arxiv: 2509.03540 · v3 · submitted 2025-08-31 · 💻 cs.CL · cs.AI

Improving Factuality in LLMs via Inference-Time Knowledge Graph Construction

Shanglin Wu , Lihui Liu , Jinho D. Choi , Kai Shu This is my paper

Pith reviewed 2026-05-18 19:23 UTC · model grok-4.3

classification 💻 cs.CL cs.AI

keywords knowledge graphsfactualityretrieval-augmented generationlarge language modelsinference-time methodsquestion answeringfactual consistency

0 comments

The pith

Dynamically constructing knowledge graphs at inference time improves factual accuracy in large language models by structuring internal and external knowledge.

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

The paper argues that LLMs can produce more factually consistent answers by building and refining knowledge graphs during inference rather than relying on unstructured text retrieval. The process begins with prompting the model to extract a seed graph from the question, expands that graph using the model's own internal knowledge, and then selectively pulls in external facts to correct gaps and errors. This structured approach is tested on three factual question-answering benchmarks where it delivers consistent accuracy gains over standard retrieval baselines. A sympathetic reader would care because organizing knowledge into graphs supports better compositional reasoning and limits the sway of irrelevant details on the final output.

Core claim

By extracting a seed knowledge graph from the question via prompting, iteratively expanding it with the LLM's internal knowledge, and then selectively refining it through external retrieval, the framework enhances factual coverage, corrects inaccuracies, and produces more consistent answers than methods that treat knowledge as unstructured text.

What carries the argument

The inference-time knowledge graph that starts as a seed extracted from the question, expands via internal prompting, and receives selective external refinement to organize facts for accurate generation.

If this is right

Structured graphs reduce the influence of irrelevant information on LLM outputs.
Compositional reasoning improves because facts are linked explicitly rather than scattered in text passages.
External retrieval becomes more targeted once a partial internal graph already exists.
The method operates without retraining and remains scalable at inference time.
Interpretability increases because the constructed graph can be inspected to trace which facts informed the answer.

Where Pith is reading between the lines

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

The same graph-construction loop could be tested on tasks beyond QA such as multi-hop reasoning or long-form generation where structure matters.
If the quality of the internally expanded graph can be measured automatically, that metric might serve as an early stopping signal before external retrieval.
Hybrid internal-external graphs might generalize to domains where parametric memory alone is insufficient but full retrieval is costly.

Load-bearing premise

Prompting the LLM to extract and iteratively expand a seed knowledge graph produces a structure accurate and complete enough that external retrieval can refine it without introducing new errors or omitting key facts.

What would settle it

Running the method on the same three factual QA benchmarks and finding no gain or a drop in factual accuracy relative to the unstructured retrieval baselines would falsify the central claim.

Figures

Figures reproduced from arXiv: 2509.03540 by Jinho D. Choi, Kai Shu, Lihui Liu, Shanglin Wu.

**Figure 2.** Figure 2: Overview of our pipeline. (A) Graph initialization, in which the input question is parsed by the LLM to extract initial triplets. (B) Graph expansion iteratively explores breadth-first relations from seed entities to build a larger KG. (C) External retrieval, search is performed (e.g., using BM25 on the content returned from wikipedia and google search) to correct or extend selected triplets, which are mer… view at source ↗

**Figure 3.** Figure 3: Accuracy and graph size for five models across different hop counts. Solid lines represent accuracy, while [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

**Figure 4.** Figure 4: Accuracy and recall across different hop [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗

read the original abstract

Large Language Models (LLMs) often struggle with producing factually consistent answers due to limitations in their parametric memory. Retrieval-Augmented Generation (RAG) paradigms mitigate this issue by incorporating external knowledge at inference time. However, such methods typically handle knowledge as unstructured text, which reduces retrieval accuracy, hinders compositional reasoning, and amplifies the influence of irrelevant information on the factual consistency of LLM outputs. To overcome these limitations, we propose a novel framework that dynamically constructs and expands knowledge graphs (KGs) during inference, integrating both internal knowledge extracted from LLMs and external knowledge retrieved from external sources. Our method begins by extracting a seed KG from the question via prompting, followed by iterative expansion using the LLM's internal knowledge. The KG is then selectively refined through external retrieval, enhancing factual coverage and correcting inaccuracies. We evaluate our approach on three diverse Factual QA benchmarks, demonstrating consistent gains in factual accuracy over baselines. Our findings reveal that inference-time KG construction is a promising direction for enhancing LLM factuality in a structured, interpretable, and scalable manner.

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 outlines a practical inference-time pipeline for building and refining KGs to boost factual QA accuracy, but the abstract leaves the error-handling claims under-supported.

read the letter

The core contribution is a three-stage inference process: prompt the LLM to pull a seed KG from the question, expand it iteratively with the model's internal knowledge, then selectively pull external facts to refine the graph before final generation. This is a distinct combination of steps even if it draws from existing RAG and KG-prompting work. The authors show consistent gains across three factual QA benchmarks, which suggests the structured approach can reduce irrelevant noise compared with plain text retrieval. That part is useful for anyone trying to make retrieval more precise without retraining the model. The motivation section also does a clean job explaining why flat text hurts compositional reasoning. The idea of keeping knowledge as an explicit, editable graph at inference time is straightforward and interpretable, which is a plus for debugging. The main weakness is that the evaluation details are thin. The abstract mentions gains in factual accuracy but gives no numbers on metrics, baseline implementations, statistical tests, or how they measured whether the expansion loop added hallucinations before external refinement could fix them. The stress-test concern about error compounding during internal expansion is reasonable given that the same LLM is doing the early iterations. Without ablations that isolate the expansion step or triple-level precision checks against gold facts, it is difficult to know whether the refinement stage actually reverses those errors or sometimes amplifies them. This work is aimed at practitioners and researchers who want incremental improvements to RAG-style systems for reliable question answering. Readers already working on structured knowledge integration or inference-time methods will find the pipeline easy to understand and potentially replicable. It is solid enough on the idea level to deserve a serious referee, though the review would need to focus on stronger empirical validation of the graph quality at each stage. I would send it for review with a request for those ablations and clearer reporting of the factual accuracy numbers.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes an inference-time framework for dynamically constructing and expanding knowledge graphs to improve factual consistency in LLMs. The method extracts a seed KG from the input question via prompting, iteratively expands it using the LLM's internal parametric knowledge, selectively refines the graph through external retrieval, and uses the resulting structure to guide generation. Evaluation on three factual QA benchmarks is reported to yield consistent gains in factual accuracy relative to baselines.

Significance. If the central empirical claims are supported by detailed metrics, ablations, and controls for error accumulation, the work could advance structured alternatives to standard RAG by combining internal and external knowledge in an interpretable graph format. The inference-time KG construction offers potential advantages for compositional reasoning and reduced influence of irrelevant information. The absence of quantitative validation for the expansion and refinement stages currently limits assessment of whether the approach reliably mitigates rather than amplifies hallucinations.

major comments (2)

[Abstract and Evaluation section] Abstract and Evaluation section: the claim of 'consistent gains in factual accuracy over baselines' on three benchmarks provides no details on the precise metrics employed (e.g., exact match, F1, or hallucination rate), baseline implementations, number of runs, or statistical significance testing. This information is load-bearing for the central empirical claim.
[§3 (Proposed Method)] §3 (Proposed Method), iterative expansion step: no triple-level precision/recall against gold facts or ablation removing the expansion loop is reported. Without such evidence it remains unclear whether LLM-driven expansion compounds hallucinations before external refinement can correct them, directly affecting the soundness of the hybrid KG construction claim.

minor comments (2)

[§3] Clarify the exact prompting templates and stopping criteria for seed extraction and iterative expansion to support reproducibility.
[Figures] Ensure all figures showing KG examples include node/edge labels and legend for readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback, which highlights important areas for improving the transparency of our empirical results and the validation of the method components. We address each major comment below and describe the revisions planned for the next version of the manuscript.

read point-by-point responses

Referee: [Abstract and Evaluation section] Abstract and Evaluation section: the claim of 'consistent gains in factual accuracy over baselines' on three benchmarks provides no details on the precise metrics employed (e.g., exact match, F1, or hallucination rate), baseline implementations, number of runs, or statistical significance testing. This information is load-bearing for the central empirical claim.

Authors: We agree that the abstract and evaluation reporting would benefit from greater specificity to support the central claims. In the revised manuscript we will update the abstract to name the metrics explicitly (Exact Match and token-level F1 for factual accuracy, plus a hallucination rate computed via an external fact-verification model). The Evaluation section will be expanded to describe all baseline implementations (including standard RAG variants and prior KG-augmented methods), report results averaged over three independent runs with standard deviations, and include paired t-test p-values for statistical significance. These additions will be made without altering the existing experimental results. revision: yes
Referee: [§3 (Proposed Method)] §3 (Proposed Method), iterative expansion step: no triple-level precision/recall against gold facts or ablation removing the expansion loop is reported. Without such evidence it remains unclear whether LLM-driven expansion compounds hallucinations before external refinement can correct them, directly affecting the soundness of the hybrid KG construction claim.

Authors: We acknowledge that triple-level precision and recall against gold facts would offer direct insight into the expansion step. Constructing such gold annotations for dynamically generated triples, however, requires exhaustive manual labeling that exceeds the scope of the current experiments. To directly address the concern about error accumulation, we will add a new ablation in the revised manuscript that disables the iterative expansion loop and reports the resulting factual accuracy on all three benchmarks. We will also insert qualitative examples and discussion in §3 illustrating how the subsequent selective refinement stage corrects inaccuracies introduced during expansion. These changes will provide quantitative evidence on the net contribution of the expansion component. revision: partial

Circularity Check

0 steps flagged

No circularity in derivation or evaluation chain

full rationale

The paper describes an inference-time procedure that extracts a seed KG by prompting, expands it internally, then refines it with external retrieval before generation. The central claim is an empirical improvement on three external factual QA benchmarks. No equations, fitted parameters, or self-referential predictions appear in the abstract or method description. No load-bearing self-citations, uniqueness theorems, or ansatzes imported from prior author work are invoked. The result is therefore not equivalent to its inputs by construction and rests on external benchmark outcomes rather than internal redefinitions.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the untested premise that LLMs can reliably perform KG extraction and expansion without introducing factual errors that external retrieval cannot fully correct.

axioms (2)

domain assumption LLMs can extract a seed knowledge graph from a question via prompting that is accurate enough to serve as a foundation for further expansion.
Invoked in the description of the first step of the method.
domain assumption Iterative expansion using the LLM's internal knowledge improves coverage without proportionally increasing hallucinations.
Stated as part of the iterative expansion phase.

pith-pipeline@v0.9.0 · 5718 in / 1342 out tokens · 28872 ms · 2026-05-18T19:23:00.714971+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

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

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Our method begins by extracting a seed KG from the question via prompting, followed by iterative expansion using the LLM's internal knowledge. The KG is then selectively refined through external retrieval...
IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We evaluate our approach on three diverse Factual QA benchmarks, demonstrating consistent gains in factual accuracy over baselines.

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.

Reference graph

Works this paper leans on

44 extracted references · 44 canonical work pages · 2 internal anchors

[1]

Shengnan An, Zexiong Ma, Zeqi Lin, Nanning Zheng, Jian-Guang Lou, and Weizhu Chen. 2023. Learning from mistakes makes llm better reasoner. arXiv preprint arXiv:2310.20689

work page arXiv 2023
[2]

Jinheon Baek, Alham Fikri Aji, and Amir Saffari. 2023. Knowledge-augmented language model prompting for zero-shot knowledge graph question answering. arXiv preprint arXiv:2306.04136

work page arXiv 2023
[3]

Xuefeng Bai, Song He, Yi Li, Yabo Xie, Xin Zhang, Wenli Du, and Jian-Rong Li. 2025. Construction of a knowledge graph for framework material enabled by large language models and its application. npj Computational Materials, 11(1):51

work page 2025
[4]

Maciej Besta, Nils Blach, Ales Kubicek, Robert Gerstenberger, Michal Podstawski, Lukas Gianinazzi, Joanna Gajda, Tomasz Lehmann, Hubert Niewiadomski, Piotr Nyczyk, and 1 others. 2024. Graph of thoughts: Solving elaborate problems with large language models. In Proceedings of the AAAI conference on artificial intelligence, volume 38, pages 17682--17690

work page 2024
[5]

Antoine Bordes, Nicolas Usunier, Sumit Chopra, and Jason Weston. 2015. Large-scale simple question answering with memory networks. In Proceedings of the 24th International Conference on Artificial Intelligence (IJCAI)

work page 2015
[6]

Roi Cohen, Mor Geva, Jonathan Berant, and Amir Globerson. 2023. Crawling the internal knowledge-base of language models. arXiv preprint arXiv:2301.12810

work page arXiv 2023
[7]

Linyi Ding, Sizhe Zhou, Jinfeng Xiao, and Jiawei Han. 2024. Automated construction of theme-specific knowledge graphs. arXiv preprint arXiv:2404.19146

work page arXiv 2024
[8]

Dheeru Dua, Yizhong Wang, Pradeep Dasigi, Gabriel Stanovsky, Sameer Singh, and Matt Gardner. 2019. Drop: A reading comprehension benchmark requiring discrete reasoning over paragraphs. NAACL

work page 2019
[9]

Mohnish Dubey, Prateek Banerjee, Ahmed Abdelkawi, and Jens Lehmann. 2019. Lc-quad 2.0: A large dataset for complex question answering over wikidata and dbpedia. The Semantic Web

work page 2019
[10]

Jiazhan Feng, Ruochen Xu, Junheng Hao, Hiteshi Sharma, Yelong Shen, Dongyan Zhao, and Weizhu Chen. 2023. Language models can be logical solvers. arXiv preprint arXiv:2311.06158

work page arXiv 2023
[11]

Zorik Gekhman, Eyal Ben David, Hadas Orgad, Eran Ofek, Yonatan Belinkov, Idan Szpektor, Jonathan Herzig, and Roi Reichart. 2025. Inside-out: Hidden factual knowledge in llms. arXiv preprint arXiv:2503.15299

work page arXiv 2025
[12]

Haoyu Han, Yu Wang, Harry Shomer, Kai Guo, Jiayuan Ding, Yongjia Lei, Mahantesh Halappanavar, Ryan A Rossi, Subhabrata Mukherjee, Xianfeng Tang, and 1 others. 2024. Retrieval-augmented generation with graphs (graphrag). arXiv preprint arXiv:2501.00309

work page internal anchor Pith review Pith/arXiv arXiv 2024
[13]

Shibo Hao, Yi Gu, Haotian Luo, Tianyang Liu, Xiyan Shao, Xinyuan Wang, Shuhua Xie, Haodi Ma, Adithya Samavedhi, Qiyue Gao, and 1 others. 2024. Llm reasoners: New evaluation, library, and analysis of step-by-step reasoning with large language models. arXiv preprint arXiv:2404.05221

work page arXiv 2024
[14]

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

work page 2025
[15]

Nikitas Karanikolas, Eirini Manga, Nikoletta Samaridi, Eleni Tousidou, and Michael Vassilakopoulos. 2023. Large language models versus natural language understanding and generation. In Proceedings of the 27th Pan-Hellenic Conference on Progress in Computing and Informatics, pages 278--290

work page 2023
[16]

Vamsi Krishna Kommineni, Birgitta K \"o nig-Ries, and Sheeba Samuel. 2024. From human experts to machines: An llm supported approach to ontology and knowledge graph construction. arXiv preprint arXiv:2403.08345

work page arXiv 2024
[17]

Pranjal Kumar. 2024. Large language models (llms): survey, technical frameworks, and future challenges. Artificial Intelligence Review, 57(10):260

work page 2024
[18]

Zhicheng Lee, Shulin Cao, Jinxin Liu, Jiajie Zhang, Weichuan Liu, Xiaoyin Che, Lei Hou, and Juanzi Li. 2025. Rearag: Knowledge-guided reasoning enhances factuality of large reasoning models with iterative retrieval augmented generation. arXiv preprint arXiv:2503.21729

work page arXiv 2025
[19]

Qian Li, Zhuo Chen, Cheng Ji, Shiqi Jiang, and Jianxin Li. 2024. Llm-based multi-level knowledge generation for few-shot knowledge graph completion. In Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence, volume 271494703

work page 2024
[20]

Jie Ma, Zhitao Gao, Qi Chai, Wangchun Sun, Pinghui Wang, Hongbin Pei, Jing Tao, Lingyun Song, Jun Liu, Chen Zhang, and 1 others. 2025. Debate on graph: a flexible and reliable reasoning framework for large language models. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 24768--24776

work page 2025
[21]

Steven Moore, Richard Tong, Anjali Singh, Zitao Liu, Xiangen Hu, Yu Lu, Joleen Liang, Chen Cao, Hassan Khosravi, Paul Denny, and 1 others. 2023. Empowering education with llms-the next-gen interface and content generation. In International Conference on Artificial Intelligence in Education, pages 32--37. Springer

work page 2023
[22]

Mohamed Nejjar, Luca Zacharias, Fabian Stiehle, and Ingo Weber. 2023. Llms for science: Usage for code generation and data analysis. Journal of Software: Evolution and Process, page e2723

work page 2023
[23]

Jixuan Nie, Xia Hou, Wenfeng Song, Xuan Wang, Xinyu Zhang, Xingliang Jin, Shuozhe Zhang, and Jiaqi Shi. 2024. Knowledge graph efficient construction: Embedding chain-of-thought into llms. Proceedings of the VLDB Endowment. ISSN, 2150:8097

work page 2024
[24]

Shirui Pan, Linhao Luo, Yufei Wang, Chen Chen, Jiapu Wang, and Xindong Wu. 2024. Unifying large language models and knowledge graphs: A roadmap. IEEE Transactions on Knowledge and Data Engineering, 36(7):3580--3599

work page 2024
[25]

Stephen Robertson, Hugo Zaragoza, and 1 others. 2009. The probabilistic relevance framework: Bm25 and beyond. Foundations and Trends in Information Retrieval , 3(4):333--389

work page 2009
[26]

Abulhair Saparov and He He. 2022. Language models are greedy reasoners: A systematic formal analysis of chain-of-thought. arXiv preprint arXiv:2210.01240

work page arXiv 2022
[27]

Priyanka Sen, Sandeep Mavadia, and Amir Saffari. 2023. Knowledge graph-augmented language models for complex question answering. In Proceedings of the 1st Workshop on Natural Language Reasoning and Structured Explanations (NLRSE), pages 1--8

work page 2023
[28]

Haitao Sun and et al. 2022. Conditionalqa: A dataset for complex conditional question answering over narratives. arXiv preprint arXiv:2202.12276

work page arXiv 2022
[29]

Jiashuo Sun, Chengjin Xu, Lumingyuan Tang, Saizhuo Wang, Chen Lin, Yeyun Gong, Lionel M Ni, Heung-Yeung Shum, and Jian Guo. 2023. Think-on-graph: Deep and responsible reasoning of large language model on knowledge graph. arXiv preprint arXiv:2307.07697

work page arXiv 2023
[30]

Alon Talmor and Jonathan Berant. 2018. The web as a knowledge-base for answering complex questions. In NAACL

work page 2018
[31]

Han Wang, Archiki Prasad, Elias Stengel-Eskin, and Mohit Bansal. 2025. Retrieval-augmented generation with conflicting evidence. arXiv preprint arXiv:2504.13079

work page arXiv 2025
[32]

Yuxia Wang, Minghan Wang, Hasan Iqbal, Georgi Georgiev, Jiahui Geng, and Preslav Nakov. 2024 a . Openfactcheck: A unified framework for factuality evaluation of llms. arXiv preprint arXiv:2405.05583

work page arXiv 2024
[33]

Yuxia Wang, Minghan Wang, Muhammad Arslan Manzoor, Fei Liu, Georgi Georgiev, Rocktim Jyoti Das, and Preslav Nakov. 2024 b . Factuality of large language models in the year 2024. CoRR

work page 2024
[34]

Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, and 1 others. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35:24824--24837

work page 2022
[35]

Jason et al. Wei. 2024. Simpleqa: Evaluating factual consistency in frontier language models. arXiv preprint arXiv:2403.12345

work page arXiv 2024
[36]

Yike Wu, Nan Hu, Sheng Bi, Guilin Qi, Jie Ren, Anhuan Xie, and Wei Song. 2023. Retrieve-rewrite-answer: A kg-to-text enhanced llms framework for knowledge graph question answering. arXiv preprint arXiv:2309.11206

work page arXiv 2023
[37]

Yao Xu, Shizhu He, Jiabei Chen, Zihao Wang, Yangqiu Song, Hanghang Tong, Guang Liu, Kang Liu, and Jun Zhao. 2024. Generate-on-graph: Treat llm as both agent and kg in incomplete knowledge graph question answering. arXiv preprint arXiv:2404.14741

work page arXiv 2024
[38]

Zhilin Yang, Peng Qi, Saizheng Zhang, Yoshua Bengio, William W Cohen, Ruslan Salakhutdinov, and Christopher D Manning. 2018. Hotpotqa: A dataset for diverse, explainable multi-hop question answering. arXiv preprint arXiv:1809.09600

work page internal anchor Pith review Pith/arXiv arXiv 2018
[39]

Yao Yao, Zuchao Li, and Hai Zhao. 2023. Beyond chain-of-thought, effective graph-of-thought reasoning in language models. arXiv preprint arXiv:2305.16582

work page arXiv 2023
[40]

Wen-tau Yih, Ming-Wei Chang, Xiaodong He, and Jianfeng Gao. 2016. The value of semantic parse labeling for knowledge base question answering. In ACL

work page 2016
[41]

Siyun Zhao, Yuqing Yang, Zilong Wang, Zhiyuan He, Luna K Qiu, and Lili Qiu. 2024. Retrieval augmented generation (rag) and beyond: A comprehensive survey on how to make your llms use external data more wisely. arXiv preprint arXiv:2409.14924

work page arXiv 2024
[42]

Ce Zheng, Lei Li, Qingxiu Dong, Yuxuan Fan, Zhiyong Wu, Jingjing Xu, and Baobao Chang. 2023. Can we edit factual knowledge by in-context learning? arXiv preprint arXiv:2305.12740

work page arXiv 2023
[43]

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...

work page
[44]

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...

work page