arxiv: 2604.24472 · v1 · submitted 2026-04-27 · 💻 cs.IR · cs.AI· cs.LG

Recognition: unknown

Modeling Behavioral Intensity and Transitions for Generative Recommendation

Hanyu Zhang, Wanqiang Xiong, Wenxuan Yang, Xiaoyang Xu, Zhaoqun Chen, Zhexuan Xu

Authors on Pith no claims yet

Pith reviewed 2026-05-08 01:49 UTC · model grok-4.3

classification 💻 cs.IR cs.AIcs.LG

keywords multi-behavior recommendationgenerative recommendationbehavioral intensitytransition patternshierarchical aggregationrelation encodinguser intent modelingsequence generation

0 comments

The pith

Explicitly modeling differences in behavioral intensity and transition patterns improves generative multi-behavior recommendation.

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

The paper contends that current generative methods for multi-behavior recommendation treat all past user actions with equal weight in their attention mechanisms, missing how some behaviors reflect stronger intent than others and how actions follow predictable sequences. It introduces BITRec to fix this by separating behaviors into distinct intensity pathways and adding explicit structures to track transitions between them. Experiments across four large e-commerce and insurance datasets show gains of 15 to 23 percent on standard ranking metrics. If the approach holds, recommendation systems could generate more accurate next-action predictions by respecting real differences in how users engage over time rather than averaging all history uniformly.

Core claim

BITRec addresses the uniform dependency assumption in prior generative models by adding Hierarchical Behavior Aggregation (HBA), which routes behaviors through separate exploration and commitment pathways to capture intensity differences, and Transition Relation Encoding (TRE), which inserts learnable relation matrices to represent how one behavior type leads to another. This structured selective activation produces consistent lifts of 15-23% across multiple metrics on RetailRocket, Taobao, Tmall, and Insurance datasets, with the largest reported gains reaching 22.79% MRR on Tmall.

What carries the argument

Hierarchical Behavior Aggregation (HBA) and Transition Relation Encoding (TRE) inside the BITRec generative framework, where HBA creates separated pathways for intensity levels and TRE supplies explicit matrices for transition structures.

Load-bearing premise

That uniform activation in attention mechanisms is the main reason existing generative models miss intensity differences and transition patterns.

What would settle it

A direct comparison on the same four datasets where a standard generative model without HBA or TRE achieves equal or higher MRR, HR@10, and NDCG@10 scores than BITRec.

Figures

Figures reproduced from arXiv: 2604.24472 by Hanyu Zhang, Wanqiang Xiong, Wenxuan Yang, Xiaoyang Xu, Zhaoqun Chen, Zhexuan Xu.

**Figure 1.** Figure 1: Overview of BITRec architecture. Phase 1: Joint Behavior-Item Embedding fuses behaviors and items into composite tokens. Phase 2: BITRec Transformer processes sequences via Hierarchical Behavior Aggregation (separating exploration/commitment representations) and Transition Relation Encoding. Phase 3: Autoregressive prediction generates recommendations. while, an item click occurred just one minute ago, de… view at source ↗

**Figure 2.** Figure 2: Performance comparison across different sequence lengths ( view at source ↗

read the original abstract

Multi-behavior recommendation aims to predict user conversions by modeling various interaction types that carry distinct intent signals. Recently, generative sequence modeling methods have emerged as an important paradigm for multi-behavior recommendation by achieving flexible sequence generation. However, existing generative methods typically treat behaviors as auxiliary token features and feed them into unified attention mechanisms. These models implicitly assume uniform activation of dependencies among historical behaviors, thereby failing to discern differences in intensity or capture transition patterns. To address these limitations, we propose BITRec, a novel generative multi-behavior recommendation framework that introduces structured behavioral modeling through selective dependency activation. BITRec incorporates (i) Hierarchical Behavior Aggregation (HBA), which explicitly models behavioral intensity differences through separated exploration and commitment pathways, and (ii) Transition Relation Encoding (TRE), which encodes transition structures through explicit learnable relation matrices. Experiments on four large-scale datasets (RetailRocket, Taobao, Tmall, Insurance Dataset) with millions of interactions achieve consistent improvements of 15-23% across multiple metrics, with peak gains of 22.79% MRR on Tmall and 17.83% HR@10, 17.55% NDCG@10 on Taobao.

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

BITRec adds HBA and TRE to generative multi-behavior models and reports 15-23% gains, but the results lack ablations or stats to show those components are responsible.

read the letter

BITRec claims to improve generative models for multi-behavior recommendation by adding Hierarchical Behavior Aggregation for intensity differences and Transition Relation Encoding for transition patterns. The main point is that it reports 15-23% gains over baselines on four datasets, but without ablations or significance tests it's hard to know if the new parts are what drives the improvement. The paper identifies that prior generative approaches feed behaviors into unified attention and assume uniform dependencies. BITRec counters this with separated pathways in HBA for exploration versus commitment behaviors, and learnable matrices in TRE to capture how one behavior type leads to another. This structured approach makes sense for data where behaviors have different strengths and sequences matter. On the positive side, the experiments use real large-scale datasets from retail and insurance domains, and the gains appear across multiple metrics like MRR, HR, and NDCG. That suggests the model has practical potential for systems that need to predict conversions from varied user actions. The weaknesses are in the validation. The abstract gives no sign of controlled experiments that turn off HBA or TRE to measure their specific contribution, nor any p-values or standard deviations from repeated runs. Baseline comparisons could also be affected by implementation differences or tuning. These are common issues but they leave the central claim open to alternative explanations like added model capacity. This paper would interest people building recommendation engines that handle multiple interaction types, particularly those already using sequence generation techniques. A practitioner might pick up the HBA and TRE ideas to experiment with in their own setups. It is worth sending for peer review. The ideas are straightforward and grounded in a real problem, and the results are on standard benchmarks. Referees can push for the missing ablations and stats, which would make the contribution clearer.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes BITRec, a generative multi-behavior recommendation framework that introduces Hierarchical Behavior Aggregation (HBA) to model behavioral intensity differences via separated exploration and commitment pathways, and Transition Relation Encoding (TRE) to capture transition patterns through explicit learnable relation matrices. It argues that prior generative methods implicitly assume uniform dependency activation in unified attention and thus fail to discern intensity or transitions. Experiments on four large-scale datasets (RetailRocket, Taobao, Tmall, Insurance Dataset) report consistent 15-23% gains across metrics, including peaks of 22.79% MRR on Tmall and 17.83% HR@10 / 17.55% NDCG@10 on Taobao.

Significance. If the performance lifts are shown to be robust and specifically attributable to HBA and TRE rather than capacity or tuning differences, the work would advance generative recommendation by providing structured mechanisms for intensity and transition modeling. The explicit separation of pathways and learnable relation matrices represent a clear technical direction that could be adopted more broadly if the empirical isolation is strengthened.

major comments (2)

[Experiments] Experiments section: the central claim that selective dependency activation via HBA and TRE produces the 15-23% gains requires controlled ablations (full model vs. HBA-removed vs. TRE-removed) and statistical significance / variance across runs; none are reported, so the improvements cannot be isolated from capacity increases, hyperparameter differences, or implementation details.
[Abstract and Experiments] Abstract and Experiments: baseline implementations, hyperparameter tuning protocols, and checks for data leakage are not described, which is load-bearing for verifying that the reported lifts on the four datasets are reproducible and not artifacts of the evaluation setup.

minor comments (1)

[Abstract] Abstract: peak gains are cited for different datasets and metrics; presenting a uniform set of metrics (e.g., HR@10, NDCG@10, MRR) for all four datasets in one table would improve comparability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which highlight important aspects of experimental rigor and reproducibility. We will revise the manuscript to incorporate the requested details and analyses, thereby strengthening the validation of our claims.

read point-by-point responses

Referee: [Experiments] Experiments section: the central claim that selective dependency activation via HBA and TRE produces the 15-23% gains requires controlled ablations (full model vs. HBA-removed vs. TRE-removed) and statistical significance / variance across runs; none are reported, so the improvements cannot be isolated from capacity increases, hyperparameter differences, or implementation details.

Authors: We agree that controlled ablations and statistical reporting are necessary to isolate the contributions of HBA and TRE. In the revised manuscript, we will add ablation studies comparing the full BITRec model against variants with HBA removed and TRE removed. We will also run each model configuration multiple times with different random seeds and report mean performance with standard deviations to establish statistical significance and robustness. These additions will directly address concerns about capacity, tuning, or implementation artifacts. revision: yes
Referee: [Abstract and Experiments] Abstract and Experiments: baseline implementations, hyperparameter tuning protocols, and checks for data leakage are not described, which is load-bearing for verifying that the reported lifts on the four datasets are reproducible and not artifacts of the evaluation setup.

Authors: We acknowledge that these implementation and evaluation details were insufficiently described. In the revised version, we will expand the Experiments section to include: (i) precise descriptions of how each baseline was re-implemented following the original papers, (ii) the full hyperparameter tuning protocol with search ranges, optimization criteria, and final selected values, and (iii) explicit verification steps confirming the absence of data leakage in the temporal or user-based splits used for the four datasets. These changes will support reproducibility of the reported improvements. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical claims rest on external dataset benchmarks

full rationale

The paper proposes BITRec with HBA (separated pathways for intensity) and TRE (learnable relation matrices for transitions) to improve over uniform-attention generative baselines. Central results are reported performance lifts (15-23%) on four held-out large-scale datasets (RetailRocket, Taobao, Tmall, Insurance). No quoted equations, self-citations, or ansatzes reduce these lifts to the fitted parameters by construction; the architecture is an independent modeling choice whose value is tested against external data splits. This is the standard non-circular pattern for empirical ML papers.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Based on abstract only; no explicit free parameters, axioms, or invented entities are named beyond the standard assumption that learnable matrices and pathways can be optimized to capture intensity and transitions.

pith-pipeline@v0.9.0 · 5520 in / 1218 out tokens · 51786 ms · 2026-05-08T01:49:04.464943+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

35 extracted references · 6 canonical work pages · 2 internal anchors

[1]

A survey of user modelling in social media websites

Ahmad Abdel-Hafez and Yue Xu. A survey of user modelling in social media websites. Computer and Information Science , 6(4):59--71, 2013

2013
[2]

Pinrec: Outcome-conditioned, multi-token generative retrieval for industry-scale recommendation systems

Anirudhan Badrinath, Prabhat Agarwal, Laksh Bhasin, Jaewon Yang, Jiajing Xu, and Charles Rosenberg. Pinrec: Outcome-conditioned, multi-token generative retrieval for industry-scale recommendation systems. arXiv preprint arXiv:2504.10507 , 2025

work page arXiv 2025
[3]

arXiv preprint arXiv:2409.12740 , year=

Junyi Chen, Lu Chi, Bingyue Peng, and Zehuan Yuan. Hllm: Enhancing sequential recommendations via hierarchical large language models for item and user modeling. arXiv preprint arXiv:2409.12740 , 2024

work page arXiv 2024
[4]

Wide & deep learning for recommender systems

Heng-Tze Cheng, Levent Koc, Jeremiah Harmsen, Tal Shaked, Tushar Chandra, Hrishi Aradhye, Glen Anderson, Greg Corrado, Wei Chai, Mustafa Ispir, Rohan Anil, Zakaria Haque, Lichan Hong, Vihan Jain, Xiaobing Liu, and Hemal Shah. Wide & deep learning for recommender systems. In Proceedings of the 1st Workshop on Deep Learning for Recommender Systems , DLRS 20...

2016
[5]

OneRec: Unifying Retrieve and Rank with Generative Recommender and Iterative Preference Alignment

Jiaxin Deng, Shiyao Wang, Kuo Cai, Lejian Ren, Qigen Hu, Weifeng Ding, Qiang Luo, and Guorui Zhou. Onerec: Unifying retrieve and rank with generative recommender and iterative preference alignment. arXiv preprint arXiv:2502.18965 , 2025

work page internal anchor Pith review arXiv 2025
[6]

Neural multi-task recommendation from multi-behavior data

Chen Gao, Xiangnan He, Dahua Gan, Xiangning Chen, Fuli Feng, Yong Li, Tat-Seng Chua, and Depeng Jin. Neural multi-task recommendation from multi-behavior data. In 2019 IEEE 35th International Conference on Data Engineering (ICDE) , pages 1554--1557, 2019

2019
[7]

Dmbin: A dual multi-behavior interest network for click-through rate prediction via contrastive learning

Tianqi He, Kaiyuan Li, Shan Chen, Haitao Wang, Qiang Liu, Xingxing Wang, and Dong Wang. Dmbin: A dual multi-behavior interest network for click-through rate prediction via contrastive learning. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval , SIGIR '23, page 1366–1375, New York, NY, USA, ...

2023
[8]

Session-based Recommendations with Recurrent Neural Networks

Bal \'a zs Hidasi, Alexandros Karatzoglou, Linas Baltrunas, and Domonkos Tikk. Session-based recommendations with recurrent neural networks. arXiv preprint arXiv:1511.06939 , 2015

work page internal anchor Pith review arXiv 2015
[9]

CORE: simple and effective session-based recommendation within consistent representation space

Yupeng Hou, Binbin Hu, Zhiqiang Zhang, and Wayne Xin Zhao. CORE: simple and effective session-based recommendation within consistent representation space. In Enrique Amig \' o , Pablo Castells, Julio Gonzalo, Ben Carterette, J. Shane Culpepper, and Gabriella Kazai, editors, SIGIR '22: The 45th International ACM SIGIR Conference on Research and Development...

2022
[10]

Multi-behavior recommendation with graph convolution networks

Bowen Jin, Chen Gao, Xiangnan He, Depeng Jin, and Yong Li. Multi-behavior recommendation with graph convolution networks. In 43nd International ACM SIGIR Conference on Research and Development in Information Retrieval , 2020

2020
[11]

Self-attentive sequential recommendation

Wang-Cheng Kang and Julian McAuley. Self-attentive sequential recommendation. In 2018 IEEE International Conference on Data Mining (ICDM) , pages 197--206, 2018

2018
[12]

Blade: A behavior-level data augmentation framework with dual fusion modeling for multi-behavior sequential recommendation

Yupeng Li, Mingyue Cheng, Yucong Luo, Yitong Zhou, Qingyang Mao, and Shijin Wang. Blade: A behavior-level data augmentation framework with dual fusion modeling for multi-behavior sequential recommendation. arXiv preprint arXiv:2512.12964 , 2025

work page arXiv 2025
[13]

Multi-behavioral sequential prediction with recurrent log-bilinear model

Qiang Liu, Shu Wu, and Liang Wang. Multi-behavioral sequential prediction with recurrent log-bilinear model. IEEE Trans. on Knowl. and Data Eng. , 29(6):1254–1267, June 2017

2017
[14]

Multi-behavior generative recommendation

Zihan Liu, Yupeng Hou, and Julian McAuley. Multi-behavior generative recommendation. In Proceedings of the 33rd ACM International Conference on Information and Knowledge Management , CIKM '24, page 1575–1585, New York, NY, USA, 2024. Association for Computing Machinery

2024
[15]

Entire space multi-task model: An effective approach for estimating post-click conversion rate

Xiao Ma, Liqin Zhao, Guan Huang, Zhi Wang, Zelin Hu, Xiaoqiang Zhu, and Kun Gai. Entire space multi-task model: An effective approach for estimating post-click conversion rate. In The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval , SIGIR '18, page 1137–1140, New York, NY, USA, 2018. Association for Computing Machinery

2018
[16]

Practice on long sequential user behavior modeling for click-through rate prediction

Qi Pi, Weijie Bian, Guorui Zhou, Xiaoqiang Zhu, and Kun Gai. Practice on long sequential user behavior modeling for click-through rate prediction. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , KDD '19, page 2671–2679, New York, NY, USA, 2019. Association for Computing Machinery

2019
[17]

Recommender systems with generative retrieval

Shashank Rajput, Nikhil Mehta, Anima Singh, Raghunandan Hulikal Keshavan, Trung Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Tran, Jonah Samost, Maciej Kula, Ed Chi, and Maheswaran Sathiamoorthy. Recommender systems with generative retrieval. In A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine, editors, Advances in Neural Information Pro...

2023
[18]

Personalized behavior-aware transformer for multi-behavior sequential recommendation

Jiajie Su, Chaochao Chen, Zibin Lin, Xi Li, Weiming Liu, and Xiaolin Zheng. Personalized behavior-aware transformer for multi-behavior sequential recommendation. In Abdulmotaleb El - Saddik, Tao Mei, Rita Cucchiara, Marco Bertini, Diana Patricia Tobon Vallejo, Pradeep K. Atrey, and M. Shamim Hossain, editors, Proceedings of the 31st ACM International Conf...

2023
[19]

Bert4rec: Sequential recommendation with bidirectional encoder representations from transformer

Fei Sun, Jun Liu, Jian Wu, Changhua Pei, Xiao Lin, Wenwu Ou, and Peng Jiang. Bert4rec: Sequential recommendation with bidirectional encoder representations from transformer. In Proceedings of the 28th ACM International Conference on Information and Knowledge Management , CIKM '19, page 1441–1450, New York, NY, USA, 2019. Association for Computing Machinery

2019
[20]

Attention is all you need

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, ukasz Kaiser, and Illia Polosukhin. Attention is all you need. In I. Guyon, U. Von Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett, editors, Advances in Neural Information Processing Systems , volume 30. Curran Associates, Inc., 2017

2017
[21]

Eager: Two-stream generative recommender with behavior-semantic collaboration

Ye Wang, Jiahao Xun, Minjie Hong, Jieming Zhu, Tao Jin, Wang Lin, Haoyuan Li, Linjun Li, Yan Xia, Zhou Zhao, and Zhenhua Dong. Eager: Two-stream generative recommender with behavior-semantic collaboration. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , KDD '24, page 3245–3254, New York, NY, USA, 2024. Association...

2024
[22]

Generative sequential recommendation via hierarchical behavior modeling

Zhefan Wang, Guokai Yan, Jinbei Yu, Siyu Gu, Jingyan Chen, Peng Jiang, Zhiqiang Guo, and Min Zhang. Generative sequential recommendation via hierarchical behavior modeling. arXiv preprint arXiv:2511.03155 , 2025

work page arXiv 2025
[23]

Hierarchically modeling micro and macro behaviors via multi-task learning for conversion rate prediction

Hong Wen, Jing Zhang, Fuyu Lv, Wentian Bao, Tianyi Wang, and Zulong Chen. Hierarchically modeling micro and macro behaviors via multi-task learning for conversion rate prediction. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval , SIGIR '21, page 2187–2191, New York, NY, USA, 2021. Associat...

2021
[24]

Modeling the sequential dependence among audience multi-step conversions with multi-task learning in targeted display advertising

Dongbo Xi, Zhen Chen, Peng Yan, Yinger Zhang, Yongchun Zhu, Fuzhen Zhuang, and Yu Chen. Modeling the sequential dependence among audience multi-step conversions with multi-task learning in targeted display advertising. In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining , KDD '21, page 3745–3755, New York, NY, USA, 2021. ...

2021
[25]

Graph meta network for multi-behavior recommendation

Lianghao Xia, Yong Xu, Chao Huang, Peng Dai, and Liefeng Bo. Graph meta network for multi-behavior recommendation. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval , SIGIR '21, page 757–766, New York, NY, USA, 2021. Association for Computing Machinery

2021
[26]

Multi-behavior hypergraph-enhanced transformer for sequential recommendation

Yuhao Yang, Chao Huang, Lianghao Xia, Yuxuan Liang, Yanwei Yu, and Chenliang Li. Multi-behavior hypergraph-enhanced transformer for sequential recommendation. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , KDD '22, page 2263–2274, New York, NY, USA, 2022. Association for Computing Machinery

2022
[27]

User behaviour modeling, recommendations, and purchase prediction during shopping festivals

Ming Zeng, Hancheng Cao, Min Chen, and Yong Li. User behaviour modeling, recommendations, and purchase prediction during shopping festivals. Electronic Markets , 29(2):263--274, 2019

2019
[28]

Actions speak louder than words: trillion-parameter sequential transducers for generative recommendations

Jiaqi Zhai, Lucy Liao, Xing Liu, Yueming Wang, Rui Li, Xuan Cao, Leon Gao, Zhaojie Gong, Fangda Gu, Jiayuan He, Yinghai Lu, and Yu Shi. Actions speak louder than words: trillion-parameter sequential transducers for generative recommendations. In Proceedings of the 41st International Conference on Machine Learning , ICML'24. JMLR.org, 2024

2024
[29]

Combinatorial optimization perspective based framework for multi-behavior recommendation

Chenhao Zhai, Chang Meng, Yu Yang, Kexin Zhang, Xuhao Zhao, and Xiu Li. Combinatorial optimization perspective based framework for multi-behavior recommendation. In Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.1 , KDD '25, page 1891–1902, New York, NY, USA, 2025. Association for Computing Machinery

1902
[30]

Sheng, Jiajie Xu, Deqing Wang, Guanfeng Liu, and Xiaofang Zhou

Tingting Zhang, Pengpeng Zhao, Yanchi Liu, Victor S. Sheng, Jiajie Xu, Deqing Wang, Guanfeng Liu, and Xiaofang Zhou. Feature-level deeper self-attention network for sequential recommendation. In Proceedings of the 28th International Joint Conference on Artificial Intelligence , IJCAI'19, page 4320–4326. AAAI Press, 2019

2019
[31]

Stock constrained recommendation in tmall

Wenliang Zhong, Rong Jin, Cheng Yang, Xiaowei Yan, Qi Zhang, and Qiang Li. Stock constrained recommendation in tmall. In Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , KDD '15, page 2287–2296, New York, NY, USA, 2015. Association for Computing Machinery

2015
[32]

Deep interest network for click-through rate prediction

Guorui Zhou, Xiaoqiang Zhu, Chenru Song, Ying Fan, Han Zhu, Xiao Ma, Yanghui Yan, Junqi Jin, Han Li, and Kun Gai. Deep interest network for click-through rate prediction. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , KDD '18, page 1059–1068, New York, NY, USA, 2018. Association for Computing Machinery

2018
[33]

Learning tree-based deep model for recommender systems

Han Zhu, Xiang Li, Pengye Zhang, Guozheng Li, Jie He, Han Li, and Kun Gai. Learning tree-based deep model for recommender systems. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , KDD '18, page 1079–1088, New York, NY, USA, 2018. Association for Computing Machinery

2018
[34]

Retailrocket recommender system dataset, 2022

Roman Zykov, Noskov Artem, and Anokhin Alexander. Retailrocket recommender system dataset, 2022

2022
[35]

write newline

" write newline "" before.all 'output.state := FUNCTION fin.entry add.period write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence output.state after.block = 'skip output.state before.all = 'skip after.sentence 'output.state := if if FUNCTION not #0 #1 if FUNCTION and 'skip pop #0 if FUNCTIO...