pith. sign in

arxiv: 2605.29940 · v1 · pith:SUUY5GLXnew · submitted 2026-05-28 · 💻 cs.AI

Make LLM Learn to Synthesize from Streaming Experiences through Feedback

Zhenlin Hu , Yan Wang , Zhen Bi , Zihao Xue , Bingyu Zhu , Longtao Huang , Xiongtao Zhang , Zeyu Yang
show 2 more authors
Zhixuan Chu Jungang Lou
This is my paper

Pith reviewed 2026-06-29 07:38 UTC · model grok-4.3

classification 💻 cs.AI
keywords synthetic data generationstreaming tasksexperience accumulationcross-task transferfeedback learningLLM synthesistask streams
0
0 comments X

The pith

LLMs improve synthetic data generation on new tasks by accumulating experience from feedback on earlier ones in sequence.

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

The paper introduces StreamSynth, a setting in which synthesis tasks arrive one after another and signals from completed tasks can inform how to handle the next. It proposes SynLearner as a framework that pushes the model to try varied synthesis patterns, incorporate feedback on what succeeded or failed, and maintain both high-quality individual samples and overall diversity in the output set. Experiments across benchmarks demonstrate that models using this approach perform better on later tasks than models that treat each task in isolation. The work therefore treats synthetic data generation as a process that can build reusable knowledge rather than restarting from scratch each time.

Core claim

SynLearner enables synthesis models to acquire reusable experience over a task stream by encouraging exploration of diverse synthesis patterns, learning from feedback, and balancing sample quality with set-level diversity, which produces consistent cross-task transferability and higher performance on later tasks.

What carries the argument

SynLearner, the framework that converts isolated synthesis tasks into an accumulating experience process by using feedback from historical tasks to guide future synthesis.

If this is right

  • Synthesis quality on subsequent tasks rises when the model retains and applies patterns learned from earlier tasks.
  • The same model exhibits transfer across different benchmarks rather than resetting its synthesis behavior for each new task.
  • Treating data generation as an experience-driven sequence reduces the need to restart learning for every new synthesis request.

Where Pith is reading between the lines

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

  • The same feedback loop could be applied to other sequential LLM tasks such as instruction following or tool use where past outcomes inform future outputs.
  • Over many tasks the accumulated experience might allow smaller models to match the output quality of larger ones trained from scratch on each task.

Load-bearing premise

Feedback collected from past synthesis tasks supplies useful signals that improve performance on future tasks without causing interference or loss of earlier gains.

What would settle it

A controlled run of SynLearner on a sequence of synthesis tasks in which later-task quality metrics show no gain or a drop relative to an isolated-task baseline that receives no historical feedback.

Figures

Figures reproduced from arXiv: 2605.29940 by Bingyu Zhu, Jungang Lou, Longtao Huang, Xiongtao Zhang, Yan Wang, Zeyu Yang, Zhen Bi, Zhenlin Hu, Zhixuan Chu, Zihao Xue.

Figure 1
Figure 1. Figure 1: Comparison of synthesis paradigms. (a) From Scratch: generation guided by fixed rules or heuristics. (b) From Seed: expansion from small seed sets. (c) Learning from Feedback: our proposed paradigm where the synthesis model progressively improves its synthesis behavior through feedback and prior experience. quality within single or isolated tasks (Yu et al., 2023b; Zhan et al., 2025). A more fundamental qu… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of SynLearner. Given a sequence of synthesis tasks, SynLearner initializes diverse synthesis conditions via dynamic prompting and evolutionary expansion, then optimizes the model through EFT and HRO with sample-level and set-level rewards. This design enables the synthesis model to progressively improve generation quality and transfer its synthesis ability across tasks. We first use dynamic prompt… view at source ↗
Figure 3
Figure 3. Figure 3: Selected qualitative analyses of synthesis diversity and cross-stage generalization. Subfigures (a) and [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Effect of task order on synthesis performance under different task stream permutations. Subfigures (a), (b), [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Effect of task position in the stream on synthesis performance. Subfigures (a) and (b) report results on two [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Full t-SNE visualization of original and synthetic samples across four tasks. [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Full heatmap analysis of cross-stage generalization under different synthesis methods. Rows denote model [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
read the original abstract

Large language models (LLMs) have been widely adopted for synthetic data generation, significantly reducing annotation costs. However, most existing studies treat synthesis as a set of isolated tasks and overlook a more fundamental question: whether a model can learn to synthesize by accumulating experience from past tasks and transferring it to future ones. In this work, we introduce StreamSynth, a new setting in which synthesis tasks arrive sequentially and experience from historical tasks provides informative signals for future synthesis. To address this setting, we propose SynLearner, a general framework that enables synthesis models to acquire reusable synthesis experience over a task stream. Instead of generating data independently for each task, SynLearner encourages the model to explore diverse synthesis patterns, learn from feedback, and balance sample quality with set-level diversity as tasks evolve. Extensive experiments across multiple benchmarks show that SynLearner effectively leverages experience from earlier tasks to improve synthesis performance on later ones, exhibiting consistent cross-task transferability. These findings provide evidence for the feasibility of StreamSynth and highlight synthetic data generation as an experience-driven process that can benefit from task streams.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

0 major / 3 minor

Summary. The manuscript introduces StreamSynth, a setting in which synthesis tasks arrive sequentially and historical experience provides signals for future tasks. It proposes SynLearner, a framework that encourages LLMs to explore diverse synthesis patterns, learn from feedback, and balance sample quality with set-level diversity across the task stream. The central empirical claim is that SynLearner produces consistent cross-task transferability, with performance on later tasks improving from experience accumulated on earlier ones, as demonstrated across multiple benchmarks.

Significance. If the experimental results hold under rigorous controls, the work would establish synthetic data generation as an experience-accumulating rather than isolated process. This could reduce annotation costs more effectively than current per-task approaches and open a new research direction on transfer in synthesis pipelines.

minor comments (3)
  1. [Abstract] Abstract: the phrase 'balance sample quality with set-level diversity' is introduced without a concrete definition or metric; a brief parenthetical or reference to the relevant section would improve clarity.
  2. [Abstract] The abstract states 'extensive experiments across multiple benchmarks' but does not name the benchmarks or report any aggregate statistics; adding these details would allow readers to assess the scope of the transferability claim.
  3. The manuscript would benefit from an explicit statement of the feedback signal (e.g., reward model, human preference, or automatic metric) used to update the synthesis policy between tasks.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary, significance assessment, and recommendation of minor revision. We are encouraged by the view that the work could open a new direction on transfer in synthesis pipelines if the results hold under rigorous controls.

Circularity Check

0 steps flagged

No circularity: empirical framework validated externally by benchmark experiments

full rationale

The paper introduces StreamSynth as a sequential task setting and SynLearner as a practical framework for accumulating synthesis experience via feedback and diversity balancing. All load-bearing claims (cross-task transferability, performance gains on later tasks) are presented as outcomes of extensive experiments on multiple benchmarks rather than derived from equations or self-referential definitions. No mathematical derivations, fitted parameters renamed as predictions, or load-bearing self-citations appear in the provided text. The central result is therefore falsifiable against external data and does not reduce to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no equations, methods, or results from which to extract free parameters, axioms, or invented entities.

pith-pipeline@v0.9.1-grok · 5742 in / 924 out tokens · 20503 ms · 2026-06-29T07:38:18.100305+00:00 · methodology

discussion (0)

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

Forward citations

Cited by 1 Pith paper

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

  1. ZooClaw-FashionSigLIP2: Distilled Fine-tuning for Robust Fashion Retrieval

    cs.CV 2026-06 unverdicted novelty 4.0

    ZooClaw-FashionSigLIP2 applies distilled full fine-tuning plus WiseFT interpolation to SigLIP2-base and reports outperforming LoRA, larger backbones, and external data on fashion retrieval benchmarks while releasing a...

Reference graph

Works this paper leans on

43 extracted references · 29 canonical work pages · cited by 1 Pith paper · 7 internal anchors

  1. [1]

    Zhen Bi, Zhenlin Hu, Xueshu Chen, Mingyang Chen, Cheng Deng, Yida Xue, Zhen Wang, Qing Shen, Ningyu Zhang, and Jungang Lou. 2026. https://arxiv.org/abs/2509.24836 Logical structure as knowledge: Enhancing llm reasoning via structured logical knowledge density estimation . Preprint, arXiv:2509.24836

    work page arXiv 2026

  2. [2]

    Younwoo Choi, Muhammad Adil Asif, Ziwen Han, John Willes, and Rahul Krishnan. 2025. https://openreview.net/forum?id=FS2nukC2jv Teaching LLM s how to learn with contextual fine-tuning . In The Thirteenth International Conference on Learning Representations

    2025

  3. [3]

    Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John Schulman. 2021. https://arxiv.org/abs/2110.14168 Training verifiers to solve math word problems . Preprint, arXiv:2110.14168

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  4. [4]

    DeepSeek-AI. 2025. https://arxiv.org/abs/2501.12948 Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning . Preprint, arXiv:2501.12948

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  5. [5]

    Yihe Deng, Weitong Zhang, Zixiang Chen, and Quanquan Gu. 2023. https://arxiv.org/abs/2311.04205 Rephrase and respond: Let large language models ask better questions for themselves . Preprint, arXiv:2311.04205

    work page arXiv 2023

  6. [6]

    Ronen Eldan and Yuanzhi Li. 2023. https://arxiv.org/abs/2305.07759 Tinystories: How small can language models be and still speak coherent english? Preprint, arXiv:2305.07759

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  7. [7]

    Aaron Grattafiori, Abhimanyu Dubey, and Abhinav Jauhri et al. 2024. https://arxiv.org/abs/2407.21783 The llama 3 herd of models . Preprint, arXiv:2407.21783

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  8. [8]

    Jinghan He, Haiyun Guo, and Kuan et al. Zhu. 2024. https://doi.org/10.18653/v1/2024.emnlp-main.190 SEEKR : Selective attention-guided knowledge retention for continual learning of large language models . In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 3254--3266, Miami, Florida, USA. Association for Computa...

    work page doi:10.18653/v1/2024.emnlp-main.190 2024

  9. [9]

    Linda He, Jue WANG, Maurice Weber, Shang Zhu, Ben Athiwaratkun, and Ce Zhang. 2025. https://openreview.net/forum?id=BkwCrIsTbR Scaling instruction-tuned LLM s to million-token contexts via hierarchical synthetic data generation . In The Thirteenth International Conference on Learning Representations

    2025

  10. [10]

    Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song, and Jacob Steinhardt. 2021. https://openreview.net/forum?id=7Bywt2mQsCe Measuring mathematical problem solving with the MATH dataset . In Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 2)

    2021

  11. [11]

    Zhenlin Hu, Zhizhi Peng, Zhen Bi, Qing Shen, Zhenfang Liu, Jungang Lou, and Xin Luo. 2025. https://doi.org/10.1109/JAS.2025.125540 Advancing healthcare with large language models: Techniques and application . IEEE/CAA Journal of Automatica Sinica, 12(12):2371--2398

    work page doi:10.1109/jas.2025.125540 2025

  12. [12]

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

    work page doi:10.1145/3703155 2025

  13. [13]

    Wei Huang, Anda Cheng, Zhao Zhang, Yinggui Wang, Lei Wang, Shoumeng Yan, and Tao Wei. 2025 b . https://openreview.net/forum?id=fHvh913U1H Mitigating catastrophic forgetting in large language models with forgetting-aware pruning

    2025

  14. [14]

    Yuan Liang, Ruobin Zhong, Haoming Xu, Chen Jiang, Yi Zhong, Runnan Fang, Jia-Chen Gu, Shumin Deng, Yunzhi Yao, Mengru Wang, Shuofei Qiao, Xin Xu, Tongtong Wu, Kun Wang, Yang Liu, Zhen Bi, Jungang Lou, Yuchen Eleanor Jiang, Hangcheng Zhu, and 30 others. 2026. https://arxiv.org/abs/2603.04448 Skillnet: Create, evaluate, and connect ai skills . Preprint, arX...

    work page arXiv 2026

  15. [15]

    Gouki Minegishi, Hiroki Furuta, Shohei Taniguchi, Yusuke Iwasawa, and Yutaka Matsuo. 2025. https://openreview.net/forum?id=Xw01vF13aV Beyond induction heads: In-context meta learning induces multi-phase circuit emergence . In Forty-second International Conference on Machine Learning

    2025

  16. [16]

    OpenAI. 2024. https://arxiv.org/abs/2303.08774 Gpt-4 technical report . Preprint, arXiv:2303.08774

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  17. [17]

    Archiki Prasad, Elias Stengel-Eskin, and Mohit Bansal. 2024. https://openreview.net/forum?id=L4nOxziGf9 Rephrase, augment, reason: Visual grounding of questions for vision-language models . In The Twelfth International Conference on Learning Representations

    2024

  18. [18]

    Qwen, :, and An Yang et al. 2025. https://arxiv.org/abs/2412.15115 Qwen2.5 technical report . Preprint, arXiv:2412.15115

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  19. [19]

    Haris Riaz, Sourav Bhabesh, Vinayak Arannil, Miguel Ballesteros, and Graham Horwood. 2025. https://arxiv.org/abs/2504.12563 Metasynth: Meta-prompting-driven agentic scaffolds for diverse synthetic data generation . Preprint, arXiv:2504.12563

    work page arXiv 2025

  20. [20]

    Haizhou Shi, Zihao Xu, Hengyi Wang, Weiyi Qin, Wenyuan Wang, Yibin Wang, Zifeng Wang, Sayna Ebrahimi, and Hao Wang. 2025. https://doi.org/10.1145/3735633 Continual learning of large language models: A comprehensive survey . ACM Comput. Surv., 58(5)

    work page doi:10.1145/3735633 2025

  21. [21]

    Yuxuan Tong, Xiwen Zhang, Rui Wang, Ruidong Wu, and Junxian He. 2024. http://papers.nips.cc/paper\_files/paper/2024/hash/0ef1afa0daa888d695dcd5e9513bafa3-Abstract-Conference.html Dart-math: Difficulty-aware rejection tuning for mathematical problem-solving . In Advances in Neural Information Processing Systems 38: Annual Conference on Neural Information P...

    2024

  22. [22]

    Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel Bowman. 2018. https://doi.org/10.18653/v1/W18-5446 GLUE : A multi-task benchmark and analysis platform for natural language understanding . In Proceedings of the 2018 EMNLP Workshop B lackbox NLP : Analyzing and Interpreting Neural Networks for NLP , pages 353--355, Brussels, Be...

    work page doi:10.18653/v1/w18-5446 2018

  23. [23]

    Chenxu Wang, Yilin Lyu, Zicheng Sun, and Liping Jing. 2025 a . https://doi.org/10.18653/v1/2025.acl-long.721 Continual gradient low-rank projection fine-tuning for LLM s . In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 14815--14829, Vienna, Austria. Association for Computational Li...

    work page doi:10.18653/v1/2025.acl-long.721 2025

  24. [24]

    Jiuqi Wang, Rohan Chandra, and Shangtong Zhang. 2025 b . https://arxiv.org/abs/2503.20018 Experience replay addresses loss of plasticity in continual learning . Preprint, arXiv:2503.20018

    work page arXiv 2025

  25. [25]

    Liyuan Wang, Xingxing Zhang, Hang Su, and Jun Zhu. 2024 a . https://doi.org/10.1109/TPAMI.2024.3367329 A comprehensive survey of continual learning: Theory, method and application . IEEE Transactions on Pattern Analysis and Machine Intelligence, 46(8):5362--5383

    work page doi:10.1109/tpami.2024.3367329 2024

  26. [26]

    Maolin Wang, Xiangyu Zhao, Ruocheng Guo, and Junhui Wang. 2025 c . https://doi.org/10.1109/ICDE65448.2025.00376 Metalora: Tensor-enhanced adaptive low-rank fine-tuning . In 2025 IEEE 41st International Conference on Data Engineering (ICDE), pages 4680--4684

    work page doi:10.1109/icde65448.2025.00376 2025

  27. [27]

    Xingjin Wang, Howe Tissue, Lu Wang, Linjing Li, and Daniel Dajun Zeng. 2025 d . https://openreview.net/forum?id=Vk1rNMl0J1 Learning dynamics in continual pre-training for large language models . In Forty-second International Conference on Machine Learning

    2025

  28. [28]

    Yifan Wang, Yafei Liu, Chufan Shi, and 1 others. 2024 b . https://doi.org/10.18653/v1/2024.naacl-long.37 I ns CL : A data-efficient continual learning paradigm for fine-tuning large language models with instructions . In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Techno...

    work page doi:10.18653/v1/2024.naacl-long.37 2024

  29. [29]

    Smith, Daniel Khashabi, and Hannaneh Hajishirzi

    Yizhong Wang, Yeganeh Kordi, Swaroop Mishra, Alisa Liu, Noah A. Smith, Daniel Khashabi, and Hannaneh Hajishirzi. 2023. https://doi.org/10.18653/v1/2023.acl-long.754 Self-instruct: Aligning language models with self-generated instructions . In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), p...

    work page doi:10.18653/v1/2023.acl-long.754 2023

  30. [30]

    Emily Xiao, Yixiao Zeng, Ada Chen, Chin-Jou Li, Amanda Bertsch, and Graham Neubig. 2025. https://arxiv.org/abs/2510.16932 Prompt-mii: Meta-learning instruction induction for llms . Preprint, arXiv:2510.16932

    work page arXiv 2025

  31. [31]

    Can Xu, Qingfeng Sun, Kai Zheng, Xiubo Geng, Pu Zhao, Jiazhan Feng, Chongyang Tao, Qingwei Lin, and Daxin Jiang. 2024. https://openreview.net/forum?id=CfXh93NDgH Wizard LM : Empowering large pre-trained language models to follow complex instructions . In The Twelfth International Conference on Learning Representations

    2024

  32. [32]

    Siqiao Xue, Zihan Liao, Jin Qin, Ziyin Zhang, Yixiang Mu, Fan Zhou, and Hang Yu. 2026 a . https://arxiv.org/abs/2605.04615 Beyond retrieval: A multitask benchmark and model for code search . arXiv preprint arXiv:2605.04615

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  33. [33]

    Siqiao Xue, Zhaoyang Zhu, Wei Zhang, Rongyao Cai, Rui Wang, Yixiang Mu, Fan Zhou, Jianguo Li, Peng Di, and Hang Yu. 2026 b . https://arxiv.org/abs/2603.26017 Quitobench: A high-quality open time series forecasting benchmark . arXiv preprint arXiv:2603.26017

    work page arXiv 2026

  34. [34]

    Zihao Xue, Zhen Bi, Long Ma, Zhenlin Hu, Yan Wang, Xueshu Chen, Zhenfang Liu, Kang Zhao, Jie Xiao, and Jungang Lou. 2026 c . https://arxiv.org/abs/2507.12314 Thought purity: A defense framework for chain-of-thought attack . Preprint, arXiv:2507.12314

    work page arXiv 2026

  35. [35]

    Shangjian Yin, Zhepei Wei, Xinyu Zhu, Wei-Lin Chen, and Yu Meng. 2025. https://arxiv.org/abs/2510.06652 Aligning large language models via fully self-synthetic data . Preprint, arXiv:2510.06652

    work page arXiv 2025

  36. [36]

    Longhui Yu, Weisen Jiang, Han Shi, Jincheng YU, Zhengying Liu, Yu Zhang, James Kwok, Zhenguo Li, Adrian Weller, and Weiyang Liu. 2024. https://openreview.net/forum?id=N8N0hgNDRt Metamath: Bootstrap your own mathematical questions for large language models . In The Twelfth International Conference on Learning Representations

    2024

  37. [37]

    Yue Yu, Yuchen Zhuang, Jieyu Zhang, Yu Meng, Alexander Ratner, Ranjay Krishna, Jiaming Shen, and Chao Zhang. 2023 a . https://arxiv.org/abs/2306.15895 Large language model as attributed training data generator: A tale of diversity and bias . Preprint, arXiv:2306.15895

    work page arXiv 2023

  38. [38]

    Ratner, Ranjay Krishna, Jiaming Shen, and Chao Zhang

    Yue Yu, Yuchen Zhuang, Jieyu Zhang, Yu Meng, Alexander J. Ratner, Ranjay Krishna, Jiaming Shen, and Chao Zhang. 2023 b . http://papers.nips.cc/paper\_files/paper/2023/hash/ae9500c4f5607caf2eff033c67daa9d7-Abstract-Datasets\_and\_Benchmarks.html Large language model as attributed training data generator: A tale of diversity and bias . In Advances in Neural...

    2023

  39. [39]

    Weihao Zeng, Can Xu, Yingxiu Zhao, Jian-Guang Lou, and Weizhu Chen. 2024. https://doi.org/10.18653/v1/2024.emnlp-main.397 Automatic instruction evolving for large language models . In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 6998--7018, Miami, Florida, USA. Association for Computational Linguistics

    work page doi:10.18653/v1/2024.emnlp-main.397 2024

  40. [40]

    Shaoxiong Zhan, Yanlin Lai, Ziyu Lu, Dahua Lin, Ziqing Yang, and Fei Tan. 2025. https://arxiv.org/abs/2508.05592 Mathsmith: Towards extremely hard mathematical reasoning by forging synthetic problems with a reinforced policy . Preprint, arXiv:2508.05592

    work page arXiv 2025

  41. [41]

    Xiang Zhang, Junbo Jake Zhao, and Yann LeCun. 2015. http://papers.nips.cc/paper/5782-character-level-convolutional-networks-for-text-classification Character-level convolutional networks for text classification . In NIPS, pages 649--657

    2015

  42. [42]

    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...

  43. [43]

    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...