similar multi-faceted preference data could be collected for other conversational tasks such as clarification or follow-up question generation.
If the three signals are partly redundant, a smaller set of dimensions might suffice, offering a route to cheaper training data.
Extending the method to include explicit user feedback signals would test whether the current automatic construction already captures most of the useful preference information.

Load-bearing premise

The automatically generated preference data from the three stages can be produced without introducing inconsistencies or biases that would confuse the optimization process.

What would settle it

An experiment in which MSPA-CQR fails to beat single-dimension preference baselines on standard CQR test sets would falsify the claim that the multi-faceted alignment is responsible for the observed gains.

Figures

Figures reproduced from arXiv: 2604.06771 by Peifeng Li, Qiaoming Zhu, Zhiyu Cao.

**Figure 2.** Figure 2: An overview of MSPA-CQR, including the two stages of Multi-Faceted Preference Data Construction and [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Figure 3: Kendall’s Tau correlation between the con [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

**Figure 4.** Figure 4: A comparison of the linguistic features of rewritten queries generated under three preference guides. “N-grams div.” refers to the diversity of n-grams (n = 2), and “Edit dist.” refers to the Levenshtein distance. Lai et al., 2025; Yoon et al., 2025), query expansion can significantly improve retrieval performance. In addition, since our method generates queries with multiple different preference dimensio… view at source ↗

**Figure 5.** Figure 5: The intersection size of passages obtained [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗

**Figure 7.** Figure 7: The trend of R@100 on TopiOCQA with the change in the number of sampled candidate rewritten queries. G.7 Correlation between Self-Consistency Scores and Downstream Tasks We measure preferences in different dimensions through self-consistency scores, based on the assumption that self-consistency is closely related to performance on downstream tasks. To verify this hypothesis, we measured the correlation be… view at source ↗

**Figure 8.** Figure 8: The trend of MRR on TopiOCQA with the change in the number of sampled candidate rewritten queries. TopiOCQA Different Values of T MRR NDCG R@10 R@100 MSPA-CQR (T = 10) 40.7 38.4 62.7 76.8 MSPA-CQR (T = 50) 41.6 39.8 64.1 78.2 MSPA-CQR (T = 100) 41.4 39.5 63.5 77.4 [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗

**Figure 9.** Figure 9: A case of merging multiple rewritten queries [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗

read the original abstract

Conversational Query Rewriting (CQR) aims to rewrite ambiguous queries to achieve more efficient conversational search. Early studies have predominantly focused on the rewriting in isolation, ignoring the feedback from query rewrite, passage retrieval and response generation in the rewriting process. To address this issue, we propose Multi-Faceted Self-Consistent Preference Aligned CQR (MSPA-CQR). Specifically, we first construct self-consistent preference alignment data from three dimensions (rewriting, retrieval, and response) to generate more diverse rewritten queries. Then we propose prefix guided multi-faceted direct preference optimization to learn preference information from three different dimensions. The experimental results show that our MSPA-CQR is effective in both in- and out-of-distribution scenarios.

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.

Desk Editor's Note private letter to a colleague

The paper adds a multi-faceted DPO variant that pulls rewriting, retrieval, and response signals into conversational query rewriting, but the self-consistency step looks under-specified.

read the letter

The core idea is to stop treating query rewriting as a standalone step and instead build preference data that stays consistent across the rewrite itself, the passages it retrieves, and the final response it enables. They generate those triples, then train with prefix-guided multi-faceted direct preference optimization. That combination is the actual novelty here; most earlier CQR work optimizes the rewrite in isolation and never closes the loop with downstream components. The prefix guidance is a small but practical touch that could keep the different facets from fighting each other during training. If the experiments show clean gains on both in-distribution and out-of-distribution test sets, this is the sort of targeted engineering improvement that production conversational systems could adopt without much extra cost. The main soft spot is exactly the one the stress-test flags: self-consistency is asserted but the abstract gives no numbers on inter-dimension agreement, no ablation on how conflicts are resolved, and no human validation of the generated preferences. Without those checks, the DPO stage risks optimizing on contradictory signals, which would undermine the out-of-distribution claims in particular. I would also want to see the precise baselines, datasets, and whether the reported improvements survive a fair comparison against recent DPO variants already used in retrieval. This is a narrow but useful piece for people already working on conversational search pipelines or preference tuning for IR tasks. A reader who needs a concrete way to incorporate retrieval and response feedback into rewriting could extract the method and try it. It is coherent enough on its own terms to deserve a serious referee who can verify the data construction and run the necessary ablations. I would send it to review rather than desk-reject, with the expectation that the consistency validation and experimental details will need tightening.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes Multi-Faceted Self-Consistent Preference Aligned CQR (MSPA-CQR) for conversational query rewriting. It first constructs self-consistent preference alignment data from three dimensions (rewriting, retrieval, and response) to generate diverse rewritten queries, then applies prefix-guided multi-faceted direct preference optimization to learn preferences across these dimensions. The central claim is that MSPA-CQR is effective in both in-distribution and out-of-distribution scenarios.

Significance. If the self-consistency of the constructed preference data can be quantitatively validated and the reported gains hold under rigorous controls, the work would meaningfully extend preference alignment techniques to conversational search by integrating multi-stage feedback signals, offering a potential improvement over isolated rewriting methods and better handling of domain shifts.

major comments (2)

[3.2] Section 3.2 (Preference Data Construction): The description of enforcing self-consistency across rewriting/retrieval/response dimensions provides no quantitative validation such as inter-dimension agreement rates, consistency thresholds, or human evaluation of the generated triples. This is load-bearing for the headline claim, as unvalidated consistency risks contradictory preference signals that could produce noisy or conflicting gradients during prefix-guided multi-faceted DPO, directly threatening the out-of-distribution generalization results.
[4] Section 4 (Experiments): No ablation studies isolate the contribution of the self-consistency mechanism (e.g., majority vote vs. sequential prompting) or the multi-faceted prefix guidance. Without these, it is impossible to confirm that performance gains stem from the proposed self-consistent alignment rather than from standard DPO or data scale, weakening the causal link to the central claim.

minor comments (2)

[Abstract] Abstract: The claim of effectiveness is stated without any mention of specific datasets, baselines, or metrics; adding a brief quantitative summary would improve clarity and allow readers to assess the strength of the results immediately.
[3.3] Notation in Section 3.3: The prefix-guided multi-faceted DPO objective would benefit from an explicit equation or pseudocode to make the integration of the three dimensions and the prefix mechanism fully reproducible.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments and recommendation for major revision. We address each point below and will incorporate the suggested changes to strengthen the manuscript.

read point-by-point responses

Referee: [3.2] Section 3.2 (Preference Data Construction): The description of enforcing self-consistency across rewriting/retrieval/response dimensions provides no quantitative validation such as inter-dimension agreement rates, consistency thresholds, or human evaluation of the generated triples. This is load-bearing for the headline claim, as unvalidated consistency risks contradictory preference signals that could produce noisy or conflicting gradients during prefix-guided multi-faceted DPO, directly threatening the out-of-distribution generalization results.

Authors: We agree that quantitative validation of self-consistency is essential to support the reliability of the preference data. In the revised manuscript, we will expand Section 3.2 to report inter-dimension agreement rates across the rewriting, retrieval, and response dimensions, the specific consistency thresholds used during triple construction, and human evaluation results on a sampled subset of the generated preference triples. These additions will directly address concerns about potential contradictory signals and provide stronger grounding for the out-of-distribution generalization claims. revision: yes
Referee: [4] Section 4 (Experiments): No ablation studies isolate the contribution of the self-consistency mechanism (e.g., majority vote vs. sequential prompting) or the multi-faceted prefix guidance. Without these, it is impossible to confirm that performance gains stem from the proposed self-consistent alignment rather than from standard DPO or data scale, weakening the causal link to the central claim.

Authors: We recognize the value of targeted ablations for establishing causality. The revised experiments section will include new ablation studies that isolate the self-consistency mechanism by comparing the full approach against variants using independent (non-majority-vote) prompting and sequential prompting alternatives. We will also ablate the multi-faceted prefix guidance against standard single-dimension DPO while controlling for data scale. Results will be reported on the same in-distribution and out-of-distribution benchmarks to clarify the source of the observed gains. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected in derivation chain

full rationale

The provided abstract and method sketch describe a two-stage process: (1) constructing self-consistent preference triples across rewriting/retrieval/response dimensions and (2) applying prefix-guided multi-faceted DPO. No equations, parameter-fitting steps, or mathematical derivations appear. The central claim is supported by experimental results rather than by any self-referential definition, fitted-input-as-prediction, or load-bearing self-citation chain. Because the derivation is purely descriptive and does not reduce any output quantity to its own inputs by construction, the paper is self-contained against the circularity criteria.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only input supplies no explicit free parameters, axioms, or invented entities; the approach appears to rest on standard direct preference optimization assumptions and the unstated claim that multi-stage feedback can be turned into consistent preference pairs.

pith-pipeline@v0.9.0 · 5420 in / 1228 out tokens · 50058 ms · 2026-05-10T18:07:04.419765+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

37 extracted references · 17 canonical work pages

[1]

Vaibhav Adlakha, Shehzaad Dhuliawala, Kaheer Suleman, Harm de Vries, and Siva Reddy. 2022. https://doi.org/10.1162/tacl_a_00471 T opi OCQA : Open-domain conversational question answering with topic switching . Transactions of the Association for Computational Linguistics, 10:468--483

work page doi:10.1162/tacl_a_00471 2022
[2]

Raviteja Anantha, Svitlana Vakulenko, Zhucheng Tu, Shayne Longpre, Stephen Pulman, and Srinivas Chappidi. 2021. https://doi.org/10.18653/v1/2021.naacl-main.44 Open-domain question answering goes conversational via question rewriting . In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Huma...

work page doi:10.18653/v1/2021.naacl-main.44 2021
[3]

Gordon V Cormack, Charles LA Clarke, and Stefan Buettcher. 2009. Reciprocal rank fusion outperforms condorcet and individual rank learning methods. In Proceedings of the 32nd international ACM SIGIR conference on Research and development in information retrieval, pages 758--759

2009
[4]

Jeffrey Dalton, Chenyan Xiong, and Jamie Callan. 2020 a . https://trec.nist.gov/pubs/trec29/papers/OVERVIEW.C.pdf Cast 2020: The conversational assistance track overview . In Proceedings of the Twenty-Ninth Text REtrieval Conference, TREC 2020 , NIST Special Publication

2020
[5]

Jeffrey Dalton, Chenyan Xiong, and Jamie Callan. 2020 b . Trec cast 2019: The conversational assistance track overview. arXiv preprint arXiv:2003.13624

work page arXiv 2020
[6]

Jeffrey Dalton, Chenyan Xiong, and Jamie Callan. 2021. https://trec.nist.gov/pubs/trec30/papers/Overview-CAsT.pdf TREC cast 2021: The conversational assistance track overview . In Proceedings of the Thirtieth Text REtrieval Conference, TREC 2021 , NIST Special Publication

2021
[7]

Luyu Gao, Xueguang Ma, Jimmy Lin, and Jamie Callan. 2023. https://doi.org/10.18653/V1/2023.ACL-LONG.99 Precise zero-shot dense retrieval without relevance labels . In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics, ACL 2023 , pages 1762--1777

work page doi:10.18653/v1/2023.acl-long.99 2023
[8]

Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen - Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen

Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen - Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. https://openreview.net/forum?id=nZeVKeeFYf9 Lora: Low-rank adaptation of large language models . In The Tenth International Conference on Learning Representations, ICLR 2022

2022
[9]

Yunah Jang, Kang-il Lee, Hyunkyung Bae, Hwanhee Lee, and Kyomin Jung. 2024. https://doi.org/10.18653/v1/2024.naacl-long.449 I ter CQR : Iterative conversational query reformulation with retrieval guidance . In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pag...

work page doi:10.18653/v1/2024.naacl-long.449 2024
[10]

Zhuoran Jin, Pengfei Cao, Yubo Chen, Kang Liu, and Jun Zhao. 2023. https://doi.org/10.18653/V1/2023.FINDINGS-EMNLP.443 Instructor: Instructing unsupervised conversational dense retrieval with large language models . In Findings of the Association for Computational Linguistics: EMNLP 2023 , pages 6649--6675

work page doi:10.18653/v1/2023.findings-emnlp.443 2023
[11]

Jeff Johnson, Matthijs Douze, and Herv \'e J \'e gou. 2019. Billion-scale similarity search with gpus. IEEE Transactions on Big Data, 7(3):535--547

2019
[12]

Maurice G Kendall. 1938. A new measure of rank correlation. Biometrika, 30(1/2):81--93

1938
[13]

Yilong Lai, Jialong Wu, Congzhi Zhang, Haowen Sun, and Deyu Zhou. 2025. https://aclanthology.org/2025.coling-main.515/ A da CQR : Enhancing query reformulation for conversational search via sparse and dense retrieval alignment . In Proceedings of the 31st International Conference on Computational Linguistics, pages 7698--7720

2025
[14]

Joon Ho Lee. 1997. Analyses of multiple evidence combination. In Proceedings of the 20th annual international ACM SIGIR conference on Research and development in information retrieval, pages 267--276

1997
[15]

Levenshtein

Vladimir I. Levenshtein. 1966. Binary codes capable of correcting deletions, insertions, and reversals. Soviet Physics Doklady, 10(8):707--710

1966
[16]

Jimmy Lin, Xueguang Ma, Sheng-Chieh Lin, Jheng-Hong Yang, Ronak Pradeep, and Rodrigo Nogueira. 2021. Pyserini: A python toolkit for reproducible information retrieval research with sparse and dense representations. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 2356--2362

2021
[17]

Sheng-Chieh Lin, Jheng-Hong Yang, Rodrigo Nogueira, Ming-Feng Tsai, Chuan-Ju Wang, and Jimmy Lin. 2020. https://arxiv.org/abs/2004.01909 Conversational question reformulation via sequence-to-sequence architectures and pretrained language models . Preprint, arXiv:2004.01909

work page arXiv 2020
[18]

Xueguang Ma, Liang Wang, Nan Yang, Furu Wei, and Jimmy Lin. 2024. https://doi.org/10.1145/3626772.3657951 Fine-tuning llama for multi-stage text retrieval . In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR 2024 , pages 2421--2425

work page doi:10.1145/3626772.3657951 2024
[19]

Kelong Mao, Zhicheng Dou, Bang Liu, Hongjin Qian, Fengran Mo, Xiangli Wu, Xiaohua Cheng, and Zhao Cao. 2023 a . https://doi.org/10.18653/v1/2023.findings-acl.256 Search-oriented conversational query editing . In Findings of the Association for Computational Linguistics: ACL 2023, pages 4160--4172

work page doi:10.18653/v1/2023.findings-acl.256 2023
[20]

Kelong Mao, Zhicheng Dou, Fengran Mo, Jiewen Hou, Haonan Chen, and Hongjin Qian. 2023 b . https://doi.org/10.18653/v1/2023.findings-emnlp.86 Large language models know your contextual search intent: A prompting framework for conversational search . In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 1211--1225

work page doi:10.18653/v1/2023.findings-emnlp.86 2023
[21]

Fengran Mo, Abbas Ghaddar, Kelong Mao, Mehdi Rezagholizadeh, Boxing Chen, Qun Liu, and Jian - Yun Nie. 2024. https://aclanthology.org/2024.emnlp-main.135 CHIQ: contextual history enhancement for improving query rewriting in conversational search . In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, EMNLP 2024 , pages...

2024
[22]

Fengran Mo, Kelong Mao, Yutao Zhu, Yihong Wu, Kaiyu Huang, and Jian-Yun Nie. 2023. https://doi.org/10.18653/v1/2023.acl-long.274 C onv GQR : Generative query reformulation for conversational search . In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics, pages 4998--5012

work page doi:10.18653/v1/2023.acl-long.274 2023
[23]

Andr \'e Mour \ a o, Fl \'a vio Martins, and Jo \ a o Magalh \ a es. 2015. Multimodal medical information retrieval with unsupervised rank fusion. Computerized Medical Imaging and Graphics, 39:35--45

2015
[24]

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. http://papers.nips.cc/paper\_files/paper/2022/hash/b1efd...

2022
[25]

Manning, Stefano Ermon, and Chelsea Finn

Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D. Manning, Stefano Ermon, and Chelsea Finn. 2023. http://papers.nips.cc/paper\_files/paper/2023/hash/a85b405ed65c6477a4fe8302b5e06ce7-Abstract-Conference.html Direct preference optimization: Your language model is secretly a reward model . In Advances in Neural Information Processing Systems 36: ...

2023
[26]

Stephen Robertson, Hugo Zaragoza, et al. 2009. The probabilistic relevance framework: Bm25 and beyond. Foundations and Trends in Information Retrieval , 3(4):333--389

2009
[27]

Liang Wang, Nan Yang, Xiaolong Huang, Linjun Yang, Rangan Majumder, and Furu Wei. 2024. https://doi.org/10.18653/V1/2024.ACL-LONG.642 Improving text embeddings with large language models . In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics, ACL 2024 , pages 11897--11916

work page doi:10.18653/v1/2024.acl-long.642 2024
[28]

Liang Wang, Nan Yang, and Furu Wei. 2023 a . https://doi.org/10.18653/V1/2023.EMNLP-MAIN.585 Query2doc: Query expansion with large language models . In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, EMNLP 2023 , pages 9414--9423

work page doi:10.18653/v1/2023.emnlp-main.585 2023
[29]

Le, Ed H

Xuezhi Wang, Jason Wei, Dale Schuurmans, Quoc V. Le, Ed H. Chi, Sharan Narang, Aakanksha Chowdhery, and Denny Zhou. 2023 b . https://openreview.net/forum?id=1PL1NIMMrw Self-consistency improves chain of thought reasoning in language models . In The Eleventh International Conference on Learning Representations, ICLR 2023

2023
[30]

Zeqiu Wu, Yi Luan, Hannah Rashkin, David Reitter, Hannaneh Hajishirzi, Mari Ostendorf, and Gaurav Singh Tomar. 2022. https://doi.org/10.18653/v1/2022.emnlp-main.679 CONQRR : Conversational query rewriting for retrieval with reinforcement learning . In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 10000--10014

work page doi:10.18653/v1/2022.emnlp-main.679 2022
[31]

Bennett, Junaid Ahmed, and Arnold Overwijk

Lee Xiong, Chenyan Xiong, Ye Li, Kwok - Fung Tang, Jialin Liu, Paul N. Bennett, Junaid Ahmed, and Arnold Overwijk. 2021. https://openreview.net/forum?id=zeFrfgyZln Approximate nearest neighbor negative contrastive learning for dense text retrieval . In 9th International Conference on Learning Representations, ICLR 2021

2021
[32]

Fanghua Ye, Meng Fang, Shenghui Li, and Emine Yilmaz. 2023. https://doi.org/10.18653/v1/2023.findings-emnlp.398 Enhancing conversational search: Large language model-aided informative query rewriting . In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 5985--6006

work page doi:10.18653/v1/2023.findings-emnlp.398 2023
[33]

Chanwoong Yoon, Gangwoo Kim, Byeongguk Jeon, Sungdong Kim, Yohan Jo, and Jaewoo Kang. 2025. https://aclanthology.org/2025.findings-naacl.328/ Ask optimal questions: Aligning large language models with retriever`s preference in conversation . In Findings of the Association for Computational Linguistics: NAACL 2025, pages 5899--5921

2025
[34]

Peitian Zhang, Shitao Xiao, Zheng Liu, Zhicheng Dou, and Jian-Yun Nie. 2023. https://arxiv.org/abs/2310.07554 Retrieve anything to augment large language models . Preprint, arXiv:2310.07554

work page arXiv 2023
[35]

Glass, and Helen M

Tianhua Zhang, Kun Li, Hongyin Luo, Xixin Wu, James R. Glass, and Helen M. Meng. 2024. https://doi.org/10.18653/v1/2024.emnlp-main.746 Adaptive query rewriting: Aligning rewriters through marginal probability of conversational answers . In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 13444--13461

work page doi:10.18653/v1/2024.emnlp-main.746 2024
[36]

online" 'onlinestring :=

ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRING...
[37]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...