pith. sign in

arxiv: 2412.04272 · v5 · submitted 2024-12-05 · 💻 cs.IR · cs.AI

PoTable: Towards Systematic Thinking via Plan-then-Execute Stage Reasoning on Tables

Qingyang Mao , Qi Liu , Zhi Li , Mingyue Cheng , Zheng Zhang , Rui Li This is my paper

Pith reviewed 2026-05-23 08:20 UTC · model grok-4.3

classification 💻 cs.IR cs.AI
keywords table reasoninglarge language modelsplan-then-executesystematic thinkingcode generationexplainabilityWikiTQTabFact
0
0 comments X

The pith

PoTable brings systematic thinking to table reasoning by using staged planning before code execution.

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

Existing LLM methods for table reasoning rely on step-by-step thinking guided by task semantics but often miss systematic structure, leading to errors in complex cases. PoTable addresses this by defining several distinct analytical stages, each with a clear objective. For each stage, it first plans an operation chain and then executes it through code generation, real-time running, and feedback. This produces reliable results in the form of accurate, step-wise commented, and fully executable programs. Experiments on WikiTQ and TabFact datasets show gains in accuracy and explainability.

Core claim

PoTable is a novel stage-oriented plan-then-execute approach for table reasoning that incorporates systematic thinking by involving several distinct analytical stages with clear objectives, planning the operation chain based on the stage objective, and executing operations sequentially through code generation, real-time running and feedback processing, resulting in reliable table reasoning results with highly accurate, step-wise commented and completely executable programs.

What carries the argument

The plan-then-execute mechanism that first plans the operation chain for each stage objective and then executes it via code generation with feedback.

If this is right

  • Produces reliable table reasoning results with highly accurate, step-wise commented and completely executable programs.
  • Mirrors the workflow of a professional data analyst.
  • Offers advantages in both accuracy and explainability.
  • Shows effectiveness, efficiency and explainability on four datasets from WikiTQ and TabFact benchmarks.

Where Pith is reading between the lines

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

  • Applying similar stage-based planning could help in other LLM reasoning domains like text or code generation.
  • The commented executable programs allow users to verify and modify reasoning steps manually.
  • Real-time feedback during execution may help correct errors early in the process.

Load-bearing premise

That breaking the task into several distinct analytical stages with clear objectives and following a plan-then-execute process will prevent omitted steps and disorganized logic while ensuring correct code execution.

What would settle it

Running PoTable on a complex table query where it produces an incomplete program or incorrect result despite the stages, or where accuracy does not exceed standard step-by-step LLM methods.

Figures

Figures reproduced from arXiv: 2412.04272 by Mingyue Cheng, Qi Liu, Qingyang Mao, Rui Li, Zheng Zhang, Zhi Li.

Figure 1
Figure 1. Figure 1: Illustrations of (a) two table reasoning tasks, (b) general step-by-step thinking in typical LLM-based table reasoning [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of our proposed PoTable, a novel LLM￾based table reasoning method that realizes systematic think￾ing. PoTable follows stage-oriented thinking including five analytical stages with relevant objectives and instructions: initialization, row selection, data type cleaning, reasoning and final answering. To achieve each stage-specific goal, PoTable integrates an LLM and a Python interpreter to con￾d… view at source ↗
Figure 3
Figure 3. Figure 3: Specifically, the prompting template contains the tabular [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Accuracy results (%) in the ablation study of the different stage division settings employed in [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: A case study of an evaluated sample from WikiTQ (T) with its generated Python program and output answer. The [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: A case study of an evaluated sample from TabFact (C) with the generated operation chains of [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
read the original abstract

In recent years, table reasoning has garnered substantial research interest, particularly regarding its integration with Large Language Models (LLMs), which have revolutionized natural language applications. Existing LLM-based studies typically achieve step-by-step thinking for table reasoning guided by task semantics. While these approaches emphasize autonomous exploration and enhance fine-grained table understanding, they often overlook systematic thinking in the reasoning process. This oversight can lead to omitted steps, disorganized logic and misleading results, especially in complex scenarios. In this paper, we propose PoTable, a novel stage-oriented plan-then-execute approach that incorporates systematic thinking into table reasoning. Specifically, PoTable involves several distinct analytical stages with clear objectives to provide adequate guidance. To accomplish stage-specific goals, PoTable employs a plan-then-execute mechanism: it first plans the operation chain based on the stage objective, and then executes operations sequentially through code generation, real-time running and feedback processing. Consequently, PoTable produces reliable table reasoning results with highly accurate, step-wise commented and completely executable programs. It mirrors the workflow of a professional data analyst, offering advantages in both accuracy and explainability. Finally, we conduct extensive experiments on four datasets from the WikiTQ and TabFact benchmarks, where the results demonstrate the effectiveness, efficiency and explainability of PoTable. Our code is available at: https://github.com/Double680/PoTable.

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 proposes PoTable, a stage-oriented plan-then-execute framework for LLM-based table reasoning. It decomposes reasoning into distinct analytical stages with explicit objectives, generates operation-chain plans per stage, and executes them via sequential code generation, runtime execution, and feedback processing. The central claim is that this produces reliable, step-wise commented, fully executable programs that improve accuracy and explainability over prior methods, mirroring professional data-analyst workflows; experiments on four WikiTQ/TabFact datasets are said to demonstrate effectiveness, efficiency, and explainability, with code released publicly.

Significance. If the empirical results hold, the work offers a concrete mechanism for injecting systematic, multi-stage planning into LLM table reasoning, with the public code release enabling direct reproducibility and the production of executable programs providing a clear explainability advantage over black-box chain-of-thought baselines.

major comments (2)
  1. [Abstract, §4] Abstract and §4 (Experiments): the claim that experiments on four WikiTQ/TabFact datasets 'demonstrate the effectiveness' is stated without any quantitative results, error analysis, or ablation on the feedback-processing component; this leaves the central reliability claim only moderately supported.
  2. [§3.2] §3.2 (Plan-then-Execute Mechanism): the description of how real-time feedback is processed to correct LLM-generated code errors is high-level; without concrete examples or quantitative breakdown of error types mitigated, it is difficult to assess whether the mechanism reliably addresses the weakest assumption that stages prevent omitted steps and disorganized logic in complex tables.
minor comments (2)
  1. [§3] Notation for stage objectives and operation chains could be formalized with a small diagram or pseudocode to improve clarity.
  2. [§4] The four datasets are referenced only by benchmark names; listing their exact names and sizes in §4 would aid reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and the recommendation for minor revision. We address each major comment below and will revise the manuscript accordingly to strengthen the presentation of results and the description of the feedback mechanism.

read point-by-point responses

  1. Referee: [Abstract, §4] Abstract and §4 (Experiments): the claim that experiments on four WikiTQ/TabFact datasets 'demonstrate the effectiveness' is stated without any quantitative results, error analysis, or ablation on the feedback-processing component; this leaves the central reliability claim only moderately supported.

    Authors: Section 4 already reports quantitative accuracy results on the four datasets with comparisons to baselines. The abstract summarizes these outcomes at a high level, which is conventional. We agree, however, that an explicit ablation isolating the feedback-processing component together with a concise error analysis would provide stronger support for the reliability claims. We will add both to the revised manuscript. revision: yes

  2. Referee: [§3.2] §3.2 (Plan-then-Execute Mechanism): the description of how real-time feedback is processed to correct LLM-generated code errors is high-level; without concrete examples or quantitative breakdown of error types mitigated, it is difficult to assess whether the mechanism reliably addresses the weakest assumption that stages prevent omitted steps and disorganized logic in complex tables.

    Authors: Section 3.2 and the accompanying figures describe the feedback loop at the architectural level, with the full implementation released in the public code repository. To improve clarity, we will insert concrete examples of error correction together with a quantitative breakdown of error categories mitigated by the feedback step in the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper proposes PoTable, a stage-oriented plan-then-execute method for LLM-based table reasoning, described directly in the abstract and method without reference to equations, fitted parameters, or predictions that reduce to inputs. It evaluates the approach on external benchmarks (WikiTQ/TabFact) and releases code. No self-citation chains, uniqueness theorems, or ansatzes are invoked as load-bearing; the derivation chain is a standard empirical description of a new workflow, self-contained against external data.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard LLM capabilities for code generation and the engineering choice of stages; no free parameters, new entities, or non-standard axioms are introduced.

axioms (1)
  • domain assumption Large language models can generate correct executable code for table operations when given a planned operation chain and feedback from execution.
    The plan-then-execute execution step depends on this capability.

pith-pipeline@v0.9.0 · 5781 in / 1439 out tokens · 31268 ms · 2026-05-23T08:20:27.997161+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. When TableQA Meets Noise: A Dual Denoising Framework for Complex Questions and Large-scale Tables

    cs.CL 2025-09 unverdicted novelty 4.0

    EnoTab is a dual denoising framework for TableQA that performs evidence-based question denoising via semantic unit decomposition and evidence tree-guided table pruning with post-order rollback to improve performance o...

Reference graph

Works this paper leans on

46 extracted references · 46 canonical work pages · cited by 1 Pith paper · 2 internal anchors

  1. [1]

    Rami Aly, Zhijiang Guo, Michael Schlichtkrull, James Thorne, Andreas Vlachos, Christos Christodoulopoulos, Oana Cocarascu, and Arpit Mittal. 2021. Feverous: Fact extraction and verification over unstructured and structured information. In NeurIPS Datasets and Benchmarks Track

    work page 2021

  2. [2]

    Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Mateusz Litwin...

    work page 2020

  3. [3]

    Yihan Cao, Shuyi Chen, Ryan Liu, Zhiruo Wang, and Daniel Fried. 2023. API- Assisted Code Generation for Question Answering on Varied Table Structures. In EMNLP. 14536–14548

    work page 2023

  4. [4]

    Wenhu Chen, Hongmin Wang, Jianshu Chen, Yunkai Zhang, Hong Wang, Shiyang Li, Xiyou Zhou, and William Yang Wang. 2020. Tabfact: A large-scale dataset for table-based fact verification. In ICLR. OpenReview.net. https://openreview.net/ forum?id=rkeJRhNYDH

    work page 2020

  5. [5]

    Xinyun Chen, Maxwell Lin, Nathanael Schärli, and Denny Zhou. 2024. Teaching Large Language Models to Self-Debug. In ICLR

    work page 2024

  6. [6]

    Yukang Chen, Shengju Qian, Haotian Tang, Xin Lai, Zhijian Liu, Song Han, and Jiaya Jia. 2024. LongLoRA: Efficient Fine-tuning of Long-Context Large Language Models. In ICLR

    work page 2024

  7. [7]

    Zhiyu Chen, Shiyang Li, Charese Smiley, Zhiqiang Ma, Sameena Shah, and William Yang Wang. 2022. ConvFinQA: Exploring the Chain of Numerical Reasoning in Conversational Finance Question Answering. In EMNLP. 6279– 6292

    work page 2022

  8. [8]

    Mingyue Cheng, Hao Zhang, Jiqian Yang, Qi Liu, Li Li, Xin Huang, Liwei Song, Zhi Li, Zhenya Huang, and Enhong Chen. 2024. Towards Personalized Evaluation of Large Language Models with An Anonymous Crowd-Sourcing Platform. In WWW. 1035–1038

    work page 2024

  9. [9]

    Zhoujun Cheng, Haoyu Dong, Zhiruo Wang, Ran Jia, Jiaqi Guo, Yan Gao, Shi Han, Jian-Guang Lou, and Dongmei Zhang. 2022. HiTab: A Hierarchical Table Dataset for Question Answering and Natural Language Generation. In ACL. 1094–1110

    work page 2022

  10. [10]

    Zhoujun Cheng, Tianbao Xie, Peng Shi, Chengzu Li, Rahul Nadkarni, Yushi Hu, Caiming Xiong, Dragomir Radev, Mari Ostendorf, Luke Zettlemoyer, et al. 2023. Binding Language Models in Symbolic Languages. In ICLR

    work page 2023

  11. [11]

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In NAACL-HLT, Jill Burstein, Christy Doran, and Thamar Solorio (Eds.). Association for Computational Linguistics, 4171–4186

    work page 2019

  12. [12]

    Haoyu Dong and Zhiruo Wang. 2024. Large language models for tabular data: Progresses and future directions. In SIGIR. 2997–3000

    work page 2024

  13. [13]

    Usama Fayyad, Gregory Piatetsky-Shapiro, and Padhraic Smyth. 1996. The KDD process for extracting useful knowledge from volumes of data. Commun. ACM 39, 11 (1996), 27–34

    work page 1996

  14. [14]

    Jonathan Herzig, Pawel Krzysztof Nowak, Thomas Mueller, Francesco Piccinno, and Julian Eisenschlos. 2020. TaPas: Weakly Supervised Table Parsing via Pre- training. In ACL. 4320–4333

    work page 2020

  15. [15]

    Congyun Jin, Ming Zhang, Weixiao Ma, Yujiao Li, Yingbo Wang, Yabo Jia, Yuliang Du, Tao Sun, Haowen Wang, Cong Fan, et al. 2024. RJUA-MedDQA: A Multimodal Benchmark for Medical Document Question Answering and Clinical Reasoning. In KDD. 5218–5229

    work page 2024

  16. [16]

    Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Veselin Stoyanov, and Luke Zettlemoyer. 2020. BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension. In ACL. 7871–7880

    work page 2020

  17. [17]

    Hongxin Li, Jingran Su, Yuntao Chen, Qing Li, and ZHAO-XIANG ZHANG. 2023. SheetCopilot: Bringing software productivity to the next level through large language models. NeurIPS 36 (2023)

    work page 2023

  18. [18]

    Qian Liu, Bei Chen, Jiaqi Guo, Morteza Ziyadi, Zeqi Lin, Weizhu Chen, and Jian-Guang Lou. 2022. TAPEX: Table Pre-training via Learning a Neural SQL Executor. In ICLR

    work page 2022

  19. [19]

    Weizheng Lu, Jiaming Zhang, Jing Zhang, and Yueguo Chen. 2025. Large language model for table processing: A survey. Frontiers of Computer Science (2025), 1–17

    work page 2025

  20. [20]

    Xinyuan Lu, Liangming Pan, Qian Liu, Preslav Nakov, and Min-Yen Kan. 2023. SCITAB: A Challenging Benchmark for Compositional Reasoning and Claim Verification on Scientific Tables. InEMNLP. 7787–7813

    work page 2023

  21. [21]

    Zeyao Ma, Bohan Zhang, Jing Zhang, Jifan Yu, Xiaokang Zhang, Xiaohan Zhang, Sijia Luo, Xi Wang, and Jie Tang. 2024. SpreadsheetBench: Towards Challenging Real World Spreadsheet Manipulation. arXiv preprint arXiv:2406.14991 (2024)

    work page arXiv 2024

  22. [22]

    Gonzalo Mariscal, Oscar Marban, and Covadonga Fernandez. 2010. A survey of data mining and knowledge discovery process models and methodologies. The Knowledge Engineering Review 25, 2 (2010), 137–166

    work page 2010

  23. [23]

    Md Nahid and Davood Rafiei. 2024. TabSQLify: Enhancing Reasoning Capabilities of LLMs Through Table Decomposition. In NAACL. 5725–5737

    work page 2024

  24. [24]

    Anusri Pampari, Preethi Raghavan, Jennifer Liang, and Jian Peng. 2018. emrQA: A Large Corpus for Question Answering on Electronic Medical Records. In EMNLP. 2357–2368

    work page 2018

  25. [25]

    Panupong Pasupat and Percy Liang. 2015. Compositional Semantic Parsing on Semi-Structured Tables. In ACL-IJCNLP. 1470–1480

    work page 2015

  26. [26]

    Mohammadreza Pourreza and Davood Rafiei. 2023. Din-sql: Decomposed in- context learning of text-to-sql with self-correction. NeurIPS 36 (2023)

    work page 2023

  27. [27]

    Brandon Smock, Rohith Pesala, and Robin Abraham. 2022. PubTables-1M: To- wards comprehensive table extraction from unstructured documents. In CVPR. 4634–4642

    work page 2022

  28. [28]

    Yuan Sui, Mengyu Zhou, Mingjie Zhou, Shi Han, and Dongmei Zhang. 2024. Table meets llm: Can large language models understand structured table data? a benchmark and empirical study. In WSDM. 645–654

    work page 2024

  29. [29]

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yas- mine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhos- ale, et al. 2023. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  30. [30]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. Attention is All you Need. In NeurIPS. 5998–6008

    work page 2017

  31. [31]

    Zhiruo Wang, Haoyu Dong, Ran Jia, Jia Li, Zhiyi Fu, Shi Han, and Dongmei Zhang

    work page

  32. [32]

    Tuta: Tree-based transformers for generally structured table pre-training. In KDD. 1780–1790

    work page

  33. [33]

    Zhiruo Wang, Daniel Fried, and Graham Neubig. 2024. Trove: Inducing veri- fiable and efficient toolboxes for solving programmatic tasks. arXiv preprint arXiv:2401.12869 (2024)

    work page arXiv 2024

  34. [34]

    Zilong Wang, Hao Zhang, Chun-Liang Li, Julian Martin Eisenschlos, Vincent Perot, Zifeng Wang, Lesly Miculicich, Yasuhisa Fujii, Jingbo Shang, Chen-Yu Lee, et al. 2024. Chain-of-Table: Evolving Tables in the Reasoning Chain for Table Understanding. In ICLR. OpenReview.net. https://openreview.net/forum?id= 4L0xnS4GQM

    work page 2024

  35. [35]

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. 2022. Chain-of-thought prompting elicits reasoning in large language models. NeurIPS 35 (2022), 24824–24837

    work page 2022

  36. [36]

    Zhiyu Yang, Zihan Zhou, Shuo Wang, Xin Cong, Xu Han, Yukun Yan, Zheng- hao Liu, Zhixing Tan, Pengyuan Liu, Dong Yu, Zhiyuan Liu, Xiaodong Shi, and Maosong Sun. 2024. MatPlotAgent: Method and Evaluation for LLM-Based Agen- tic Scientific Data Visualization. In ACL Findings. Association for Computational Linguistics, 11789–11804

    work page 2024

  37. [37]

    Yunhu Ye, Binyuan Hui, Min Yang, Binhua Li, Fei Huang, and Yongbin Li. 2023. Large language models are versatile decomposers: Decomposing evidence and questions for table-based reasoning. In SIGIR. 174–184

    work page 2023

  38. [38]

    Pengcheng Yin, Graham Neubig, Wen-tau Yih, and Sebastian Riedel. 2020. TaBERT: Pretraining for Joint Understanding of Textual and Tabular Data. In ACL. 8413–8426

    work page 2020

  39. [39]

    Tao Yu, Rui Zhang, Kai Yang, Michihiro Yasunaga, Dongxu Wang, Zifan Li, James Ma, Irene Li, Qingning Yao, Shanelle Roman, et al. 2018. Spider: A Large-Scale Human-Labeled Dataset for Complex and Cross-Domain Semantic Parsing and Text-to-SQL Task. In EMNLP. 3911–3921

    work page 2018

  40. [40]

    ChengXiang Zhai. 2024. Large language models and future of information retrieval: opportunities and challenges. In SIGIR. 481–490

    work page 2024

  41. [41]

    Shuo Zhang, Zhuyun Dai, Krisztian Balog, and Jamie Callan. 2020. Summarizing and exploring tabular data in conversational search. In SIGIR. 1537–1540

    work page 2020

  42. [42]

    Tianshu Zhang, Xiang Yue, Yifei Li, and Huan Sun. 2024. TableLlama: Towards Open Large Generalist Models for Tables. In NAACL-HLT. 6024–6044

    work page 2024

  43. [43]

    Yunjia Zhang, Jordan Henkel, Avrilia Floratou, Joyce Cahoon, Shaleen Deep, and Jignesh M Patel. 2024. ReAcTable: Enhancing ReAct for Table Question Answering. VLDB 17, 8 (2024), 1981–1994

    work page 2024

  44. [44]

    Weichao Zhao, Hao Feng, Qi Liu, Jingqun Tang, Shu Wei, Binghong Wu, Lei Liao, Yongjie Ye, Hao Liu, Wengang Zhou, et al. 2024. Tabpedia: Towards com- prehensive visual table understanding with concept synergy. arXiv preprint arXiv:2406.01326 (2024)

    work page arXiv 2024

  45. [45]

    Wayne Xin Zhao, Kun Zhou, Junyi Li, Tianyi Tang, Xiaolei Wang, Yupeng Hou, Yingqian Min, Beichen Zhang, Junjie Zhang, Zican Dong, et al. 2023. A survey of large language models. arXiv preprint arXiv:2303.18223 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  46. [46]

    Fengbin Zhu, Wenqiang Lei, Youcheng Huang, Chao Wang, Shuo Zhang, Jiancheng Lv, Fuli Feng, and Tat-Seng Chua. 2021. TAT-QA: A Question An- swering Benchmark on a Hybrid of Tabular and Textual Content in Finance. In ACL-IJCNLP. 3277–3287

    work page 2021