pith. machine review for the scientific record. sign in

arxiv: 2605.14169 · v1 · submitted 2026-05-13 · 💻 cs.CL

Recognition: no theorem link

BOOKMARKS: Efficient Active Storyline Memory for Role-playing

Letian Peng , Ziche Liu , Yiming Huang , Longfei Yun , Kun Zhou , Yupeng Hou , Jingbo Shang
Authors on Pith no claims yet

Pith reviewed 2026-05-15 04:48 UTC · model grok-4.3

classification 💻 cs.CL
keywords role-playing agentsmemory systemssearch-based memorystoryline consistencybookmarksactive groundingsynchronizationRPA memory
0
0 comments X

The pith

BOOKMARKS replaces recurrent summarization with search-based bookmarks that keep task-relevant details alive across long role-play storylines.

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

Existing RPA memory methods rely on recurrent summarization that compresses the story and inevitably drops important details over long horizons. BOOKMARKS instead treats memory as a set of bookmarks, each defined as the answer to a task-relevant question fixed at a particular point in the storyline. For any current task the system selects reusable bookmarks or initializes fresh ones at the start, then synchronizes their answers forward to the present story point so they remain accurate and ready for reuse. This design supplies active grounding for task-specific facts while performing only passive updates, avoiding full recomputation each turn. Experiments on 85 characters drawn from 16 artifacts show clear gains over standard baselines, confirming that search-based memory improves consistency without the compression penalty.

Core claim

BOOKMARKS is a search-based memory framework that actively initializes, maintains, and updates task-relevant bookmarks for role-playing agents. Each bookmark is structured as the answer to a question at a specific storyline point. For the current task the framework selects existing reusable bookmarks or creates new ones at the storyline beginning, then applies synchronization to bring their answers up to the present story point for efficient reuse in future grounding rounds. The implementation supports concept, behavior, and state searches, each driven by its own efficient synchronization method.

What carries the argument

A bookmark, defined as the answer to a question at a specific point in the storyline, which carries active task grounding and enables passive forward synchronization.

If this is right

  • Role-playing agents can sustain longer storylines while retaining task-specific details that summarization discards.
  • Computation is reduced because bookmarks are updated passively rather than regenerated recurrently.
  • Bookmarks initialized for one task become reusable across later grounding rounds without re-initialization.
  • Separate support for concept, behavior, and state searches gives the memory system flexibility to handle different kinds of character information.

Where Pith is reading between the lines

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

  • The same bookmark-and-synchronization pattern could be applied to other long-context agents that need to track evolving facts without full context recompression.
  • If question selection or synchronization occasionally introduces small errors, those errors could compound over hundreds of turns and would need explicit correction mechanisms.
  • Evaluating the system on live user-driven conversations rather than static artifacts would test whether the performance gains hold outside curated storylines.

Load-bearing premise

Task-relevant questions for bookmarks can be reliably selected or initialized and synchronization methods can accurately update their answers without introducing errors or missing critical context as the storyline advances.

What would settle it

If BOOKMARKS fails to produce higher consistency scores than recurrent summarization baselines when evaluated on the same 85 characters from 16 artifacts, the claim that search-based memory is more effective would be refuted.

Figures

Figures reproduced from arXiv: 2605.14169 by Jingbo Shang, Kun Zhou, Letian Peng, Longfei Yun, Yiming Huang, Yupeng Hou, Ziche Liu.

Figure 1
Figure 1. Figure 1: BOOKMARKS grounds role-playing by ac￾tively searching for useful information from the preced￾ing storyline, while passively updating previous search results for efficiency. that only a partial preceding storyline can reach the grounding stage, either due to the filtering mech￾anism in retrieval or the compression of details in profiling. In contrast, search-based grounding (Jin et al., 2025) can utilize th… view at source ↗
Figure 2
Figure 2. Figure 2: The role-playing grounding workflow of B [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The hit rate (matching an existing bookmark) and efficiency analysis of B [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Case study on comparing BOOKMARKS and conventional profiling, based on multiple action prediction. involved, as shown in [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
read the original abstract

Memory systems are critical for role-playing agents (RPAs) to maintain long-horizon consistency. However, existing RPA memory methods (e.g., profiling) mainly rely on recurrent summarization, whose compression inevitably discards important details. To address this issue, we propose a search-based memory framework called BOOKMARKS, which actively initializes, maintains, and updates task-relevant pieces of bookmarks for the current task (e.g., character acting). A bookmark is structured as the answer to a question at a specific point in the storyline. For each current task, BOOKMARKS selects reusable existing bookmarks or initializes new ones (at storyline beginning) with useful questions. These bookmarks are then synchronized to the current story point, with their answers updated accordingly, so they can be efficiently reused in future grounding rounds. Compared with recurrent summarization, BOOKMARKS offers (1) active grounding for capturing task-specific details and (2) passive updating to avoid unnecessary computation. In implementation, BOOKMARKS supports concept, behavior, and state searches, each powered by an efficient synchronization method. BOOKMARKS significantly outperforms RPA memory baselines on 85 characters from 16 artifacts, demonstrating the effectiveness of search-based memory for RPAs.

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 / 1 minor

Summary. The manuscript proposes BOOKMARKS, a search-based memory framework for role-playing agents that actively selects or initializes task-relevant bookmarks (structured as question-answer pairs tied to specific storyline points), synchronizes their answers to the current narrative state via concept/behavior/state searches, and reuses them for efficient grounding. It claims this avoids the detail loss of recurrent summarization, offering active grounding and passive updating, with significant outperformance over RPA memory baselines demonstrated on 85 characters from 16 artifacts.

Significance. If the empirical claims are supported by detailed metrics, controls, and ablations, the approach could meaningfully advance long-horizon consistency in RPAs by providing a non-compressive alternative to profiling methods, with potential for broader use in interactive narrative systems where task-specific details must be preserved without full history recomputation.

major comments (2)
  1. [Abstract] Abstract: The central claim that BOOKMARKS 'significantly outperforms RPA memory baselines' on 85 characters from 16 artifacts is asserted without any reported metrics (e.g., accuracy, consistency scores), baseline descriptions, statistical tests, or ablation results, rendering the empirical contribution impossible to evaluate from the provided text.
  2. [Method] Method (synchronization description): The synchronization methods for updating bookmark answers as the storyline evolves are outlined at a high level for concept, behavior, and state searches, but no error-rate analysis, failure-case examination (e.g., narrative contradictions or context loss), or fidelity comparison versus recurrent summarization is supplied; this directly undermines validation of the claimed advantage over compression-based methods.
minor comments (1)
  1. [Abstract] Abstract: The term 'artifacts' is introduced without definition or reference to a specific dataset or collection of stories; this should be clarified with a citation or brief description for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and will revise the manuscript to improve clarity and completeness.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that BOOKMARKS 'significantly outperforms RPA memory baselines' on 85 characters from 16 artifacts is asserted without any reported metrics (e.g., accuracy, consistency scores), baseline descriptions, statistical tests, or ablation results, rendering the empirical contribution impossible to evaluate from the provided text.

    Authors: We agree that the abstract would be strengthened by including specific quantitative results. The full manuscript reports these details in the Experiments section, including consistency and accuracy metrics, baseline descriptions (e.g., recurrent summarization and profiling methods), ablation studies, and statistical tests. We will revise the abstract to incorporate key performance numbers, baseline references, and significance indicators. revision: yes

  2. Referee: [Method] Method (synchronization description): The synchronization methods for updating bookmark answers as the storyline evolves are outlined at a high level for concept, behavior, and state searches, but no error-rate analysis, failure-case examination (e.g., narrative contradictions or context loss), or fidelity comparison versus recurrent summarization is supplied; this directly undermines validation of the claimed advantage over compression-based methods.

    Authors: The synchronization procedures are specified in the Method section, including the three search types with algorithmic descriptions. We acknowledge the absence of dedicated error-rate analysis and failure-case discussion. We will add a new subsection providing quantitative error rates, examples of edge cases such as narrative contradictions, and fidelity comparisons (e.g., information retention metrics) against recurrent summarization. revision: yes

Circularity Check

0 steps flagged

No significant circularity: empirical claims rest on external benchmarks

full rationale

The paper introduces BOOKMARKS as a search-based memory framework for role-playing agents, with bookmarks defined as question-answer pairs synchronized across storyline points. All load-bearing claims are empirical comparisons against RPA baselines on 85 characters from 16 artifacts. No equations, derivations, fitted parameters renamed as predictions, or self-citation chains appear in the provided text. The synchronization and initialization procedures are presented as engineering choices evaluated by performance metrics, not as results forced by definition or prior self-work. The derivation chain is therefore self-contained against external data.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The framework rests on the domain assumption that useful questions can be chosen to capture task-relevant details and that synchronization preserves accuracy; the bookmark entity itself is newly introduced without external validation.

axioms (1)
  • domain assumption Task-relevant bookmarks can be initialized at storyline beginnings and synchronized to current points without critical information loss.
    Invoked to justify passive updating and reuse across grounding rounds.
invented entities (1)
  • Bookmark no independent evidence
    purpose: Structured memory unit consisting of the answer to a question at a specific storyline point.
    Newly postulated memory primitive enabling search-based rather than summarization-based retention.

pith-pipeline@v0.9.0 · 5526 in / 1149 out tokens · 42048 ms · 2026-05-15T04:48:45.692385+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

136 extracted references · 136 canonical work pages · 14 internal anchors

  1. [1]

    Proceedings of the Thirty-Ninth Annual Conference on Neural Information Processing Systems (NeurIPS 2025) Poster Session , year =

    Codifying Character Logic in Role-Playing , author =. Proceedings of the Thirty-Ninth Annual Conference on Neural Information Processing Systems (NeurIPS 2025) Poster Session , year =

    work page 2025

  2. [2]

    Computational linguistics , volume=

    Class-based n-gram models of natural language , author=. Computational linguistics , volume=

    work page

  3. [3]

    Second Conference on Language Modeling , year=

    Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learning , author=. Second Conference on Language Modeling , year=

    work page

  4. [4]

    Character-LLM:

    Yunfan Shao and Linyang Li and Junqi Dai and Xipeng Qiu , editor =. Character-LLM:. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing,. 2023 , url =. doi:10.18653/V1/2023.EMNLP-MAIN.814 , timestamp =

    work page doi:10.18653/v1/2023.emnlp-main.814 2023

  5. [5]

    SimCSE: Simple Contrastive Learning of Sentence Embeddings , booktitle =

    Tianyu Gao and Xingcheng Yao and Danqi Chen , editor =. SimCSE: Simple Contrastive Learning of Sentence Embeddings , booktitle =. 2021 , url =. doi:10.18653/V1/2021.EMNLP-MAIN.552 , timestamp =

    work page doi:10.18653/v1/2021.emnlp-main.552 2021

  6. [6]

    Learning to Retrieve In-Context Examples for Large Language Models

    Wang, Liang and Yang, Nan and Wei, Furu. Learning to Retrieve In-Context Examples for Large Language Models. Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers). 2024. doi:10.18653/v1/2024.eacl-long.105

    work page doi:10.18653/v1/2024.eacl-long.105 2024

  7. [7]

    Human-AI Interaction in the Age of Large Language Models , booktitle =

    Diyi Yang , editor =. Human-AI Interaction in the Age of Large Language Models , booktitle =. 2024 , url =. doi:10.1609/AAAISS.V3I1.31183 , timestamp =

    work page doi:10.1609/aaaiss.v3i1.31183 2024

  8. [8]

    Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models

    Yanzhao Zhang and Mingxin Li and Dingkun Long and Xin Zhang and Huan Lin and Baosong Yang and Pengjun Xie and An Yang and Dayiheng Liu and Junyang Lin and Fei Huang and Jingren Zhou , title =. CoRR , volume =. 2025 , url =. doi:10.48550/ARXIV.2506.05176 , eprinttype =. 2506.05176 , timestamp =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2506.05176 2025

  9. [9]

    Qwen3 Technical Report

    An Yang and Anfeng Li and Baosong Yang and Beichen Zhang and Binyuan Hui and Bo Zheng and Bowen Yu and Chang Gao and Chengen Huang and Chenxu Lv and Chujie Zheng and Dayiheng Liu and Fan Zhou and Fei Huang and Feng Hu and Hao Ge and Haoran Wei and Huan Lin and Jialong Tang and Jian Yang and Jianhong Tu and Jianwei Zhang and Jian Yang and Jiaxi Yang and Ji...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2505.09388 2025

  10. [10]

    SweetieChat:

    Jing Ye and Lu Xiang and Yaping Zhang and Chengqing Zong , editor =. SweetieChat:. Proceedings of the 31st International Conference on Computational Linguistics,. 2025 , url =

    work page 2025

  11. [11]

    CoRR , volume =

    Alexander Gurung and Mirella Lapata , title =. CoRR , volume =. 2025 , url =. doi:10.48550/ARXIV.2503.22828 , eprinttype =. 2503.22828 , timestamp =

    work page doi:10.48550/arxiv.2503.22828 2025

  12. [12]

    BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

    Bert: Pre-training of deep bidirectional transformers for language understanding , author=. arXiv preprint arXiv:1810.04805 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  13. [13]

    Advances in neural information processing systems , volume=

    A neural probabilistic language model , author=. Advances in neural information processing systems , volume=

    work page

  14. [14]

    CoRR , volume =

    Pengfei Yu and Dongming Shen and Silin Meng and Jaewon Lee and Weisu Yin and Andrea Yaoyun Cui and Zhenlin Xu and Yi Zhu and Xingjian Shi and Mu Li and Alex Smola , title =. CoRR , volume =. 2025 , url =. doi:10.48550/ARXIV.2502.00595 , eprinttype =. 2502.00595 , timestamp =

    work page doi:10.48550/arxiv.2502.00595 2025

  15. [15]

    Advances in neural information processing systems , volume=

    Language models are few-shot learners , author=. Advances in neural information processing systems , volume=

    work page

  16. [16]

    The Curious Case of Neural Text Degeneration

    The curious case of neural text degeneration , author=. arXiv preprint arXiv:1904.09751 , year=

    work page internal anchor Pith review Pith/arXiv arXiv 1904

  17. [17]

    arXiv preprint arXiv:2402.17532 , archivePrefix=

    Retrieval is Accurate Generation , author=. arXiv preprint arXiv:2402.17532 , archivePrefix=. 2024 , eprint=

    work page arXiv 2024

  18. [18]

    arXiv preprint arXiv:2004.04906 , year=

    Dense passage retrieval for open-domain question answering , author=. arXiv preprint arXiv:2004.04906 , year=

    work page arXiv 2004

  19. [19]

    arXiv preprint arXiv:1909.01066 , year=

    Language models as knowledge bases? , author=. arXiv preprint arXiv:1909.01066 , year=

    work page arXiv 1909

  20. [20]

    2022 , eprint=

    A Review on Language Models as Knowledge Bases , author=. 2022 , eprint=

    work page 2022

  21. [21]

    arXiv preprint arXiv:2206.14268 , year=

    BertNet: Harvesting knowledge graphs with arbitrary relations from pretrained language models , author=. arXiv preprint arXiv:2206.14268 , year=

    work page arXiv

  22. [22]

    Llama 2: Open Foundation and Fine-Tuned Chat Models

    Llama 2: Open foundation and fine-tuned chat models , author=. arXiv preprint arXiv:2307.09288 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  23. [23]

    The Llama 3 Herd of Models

    The llama 3 herd of models , author=. arXiv preprint arXiv:2407.21783 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  24. [24]

    GPT-4 Technical Report

    Gpt-4 technical report , author=. arXiv preprint arXiv:2303.08774 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  25. [25]

    Gemma: Open Models Based on Gemini Research and Technology

    Gemma: Open models based on gemini research and technology , author=. arXiv preprint arXiv:2403.08295 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  26. [26]

    Proceedings of the 2003 Human Language Technology Conference of the North American Chapter of the Association for Computational Linguistics , pages=

    Greedy decoding for statistical machine translation in almost linear time , author=. Proceedings of the 2003 Human Language Technology Conference of the North American Chapter of the Association for Computational Linguistics , pages=

    work page 2003

  27. [27]

    Self-Consistency Improves Chain of Thought Reasoning in Language Models

    Self-consistency improves chain of thought reasoning in language models , author=. arXiv preprint arXiv:2203.11171 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  28. [28]

    arXiv preprint arXiv:2402.10200 , year=

    Chain-of-thought reasoning without prompting , author=. arXiv preprint arXiv:2402.10200 , year=

    work page arXiv

  29. [29]

    arXiv preprint arXiv:2007.00808 , year=

    Approximate nearest neighbor negative contrastive learning for dense text retrieval , author=. arXiv preprint arXiv:2007.00808 , year=

    work page arXiv 2007

  30. [30]

    ArXiv , year=

    Copy is All You Need , author=. ArXiv , year=

    work page

  31. [31]

    arXiv preprint arXiv:2404.10877 , year=

    Incubating Text Classifiers Following User Instruction with Nothing but LLM , author=. arXiv preprint arXiv:2404.10877 , year=

    work page arXiv

  32. [32]

    arXiv preprint arXiv:2202.07922 , year=

    Zerogen: Efficient zero-shot learning via dataset generation , author=. arXiv preprint arXiv:2202.07922 , year=

    work page arXiv

  33. [33]

    arXiv preprint arXiv:2210.12329 , year=

    Progen: Progressive zero-shot dataset generation via in-context feedback , author=. arXiv preprint arXiv:2210.12329 , year=

    work page arXiv

  34. [34]

    Advances in neural information processing systems , volume=

    Chain-of-thought prompting elicits reasoning in large language models , author=. Advances in neural information processing systems , volume=

    work page

  35. [35]

    arXiv preprint arXiv:2004.13897 , year=

    Empower entity set expansion via language model probing , author=. arXiv preprint arXiv:2004.13897 , year=

    work page arXiv 2004

  36. [36]

    , author=

    Visualizing data using t-SNE. , author=. Journal of machine learning research , volume=

    work page

  37. [37]

    IEEE Trans

    Least squares quantization in PCM , author=. IEEE Trans. Inf. Theory , year=

    work page

  38. [38]

    arXiv preprint arXiv:2404.10877 , year=

    Letian Peng and Jingbo Shang , title =. CoRR , volume =. 2024 , url =. doi:10.48550/ARXIV.2404.10877 , eprinttype =. 2404.10877 , timestamp =

    work page doi:10.48550/arxiv.2404.10877 2024

  39. [39]

    ZeroGen: Efficient Zero-shot Learning via Dataset Generation , booktitle =

    Jiacheng Ye and Jiahui Gao and Qintong Li and Hang Xu and Jiangtao Feng and Zhiyong Wu and Tao Yu and Lingpeng Kong , editor =. ZeroGen: Efficient Zero-shot Learning via Dataset Generation , booktitle =. 2022 , url =. doi:10.18653/V1/2022.EMNLP-MAIN.801 , timestamp =

    work page doi:10.18653/v1/2022.emnlp-main.801 2022

  40. [40]

    N-gram Counts and Language Models from the Common Crawl , booktitle =

    Christian Buck and Kenneth Heafield and Bas van Ooyen , editor =. N-gram Counts and Language Models from the Common Crawl , booktitle =. 2014 , url =

    work page 2014

  41. [41]

    Mark Everingham and Luc Van Gool and Christopher K. I. Williams and John M. Winn and Andrew Zisserman , title =. Int. J. Comput. Vis. , volume =. 2010 , url =. doi:10.1007/S11263-009-0275-4 , timestamp =

    work page doi:10.1007/s11263-009-0275-4 2010

  42. [42]

    Diversity of Thought Improves Reasoning Abilities of Large Language Models , journal =

    Ranjita Naik and Varun Chandrasekaran and Mert Y. Diversity of Thought Improves Reasoning Abilities of Large Language Models , journal =. 2023 , url =. doi:10.48550/ARXIV.2310.07088 , eprinttype =. 2310.07088 , timestamp =

    work page doi:10.48550/arxiv.2310.07088 2023

  43. [43]

    Training Verifiers to Solve Math Word Problems

    Karl Cobbe and Vineet Kosaraju and Mohammad Bavarian and Mark Chen and Heewoo Jun and Lukasz Kaiser and Matthias Plappert and Jerry Tworek and Jacob Hilton and Reiichiro Nakano and Christopher Hesse and John Schulman , title =. CoRR , volume =. 2021 , url =. 2110.14168 , timestamp =

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  44. [44]

    Arkil Patel and Satwik Bhattamishra and Navin Goyal , editor =. Are. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,. 2021 , url =. doi:10.18653/V1/2021.NAACL-MAIN.168 , timestamp =

    work page doi:10.18653/v1/2021.naacl-main.168 2021

  45. [45]

    2017 , howpublished =

    Google DeepMind , title =. 2017 , howpublished =

    work page 2017

  46. [46]

    Manning and Andrew Y

    Richard Socher and Alex Perelygin and Jean Wu and Jason Chuang and Christopher D. Manning and Andrew Y. Ng and Christopher Potts , title =. Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing,. 2013 , url =

    work page 2013

  47. [47]

    Character-level Convolutional Networks for Text Classification , booktitle =

    Xiang Zhang and Junbo Jake Zhao and Yann LeCun , editor =. Character-level Convolutional Networks for Text Classification , booktitle =. 2015 , url =

    work page 2015

  48. [48]

    CARER : Contextualized Affect Representations for Emotion Recognition

    Saravia, Elvis and Liu, Hsien-Chi Toby and Huang, Yen-Hao and Wu, Junlin and Chen, Yi-Shin. CARER : Contextualized Affect Representations for Emotion Recognition. Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. 2018. doi:10.18653/v1/D18-1404

    work page doi:10.18653/v1/d18-1404 2018

  49. [49]

    Abdelzaher and Heng Ji , editor =

    Chi Han and Jialiang Xu and Manling Li and Yi Fung and Chenkai Sun and Nan Jiang and Tarek F. Abdelzaher and Heng Ji , editor =. Word Embeddings Are Steers for Language Models , booktitle =. 2024 , url =. doi:10.18653/V1/2024.ACL-LONG.864 , timestamp =

    work page doi:10.18653/v1/2024.acl-long.864 2024

  50. [50]

    Answer is All You Need: Instruction-following Text Embedding via Answering the Question , booktitle =

    Letian Peng and Yuwei Zhang and Zilong Wang and Jayanth Srinivasa and Gaowen Liu and Zihan Wang and Jingbo Shang , editor =. Answer is All You Need: Instruction-following Text Embedding via Answering the Question , booktitle =. 2024 , url =. doi:10.18653/V1/2024.ACL-LONG.27 , timestamp =

    work page doi:10.18653/v1/2024.acl-long.27 2024

  51. [51]

    Morris and Richard Antonello and Ion Stoica and Alexander G

    Vinamra Benara and Chandan Singh and John X. Morris and Richard Antonello and Ion Stoica and Alexander G. Huth and Jianfeng Gao , title =. CoRR , volume =. 2024 , url =. doi:10.48550/ARXIV.2405.16714 , eprinttype =. 2405.16714 , timestamp =

    work page doi:10.48550/arxiv.2405.16714 2024

  52. [52]

    RoBERTa: A Robustly Optimized BERT Pretraining Approach

    Yinhan Liu and Myle Ott and Naman Goyal and Jingfei Du and Mandar Joshi and Danqi Chen and Omer Levy and Mike Lewis and Luke Zettlemoyer and Veselin Stoyanov , title =. CoRR , volume =. 2019 , url =. 1907.11692 , timestamp =

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  53. [53]

    7th International Conference on Learning Representations,

    Ilya Loshchilov and Frank Hutter , title =. 7th International Conference on Learning Representations,. 2019 , url =

    work page 2019

  54. [54]

    Dirk Groeneveld and Iz Beltagy and Evan Pete Walsh and Akshita Bhagia and Rodney Kinney and Oyvind Tafjord and Ananya Harsh Jha and Hamish Ivison and Ian Magnusson and Yizhong Wang and Shane Arora and David Atkinson and Russell Authur and Khyathi Raghavi Chandu and Arman Cohan and Jennifer Dumas and Yanai Elazar and Yuling Gu and Jack Hessel and Tushar Kh...

    work page doi:10.18653/v1/2024.acl-long.841 2024

  55. [55]

    Stolen Probability:

    David Demeter and Gregory Kimmel and Doug Downey , editor =. Stolen Probability:. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics,. 2020 , url =. doi:10.18653/V1/2020.ACL-MAIN.198 , timestamp =

    work page doi:10.18653/v1/2020.acl-main.198 2020

  56. [56]

    Low-Rank Softmax Can Have Unargmaxable Classes in Theory but Rarely in Practice , booktitle =

    Andreas Grivas and Nikolay Bogoychev and Adam Lopez , editor =. Low-Rank Softmax Can Have Unargmaxable Classes in Theory but Rarely in Practice , booktitle =. 2022 , url =. doi:10.18653/V1/2022.ACL-LONG.465 , timestamp =

    work page doi:10.18653/v1/2022.acl-long.465 2022

  57. [57]

    Cohen , title =

    Zhilin Yang and Zihang Dai and Ruslan Salakhutdinov and William W. Cohen , title =. 6th International Conference on Learning Representations,. 2018 , url =

    work page 2018

  58. [58]

    Softmax Bottleneck Makes Language Models Unable to Represent Multi-mode Word Distributions , booktitle =

    Haw. Softmax Bottleneck Makes Language Models Unable to Represent Multi-mode Word Distributions , booktitle =. 2022 , url =. doi:10.18653/V1/2022.ACL-LONG.554 , timestamp =

    work page doi:10.18653/v1/2022.acl-long.554 2022

  59. [59]

    Contrastive Decoding: Open-ended Text Generation as Optimization , booktitle =

    Xiang Lisa Li and Ari Holtzman and Daniel Fried and Percy Liang and Jason Eisner and Tatsunori Hashimoto and Luke Zettlemoyer and Mike Lewis , editor =. Contrastive Decoding: Open-ended Text Generation as Optimization , booktitle =. 2023 , url =. doi:10.18653/V1/2023.ACL-LONG.687 , timestamp =

    work page doi:10.18653/v1/2023.acl-long.687 2023

  60. [60]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Interactive narrative: A novel application of artificial intelligence for computer games , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    work page

  61. [61]

    International Workshop on Intelligent Virtual Agents , pages=

    An objective character believability evaluation procedure for multi-agent story generation systems , author=. International Workshop on Intelligent Virtual Agents , pages=. 2005 , organization=

    work page 2005

  62. [62]

    Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing , pages=

    Character-LLM: A Trainable Agent for Role-Playing , author=. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing , pages=

    work page 2023

  63. [63]

    Transactions on Machine Learning Research , year=

    From Persona to Personalization: A Survey on Role-Playing Language Agents , author=. Transactions on Machine Learning Research , year=

    work page

  64. [64]

    Larp: Language-agent role play for open-world games

    Larp: Language-agent role play for open-world games , author=. arXiv preprint arXiv:2312.17653 , year=

    work page arXiv

  65. [65]

    Findings of the Association for Computational Linguistics: ACL 2024 , pages=

    Rolellm: benchmarking, eliciting, and enhancing role-playing abilities of large language models , author=. Findings of the Association for Computational Linguistics: ACL 2024 , pages=. 2024 , publisher=

    work page 2024

  66. [66]

    Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , pages=

    Neeko: Leveraging Dynamic LoRA for Efficient Multi-Character Role-Playing Agent , author=. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , pages=

    work page 2024

  67. [67]

    arXiv preprint arXiv:2506.01748 , year=

    Thinking in Character: Advancing Role-Playing Agents with Role-Aware Reasoning , author=. arXiv preprint arXiv:2506.01748 , year=

    work page arXiv

  68. [68]

    arXiv preprint arXiv:2505.12814 , year=

    PsyMem: Fine-grained psychological alignment and Explicit Memory Control for Advanced Role-Playing LLMs , author=. arXiv preprint arXiv:2505.12814 , year=

    work page arXiv

  69. [69]

    Information and control , volume=

    Finite state languages , author=. Information and control , volume=. 1958 , publisher=

    work page 1958

  70. [70]

    Proceedings of the 22nd international conference on Software engineering , pages=

    Bandera: Extracting finite-state models from Java source code , author=. Proceedings of the 22nd international conference on Software engineering , pages=

    work page

  71. [71]

    2013 IEEE Global Engineering Education Conference (EDUCON) , pages=

    Teaching finite state machines with case method and role play , author=. 2013 IEEE Global Engineering Education Conference (EDUCON) , pages=. 2013 , organization=

    work page 2013

  72. [72]

    arXiv preprint arXiv:2310.05161 , year=

    Recurrent neural language models as probabilistic finite-state automata , author=. arXiv preprint arXiv:2310.05161 , year=

    work page arXiv

  73. [73]

    Findings of the Association for Computational Linguistics: EMNLP 2024 , pages=

    GraphReader: Building Graph-based Agent to Enhance Long-Context Abilities of Large Language Models , author=. Findings of the Association for Computational Linguistics: EMNLP 2024 , pages=

    work page 2024

  74. [74]

    arXiv preprint arXiv:2407.19412 , year=

    Identity-driven hierarchical role-playing agents , author=. arXiv preprint arXiv:2407.19412 , year=

    work page arXiv

  75. [75]

    International Conference on Machine Learning , pages=

    Pal: Program-aided language models , author=. International Conference on Machine Learning , pages=. 2023 , organization=

    work page 2023

  76. [76]

    Science , volume=

    Competition-level code generation with alphacode , author=. Science , volume=. 2022 , publisher=

    work page 2022

  77. [77]

    WizardCoder: Empowering Code Large Language Models with Evol-Instruct , author=

    work page

  78. [78]

    International Conference on Learning Representations (ICLR) , year=

    React: Synergizing reasoning and acting in language models , author=. International Conference on Learning Representations (ICLR) , year=

    work page

  79. [79]

    Autonomous Robots , volume=

    Tidybot: Personalized robot assistance with large language models , author=. Autonomous Robots , volume=. 2023 , publisher=

    work page 2023

  80. [80]

    Code as Policies: Language Model Programs for Embodied Control

    Code as policies: Language model programs for embodied control , author=. arXiv preprint arXiv:2209.07753 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

Showing first 80 references.