pith. sign in

arxiv: 2606.08018 · v1 · pith:OOCVMA6Anew · submitted 2026-06-06 · 💻 cs.AI

UniQL: Towards Dialect-Universal Benchmarking for Text-to-SQL

Jianling Gao , Chongyang Tao , Jiayuan Bai , Liu Yang , Xuanguang Pan , Jinrui Liu , Shihao Xing , Xiaohan Xu
show 2 more authors
Jie Liang Shuai Ma
This is my paper

Pith reviewed 2026-06-27 19:51 UTC · model grok-4.3

classification 💻 cs.AI
keywords text-to-SQLSQL dialectscross-dialect benchmarkLLM evaluationdatabase migrationsemantic equivalencedialect generalization
0
0 comments X

The pith

UniQL aligns 1534 questions with executable SQL across 16 dialects to measure cross-dialect generalization.

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

Text-to-SQL benchmarks have focused almost exclusively on SQLite, yet real database systems use many dialects that differ in syntax, functions, and execution rules. The paper introduces UniQL, a dataset that pairs the same natural language questions with verified SQL queries in 16 dialects while keeping schemas and data contents identical. A hybrid construction process migrates databases, translates queries, verifies execution, summarizes translation rules, and applies human checks to ensure semantic equivalence. Experiments with open- and closed-source models show large performance gaps across dialects and weak transfer from SQLite results to others. The work establishes that current models are not dialect-universal and that aligned multi-dialect benchmarks are required to expose this gap.

Core claim

UniQL supplies 1534 natural language questions aligned to 24544 dialect-specific SQL queries spanning 16 SQL dialects. All queries share identical intents, schemas, and database contents, and were produced by a pipeline of database migration, SQL translation, execution-guided verification, iterative rule summarization, and human validation. Evaluation on multiple LLMs reveals substantial accuracy variation across dialects together with limited transfer from success on SQLite to other systems.

What carries the argument

The hybrid pipeline of database migration, SQL translation, execution-guided verification, iterative rule summarization, and human validation that produces semantically equivalent queries across dialects.

If this is right

  • Model accuracy on SQLite does not reliably predict accuracy on other dialects.
  • Text-to-SQL methods must incorporate explicit dialect handling to reach uniform performance.
  • Future benchmarks should maintain aligned schemas and contents across multiple dialects rather than treating each dialect in isolation.

Where Pith is reading between the lines

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

  • Training or adaptation procedures that expose models to multiple dialects simultaneously may reduce the observed performance variation.
  • Real-world text-to-SQL systems deployed against heterogeneous database backends will require dialect-specific post-processing or routing until generalization improves.

Load-bearing premise

The hybrid pipeline produces queries that are semantically equivalent across dialects without systematic translation errors or selection bias.

What would settle it

Demonstrating a non-negligible fraction of UniQL query pairs that produce different results when executed on their respective dialect databases would falsify the claimed equivalence.

Figures

Figures reproduced from arXiv: 2606.08018 by Chongyang Tao, Jianling Gao, Jiayuan Bai, Jie Liang, Jinrui Liu, Liu Yang, Shihao Xing, Shuai Ma, Xiaohan Xu, Xuanguang Pan.

Figure 1
Figure 1. Figure 1: UNIQL construction pipeline. UNIQL extends the widely-used BIRD SQLite development set to 16 SQL dialects through database migration, hybrid SQL translation, execution-based verification, iterative translation rule evolution, and human validation. is presented in [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Construction pipeline statistics for the 15 [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Stratified execution accuracy (%) of the 13 [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
read the original abstract

Existing text-to-SQL benchmarks are largely centered on SQLite, making it difficult to evaluate whether models can generalize across heterogeneous SQL dialects. However, real-world database systems differ substantially in syntax, functions, type systems, and execution semantics, so the same natural language intent often requires dialect-specific SQL realizations. We introduce UniQL, a human-verified benchmark for cross-dialect text-to-SQL evaluation. UniQL aligns 1,534 natural language questions with executable SQL annotations across 16 SQL dialects, yielding 24,544 dialect-specific queries. All dialects share the same intents, aligned schemas and database contents, enabling controlled evaluation of dialect generalization. UniQL is constructed through a hybrid pipeline combining database migration, SQL translation, execution-guided verification, iterative rule summarization, and human validation. Experiments on both open-source and closed-source LLMs show that current models remain far from dialect-universal, with substantial performance variation across database systems and limited transfer from SQLite success to other dialects. These findings highlight the need for aligned cross-dialect benchmarks and more dialect-aware text-to-SQL methods. Code and data are available at https://github.com/JerryGao818/UniQL

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

2 major / 2 minor

Summary. The paper introduces UniQL, a benchmark aligning 1,534 natural language questions to executable SQL across 16 dialects (yielding 24,544 queries) via a hybrid pipeline of database migration, SQL translation, execution-guided verification, iterative rule summarization, and human validation. All dialects share identical intents, schemas, and data. Experiments on open- and closed-source LLMs demonstrate substantial cross-dialect performance variation and limited transfer from SQLite success, arguing that current models are far from dialect-universal and that aligned cross-dialect benchmarks are needed.

Significance. If the alignments hold, the work is significant for exposing a real gap in text-to-SQL evaluation: most existing benchmarks are SQLite-centric while production systems differ in syntax, functions, and semantics. The public code and data release supports reproducibility and enables follow-on work on dialect-aware methods.

major comments (2)
  1. [§3 (pipeline description)] Construction pipeline (abstract and §3): the central claim that the 24,544 queries are semantically equivalent across dialects rests on the hybrid pipeline, yet no inter-annotator agreement, human-validation rejection rates, translation error rates, or post-validation execution-match statistics are reported. Without these quantities the observed performance gaps could arise from undetected alignment artifacts rather than model limitations.
  2. [§4 (experiments)] Experimental results (§4): the headline finding of 'limited transfer from SQLite success to other dialects' requires that every NL question maps to executable, semantically identical SQL in all 16 dialects on identical schemas; the absence of the validation metrics above makes this transfer claim difficult to interpret.
minor comments (2)
  1. [Abstract] Abstract: the arithmetic 1,534 × 16 = 24,544 is correct and should be stated explicitly when first introducing the query count.
  2. [Results tables] Table/figure captions: ensure every table or figure reporting per-dialect accuracies also states the exact number of queries evaluated per dialect so readers can assess statistical power.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback, which identifies a key area for strengthening the presentation of our validation process. We address each major comment below and commit to revisions that will incorporate the requested quantitative details without altering the core claims or methodology.

read point-by-point responses

  1. Referee: [§3 (pipeline description)] Construction pipeline (abstract and §3): the central claim that the 24,544 queries are semantically equivalent across dialects rests on the hybrid pipeline, yet no inter-annotator agreement, human-validation rejection rates, translation error rates, or post-validation execution-match statistics are reported. Without these quantities the observed performance gaps could arise from undetected alignment artifacts rather than model limitations.

    Authors: We agree that the absence of these specific quantitative metrics limits the ability to fully assess alignment quality. The original manuscript describes the hybrid pipeline (database migration, SQL translation, execution-guided verification, iterative rule summarization, and human validation) but does not report the requested statistics. In the revised version we will add a new subsection (or appendix) providing: (1) inter-annotator agreement computed on a sampled subset of annotations, (2) human-validation rejection rates, (3) translation error rates observed during the automated steps, and (4) post-validation execution-match statistics across dialects. These additions will directly support the claim of semantic equivalence and reduce the possibility that performance gaps arise from undetected artifacts. revision: yes

  2. Referee: [§4 (experiments)] Experimental results (§4): the headline finding of 'limited transfer from SQLite success to other dialects' requires that every NL question maps to executable, semantically identical SQL in all 16 dialects on identical schemas; the absence of the validation metrics above makes this transfer claim difficult to interpret.

    Authors: We concur that the transfer claim is more interpretable when accompanied by explicit validation metrics. Once the statistics outlined in our response to the first comment are added, we will update §4 to reference these numbers when discussing the limited transfer from SQLite. We will also add a brief clarification in the experimental setup that the pipeline (including execution-guided verification and human validation) was applied uniformly to ensure every question has executable, semantically aligned SQL across all 16 dialects on identical schemas. This will make the headline finding more robust. revision: yes

Circularity Check

0 steps flagged

No circularity: benchmark construction and empirical measurement are self-contained

full rationale

The paper presents a dataset construction pipeline (migration, translation, verification, human validation) followed by empirical evaluation of LLMs on the resulting benchmark. No equations, fitted parameters, predictions, or first-principles derivations are claimed. The central claim (models show cross-dialect performance gaps) is a direct measurement on externally produced model outputs, not a reduction to the construction inputs by definition or self-citation. No load-bearing self-citations or ansatzes are present in the provided text. This matches the default case of an empirical benchmark paper with independent content.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central contribution is a new dataset; it rests on the domain assumption that semantic equivalence can be maintained across dialects via the described pipeline, with no free parameters or invented entities.

axioms (1)
  • domain assumption SQL translation and migration can preserve query semantics and execution results for the selected natural language intents across the 16 dialects
    Invoked in the hybrid pipeline description to justify alignment of queries.

pith-pipeline@v0.9.1-grok · 5762 in / 1110 out tokens · 16616 ms · 2026-06-27T19:51:49.279650+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

246 extracted references · 2 canonical work pages

  1. [1]

    基于深度学习的数据库自然语言接口综述 , year =

  2. [2]

    2024 , url =

    Eosphoros-Ai/. 2024 , url =

    2024

  3. [3]

    , year =

    Abhyankar, Nikhil and Gupta, Vivek and Roth, Dan and Reddy, Chandan K. , year =. H-. arXiv , url =

  4. [4]

    2025 , journal =

    Semantic. 2025 , journal =

    2025

  5. [5]

    , year =

    Ascoli, Benjamin and Kandikonda, Ram and Choi, Jinho D. , year =. arXiv , url =

  6. [6]

    AAAI-25 , year =

    Arian Askari and Christian P. AAAI-25 , year =

  7. [7]

    arXiv , url =

    Bai, Qiushi and Alsudais, Sadeem and Li, Chen , year =. arXiv , url =

  8. [8]

    and Guestrin, Carlos and Zaharia, Matei , year =

    Biswal, Asim and Patel, Liana and Jha, Siddarth and Kamsetty, Amog and Liu, Shu and Gonzalez, Joseph E. and Guestrin, Carlos and Zaharia, Matei , year =. arXiv , url =

  9. [9]

    Chafik, Salmane and Ezzini, Saad and Berrada, Ismail , year =

  10. [10]

    Chang, Shuaichen and. How to. 2023 , journal =

    2023

  11. [11]

    Selective

    Chang, Shuaichen and. Selective. 2023 , journal =

    2023

  12. [12]

    Text-to-

    Chen, Ziru and Chen, Shijie and White, Michael and Mooney, Raymond and Payani, Ali and Srinivasa, Jayanth and Su, Yu and Sun, Huan , year =. Text-to-. arXiv , url =

  13. [13]

    2024 , journal =

    Chen, Peter Baile and Wenz, Fabian and Zhang, Yi and Kayali, Moe and Tatbul, Nesime and Cafarella, Michael and Demiralp,. 2024 , journal =

    2024

  14. [14]

    Chen, Si-An and Miculicich, Lesly and Eisenschlos, Julian Martin and Wang, Zifeng and Wang, Zilong and Chen, Yanfei and Fujii, Yasuhisa and Lin, Hsuan-Tien and Lee, Chen-Yu and Pfister, Tomas , year =

  15. [15]

    Chen, Jikai and Gan, Leilei , year =

  16. [16]

    2025 , journal =

    Pi-. 2025 , journal =

    2025

  17. [17]

    Dai, Yaxun and Xie, Wenxuan and Zhuang, Xialie and Yang, Tianyu and Yang, Yiying and Yang, Haiqin and Zhao, Yuhang and Chao, Pingfu and Jiang, Wenhao , year =

  18. [18]

    Deng, Naihao and Chen, Yulong and Zhang, Yue , year =. Recent

  19. [19]

    Deng, Minghang and Ramachandran, Ashwin and Xu, Canwen and Hu, Lanxiang and Yao, Zhewei and Datta, Anupam and Zhang, Hao , year =

  20. [20]

    Dom. Blar-. 2024 , publisher =

    2024

  21. [21]

    , year =

    Dong, Xuemei and Zhang, Chao and Ge, Yuhang and Mao, Yuren and Gao, Yunjun and et al. , year =. C3:

  22. [22]

    Dong, Mingwen and Kumar, Nischal Ashok and Hu, Yiqun and Chauhan, Anuj and Hang, Chung-Wei and Chang, Shuaichen and Pan, Lin and Lan, Wuwei and Zhu, Henghui and Jiang, Jiarong and Ng, Patrick and Wang, Zhiguo , year =

  23. [23]

    Dou, Longxu and Gao, Yan and Liu, Xuqi and Pan, Mingyang and Wang, Dingzirui and Che, Wanxiang and Zhan, Dechen and Kan, Min-Yen and Lou, Jian-Guang , year =. Towards

  24. [24]

    Fan, Yuankai and He, Zhenying and Ren, Tonghui and Guo, Dianjun and Chen, Lin and Zhu, Ruisi and Chen, Guanduo and Jing, Yinan and Zhang, Kai and Wang, X.Sean , year =. Gar:. 2023

    2023

  25. [25]

    Fan, Yuankai and Ren, Tonghui and He, Zhenying and Wang, X.Sean and Zhang, Ye and Li, Xingang , year =. 2023

    2023

  26. [26]

    Combining

    Fan, Ju and Gu, Zihui and Zhang, Songyue and Zhang, Yuxin and Chen, Zui and Cao, Lei and Li, Guoliang and Madden, Samuel and Du, Xiaoyong and Tang, Nan , year =. Combining. Proceedings of the VLDB Endowment , volume =

  27. [27]

    Sean , year =

    Fan, Yuankai and He, Zhenying and Ren, Tonghui and Huang, Can and Jing, Yinan and Zhang, Kai and Wang, X. Sean , year =. Metasql:

  28. [28]

    Sean , year =

    Fan, Yuankai and Ren, Tonghui and Huang, Can and He, Zhenying and Wang, X. Sean , year =. Grounding

  29. [29]

    Feng, Yuxi and Li, Raymond and Fan, Zhenan and Carenini, Giuseppe and Pourreza, Mohammadreza and Zhang, Weiwei and Zhang, Yong , year =

  30. [30]

    Evaluating the

    F. Evaluating the. 2024 , publisher =

    2024

  31. [31]

    , year =

    Gan, Yujian and Purver, Matthew and Woodward, John R. , year =. A. Proceedings of the 1st

  32. [32]

    Ganesan, Poojah and Jha, Rajat Aayush and Roth, Dan and Gupta, Vivek , year =

  33. [33]

    Text-to-

    Gao, Dawei and Wang, Haibin and Li, Yaliang and Sun, Xiuyu and Qian, Yichen and Ding, Bolin and Zhou, Jingren , year =. Text-to-

  34. [34]

    2025 , eprint=

    XiYan-SQL: A Novel Multi-Generator Framework For Text-to-SQL , author=. 2025 , eprint=

    2025

  35. [35]

    Automatic Database Description Generation for

    Gao, Yingqi and Luo, Zhiling , year =. Automatic Database Description Generation for

  36. [36]

    Extractive

    Glass, Michael and Eyceoz, Mustafa and Subramanian, Dharmashankar and Rossiello, Gaetano and Vu, Long and Gliozzo, Alfio , year =. Extractive

  37. [37]

    Gong, Yue and Lei, Chuan and Qin, Xiao and Vaidya, Kapil and Narayanaswamy, Balakrishnan and Kraska, Tim , year =

  38. [38]

    2024 , publisher =

    Gorti, Satya Krishna and Gofman, Ilan and Liu, Zhaoyan and Wu, Jiapeng and Vouitsis, No. 2024 , publisher =

    2024

  39. [39]

    Guo, Jiaqi and Si, Ziliang and Wang, Yu and Liu, Qian and Fan, Ming and Lou, Jian-Guang and Yang, Zijiang and Liu, Ting , year =. Chase:. Proceedings of the 59th

  40. [40]

    Findings of the Association for Computational Linguistics: ACL 2025 , pages=

    SQLForge: Synthesizing Reliable and Diverse Data to Enhance Text-to-SQL Reasoning in LLMs , author=. Findings of the Association for Computational Linguistics: ACL 2025 , pages=

    2025

  41. [41]

    ST a R - SQL : Self-Taught Reasoner for Text-to- SQL

    He, Mingqian and Shen, Yongliang and Zhang, Wenqi and Peng, Qiuying and Wang, Jun and Lu, Weiming. ST a R - SQL : Self-Taught Reasoner for Text-to- SQL. ACL. 2025

    2025

  42. [42]

    Knowledge-to-

    Hong, Zijin and Yuan, Zheng and Chen, Hao and Zhang, Qinggang and Huang, Feiran and Huang, Xiao , year =. Knowledge-to-

  43. [43]

    Hong, Zijin and Yuan, Zheng and Zhang, Qinggang and Chen, Hao and Dong, Junnan and Huang, Feiran and Huang, Xiao , year =. Next-

  44. [44]

    Service-Oriented

    Hu, Wangsu and Tian, Jilei , year =. Service-Oriented. Findings of the

  45. [45]

    Exploring the

    Huang, Yiming and Guo, Jiyu and Mao, Wenxin and Gao, Cuiyun and Han, Peiyi and Liu, Chuanyi and Ling, Qing , year =. Exploring the

  46. [46]

    Huo, Siyu and Ma, Tengfei and Chen, Jie and Chang, Maria and Wu, Lingfei and Witbrock, Michael , year =. Graph. Proceedings of the

  47. [47]

    Proceedings of the 2024

    Katsimpras, Georgios and Paliouras, Georgios , year =. Proceedings of the 2024

    2024

  48. [48]

    2023 , journal =

    A Survey on Deep Learning Approaches for Text-to-. 2023 , journal =

    2023

  49. [49]

    Kobayashi, Hideo and Lan, Wuwei and Shi, Peng and Chang, Shuaichen and Guo, Jiang and Zhu, Henghui and Wang, Zhiguo and Ng, Patrick , year =. You

  50. [50]

    Bridging the

    Kong, Yonghui and Hu, Hongbing and Zhang, Dan and Chai, Siyuan and Zhang, Fan and Wang, Wei , year =. Bridging the

  51. [51]

    Kothyari, Mayank and Dhingra, Dhruva and Sarawagi, Sunita and Chakrabarti, Soumen , year =

  52. [52]

    Reinforcing

    Kulkarni, Atharv and Srikumar, Vivek , year =. Reinforcing

  53. [53]

    Kumar, Rahul and Dibbu, Amar Raja and Harsola, Shrutendra and Subrahmaniam, Vignesh and Modi, Ashutosh , year =

  54. [54]

    Lan, Wuwei and Wang, Zhiguo and Chauhan, Anuj and Zhu, Henghui and Li, Alexander and Guo, Jiang and Zhang, Sheng and Hang, Chung-Wei and Lilien, Joseph and Hu, Yiqun and Pan, Lin and Dong, Mingwen and Wang, Jun and Jiang, Jiarong and Ash, Stephen and Castelli, Vittorio and Ng, Patrick and Xiang, Bing , year =

  55. [55]

    Lee, Jinhyuk and Chen, Anthony and Dai, Zhuyun and Dua, Dheeru and Sachan, Devendra Singh and Boratko, Michael and Luan, Yi and Arnold, S. Can. 2024 , publisher =

    2024

  56. [56]

    Proceedings of the 31st International Conference on Computational Linguistics , pages=

    Mcs-sql: Leveraging multiple prompts and multiple-choice selection for text-to-sql generation , author=. Proceedings of the 31st International Conference on Computational Linguistics , pages=

  57. [57]

    Overview of the

    Lee, Gyubok and Kweon, Sunjun and Bae, Seongsu and Choi, Edward , year =. Overview of the

  58. [58]

    Lee, Jimin and Baek, Ingeol and Kim, Byeongjeong and Lee, Hwanhee , year =

  59. [59]

    The Thirteenth International Conference on Learning Representations , year=

    Spider 2.0: Evaluating Language Models on Real-World Enterprise Text-to-SQL Workflows , author=. The Thirteenth International Conference on Learning Representations , year=

  60. [60]

    Li, Haoyang and Zhang, Jing and Li, Cuiping and Chen, Hong , year =

  61. [61]

    Li, Minghan and Zhuang, Honglei and Hui, Kai and Qin, Zhen and Lin, Jimmy and Jagerman, Rolf and Wang, Xuanhui and Bendersky, Michael , year =. Can. Proceedings of the 47th

  62. [62]

    Li, Haoyang and Zhang, Jing and Liu, Hanbing and Fan, Ju and Zhang, Xiaokang and Zhu, Jun and Wei, Renjie and Pan, Hongyan and Li, Cuiping and Chen, Hong , year =

  63. [63]

    Li, Boyan and Luo, Yuyu and Chai, Chengliang and Li, Guoliang and Tang, Nan , year =. The

  64. [64]

    Li, Zhaodonghui and Yuan, Haitao and Wang, Huiming and Cong, Gao and Bing, Lidong , year =

  65. [65]

    Findings of the

    Li, Chunhui and Wang, Yifan and Wu, Zhen and Yu, Zhen and Zhao, Fei and Huang, Shujian and Dai, Xinyu , year =. Findings of the

  66. [66]

    Li, Zhishuai and Wang, Xiang and Zhao, Jingjing and Yang, Sun and Du, Guoqing and Hu, Xiaoru and Zhang, Bin and Ye, Yuxiao and Li, Ziyue and Zhao, Rui and Mao, Hangyu , year =

  67. [67]

    Li, Chaofan and Shao, Yingxia and Liu, Zheng , year =

  68. [68]

    Tapilot-

    Li, Jinyang and Huo, Nan and Gao, Yan and Shi, Jiayi and Zhao, Yingxiu and Qu, Ge and Wu, Yurong and Ma, Chenhao and Lou, Jian-Guang and Cheng, Reynold , year =. Tapilot-

  69. [69]

    Li, Zhenwen and Xie, Tao , year =. Using

  70. [70]

    Li, Boyan and Zhang, Jiayi and Fan, Ju and Xu, Yanwei and Chen, Chong and Tang, Nan and Luo, Yuyu , year =. Alpha-

  71. [71]

    2025 , issue_date =

    Li, Haoyang and Wu, Shang and Zhang, Xiaokang and Huang, Xinmei and Zhang, Jing and Jiang, Fuxin and Wang, Shuai and Zhang, Tieying and Chen, Jianjun and Shi, Rui and Chen, Hong and Li, Cuiping , title =. 2025 , issue_date =. doi:10.14778/3749646.3749723 , journal =

    work page doi:10.14778/3749646.3749723 2025

  72. [72]

    Li, Jinyang and Li, Xiaolong and Qu, Ge and Jacobsson, Per and Qin, Bowen and Hui, Binyuan and Si, Shuzheng and Huo, Nan and Xu, Xiaohan and Zhang, Yue and Tang, Ziwei and Li, Yuanshuai and Widjaja, Florensia and Zhu, Xintong and Zhou, Feige and Huang, Yongfeng and Papakonstantinou, Yannis and Ozcan, Fatma and Ma, Chenhao and Cheng, Reynold , year =

  73. [73]

    SCIENTIA SINICA Informationis , volume =

    梁, 清源 and 朱, 琪豪 and 孙, 泽宇 and 张, 路 and 张, 文杰 and 熊, 英飞 and 梁, 广泰 and 郁, 莲 , year =. SCIENTIA SINICA Informationis , volume =

  74. [74]

    Augmenting

    Liu, Qi and Ye, Zihuiwen and Yu, Tao and Song, Linfeng and Blunsom, Phil , year =. Augmenting. Findings of the

  75. [75]

    , year =

    Liu, Aiwei and Hu, Xuming and Wen, Lijie and Yu, Philip S. , year =. A Comprehensive Evaluation of

  76. [76]

    Divide and

    Liu, Xiping and Tan, Zhao , year =. Divide and

  77. [77]

    Uncovering and

    Liu, Yan and Gao, Yan and Su, Zhe and Chen, Xiaokang and Ash, Elliott and Lou, Jian-Guang , year =. Uncovering and

  78. [78]

    Liu, Xiping and Tan, Zhao , year =

  79. [79]

    IEEE Transactions on Knowledge and Data Engineering , year=

    A Survey of Text-to-SQL in the Era of LLMs: Where are we, and where are we going? , author=. IEEE Transactions on Knowledge and Data Engineering , year=

  80. [80]

    Liu, Tao and Zan, Hongying and Li, Yifan and Zhang, Dixuan and Kong, Lulu and Liu, Haixin and Hou, Jiaming and Zheng, Aoze and Li, Rui and et al. , year=

Showing first 80 references.